Metrologia: Dalla Fisica Fondamentale Alla Qualit' A Della Vita

` ITALIANA DI FISICA
SOCIETA
RENDICONTI
DELLA
SCUOLA INTERNAZIONALE DI FISICA

“ENRICO FERMI”
CXCVI Corso
a cura di P. Tavella, M. J. T. Milton e M. Inguscio
Direttori del Corso
e di
N. De Leo
VARENNA SUL LAGO DI COMO
VILLA MONASTERO
27 Giugno – 6 Luglio 2016
Metrologia: dalla fisica

fondamentale alla qualità
della vita
Copyright © 2018. IOS Press, Incorporated. All rights reserved.
2017
` ITALIANA DI FISICA
SOCIETA
BOLOGNA-ITALY
Metrology: from Physics Fundamentals to Quality of Life, edited by P. Tavella, et al., IOS Press, Incorporated, 2018. ProQuest Ebook Central,
ITALIAN PHYSICAL SOCIETY
PROCEEDINGS
OF THE
INTERNATIONAL SCHOOL OF PHYSICS

“ENRICO FERMI”
Course 196
edited by P. Tavella, M. J. T. Milton and M. Inguscio
Directors of the Course
and
N. De Leo
VARENNA ON LAKE COMO
VILLA MONASTERO
27 June – 6 July 2016
Metrology: From Physics

Fundamentals to Quality
of Life
2017
AMSTERDAM, OXFORD, TOKIO, WASHINGTON DC
Copyright
c 2017 by Società Italiana di Fisica
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or any means, electronic, mechanical, photocopying,
recording or otherwise, without the prior permission of the copyright owner.
ISSN 0074-784X (print)

ISSN 1879-8195 (online)
ISBN 978-1-61499-817-4 (print) (IOS Press)
ISBN 978-1-61499-818-1 (online) (IOS Press)
ISBN 978-88-7438-109-8 (SIF)
LCCN 2017962233
jointly published and distributed by:
IOS PRESS SOCIETÀ ITALIANA DI FISICA

Nieuwe Hemweg 6B Via Saragozza 12
1013 BG Amsterdam 40123 Bologna
The Netherlands Italy
fax: +31 20 687 0019 fax: +39 051 581340
info@iospress.nl order@sif.it
Distributor in the USA and Canada

IOS Press, Inc.
4502 Rachael Manor Drive
Fairfax, VA 22032
USA
fax: +1 703 323 3668
sales@iospress.com
Supported by
Camera di Commercio di Lecco
Istituto Nazionale di Fisica Nucleare (INFN)
Istituto Nazionale di Geofisica e Vulcanologia (INGV)
UNIVERLECCO
Museo Storico della Fisica e Centro Studi e Ricerche “Enrico Fermi”
Istituto Nazionale di Ricerca Metrologica (INRIM)

FEI Italia s.r.l.
Cal Power s.r.l.
Bureau International des Poids et Mesures (BIPM)
Produced by the SIF Editorial Staff
Production Editor: Marcella Missiroli
Cover: see Bernd Güttler, Detlef Schiel, Axel Pramann and Olaf Rienitz, Amount of
substance — the Avogadro constant and the SI unit “mole”, p. 83
Source: Leibniz-Institute for Crystal Growth (IKZ), Berlin, Germany
Graphic elaboration by Simona Oleandri

Proprietà Letteraria Riservata
Printed in Italy by Labanti e Nanni Industrie Grafiche S.r.l. - Crespellano (BO)
CONTENTS
P. Tavella, M. J. T. Milton and M. Inguscio – Preface . . . . . . . . . . . . p. XIX
Course group shot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XXII
MODULE I. METROLOGY FOR QUALITY OF LIFE
Vincent Delatour, Gustavo Martos, Julie Cabillic, Maryline

Peignaux, Catherine Perrot, Carine Fallot, Béatrice Lalere and
Sophie Vaslin-Reimann – Reference methods and commutable reference
materials for clinical measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
.
1 1. Importance of medical tests reliability . . . . . . . . . . . . . . . . . . . . . . . . . 1
.
1 2. Role of a national metrology institute in bioanalysis . . . . . . . . . . . . . . 2
.
1 3. Examples of projects undertaken in LNE (France) . . . . . . . . . . . . . . . 3
.
1 4. Regulatory drivers and traceability chains in laboratory medicine . . . 4
.
1 5. JCTLM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
.
1 6. Accreditation according to ISO 15189 . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2. Importance of reference methods in EQAS . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3. Importance of commutability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
.
3 1. Why commutability matters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

.
3 2. Principle of commutability assessment . . . . . . . . . . . . . . . . . . . . . . . . . 6
4. Conclusions and perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Vincent Delatour, Gustavo Martos, François Gaie-Levrel,

Noémie Clouet-Foraison, Pauline Bros, Maud Heuillet, Béatrice
Lalere and Sophie Vaslin-Reimann – Reference measurement systems
for biomarkers: Towards biometrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2. Metabolites and small molecules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
VII
VIII contents
.
2 1. Kidney disease: creatinine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
.
2 2. Diabetes mellitus: glucose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3. Peptides and proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
.
3 1. Absolute quantification of peptides and proteins . . . . . . . . . . . . . . . . . 12
.
3 2. Diabetes mellitus: HbA1c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
.
3 3. Sepsis and antimicrobial resistance: Procalcitonine . . . . . . . . . . . . . . . 14
.
3 4. Alzheimer’s disease: amyloid beta & tau . . . . . . . . . . . . . . . . . . . . . . . 15
.
3 5. Iron-related disorders: hepcidin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4. Lipids and lipoproteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
.
4 1. Cardiovascular diseases: cholesterol, triglycerides, LDL-C and HDL-C 17
.
4 2. Lipoprotein particle concentration: beyond LDL-C in CVD risk
assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
N. P. Fox – SI traceable measurements of the Earth from space to monitor

and mitigate against climate change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
.
1 1. Climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
.
1 2. Earth Observation data quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
.
1 3. Sensor post launch Calibration and Validation (Cal/Val) . . . . . . . . . . 31
.
1 4. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2. Key climate parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
.
2 1. Essential Climate Variables (ECV) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
.
2 2. Earth Radiation Budget (ERB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
.
2 3. Solar variability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
.
2 3.1. Total Solar Irradiance (TSI) . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
.
2 3.2. Solar spectral irradiance (SSI) . . . . . . . . . . . . . . . . . . . . . . . . . . 38
.
2 4. Climate feedbacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
.
2 4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
.
2 4.2. Cloud feedback on climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3. Establishing SI traceability for the Earth observing system . . . . . . . . . . . . . 46
.
3 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
.
3 2. Near simultaneous overpass calibrations (SNO) . . . . . . . . . . . . . . . . . . 47
.
3 3. Reference standard calibration test sites . . . . . . . . . . . . . . . . . . . . . . . . 47
.
3 4. Lunar calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
.
3 5. Dominant sources of uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
.
3 6. Radiometric accuracy and traceability to SI . . . . . . . . . . . . . . . . . . . . . 50
4. Traceable Radiometry Underpinning Terrestrial- and Helio-Studies (TRUTHS):
an NMI in space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
.
4 1. Mission requirements and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
.
4 2. TRUTHS instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
.
4 2.1. Calibration system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
.
4 3. On-board calibration methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
.
4 3.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
contents IX
.
4 3.2. Step 1: Calibration of TR against CSAR using LDs . . . . . . . . 62
.
4 3.3. Step 2: HIS (Earth radiance view) calibrated against the TR
at each LD wavelength (radiance mode) . . . . . . . . . . . . . . . . . . 63
.
4 3.4. Step 3: HIS (Earth radiance view) calibrated at intermediate
wavelengths with a lamp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
.
4 3.5. Step 4: Measurements of the Earth and Sun . . . . . . . . . . . . . . 65
.
4 3.6. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Bernd Güttler, Detlef Schiel, Axel Pramann and Olaf Rienitz –

Amount of substance – the Avogadro constant and the SI unit “mole” . . . . . 73
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2. History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3. The mole as an SI unit in chemistry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4. Realization and dissemination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
.
4 1. Primary standards as reference points . . . . . . . . . . . . . . . . . . . . . . . . . . 76
.
4 2. Example: National standards for the determination of element con-
centrations in solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
.
4 3. Examples of primary measurement procedures . . . . . . . . . . . . . . . . . . 78
.
4 4. International comparability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5. Redefinitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
.
5 1. General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
.
5 2. Silicon single crystal and the Avogadro constant . . . . . . . . . . . . . . . . . 81
.
5 3. Realization and dissemination in accordance with the redefinitions . . 85
6. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
R. I. Wielgosz, J. Viallon, E. Flores, P. Moussay, F. Idrees and

T. Choteau – Comparisons of gas standards for climate change and air
quality monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
1. Introduction to air quality and greenhouse gas monitoring . . . . . . . . . . . . . . 89
2. Methods for gas standard preparation and verification . . . . . . . . . . . . . . . . . 91

3. Standards produced by static gravimetric methods (CH4 and NO) . . . . . . . 91
.
3 1. Methane in air standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
.
3 2. Nitrogen monoxide in nitrogen standards . . . . . . . . . . . . . . . . . . . . . . . 96
4. Dynamic methods (NO2 and HCHO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5. Spectroscopic methods (O3 and FTIR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
.
5 1. Ozone standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
.
5 2. FTIR for the comparison and analysis of gas standards . . . . . . . . . . . 106
.
5 3. FTIR measurements of isotope ratios in CO2 . . . . . . . . . . . . . . . . . . . 109
6. Manometric methods (CO2 and O3 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
.
6 1. Manometric reference method for CO2 mole fraction value assignment 111
.
6 2. Manometric method for ozone cross-section measurements . . . . . . . . . 114
X contents
R. I. Wielgosz, S. Westwood, R. D. Josephs, N. Stoppacher,

A. Daireaux and T. Choteau – Chemical primary reference materials:
From valine to C-peptide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
1. Introduction: Organic primary reference materials for analytical chemistry 121
2. The Mass balance purity assignment method applied to valine . . . . . . . . . . 122
.
2 1. Measurement of mass fraction of related structure impurities (wRS )
in valine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
.
2 1.1. Experimental and method description . . . . . . . . . . . . . . . . . . . 123
.
2 1.2. Metrological uncertainty and SI-traceability of mass fraction
of related structure impurities . . . . . . . . . . . . . . . . . . . . . . . . . . 126
.
2 2. Measurement of mass fraction of water (wW ) in valine . . . . . . . . . . . . 127
.
2 3. Measurement of the mass fraction of residual organic solvent (wOS )
for valine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
.
2 4. Measurements of the mass fraction of non-volatile materials (wN V ) in
valine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
.
2 5. Assignment of the mass fraction content of valine . . . . . . . . . . . . . . . . 129
3. qNMR applied to pure material standards of folic acid . . . . . . . . . . . . . . . . . 129
.
3 1. Introduction to qNMR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
.
3 2. Application of qNMR for purity measurements of folic acid . . . . . . . . 132
4. Methods for large organics: peptides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
.
4 1. High-resolution mass spectrometry, identification and quantification
of impurities in peptides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
.
4 1.1. Typical configuration of a mass spectrometer . . . . . . . . . . . . . 134
.
4 1.2. Electrospray ionization (ESI) . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
.
4 1.3. Mass analysers and mass resolution . . . . . . . . . . . . . . . . . . . . . 136
.
4 1.4. The Orbitrap mass analyser . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
.
4 2. High-resolution mass spectrometry of Angiotensin I . . . . . . . . . . . . . . 137
.
4 2.1. Sequencing of Angiotensin I using MS/MS . . . . . . . . . . . . . . . 138
.
4 2.2. Identification of impurities in ANG I peptide material . . . . . . 140
.
4 2.3. PICAA analysis of ANG I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
.
4 3. Characterization of C-peptide primary reference material . . . . . . . . . 143
.
4 3.1. Peptide impurity identification and quantification . . . . . . . . . 143
.
4 3.2. PICCA analysis of C-peptide material . . . . . . . . . . . . . . . . . . . 145
MODULE II. FUNDAMENTALS OF METROLOGY
W. Bich – Uncertainty of measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Disclaimer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2. Physical quantities and their values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
.
2 1. Quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
.
2 2. Quantity values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
.
2 3. Physical constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
.
2 4. Other quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
contents XI
3. Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
.
3 1. Measurement and measurand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
.
3 2. Measurement result and measurand estimate . . . . . . . . . . . . . . . . . . . . 154
4. Measurement uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
.
4 1. General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
.
4 2. Definitional uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5. Modelling a measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6. Propagating uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7. Evaluating uncertainties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
.
7 1. General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
.
7 2. Random and systematic effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
.
7 3. Frequentist and Bayesian (or subjective) approaches . . . . . . . . . . . . . 163
.
7 4. Change of paradigm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8. Coverage intervals – propagation of PDFs . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
9. Bayesian inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
10. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
Andy Henson – International recognition of NMI calibration and measure-

ment capabilities: The CIPM MRA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
2. The origins of the CIPM MRA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
3. Launch of the CIPM MRA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4. Structure and mechanisms of the CIPM MRA . . . . . . . . . . . . . . . . . . . . . . . . 176
.
4 1. Key and supplementary comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . 178
.
4 2. Calibration and Measurement Capabilities (CMCs) . . . . . . . . . . . . . . 180
5. The CIPM MRA today . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
6. The CIPM MRA review and the way forward . . . . . . . . . . . . . . . . . . . . . . . . 183
7. In conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Martin J. T. Milton – On the proposed re-definition of the SI: “for all

people for all time” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
1. Metrology and the role of the SI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2. The development of the SI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
3. Establishing units for electrical quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4. The 1990 convention for the electrical units . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5. The Kibble balance and the silicon Avogadro experiment . . . . . . . . . . . . . . 190
6. A method for establishing a mass unit from the atomic mass unit . . . . . . . 191
7. The 2005 proposal to redefine the kilogram . . . . . . . . . . . . . . . . . . . . . . . . . . 191
8. The 2006 proposals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
9. Articulating the definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
10. The motivation for re-definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
XII contents
.
10 1. Eliminating the dependence on artefacts . . . . . . . . . . . . . . . . . . . . . . . 196
.
10 2. Widening access to the realization of the base units . . . . . . . . . . . . . . 198
11. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
T. J. Quinn – The Metre Convention and the creation of the BIPM . . . . . . 203
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
2. The call by geodesists in 1867 for a better standard of the metre and for a
European international Bureau of weights and measures . . . . . . . . . . . . . . . 204
3. The International Metre Commission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4. The Metre Convention of 1875 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
5. The creation of the BIPM and early scientific work . . . . . . . . . . . . . . . . . . . 220
6. The BIPM since 1921 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7. The CIPM MRA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
8. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
T. J. Quinn – The development of units of measurement from the origin of

the metric system in the 18th century to proposals for redefinition of the SI
in 2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
2. The origin of the metric system in 1791 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
3. The Metre and Kilogram of the Archives 1796 . . . . . . . . . . . . . . . . . . . . . . . . 229
4. The Berlin Conference of 1867 and proposals for an international European
bureau of weights and measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5. The construction of the new prototypes of the Metre and Kilogram . . . . . . 231
6. Next steps: Maxwell, Michelson and successive definitions of the Metre . . . 233
7. Electrical standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
8. Developments in other base units until 1971 . . . . . . . . . . . . . . . . . . . . . . . . . 237

9. The changes since 1971 that finally opened the way to units based on con-
stants of physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
10. The Kibble balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
11. The x-ray crystal density method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
12. Comparison of the two methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
13. Present plans for a fully constants-based SI for 2018 . . . . . . . . . . . . . . . . . . . 243
.
13 1. What does it mean to fix the numerical value of a constant of nature? 243
.
13 2. The form of the proposed new definition of the SI and its base units 244
.
13 3. Advantages of units defined in terms of constants of nature: explicit
constant definitions rather than an explicit unit definitions . . . . . . . . 245
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
contents XIII
Th. Udem and F. Riehle – Frequency combs applications and optical

frequency standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
2. Frequency combs from mode locked lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
.
2 1. Derivation from cavity boundary conditions . . . . . . . . . . . . . . . . . . . . . 254
.
2 2. Derivation from the pulse train . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
.
2 3. Linewidth of a single mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
.
2 4. Generating an octave spanning comb . . . . . . . . . . . . . . . . . . . . . . . . . . 262
3. Self-referencing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
4. Scientific applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
.
4 1. Hydrogen and drifting constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
.
4 2. Fine structure constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
.
4 3. Optical frequencies in astronomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
.
4 4. Reconstructing pulse transients and generating attosecond pulses . . . 279
.
4 5. Frequency comb spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
5. All optical clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
6. Optical frequency standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
.
6 1. Optical frequency standards based on single ions . . . . . . . . . . . . . . . . 289
.
6 1.1. Clock interrogation by use of quantum jumps . . . . . . . . . . . . . 289
.
6 1.2. Yb+ and Hg+ single-ion standards . . . . . . . . . . . . . . . . . . . . . . 289
.
6 1.3. 88 Sr+ single-ion standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
.
6 2. Neutral atom optical frequency standards . . . . . . . . . . . . . . . . . . . . . . 291
.
6 2.1. Atomic beam standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
.
6 2.2. Neutral atom standards based on ballistic atoms . . . . . . . . . . 291
.
6 2.3. Optical lattice clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
7. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Carl J. Williams – The SI and quantum metrology . . . . . . . . . . . . . . . . . . 301

1. An old new idea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
2. Quantum standards and redefinition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
3. Quantum standards and quantum metrology today . . . . . . . . . . . . . . . . . . . 304
.
3 1. Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
.
3 2. Electrical units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
.
3 3. From instrumenting the SI to embedded measurements . . . . . . . . . . . 306
Carl J. Williams – The future of quantum-based measurements and the SI 309

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
2. Single-photon sources & detectors and quantum communication . . . . . . . . . 313
3. Photon pressure and small mass and force . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
4. Quantum-based pressure and vacuum standards . . . . . . . . . . . . . . . . . . . . . . 315
5. Quantum-based approaches to temperature . . . . . . . . . . . . . . . . . . . . . . . . . . 316
XIV contents
6. Other atom-based sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

7. Quantum transduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
8. Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
MODULE III. PHYSICAL METROLOGY AND FUNDAMENTAL CONSTANTS
R. Battiston – Space metrology and space-time warping . . . . . . . . . . . . . . 321

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
2. Theoretical landscape of the 20th century . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
3. Testing General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
4. Deep-space navigation and the investigation of gravity . . . . . . . . . . . . . . . . . 329
.
4 1. Observables used in deep-space navigation . . . . . . . . . . . . . . . . . . . . . . 329
.
4 2. Radioscience GR tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
.
4 3. Satellites orbit anomalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
.
4 4. Future deep-space satellites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
5. Gravitational waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
.
5 1. The basic concept of a space-borne GW detector . . . . . . . . . . . . . . . . 333
.
5 2. LISA Pathfinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
6. Testing GR within the Earth gravity field . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
.
6 1. Satellite geodesy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
.
6 2. Very Long Baseline Interferometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
.
6 3. Laser Lunar Ranging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
.
6 4. Frame dragging and Gravitomagnetism in the Earth gravity field . . . 339
7. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Giorgio Brida – Measuring light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
2. Primary standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
.
2 1. Blackbody source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
.
2 2. Synchrotron radiation source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
.
2 3. Electrical Substitution Radiometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
3. The state of the art in primary standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
.
3 1. Carbon nanotube for radiometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
.
3 2. Predictable Quantum Efficiency Detector . . . . . . . . . . . . . . . . . . . . . . . 350
4. Counting photons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
.
4 1. Spontaneous Parametric Down-Conversion (SPDC) . . . . . . . . . . . . . . 352
.
4 2. Single-Photon Source (SPS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
5. Photometry – The realization of the candela . . . . . . . . . . . . . . . . . . . . . . . . . 354
contents XV
Michael de Podesta – Why redefine the kelvin? . . . . . . . . . . . . . . . . . . . . . 357

1. Measurement and unit quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
.
1 1. Measurement and the role of a unit definition . . . . . . . . . . . . . . . . . . . 357
.
1 2. The International System of Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
.
1 3. What makes a “good” unit definition? . . . . . . . . . . . . . . . . . . . . . . . . . . 359
.
1 4. Historical Decay: The kilogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
.
1 5. Technological irrelevance: the ampere . . . . . . . . . . . . . . . . . . . . . . . . . . 363
.
1 6. The definition of the kelvin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
2. A new type of unit definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
.
2 1. Unit definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
.
2 2. Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
.
2 3. Making the change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
3. The kelvin redefinition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
.
3 1. The new definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
.
3 2. Impact of the kelvin redefinition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
.
3 2.1. Impact on ITS-90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
.
3 2.2. Impact on thermometry at extreme temperatures . . . . . . . . . . 372
.
3 3. Why has this taken so long? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
H. Fleurbaey, C. Courvoisier, P. Cladé, S. Guellati-Khélifa,

L. Julien, F. Biraben and F. Nez – Contributions of fundamental con-
stants from atomic physics to the redefinition of kilogram . . . . . . . . . . . . . . . 381
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
2. The Rydberg constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
3. The relative atomic mass of the electron . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
.
3 1. Ar (e) from Penning trap mass spectrometer . . . . . . . . . . . . . . . . . . . . . 386
.
3 2. Ar (e) from antiprotonic helium spectroscopy . . . . . . . . . . . . . . . . . . . . 386
.
3 3. Ar (e) from the Larmor frequency to cyclotron frequency ratio of ions 386
4. The fine-structure constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
.
4 1. Atomic recoil experiment, determination of h/m . . . . . . . . . . . . . . . . . 388
.
4 2. Anomaly of the gyromagnetic factor of the electron . . . . . . . . . . . . . . 389
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
K. Helmerson and W. D. Phillips – Cooling, trapping and manipulation

of atoms and Bose-Einstein condensates: Applications to metrology . . . . . . . 395
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
2. Radiative forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
.
2 1. The scattering force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
.
2 2. The dipole force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
.
2 3. Doppler cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
XVI contents
3. Deceleration and cooling of an atomic beam . . . . . . . . . . . . . . . . . . . . . . . . . 403

.
3 1. Chirp cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
.
3 2. Zeeman tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
.
3 3. Optical pumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
4. Traps for neutral atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
.
4 1. Dipole force traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
.
4 2. Radiation pressure force traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
.
4 3. Magneto-static traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
5. Sub-Doppler laser cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
.
5 1. Observation of sub-Doppler temperatures . . . . . . . . . . . . . . . . . . . . . . . 414
.
5 2. New cooling mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
6. Metrology with cold atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
.
6 1. Atomic fountain clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
.
6 2. Atom interferometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
7. Coherent manipulation of Bose-Einstein condensates with light . . . . . . . . . 421
.
7 1. The triaxial TOP trap for sodium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
.
7 2. BEC of sodium in a TOP trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
.
7 3. Diffraction of atoms by a standing wave . . . . . . . . . . . . . . . . . . . . . . . . 425
.
7 4. Bragg diffraction of atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
.
7 5. Raman output coupling: Demonstration of a CW atom laser . . . . . . 428
8. Nonlinear atom optics with Bose-Einstein condensates . . . . . . . . . . . . . . . . . 430
.
8 1. Four-wave mixing with matter-waves . . . . . . . . . . . . . . . . . . . . . . . . . . 431
.
8 2. Quantum phase engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
.
8 3. Solitons in a BEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
.
8 4. Observation of solitons in a BEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
.
8 5. Quantum atom optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
Carl J. Williams – From the mise-en-pratique for mass to the future of

metrology for the SI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
1. Problems with the current SI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450

2. The redefinition: a quantum SI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
3. Primary realization of the definition of the kilogram . . . . . . . . . . . . . . . . . . . 450
4. The mise-en-pratique of the kilogram at NIST . . . . . . . . . . . . . . . . . . . . . . . . 451
.
4 1. Realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
.
4 2. Vacuum-to-air transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
.
4 3. Storage of artifacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
.
4 4. Dissemination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
5. Small mass and force realization and metrology at NIST . . . . . . . . . . . . . . . 454
6. The quantum SI, quantum information, and the Standard Model . . . . . . . . 456
7. Future metrology and quantum information science . . . . . . . . . . . . . . . . . . . 458
8. Ubiquitous embedded sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
contents XVII
Patrizia Tavella – Precise time scales and navigation systems . . . . . . . . . 463

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
2. Time inside a Global Navigation Satellite System . . . . . . . . . . . . . . . . . . . . . 464
3. Time from a Global Navigation Satellite System . . . . . . . . . . . . . . . . . . . . . . 466
4. The universal coordinated time and the definition of the seconds in the
centuries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
.
4 1. Universal time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
.
4 2. Ephemeris time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
.
4 3. Atomic time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
.
4 4. UTC: the trade-off . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
.
4 5. The future of leap second . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
5. Will time scales return in space? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
Patrizia Tavella – Statistical tools for time varying quantities . . . . . . . . . 477

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
2. Estimating time varying behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
3. Stochastic processes to model time varying quantities . . . . . . . . . . . . . . . . . 481
4. The need of preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
5. The detection of anomalous behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
6. Dynamic noise analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
7. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
List of participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495

The electronic version of volumes from 124 to 194 is available online at the IOS Press
web site
http://ebooks.iospress.nl/bookseries/proceedings-of-the-international
-school-of-physics-enrico-fermi
Figures with colour source files will appear in colour in the online version.
SIF Members are granted free access to these volumes at

http://members.sif.it
For a complete list of published courses see

https://en.sif.it/books/series/rendiconti_fermi
Preface
We are delighted to introduce the proceedings of the 6th of the Enrico Fermi Summer
Schools that has addressed the topic of Metrology. The school was held in Varenna from
June 26th to July 6th, 2016, and was organised by the Italian Physical Society (SIF), the
Bureau Internationals des Poids et Mesures (BIPM), and the Italian Istituto Nazionale
di Ricerca Metrologica (INRIM).
Metrology is an active field that has developed in many ways since the first of these
schools devoted to metrology was held in 1976. The successive schools have reflected
these developments. The sixth one is no exception, and may in the course of time be
considered to have been the one that has signalled the greatest changes.
The programme addressed two major new directions for metrology. The first was
the work done in preparation for a possible re-definition of four of the base units of
the SI in 2018. The main principles of the possible redefinition of the base units were
explained, the challenges involved in the accurate measurement of fundamental constants
were described, and attention was brought to the proposed changes in the dissemination
of the units following the new definitions.
Secondly, the lectures included a deep reflection on the impact of the application
of metrology to chemical measurements on science, citizens and society. The need for
precise and traceable measurements in chemistry was exemplified by their application to
issues addressing quality of life, global climate change, clinical and food analysis.
For the first time, the lectures were grouped into three modules held over 10 days.
All students took part in a core module on the Fundamentals of Metrology. They were
then able to choose to attend one or both of the optional modules that addressed:
Metrology for Quality of Life and Physical Metrology and Fundamental Constants. In
this way, students could choose to be part of the school for the whole 10 days, or only
for 7 days according to the topics of greatest interest to them. The topics addressed in
the modules were:
Module I: Metrology for Quality of Life (26th June to 29th June 2016)
Metrology in chemistry, Food supply and safety, Biomarkers, Methods and materi-
als for clinical measurements, Climate and air quality monitoring, Redefinition of
the mole.
XIX
XX Preface
Module II: Fundamentals of Metrology (30th June to 2nd July 2016)

The New International System of Units, Fundamental constants, Quantum metrol-
ogy, Nanotechnology for metrology, Measurement uncertainty, the Organization of
international metrology.
Module III: Physical Metrology and Fundamental Constants (3rd July to 6th July
2016)
Electrical metrology, the Future of the standard of mass, Temperature metrology,
Optical frequency standards, Metrology in space, Photometry and light metrology.
As always, the success of a school is due to the dedication and effort of many people
and institutions. Our special thanks go to:
– the lecturers who prepared their lessons and manuscripts and were available to
spend time with the students, visiting their posters sessions, discussing ideas and
participating in the award jury.
– the 77 students and observers who were selected from 120 applications and who
created a stimulating and warm atmosphere throughout the period of the school.
– the scientific secretary, Natascia De Leo from INRIM, who devoted a lot of her
time, energy, and creativity to the success of the school.
We particularly want to thank the sponsors: the Fondazione Cariplo, CalPower, FEI,
Europhysics Letter, METAS, SIM, and NIST. They supported the participation of stu-
dents who did not have access to sufficient funding of their own and also the best poster
award which stimulated students in the clear and efficient presentation of their work.
Finally, we wish to thank Prof. Luisa Cifarelli, SIF President, for promoting the 6th
edition of a school devoted to metrology, Mrs Barbara Alzani (SIF), Ramona Brigatti
and Arianna Minonzio for their invaluable support in the management and organisation
of the school.
P. Tavella, M. J. T. Milton and M. Inguscio
This page intentionally left blank
Italian Physical Society

INTERNATIONAL SCHOOL OF PHYSICS «E. FERMI»
COURSE 196
27 June - 6 July 2016
VILLA MONASTERO − VARENNA, LAKE COMO
68 69 70 71 72 73 74 75 76 78 79 80 81 82
77
83 84 85 86
63 64 65 40 66
67 43 49 55 56 57 58
36 61 38 44 45 46 47 48 51 52 27 59
37 62 39 41 42 50 54 32
53 28 29 30 60 31
8 9 10 11 12 13 18 19 20 21 22 23 24 25 26
14
35 7
33 34 16
3 4 6 15 17
1 2 5
1) Arianna Minonzio 16) Mostafa Tarboosh 31) Faisal Alqahtani 46) Hideaki Nozato 59) Natalie Belsey 74) Davide D’Ambrosio
2) Ramona Brigatti 17) Huang Che 32) Leon Chao 47) Kuniaki Amemiya 60) Steven Wang 75) Aaron Whittam
3) Benjamin Piper 18) Patrizia Tavella 33) Alejandro Savarin 48) Dario Imbraguglio 61) Paolo Durandetto 76) Daniel Stokes
4) Yu-Shan Yehe 19) Massimo Inguscio 34) Sebastian Galer 49) Alessandro Cultrera 62) Fang Fang 77) Sara Orski
5) Anna Tampellini 20) Martin Milton 35) Luisa Mandrile 50) Jinjie Wu 63) Daniel Beach 78) Wilson Ombati
6) Matteo Barbiero 21) Jürgen Mlynek 36) Marc Olivier Andrè 51) Pinh Yang 64) Yljon Seferi 79) Zachary Reed
7) Filippo Bregolin 22) Natascia De Leo 37) Valerio Formichella 52) Dennis Mkhize 65) Florian Stuker 80) Michael Moawad
8) Faranak Pourdanesh 23) Daniel Sarno 38) Ghislain Granger 53) Dorela Mirabela Albu 66) Marco Di Luzio 81) Meirbek Mussatayev
9) Thomas Udem 24) Edward C. Muhlern 39) Osamu Sato 54) Andreia De Lima 67) Dinesh Indika Dias 82) Rasmus Havelund
10) Walter Bich 25) Eleonora Cara 40) Masoud Dialameh Fioravante 68) Riccardo Ferrero 83) Adrian Vicbent Claramunt
11) Terry J. Quinn 26) Irma Kruger 41) Ilaria Finardi 55) Ionut Rasu 69) Elia Perego 84) Paul Green
12) Carl Williams 27) Carlos Luis Nobre 42) Enrica Pessana 56) Alice Lorenzati 70) Martin Parker 85) Kevin Thomson
13) Andy Henson 28) Mihaela Ispas 43) Moto Kinoshita 57) Valentin-Constantin 71) Peter Davis 86) Shahin Tabandeh
14) William Phillips 29) Takamitsu Otake 44) Ryunosuke Kuroda Mihalcea 72) Fabian Heisse
15) Barbara Alzani 30) Chien-Chih Yeh 45) Hironobu Takahashi 58) Nicholas Vlajic 73) Vicentiu Iancu
MODULE I. METROLOGY FOR QUALITY OF LIFE
Proceedings of the International School of Physics “Enrico Fermi”
Course 196 “Metrology: from physics fundamentals to quality of life”,
edited by P. Tavella, M. J. T. Milton, M. Inguscio and N. De Leo
(IOS, Amsterdam; SIF, Bologna) 2017
DOI 10.3254/978-1-61499-818-1-1
Reference methods and commutable reference

materials for clinical measurements
Vincent Delatour, Gustavo Martos, Julie Cabillic, Maryline Peignaux,
Catherine Perrot, Carine Fallot, Béatrice Lalere
and Sophie Vaslin-Reimann
Laboratoire National de Métrologie et d’Essais - 1 rue Gaston Boissier, 75015 Paris, France
Summary. — Reliability of medical tests is a major public-health challenge to

establish the correct diagnosis and adjust treatments. In order to evaluate and
improve comparability and reliability of medical tests, the input of metrology is
necessary. As imposed by the Directive 98/79/EC on in vitro diagnostic medical
devices establishing results traceability to higher-order reference methods and ma-
terials helps harmonizing results obtained in different laboratories, even if they do
not use the same method. Then, using reference methods to value assign quality
control materials makes it possible to assess trueness of routine methods. However,
a prerequisite is that quality control materials are commutable, that is to say that
they should mimic the behavior of patient samples that were not modified. In this
lecture, the importance of reference methods and commutability will be explained

through different examples.
1. – Introduction
.
1 1. Importance of medical tests reliability. – Sixty to seventy percent of medical
decisions being based on the result of in vitro diagnostic test [1], having reliable mea-
surements is critical to ensure that patients receive the correct diagnosis and treatment.
c Società Italiana di Fisica
1
2 Vincent Delatour, Gustavo Martos, Julie Cabillic, etc.
However, different methods are available to measure the same biomarker. Since results
are not always traceable to internationally recognized references, results will not always
be the same depending on the method used. Also, measurement uncertainties are not
always assessed and/or are too high. As a consequence, there can be a doubt on results,
which can have very serious consequences. From a health point of view, unreliable test
results induce a risk of inappropriate medical decision: while some patient will not be
given the needed treatment, some others will receive an unnecessary treatment that could
result in iatrogenic illness. From an economic point of view, mistreatment will worsen
the quality of life of patients and generate additional costs. Unreliable diagnostic tests
will also cause unnecessary repetition of measurements that represents a waste of money.
It has been estimated that 25 to 30% of costs are due to test repetitions, prevention and
error detection instead of diagnosis itself [2], which represents 15–30 billion $ per year
in the USA. In Germany, the repetition of analyses has an estimated cost of 1.5 billion
¤ per year. Last but not least, unreliable diagnostic tests can cause a lack of reliable
data for epidemiological studies and clinical trials, which hampers the understanding of
human pathologies and represents a break for the development of new treatments. As a
conclusion, it is very important to have reliable diagnostic tests and metrology has an
important role to play.
.
1 2. Role of a national metrology institute in bioanalysis. – National Metrology In-
stitutes (NMIs) are in charge of developing higher-order reference methods that have 2
main utilities. First, these methods can be used to assign target values to quality con-
trol materials used in External Quality Assessment Schemes (EQAS) that help medical
laboratories and manufacturers of in vitro diagnostic products to monitor the accuracy
of routine methods. Second, reference methods can be used to produce matrix Certified
Reference Materials (CRMs) with which routine methods will be calibrated. CRMs help
harmonizing tests results among different laboratories even if they do not use the same
method [3]. CMRs also help controlling the accuracy of routine methods.
An important challenge for NMIs is that hundreds of clinically relevant biomarkers ex-
ist and to date, only a few CRMs and reference methods are available because validating
reference methods is highly time consuming. To fill this gap, it is necessary to prioritize
biomarkers for which the need is the highest. Different criteria can be used. First, it
makes sense to consider the need for a metrological input. For example, results of Exter-
nal Quality Assessment Schemes are very helpful to identify biomarkers for which there is
poor agreement between the different available methods, suggesting that standardization
efforts are needed or that some methods need to be improved. Insufficient performance
and/or comparability of routine methods is not necessarily due to the absence of ref-
erence systems and can arise from an incorrect implementation of available standards.
Also, it should be kept in mind that developing new or alternative reference systems can
create confusion if results of the different available reference methods are not equivalent.
Another criterion to prioritize project is of course the prevalence of the concerned
pathology: although rare diseases should not be neglected, establishing traceability chains
for the most frequently measured biomarkers and/or biomarkers related with the most
Reference methods and commutability 3
frequent and/or costly pathologies will logically have a greater impact. In certain dis-
eases, the lack of reliable measurements preclude any attempt to examine the effectiveness
of emerging drugs, with the risk of overlooking potential benefits due to the poor quality
of measurement results. Then, other criteria can be used to prioritize projects like part-
ners susceptible to maximize impact or success of the project. The analytical challenge
is also to be considered because all NMIs do not have the capabilities needed to handle
challenging measurands: usually, NMIs progressively develop their skills and engage in
projects of growing complexity.
.
1 3. Examples of projects undertaken in LNE (France). – In bioanalysis, LNE (the
French NMI) is focused on the measurement of clinically relevant biomarkers routinely
measured in medical laboratories. The corresponding pathologies and biomarkers include:
– Diabetes Mellitus: Glucose & HbA1c
– Kidney Disease: creatinine, cystatin C, urea
– Cardio-vascular Disease: total cholesterol, triglycerides, HDL-C, LDL-C, non-HDL-

P, LDL-P
– Alzheimer’s Disease: Aβ & Tau
– Gout, metabolic syndrome: uric acid
– Electrolyte disorders: Na, K, Cl, Ca
– Heavy metal intoxication: Pb, As, Hg, Cd
– Anemia and iron-related disorders: Serum Iron, total hemoglobin, Hepcidin
Collection of needs and prioritization of projects is done in close cooperation with:
– public health authorities like the CDC (USA), the French Ministry of health (DGS),
the French Health Products Safety Agency (ANSM) and the French Health Au-
thority (HAS);
– international organizations: IFCC, AACC, BIPM, JCTLM
– National associations: SFBC, LABAC
– Clinicians and scientists from university hospitals and research institutes
– EQAS providers: ANSM, CTCB, Asqualab, ProBioQual, CSCQ, . . .
– National Metrology Institutes: LGC, NPL & NIBSC (UK), PTB & BAM (Ger-
many), INRIM (Italy), NIST (USA), HSA (Singapore), NMIJ (Japan), KRISS
(South Korea), NIM (China), INMETRO (Brazil), TUBITAK (Turkey)
.
1 4. Regulatory drivers and traceability chains in laboratory medicine. – Metrological
traceability is a property of a measurement result or of the value of a calibrator, whereby
it can be linked to the value of a stated reference (e.g. Certified Reference Material or
Reference Measurement Procedure) through a measurement or a series of measurements
all with known and stated uncertainties [4].
Manufacturers of diagnostic test kits sold in Europe must comply with the European
Community Directive 98/79/EC on in vitro diagnostic medical devices, including the
requirement according to which “the traceability of values assigned to calibrators and/or
control materials must be assured through available reference measurement procedures
and/or available reference materials of a higher order” [5].
According to ILAC (International Laboratory Accreditation Cooperation) elements
needed to confirm metrological traceability of results include: an unbroken metrological
traceability chain to an international measurement standard or a national measurement
standard, a documented measurement uncertainty, a documented measurement proce-
dure, accredited technical competence, metrological traceability to the SI, calibration
intervals [6]. The example of cholesterol was taken to explain the difference between
primary calibrators (purified compound of certified purity) and secondary reference ma-
terials (matrix materials in which the concentration is certified) and how results of routine
methods are made traceable to the higher order reference materials.
.
1 5. JCTLM . – Founded in 2002, the Joint Committee for Traceability in Laboratory
Medicine (JCTLM) is an international consortium that promotes the global standardiza-
tion of clinical laboratory test results, and provides information on Reference Materials,
Measurement Methods and Services that are available from around the world [7]. In or-
der to help manufacturers fulfil the requirements of the IVD Directive, the JCTLM has
developed and maintains a database of available certified reference materials (CRMs),
reference measurement methods that can be used by manufacturers to meet the trace-
ability requirements of the EC Directive. More than 200 certified reference materials
(CRMs) for about 130 measurands together with more than 125 reference measurement
methods/procedures for 76 different health markers are presently listed in the JCTLM
database. These have been reviewed and internationally agreed to be consistent with
the criteria described in the standard ISO 15194 [8]. and 15193 [9], respectively. The
JCTLM has also developed and maintains a database of laboratories worldwide pro-
viding reference measurement services for the value assignment of calibrators for the
IVD industry. Reference Measurement Services listed in the JCTLM database are from
laboratories that use a reference method published in the JCTLM database, will be
accredited as a calibration laboratory according to ISO 17025 [10] and ISO 15195 [11]
by an ILAC full member, and have participated regularly in inter-laboratory compar-
isons. This allows manufacturers to establish appropriate calibration hierarchies for their
measurement kits, and enables accurate and comparable results for patient care to be
achieved.
More recently, the JCTLM Working Group on Education and Promotion of Trace-
ability (WG-TEP), aims to produce and promote educational materials to demonstrate
the value of traceability in laboratory medicine as a means to reduce between method

variability in the interests of improved clinical outcomes and patient safety [12].
.
1 6. Accreditation according to ISO 15189 . – Following the reform of medical biology
operated in France, accreditation according to ISO 15189 [13] will become mandatory
for all clinical laboratories (both public and private) by 2020. The ISO 15189 standard
defines particular requirements for quality and competence of medical laboratories and
basically contains quality management requirements that are similar to those from ISO
9001 [14] and technical requirements that are similar to those from ISO 17025 standard.
Among the technical requirements, chapter 5.5.3 states that procedures should be
documented and that documentation should include —among others— calibration proce-
dures (metrological traceability) and performance specifications like “linearity, precision,
accuracy expressed as uncertainty of measurement, detection limit, measuring interval,
trueness of measurement, analytical sensitivity and analytical specificity”. Although it is
the responsibility of the IVD manufacturer to ensure that the calibrators they provide
are traceable to available and appropriate reference methods and/or materials, medical
laboratories also have the responsibility to document which reference their results are
traceable to.
Additional technical requirements describe how quality of examination procedures
should be assured and key concepts of metrology are well covered like measurement un-
certainties and results traceability to the SI. Chapter 5.6.2 states that medical laborato-
ries should determine results uncertainty and that uncertainty components of importance
should be taken into account, sampling, sample preparation, sample portion selection,
calibrators, reference materials. Chapter 5.6.3 states that “a programme for calibration
of measuring systems and verification of trueness shall be designed and performed so
as to ensure that results are traceable to SI units or by reference to a natural constant
or other stated reference”. Chapter 5.6.4 states that “the laboratory shall participate
in interlaboratory comparisons such as those organized by external quality assessment
schemes”. In the next chapter, the importance of reference methods in External Quality
Assessment Schemes will be explained.
2. – Importance of reference methods in EQAS
To monitor the performance of their analytical methods, medical laboratories have

different possibilities that are complementary to each other. First, they can use quality
control materials that are provided to medical laboratories by assay manufacturers or
producers of quality control materials. These samples can be used to estimate method
precision and repeating measurements over time can help identifying potential drifts.
Results obtained on quality control materials can also be compared with those obtained
in laboratories that are using the same method to ensure that the method is performing
according to manufacturer’s specifications. However, this does not allow estimating mea-
surement trueness unless target values have been assigned with a reference method. The
most reliable way to assess trueness is to use Certified Reference Materials but only a few
CRMs exist in laboratory medicine. The most popular strategy consists in participating
in External Quality Assessment schemes
However, a major limitation is that target values are generally consensus values corre-
sponding to the mean of results obtained by all participants and/or participants using the
same method. Consensual values can thus be biased, which could lead to interpretation
errors. For example, if most methods and/or the most frequently used method have a
positive systematic bias, the positive bias will not be detected and since the target value
will be positively biased, a non-biased method could be unduly charged to underestimate
results [15]. As a result, most of the current External Quality Assessment Schemes allow
assessing the ability of a laboratory to successfully implement a method but not to assess
its trueness. Only accuracy-based programs allow doing this.
3. – Importance of commutability
.
3 1. Why commutability matters. – Although reference method target values are neces-
sary to estimate methods trueness, another prerequisite is needed: quality control mate-
rials must be commutable, i.e. mimic the behavior of patient samples [16, 17]. Contrary
to authentic clinical specimens (that are commutable by definition), External Quality
Assessment materials can be subject to matrix effects that are due to the different treat-
ments applied to the materials during their preparation (e.g. lyophilization, freezing,
addition of preservatives and/or cryoprotectants, pooling, spiking with exogenous chem-
icals or purified compounds to increase concentration. . . ). Interestingly, chapter 5.6.4
of the ISO 15189 standard states that external quality assessment programmes should
provide clinically relevant challenges that mimic patient samples. Non-commutability of
EQA materials may skew trueness assessment because the difference between the refer-
ence method target value and the measured value will actually correspond to the sum of
the analytical bias (that is due to the method itself) and the bias due to matrix effects
(that is due to sample non-commutability). Given that non-commutability can affect
some methods more than others (a material could be commutable for one method but
not for another one), non-commutability of EQA materials can also compromise the abil-
ity to evaluate the agreement between the different methods [18]. Commutability is also
an important property for calibrators in the sense that using a non-commutable calibra-
tor will induce a calibration bias and a systematic error [19]. Using a non-commutable
material as international standard for harmonization purposes could also decrease the
agreement between methods instead of improving it [20].
.
3 2. Principle of commutability assessment. – VIM defines commutability as a prop-
erty of a reference material, demonstrated by the closeness of agreement between the
relation among the measurement results for a stated quantity in this material, obtained
according to two given measurement procedures, and the relation obtained among the
measurement results for other specified materials [4].
Conventional approaches for commutability assessment like those described in CLSI

EP 14 [21] and CLSI C53A [22] guidelines, samples whose commutability should be
assessed and at least 20 patient samples should be measured in the same analytical
sequence by at least 2 different methods. Using results of the patient samples, a Deming
regression line is then established along with a 95% prediction interval. The commonly
used prediction interval approach classifies a tested material as commutable when its
results are within a 95% prediction interval that represents the statistical distribution of
patient sample results within which future observations will fall.
A first limitation of the prediction interval approach is that the uncertainty is ne-
glected, which is equivalent to testing the hypothesis of non-commutability on the 50%
level of significance. A second limitation of the prediction interval approach is that ac-
ceptance criteria vary for different pairwise comparisons, making it difficult to define
generally applicable criteria. A third limitation of the prediction interval approach is
its inability to use criteria on the basis of intended use, because the difference in bias
between a RM and patient samples is not estimated. For these reasons, new statistic
designs were developed among the IFCC working group on commutability [23]. The dif-
ference in bias approach consists in measuring the bias between pairs of methods for a
panel of fresh unadulterated clinical specimens (that are commutable by definition) and
then compare it with the bias observed for the materials whose commutability is to be
assessed. The difference in bias DRM is used as a quantitative estimate of commutabilty:
it somehow corresponds to the level of agreement between the behavior of the material
under investigation and the set of fresh clinical specimens. As uncertainties are taken
into account in the difference in bias approach, a material can be found commutable,
non-commutable but also, the statistical analysis can be inconclusive when uncertainties
overlap with the acceptance limits. Acceptance criteria can also be chosen depending
on the intended use of materials: more stringent acceptance criteria are required for
calibrators than for trueness verifiers and EQA materials.
4. – Conclusions and perspectives
Metrology has a key role to play to help monitoring and improving reliability of med-
ical tests. Reference methods can be used to produce certified reference materials that
will help harmonizing tests results across different laboratories that do not use the same
method. Reference methods can also be used to assign target values to quality control
materials aimed at estimating methods accuracy. However, an important prerequisite is
to have commutable materials.
REFERENCES
[1] Forsman R. W., Why is the laboratory an afterthought for managed care organizations?
Clin Chem., 42 (1996) 813.
[2] https://www.nist.gov/director/e-health-and-technology-empowering-consumers-
simpler-more-cost-effective-health-care-system.
[3] Miller W. G., Tate J. R., Barth J. H. and Jones G. R., Harmonization: the sample,
the measurement, and the report. Ann. Lab. Med., 34 (2014) 187.
[4] ISO - International Organization for Stardardization, NF ISO/CEI Guide 99: International
vocabulary of metrology — Basic and general concepts and associated terms (VIM).
[5] ILAC Document ILAC P10:2002 “Policy on Traceability of Measurement Results” (2002).
[6] Directive 98/79/EC of the European Parliament and of the Council of 27 October 1998 on
In Vitro Diagnostic Medical Devices.
[7] http://www.bipm.org/jctlm/.
[8] ISO - International Organization for Standardization, ISO 15194 (2009) In vitro diagnostic
medical devices: Measurement of quantities in samples of biological origin: requirements
for certified reference materials and the content of supporting documentation.
[9] ISO - International Organization for Standardization, ISO 15193 (2009) In vitro diagnostic
medical devices: Measurement of quantities in samples of biological origin: requirements
for content and presentation of reference measurement procedures.
[10] ISO - International Organization for Standardization, ISO 15195 (2003) Laboratory
medicine: requirements for reference measurement laboratories.
[11] ISO - International Organization for Standardization, ISO 17025 (2005) General
requirements for the competence of testing and calibration laboratories.
[12] http://www.bipm.org/en/committees/cc/wg/jctlm-wg-tep.html.
[13] ISO - International Organization for Standardization, ISO 15189 (2012) Medical
laboratories: Requirements for quality and competence.
[14] ISO - International Organization for Standardization, ISO 9001 (2015) Quality
management systems - Requirements.
[15] Heuillet M., Lalere B., Peignaux M., De Graeve J., Vaslin-Reimann S., Pais De
Barros J. P., Gambert P., Duvillard L. and Delatour V., Validation of a reference
method for total cholesterol measurement in human serum and assignation of reference
values to proficiency testing samples. Clin Biochem., 46 (2013) 359.
[16] Miller W. G., Myers G. L. and Rej R., Why commutability matters. Clin Chem., 52
(2006) 553.
[17] Miller W. G. and Myers G. L., Commutability still matters. Clin Chem., 59 (2013)
1291.
[18] Miller W. G., Jones G. R., Horowitz G. L. and Weykamp C., Proficiency
testing/external quality assessment: current challenges and future directions. Clin Chem.,
57 (2011) 1670.
[19] Vesper H. W., Miller W. G. and Myers G. L., Reference materials and commutability.
Clin Biochem. Rev., 28 (2007) 139.
[20] Zegers I., Beetham R., Keller T., Sheldon J., Bullock D., MacKenzie F.,
Trapmann S., Emons H. and Schimmel H., The importance of commutability of

reference materials used as calibrators: the example of ceruloplasmin. Clin Chem., 59
(2013) 1322.
[21] Clinical & Laboratory Standards Institute, EP14-A2: Evaluation of Matrix Effects.
[22] Clinical & Laboratory Standards Institute, C53-A - Characterization and Qualification of
Commutable Reference Materials for Laboratory Medicine.
[23] http://www.ifcc.org/ifcc-scientific-division/sd-working-groups/
commutability-wg-c/.
DOI 10.3254/978-1-61499-818-1-9
Reference measurement systems for biomarkers:

Towards biometrology
Vincent Delatour, Gustavo Martos, François Gaie-Levrel,
Noémie Clouet-Foraison, Pauline Bros, Maud Heuillet,
Béatrice Lalere and Sophie Vaslin-Reimann
Laboratoire National de Métrologie et d’Essais - 1 rue Gaston Boissier, 75015 Paris, France
Summary. — Reference methods and Certified Reference Materials are neces-

sary to assess and improve comparability and reliability of medical tests. However,
development of reference measurement systems is more complex for some biomark-
ers than for others. While the state of the art in metrology is mature enough to
address the case of most metabolites and electrolytes quite easily, absolute quantifi-
cation of protein biomarker is much more difficult due to many factors. Hyphenated
quantification techniques are often necessary to measure low concentrations of large
proteins in complex matrices. As different isoforms can have different clinical sig-
nificance, structural heterogeneity of proteins complicates the situation even further
and requires to carefully define the measurand. Due to the relationship between
folding and activity, absolute quantification of some complexity proteins remains
a challenge and some biomarkers are so large that they can even be considered as
bionanoparticles. Through different case studies, this presentation will highlight a

number of challenges associated with absolute quantification of biological entities
and establishment of traceability chains in clinical diagnostics.
1. – Introduction
At LNE (the French National Metrology Institute), reference methods have been and
are being developed for different types of clinically relevant biomarkers that are related
to the most common human diseases. Metabolites include glucose (diabetes mellitus),
9
10 Vincent Delatour, Gustavo Martos, François Gaie-Levrel, etc.
creatinine and urea (kidney disease) and uric acid (gout and metabolic syndrome). Elec-
trolytes include sodium, potassium, chloride, calcium (electrolyte disorders) and iron
(anemia). Heavy metals include lead, arsenic, mercury, cadmium (heavy metals in-
toxication). Proteins include HbA1c (diabetes mellitus), cystatin C (kidney disease),
amyloid beta and tau (Alzheimer’s disease), Procalcitonin (Sepsis) total hemoglobin and
hepcidin (anemia and iron-related disorders). Lipids and lipoproteins include total choles-
terol, triglycerides, HDL-C, LDL-C, LDL-P and non-HDL-P (cardiovascular diseases).
Depending on the type of biomarker, absolute quantification relies on different method-
ologies and equipment: while metabolites and lipids are generally measured by GC-IDMS
or LC-IDMS (isotope dilution mass spectrometry coupled to gas or liquid chromatogra-
phy), electrolytes are usually measured by ICP/MS (Inductively Coupled Plasma Mass
Spectrometry). Peptides and proteins are usually measured by LC-IDMS with or with-
out a preliminary purification step and/or a digestion with an enzyme. Larger entities
like protein aggregates or lipoproteins can be measured with methods usually applied
to nanoparticle analysis. In this presentation, different examples of projects conducted
at LNE will be taken to illustrate the different methodologies and equipment that are
used to perform absolute quantification of clinically relevant biomarkers. The clinical
relevance of each biomarker will be explained in order to justify why traceability chains
are needed and how this work is connected to standardization initiatives and medical
needs.
2. – Metabolites and small molecules

.
2 1. Kidney disease: creatinine. – In France, 2 million people are suffering from chronic
kidney disease. Early detection can help preventing the progression of chronic kidney
disease to kidney failure, which has an estimated cost of 1.5 billion ¤ per year in France.
Creatinine is the main biomarker used for the diagnosis and follow up of kidney disease.
In 2015, creatinine measurement was the 6th most frequent diagnostic test performed in
France with about 19 million measurements [1].
Reimbursement of these tests represented a cost of about 28 million euros for the
French healthcare system. Different routine methods are used to measure serum crea-
tinine in medical laboratories and 3 main principles can be distinguished: colorimetric
methods based on the Jaffe reaction, enzymatic methods and dry chemistry methods. As
pointed out in different quality assessment schemes, results obtained with these different
methods are not always equivalent. Reference methods can help improving this situation
and also help determining which method(s) is/are the most accurate. Different studies
showed that enzymatic methods should be preferred because they provide more accurate
results regardless of the creatinine concentration and sample composition [2-4].
From a metrological point of view, reference methods listed in the database of the
Joint Committee for Traceability in Laboratory Medicine (JCTLM) mostly rely on GC-
IDMS and LC-IDMS [5]. In the case of GC/MS, a preliminary purification step is needed
to separate creatinine from creatine because these compounds cannot be distinguished
during the mass spectrometry analysis performed after derivatization. In the case of
Towards biometrology 11
LC/MS, separation of creatine and creatinine is directly performed online by liquid chro-
matography. In both cases, sample preparation is quite straightforward and very low
uncertainties can be obtained. From a medical point of view the National Kidney Disease
Education Program recommends that bias of routine methods for serum creatinine should
not exceed 3% and imprecision should not exceed 3%, so that inaccuracy of creatinine
measurement don’t induce a bias of more than 10% of the glomerular filtration rate [6]
(that is calculated with equations that include creatinine concentration). A common
consensus being that relative expanded uncertainty of a reference method should range
between one third and one half of the target uncertainty of routine methods, reaching
uncertainties around 1% is actually needed for reference methods for creatinine. Results’
traceability to the International System of Units (SI) is achieved through gravimetric
preparation of calibrators consisting in solutions of labelled creatinine and non-labelled
creatinine of thoroughly characterized purity (e.g. NIST SRM 914a). It thus appears that
all metrology requirements are met to have reliable traceability chains but other issues
like commutability of calibrators can hamper proper implementation of reference systems.
.
2 2. Diabetes mellitus: glucose. – Glucose is another example of metabolite for which
reference methods have successfully been validated in a number of National Metrology
Institutes (NMIs) and for which relative expanded uncertainties at the 1% level can be
obtained [7]. Glucose is the main biomarker used for diagnosis and monitoring of diabetes
mellitus. The number of people with diabetes has risen from 108 million in 1980 to 422
million in 2014 and this number is expected to rise to 642 million by 2040 [8]. In
2012, an estimated 1.5 million deaths were directly caused by diabetes and another 2.2
million deaths were attributable to high blood glucose. Diabetes can be treated and
its consequences avoided or delayed with diet, physical activity, medication and regular
screening and treatment for complications. Accurate tests results are therefore crucial
to identify patients with diabetes and adjust treatment of already identified patients. In
2015, glucose measurement was the second most frequent diagnostic test performed in
France with about 25 million measurements. Reimbursement of these tests represented
a cost of about 25 million euros for the French healthcare system [1].
Like creatinine and other metabolites, reference methods for glucose listed in JCTLM
database mostly rely on GC-IDMS and LC-IDMS [5]. In both cases, sample prepara-
tion is quite straightforward (derivatization and/or protein precipitation) and very low
uncertainties can be obtained. Results’ traceability to the SI units is achieved through
gravimetric preparation of calibrators consisting in solutions of labelled glucose and non-
labelled glucose whose purity was thoroughly characterized (e.g. NIST SRM 917c).
3. – Peptides and proteins
While higher-order reference methods and reference materials are available for a num-
ber of metabolites measured in laboratory medicine an important gap remains to be filled
for many clinically relevant biomarkers and especially proteins.
.
3 1. Absolute quantification of peptides and proteins. – In metrology institutes, like
in research and routine laboratories, quantification of proteins by LC/MS/MS is usually
performed with the bottom-up approach, which consists in measuring signature peptides
resulting from the enzymatic cleavage of the protein of interest. The chosen peptides
should be specific of the target protein, which means that they should not be found in
other proteins present in the biological sample. When the size of the protein of interest is
sufficiently small and/or if instrumentation allows it, direct analysis of the intact protein
can be performed. The so-called top-down approach is often used to identify proteins [9]
while the bottom-up approach is mainly used for quantification purposes.
Given the usual low concentration of protein biomarkers and the presence of com-
peting proteins or peptides in the ionization source, purification techniques are often
necessary to measure proteins in complex matrices. Different methods can be used and
potentially combined to simplify the matrix: Solid Phase Extraction, selective precipita-
tion, immuno-affinity, preparative HPLC, etc.
In order to compensate for both potential material loss during sample preparation and
variability of the different measurement steps, an internal standard is generally added to
the sample, ideally at the beginning of sample preparation [10]. In the ideal situation,
the internal standard should consist in a labelled homologue of the protein of interest.
Labelled proteins are generally recombinant proteins that are produced in a medium
enriched with stable isotopes but new approaches label recombinant proteins afterwards.
The main advantage of using labelled proteins is that the behavior of the internal stan-
dard will be very close to that of the endogenous protein and potential incompleteness
of the digestion will be compensated. The drawback is that recombinant proteins are
usually very expensive and sometimes difficult to produce. For that reason, the most
popular strategy consists in using labelled peptides as internal standards. This is a
much more cost effective approach but in this case, completeness of the digestion is crit-
ical to obtain accurate results. To better account for digestion incompleteness without
requiring full length proteins, peptide concatamers are sometimes used as internal stan-
dards. These standards consist in peptides containing the peptide(s) of interest, whose
sequence is bracketed by (trypsin) cleavage sites and other reporter peptides. Although
this approach helps taking into account digestion incompleteness, it is assumed that the
digestion yield will be the same for the peptide concatamer and for the full length pro-
tein. This assumption is probably reasonable for linear or unstructured proteins, but
probably much less valid for folded and/or cross-linked proteins from which the target
peptides are not easily accessible.
Although using full length proteins as standards certainly has many advantages, it
still introduces a major challenge for National Metrology Institutes. Establishing re-
sults traceability to the SI Units requires accurate determination of primary calibrators’
purity [11]. While the state of the art in metrology is sufficient to address purity as-
sessment of metabolites and electrolytes — but still challenging [12] — purity assess-
ment of long peptides and proteins is currently a bottleneck in absolute quantification of
biomolecules [13, 14]. Due to the strong structural heterogeneity of proteins, even highly
purified protein preparations contain a very large number of impurities and the mass bal-
ance approach can hardly be used. In addition, purity depends on measurand definition
and a lot of care should be taken to clearly define what the clinically relevant form of the
protein is. How clinical decision limits have been established should also be considered.
When peptide calibrators are used, purity assessment is easier than for proteins be-
cause peptides are usually much shorter and impurity identification is simplified due to a
less ambiguous measurand definition. However, even highly purified peptides of reason-
able size contain a large number of impurities and purity assessment remains challenging.
Different approaches exist for the purity value assignment: the mass balance approach,
the peptide-impurity-corrected amino acids (PICAA) analysis and the peptide-impurity-
corrected quantitative NMR. High-resolution mass spectrometry (HRMS) is a key tech-
nique required in all cases. For mass balance, all the impurities of powder standards are
identified and quantified. In the PICAA, the standard is in solution and all the peptide-
related impurities are identified and quantified by HRMS. Given that the mass balance
approach requires very large amounts of material and requires considering all forms of
impurities, PICAA is the most popular strategy. The concentration of the peptide of
interest is determined by amino acid analysis (i.e. absolute quantification of amino acids
released from peptide hydrolysis.) after the subtraction of the peptide related impuri-
ties contribution. Different prerequisites are necessary to obtain accurate results. First,
determination of peptide concentration by amino acid analysis implies completeness of
hydrolysis of peptides into amino acids. Hydrolysis is generally performed during pro-
longed periods of time (40 hours) in strong acidic conditions (6 mol/L HCl), either in
liquid phase or in gas phase with or without microwave assistance. Not all amino acid
are stable enough to withstand such harsh conditions and only some amino acids are
appropriate: valine, isoleucine, leucine, proline phenylalanine, arginine and alanine are
the most commoly used amino acids. A first prerequisite is therefore that the peptide
of interest contains some of these amino acids (at least 3, ideally). Another prerequisite
is that the amino acids measured after hydrolysis only come from the peptide of inter-
est: in the presence of free amino acids and/or peptide impurities that contain some
of the target amino acids, concentration of the target peptide will be overestimated. It
is therefore necessary to identify impurities that are present along with the peptide of
interest. Depending on the intended use, a decision must be made concerning the neces-
sity to quantify impurities and subtract their contribution from the measured amino acid
concentrations. In some cases, iterative peptide purification can potentially make it un-
necessary to correct the contribution of remaining impurities but the cost of high-purity
peptides is often very high. On the contrary, using peptides of poor purity is still possible
but in that case, identifying and quantifying peptide impurities is necessary. The main
advantage of this approach compared with mass balance is that only impurities contain-
ing amino acids need to be considered. The drawback is that measurement uncertainties
are often larger than with the mass balance approach and might not fulfill the medical
need in cases where very low uncertainties are needed.
.
3 2. Diabetes mellitus: HbA1c. – HbA1c is one of the most frequently measured
protein biomarker in medical laboratories with about 10 million measurements per year
in France [1]. Reimbursement of these tests represented a cost of about 68 million euros
for the French healthcare system in 2015 (6th most expensive analysis for French health
insurance). Compared with glucose concentration in serum, HbA1c concentration is less
sensitive to transitory variations and reflects average blood glucose during the 4 to 6
weeks preceding the assay. French Health authorities encourage using this biomarker as
a diabetes follow-up tool and there is also growing evidence that HbA1c can be used for
diagnosis of diabetes mellitus.
HbA1c is a very good example of biomarker for which establishing traceability chains
greatly improved comparability between the different methods [15]. The current refer-
ence system is maintained by the IFCC HbA1c network and relies on the IFCC reference
method published in 2002 [16]. After a digestion step with endoproteinase GluC, separa-
tion and quantification of the glycated and non-glycated N-terminal hexapeptides of the
beta-chain is done by LC/MS or capillary electrophoresis coupled with UV-detection.
Results are expressed in mmol/mol as the ratio between concentrations of the glycated
hexapeptides and the sum of the non-glycated hexapeptide and the glycated hexapep-
tides. Initially, calibrators used for the reference method consisted in mixtures of highly
purified HbA1c (IRMM-IFCC 466) and HbA0 (IRMM-IFCC 467) of certified purity.
IRMM-IFCC 466 and 467 now being out of stock, the calibration of the reference method
now relies on a set of already prepared mixtures of purified HbA1c and HbA0 of certified
concentration.
New methods relying on LC-IDMS have been validated [17, 18] but are not yet listed
in the JCTLM database so the official reference method for HbA1c remains the IFCC
method. IDMS methods rely on peptide calibrators instead of purified proteins and quan-
tification is done by LC/MS/MS instead of LC/MS. Equivalence of the IDMS and the
IFCC methods remains an open question and ongoing projects are currently addressing
this important issue. Indeed, switching from a reference system to another one is not
trivial because it would potentially necessitate redefining clinical-decision limits and ed-
ucate clinicians in case the different reference methods do not provide equivalent results:
strong arguments must be provided to justify why changing an existing reference system
would be beneficial and how results will be affected [19].
.
3 3. Sepsis and antimicrobial resistance: Procalcitonine. – Early identification di-
agnosis and appropriate therapy of severe sepsis remains a real challenge in intensive
care units. Despite new therapies and guidelines developed over the past decade, sep-
sis incidence and mortality remain high. Every year, severe sepsis strikes an estimated
population of 750 000 people in the US alone resulting in 100 000 deaths. The intensive
care specialists took the challenge to overcome the current situation and to reduce sepsis
mortality significantly by implementing evidence-based clinical standards for diagnosis
and treatment of sepsis worldwide. Procalcitonin (PCT) is increasingly recognized as
the most promising marker for risk stratification to progression to severe sepsis or septic
shock. PCT is an early, sensitive and very specific marker of severe bacterial infections
(sensitivity 93%, specificity 96%). High PCT concentrations allow the early identification
of patients prone to develop severe sepsis or septic shock. The medical area of application
of PCT measurement is rather broad as it includes discrimination between inflammatory

disease and infectious complication or between bacterial/parasitic and viral infection, but
also allows assessing antibiotic treatment efficiency. PCT measurements have a key role
to play in the reduction of the excessive use of antibiotics, which is essential to combat
the increase of antibiotic-resistant microorganisms [20, 21]. In a clinical trial involving
more than 200 patients with suspected infections of the lower respiratory tract, utilizing
the biomarker PCT as decision maker allowed reducing the antibiotic prescriptions by
50% [22]. When an antibiotic-based treatment is needed, PCT demonstrated its interest
to monitor its efficiency and adapt posology. Persisting high or further increasing PCT
levels —indication for non-controlled infectious process— justifies a re-assessment of the
therapeutic strategy. Overall, having reliable PCT measurements is of major interest for
the accurate identification of bacterial infections in patients presenting at the emergency
department to allow early and rational antibiotic treatment.
While PCT concentrations in serum are elevated in severe sepsis or septic shock
(> 2 μg/L), PCT concentrations between 0.1 to 0.5 μg/L are indicative of bacterial infec-
tion associated with a moderate risk of sepsis. Ensuring accuracy of PCT measurements
in this concentration range is thus very important. In the context of the EMPIR Project
AntiMicroResist [23], a reference method and certified reference materials for PCT will
be developed with the objective to assess and improve accuracy and comparability of
PCT measurements conducted in routine. Measuring low PCT concentrations is a chal-
lenge that requires the development of hyphenated purification techniques. In addition,
certification of primary calibrators consisting in solutions of purified PCT represents an
interesting case study on purity assessment of large recombinant proteins. As differ-
ent isoforms of PCT exist, considerations related with measurand definitions are to be
addressed at the earliest stages of the standardization project.
.
3 4. Alzheimer’s disease: amyloid beta & tau. – Along with imaging and neuropsycho-
logical tests, concentrations of amyloid beta and tau in Cerebrospinal Fluid (CSF) are
the main biomarkers used for the diagnosis of Alzheimer’s disease [24]. However, devel-
opment of traceability chains for the analysis of these biomarkers is not straightforward,
as explained hereafter.
It is now well established that Aβ1-42 concentration in CSF is significantly reduced
in patients with Alzheimer’s disease compared with healthy controls [25]. It is hypoth-
esized that formation of amyloid plaques results in reduced amounts in the CSF [26].
From a metrological point of view, development of reference methods for amyloid beta
is difficult for different reasons. First, concentrations of amyloid beta in CSF are very
low and quantification by mass spectrometry implies a sample preparation step to reduce
matrix complexity. Second, more than 40 different endogenous amyloid beta have been
identified in CSF, including species with truncations and/or post translational modifi-
cations. It is therefore needed that analytical methods are able to discriminate these
different compounds that have a similar structure but potentially very different clini-
cal significance. Third, it is well known that amyloid beta are prone to non-specific
adsorption and aggregation. This imposes taking precautions during the pre-analytical
phase but also during preparation and certification of primary calibrators. The reason is
that synthesis of peptides used as primary calibrators can result in large aggregates that
will not be detected by mass spectrometry but still release amino acids during the acid
hydrolysis that is performed to value assign peptide concentration by amino acid anal-
ysis. Aggregation of amyloid beta in calibration solutions could thus induce a positive
calibration bias and precautions should be taken to account for this.
An elevated concentration of tau protein in CSF is considered as a marker for the
severity of the neuronal degeneration and several clinical studies demonstrated a signifi-
cant increase of tau concentration in CSF of patients having Alzheimer’s disease [27-29].
Compared with total tau, phosphorylated forms of tau are a more specific biomarker
for Alzheimer’s disease [30, 31]. Tau phosphorylated at Thr181 is the most commonly
analyzed phospho-form of tau in CSF [32] and is consistently increased in Alzheimer’s
disease [33]. Several studies shown that measuring p-tau in CSF provides a diagnostic
specificity of 80–100% [34, 35]. Absolute quantification of tau by mass spectrometry is
complex due to its low concentration in CSF, its important size and the presence of six
different isoforms ranging from 352 to 441 amino acids with molecular weights between
50 and 65 kDa [36, 37]. Another difficulty is related to the strong structural heterogene-
ity of Tau, which includes approximately 80 potential phosphorylation sites [38]. Here
again, measurand definition is a challenging issue that should be carefully addressed
before conducting a standardization initiative.
The developement of fully automated analytical platforms is expected to facilitate the
introduction of CSF biomarkers in clinical routine but several studies highlighted a large
measurement variability of Alzheimer’s disease biomarkers performed in different labo-
ratories [39]. This suggests that there is a need for certified reference materials (CRM)
and reference measurement procedures (RMP) to harmonize the different assays used so
that absolute values can be compared worldwide. A candidate reference method Aβ1–42
has recently been published [40] and candidate CRMs are currently being developed by
the IFCC working group on CSF proteins [41] in conjunction with the EMPIR project
NeuroMet [42].
.
3 5. Iron-related disorders: hepcidin. – Although hepcidin is still an emerging
biomarker, recent studies suggest that it could play a role in iron perturbation occur-
ring in some iron-related disorders and potentially neurodegenerative disorders such as

Alzheimer’s disease and Parkinson’s disease. Hepcidin quantification is challenging be-
cause it is prone to aggregation and non-specific adsorption. In addition, hepcidin con-
tains 4 disulfide bridges [43] and the endogenous connectivity is difficult to reproduce for
custom-synthesis of primary calibrators used for hepcidin quantification and for SI trace-
ability. Since ELISA tests are based on epitope recognition, using misfolded hepcidin
to calibrate immunoassays could induce a bias in quantification. Therefore, a suitable
primary reference material of well characterized purity is needed.
Although the PICAA approach can be used to identify and quantify hepcidin related
peptide impurities, the different hepcidin foldings cannot be discriminated by classical
HRMS. In this case, ion mobility is a helpful tool to discriminate different forms of
hepcidin having the same mass but different folding states. However, the quantification
of the different hepcidin foldings is relative only because ion mobility did not prove yet its
ability to quantify the ratio between mobiloformes in an absolute manner. The primary
calibrator consisting in purified hepcidin that will be made available soon can only be
used to calibrate mass spectrometry methods in which detection is based on hepcidin
molecular weight and in which hepdicin folding is not considered. Standardization of
immunoassays can still be performed through matrix reference materials that are value
assigned with well-calibrated mass spectrometry methods [44].
4. – Lipids and lipoproteins

.
4 1. Cardiovascular diseases: cholesterol, triglycerides, LDL-C and HDL-C . – It is
well established that high concentrations of blood cholesterol are a major risk factor
for coronary heart disease [45, 46] and that lowering blood cholesterol concentration de-
creases morbidity and mortality [47, 48]. It was evidenced that a reduction of 1 mmol/L
of the total cholesterol concentration was associated with a 50% reduction of the coro-
nary mortality for a middle-aged and mixed-gender population. In 1985, the National
Cholesterol Education Program (NCEP) began an unprecedented campaign to educate
the medical community and the general public about the risk factors for coronary artery
disease. To identify people at risk for coronary artery disease, the NCEP recommended
initial classification using the medical decision points of 200 and 240 mg/dL (5.17 and
6.21 mmol/L) for total cholesterol [49]. Proper classification of a patient’s risk based
on these medical decision points requires reliable and standardized cholesterol measure-
ments. Therefore, the NCEP also developed recommendations for accuracy and preci-
sion [50]. Thus, cholesterol became the first, and for many years was the only, clinical
analyte for which national performance standards were available. A practical approach to
standardizing cholesterol measurements is to ensure that diagnostic products are properly
calibrated by the manufacturers. Historically, cholesterol measurements were standard-
ized with the Abell-Kendall (AK) colorimetric method [51]. However, some compounds
such as non-cholesterol sterols, cholesterol precursors and oxidation products were also
measured [52]. Due to this lack of specificity, most recent reference methods listed in the
JCTLM database are based on IDMS. Like glucose and creatinine, IDMS-based reference
methods for absolute quantification of total cholesterol in serum samples allow reaching
relative expanded uncertainties around 1% and establishing results traceability to the SI

units through the use of primary calibrators of well characterized purity [53]. In 1990,
CDC established the Cholesterol Reference Method Laboratory Network (CRMLN) to
provide reference services to manufacturers. Working with the manufacturers is the most
effective means, with the greatest impact, of standardizing the measurement of lipids and
lipoproteins within the clinical laboratory community and the Food and Drug Admin-
istration (FDA) has recognized the value of the CRMLN’s certification program and
encourages manufacturers to certify their products through this network [54].
Although measurement of total cholesterol concentration has long been the most used
diagnostic test to assess the risk of cardiovascular diseases (CVD), CVD risk assessment
is now performed by measuring the concentration of triglycerides, total cholesterol but
most importantly cholesterol associated with Low Density Lipoproteins (LDL-C, some-
times referred to as “bad cholesterol”) and cholesterol associated with High Density
Lipoproteins (HDL-C, sometimes referred to as “good cholesterol”). The NCEP issued
different reports emphasizing the importance of HDL-C and LDL-C in identifying people
at risk for coronary heart disease [49, 55, 56]. It was evidenced that reduction of LDL-C
was associated with a reduction of major cardiovascular events while an inverse correla-
tion between HDL-C and CVD events was observed. Determination of the lipid profile,
including total cholesterol (TC), HDL-C, LDL-C and triglycerides (TG) measurements,
was the third most costly medical test in France in 2015 with an estimated expenditure
of ¤ 94 M related to the reimbursement of the medical tests [1]. In addition, the costs
associated with the reimbursement of cholesterol lowering drugs (e.g. statins) exceeded
1 billion euros in 2015 in France. With the advent of homogeneous reagents for direct
measurements of HDL-C and LDL-C, traceability of diagnostic systems to the accuracy
bases becomes even more critical [57]. The NCEP’s Working Group on Lipoprotein Mea-
surement issued performance guidelines for HDL-C [55] and LDL-C [58]. HDL-C should
be measured with a bias from the reference method ≤ 5%. The methods should report
CVs ≤ 4% at ≥ 42 mg/dL (1.09 mmol/L) and a standard deviation of ≤ 1.7 mg/dL
(0.044 mmol/L) at < 42 mg/dL (1.09 mmol/L). Clinical laboratories should measure
LDL-C with a bias from the reference method ≤ 4% and perform with a CV ≤ 4%.
The CRMLN now certifies manufacturers of clinical diagnostic products that mea-
sure total cholesterol, TG, HDL-C, and LDL-C. The CRMLN laboratories use reference
methods or designated comparison methods that are rigorously standardized to the CDC
reference methods [59]. These methods are quite challenging because imprecision is much
larger than other reference methods for lipids like total cholesterol. The reason is that
the different lipoprotein fractions should be separated (by selective precipitation) before
cholesterol is measured by AK or IDMS. From a metrological point of view, measur-
and definition is very questionable because lipoproteins are defined as function of den-
sity, which means that only ultracentrifugation can be used to discriminate the different
lipoprotein fractions. Although other analytical methods are now available, the fact that
clinical-decision limits were established on the basis of the beta quantification reference
method makes it very complicated to establish new traceability chains in this field.
.
4 2. Lipoprotein particle concentration: beyond LDL-C in CVD risk assessment. –
To date, CVD risk assessment mainly relies on lipid tests that consist in measuring the
concentration of TC, LDL-C, HDL-C and TG in serum. However, cholesterol-lowering
therapies only decrease the risk by about a half and the remaining residual risk for acute
CVD events remains important [60]. In the search of more predictive biomarkers for
CVD risk, the importance of lipoprotein particles themselves in the atherosclerosis path-
way rather than their lipid load was evidenced and new biomarkers emerged. Although
a strong correlation between large concentrations of apolipoprotein B (apoB) and CVD
risk was shown [61-63], an important finding is that all lipoproteins containing apoB do
not have the same atherogenicity. Increased concentrations of lipoprotein(a) were found
to be causally related to premature CVD [64-66] and LDL particle number concentration
(LDL-P) was also found to be a good predictor of patient’s risk of developing atheroscle-
rosis [67, 68]. Additionally, it has been demonstrated that LDL particles and especially
small-dense-LDL (sd-LDL) particles tend to be more atherogenic and that elevated con-
centrations of sd-LDL particles were associated with premature CVD events [69-73]. The
growing interest in hyphenated methods for lipoprotein analysis resulted in the develop-
ment of Advanced Lipoprotein Testing (ALT) methods that exploit the different ways
to separate and detect lipoproteins. Historically, lipoprotein classes and sub-classes were
defined according to their density (using ultracentrifugation) or their electrophoretic
mobility (using electrophoresis) [74]. More recent methods rely on other separation and
detection principles based on other characteristics of lipoproteins such as size (e.g. SEC),
protein content (e.g. LC/MS/MS, immuno-nephelometry/turbidimetry), electrical mo-
bility (e.g. ES-DMA) or proton resonance of specific lipids (e.g. NMR). Since measurand
definition is not the same, it should be expected that these methods do not provide com-
parable results and do not provide the same clinical information. A number of studies
have highlighted the poor comparability and the need for standardization for lipoprotein
subclass distribution and quantitation methods [75-77]. As stated in Chung et al. [78],
“the lack of standardization among analytical methods and reference material for stan-
dardizing each analytical method may be the major contributing factors accounting for
the wide range of agreements observed in the literature. Lack of harmonization and stan-
dardization precludes efforts to determine whether any biases identified are artefacts of
the comparison as a result of an inadequate reference material, or are genuine biases
among the methods examined. [. . . ] Development of reference materials that are widely
accepted as appropriate, accurate and reliable is necessary”. As indirectly discussed in
Master and Rader [79] and explicitly suggested by Otvos et al. [80] “collaborative stan-
dardization efforts between groups that perform particle-concentration measurements will
be required to enable its broader use in clinical practice”.
One key objective of the BioSITrace project was to produce reference materials for
ALT methods and especially for methods measuring particle concentrations. A prereq-
uisite was to evaluate how different lipoprotein enumeration methods compare with each
other to determine what method(s) should be used to value assign the reference materials.
The considered methods included ES-DMA, NMR, LC/MS/MS and ApoB quantification.
Since these methods have different measurands / specificities, it was anticipated that the
correlation would sometimes be poor but it turned out that agreement between the dif-
ferent methods was better than expected. The reason for this is that several methods are
to some extent harmonized through calibration with WHO reference reagents that are
value assigned with immuno-nephelometry. This work will be pursued within the new
IFCC working group on apolipoprotein standardization that will be launched in 2017.
5. – Conclusion
Development of Reference measurement systems for biomarkers is key to assess and

improve comparability and reliability of medical tests. However, some biomarkers are
more challenging than others and reference methods and Certified Reference materials
are still missing for a number of major biomarkers. Although harmonization still helps
improving agreement between the different methods in the short term, standardization
makes it easier to maintain traceability chains over time because SI-traceable materi-
als meet the highest-accuracy requirements and are less likely to fluctuate when new
calibrator lots are produced. Due to reasonable size, lower structural heterogeneity,
unambiguous measurand definition and easier purity assessment of primary calibrators,
standardization is much easily achieved for metabolites and electrolytes than for large
proteins and complex analytes. To this date, a major limitation is that folding and ac-
tivity of proteins is rarely considered because the current state of the art does not allow
addressing this challenging issue. It is expected that in the future, NMIs capabilities will
improve so as to cope with increasingly challenging projects and fill the gap.
REFERENCES
[1] http://www.ameli.fr/l-assurance-maladie/statistiques-et-publications/donnees
-statistiques/actes-de-biologie-medicale/biolam-2013-2015.php.
[2] Piéroni L., Delanaye P., Boutten A., Bargnoux A. S., Rozet E., Delatour V.,
Carlier M. C., Hanser A. M., Cavalier E., Froissart M. and Cristol J. P., A
multicentric evaluation of IDMS-traceable creatinine enzymatic assays, Clin. Chim. Acta.,
412 (2011) 2070.
[3] Boutten A., Bargnoux A. S., Carlier M. C., Delanaye P., Rozet E., Delatour
V., Cavalier E., Hanser A. M., Froissart M., Cristol J. P. and Piéroni L.,
Enzymatic but not compensated Jaffe methods reach the desirable specifications of
NKDEP at normal levels of creatinine. Results of the French multicentric evaluation, Clin.
Chim. Acta., 419 (2013) 132.
[4] Cobbaert C. M., Baadenhuijsen H. and Weykamp C. W., Prime time for enzymatic
creatinine methods in pediatrics, Clin. Chem., 55 (2009) 549.
[5] http://www.bipm.org/jctlm/.
[6] Myers G. L., Miller W. G., Coresh J., Fleming J., Greenberg N., Greene T.,
Hostetter T., Levey A. S., Panteghini M., Welch M. and Eckfeldt J. H., National
Kidney Disease Education Program Laboratory Working Group, Recommendations for
improving serum creatinine measurement: a report from the Laboratory Working Group
of the National Kidney Disease Education Program, Clin. Chem., 52 (2006) 5.
[7] Welch M. J., Sniegoski L. T., Parris R. M., May W. E., Heo G. S. and Henrion
A., CCQM-K11: The determination of glucose in serum, Metrologia, 40 (2003) 08003.

[8] http://www.diabetesatlas.org/.
[9] Catherman A. D., Skinner O. S. and Kelleher N. L., Top Down proteomics: facts
and perspectives, Biochem. Biophys. Res. Commun., 445 (2014) 683.
[10] Brun V., Masselon C., Garin J. and Dupuis A., Isotope dilution strategies for absolute
quantitative proteomics, J. Proteomics., 72 (2009) 740.
[11] Josephs R. D., Stoppacher N., Daireaux A., Choteau T., Westwood S. and
Wielgosz R. I., SI-Traceable Purity Determination of Peptides, 13th International
Symposium on Biological and Environmental Reference Materials, Book of Synopses IAEA-
CN-195/103 (2012) 135-137.
[12] Westwood S., Choteau T., Daireaux A., Josephs R. D. and Wielgosz R. I., Mass
balance method for the SI value assignment of the purity of organic compounds, Anal.
Chem., 85 (2013) 3118.
[13] Stoppacher N., Josephs R. D., Daireaux A., Choteau T., Westwood S. W.
and Wielgosz R. I., Impurity identification and determination for the peptide hormone
angiotensin I by liquid chromatography-high-resolution tandem mass spectrometry and the
metrological impact on value assignments by amino acid analysis, Anal. Bioanal. Chem.,
405 (2013) 8039.
[14] Stoppacher N., Josephs R. D., Daireaux A., Choteau T., Westwood S. and
Wielgosz R. I., Accurate quantification of impurities in pure peptide material—
angiotensin I: Comparison of calibration requirements and method performance
characteristics of liquid chromatography coupled to hybrid tandem mass spectrometry
and linear ion trap high-resolution mass spectrometry, Rapid Commun. Mass Spectrom.,
29 (2015) 1651.
[15] Weykamp C., John W. G., Mosca A., Hoshino T., Little R., Jeppsson J. O.,
Goodall I., Miedema K., Myers G., Reinauer H., Sacks D. B., Slingerland R. and
Siebelder C., The IFCC Reference Measurement System for HbA1c: a 6-year progress
report, Clin. Chem., 54 (2008) 240.
[16] Jeppsson J. O., Kobold U., Barr J., Finke A., Hoelzel W., Hoshino T., Miedema
K., Mosca A., Mauri P., Paroni R., Thienpont L., Umemoto M. and Weykamp C.,
International Federation of Clinical Chemistry and Laboratory Medicine (IFCC). Approved
IFCC reference method for the measurement of HbA1c in human blood, Clin. Chem. Lab.
Med., 40 (2002) 78.
[17] Kaiser P., Akerboom T., Ohlendorf R. and Reinauer H., Liquid chromatography-
isotope dilution-mass spectrometry as a new basis for the reference measurement procedure
for hemoglobin A1c determination, Clin. Chem., 56 (2010) 750.
[18] Liu H., Wong L., Yong S., Liu Q. and Lee T. K., Achieving comparability with IFCC
reference method for the measurement of hemoglobin A1c by use of an improved isotope-
dilution mass spectrometry method, Anal. Bioanal. Chem., 407 (2015) 7579.
[19] Wong L., Liu H., Yong S., Liu Q. and Lee T. K., Improved reference measurement
method for hemoglobin A1c by use of liquid chromatography-isotope dilution-tandem mass
spectrometry, Clin. Chem., 61 (2015) 435.
[20] Gonzales R., Steiner J. F., Lum A. and Barrett P. H., Decreasing antibiotic use
in ambulatory practice: impact of a multidimensional intervention on the treatment of
uncomplicated acute bronchitis in adults, JAMA, 281 (1999) 1512.
[21] Guillemot D. and Courvalin P., French Working Party to Promote Research to Control
Bacterial Resistance. Better control of antibiotic resistance, Clin. Infect Dis., 33 (2001)
542.
[22] Christ-Crain M., Jaccard-Stolz D., Bingisser R., Gencay M. M., Huber P. R.,
Tamm M. and Müller B., Effect of procalcitonin-guided treatment on antibiotic use
and outcome in lower respiratory tract infections: cluster-randomised, single-blinded

intervention trial, Lancet, 363 (2004) 600.
[23] https://www.euramet.org/research-innovation/search-research-projects/details
/?eurametCtcp project show%5Bproject%5D=1421&eurametCtcp project listResearch
%5Btitle%5D=antimicro.
[24] Höglund K., Fourier A., Perret-Liaudet A., Zetterberg H., Blennow K. and
Portelius E., Alzheimer’s disease - Recent biomarker developments in relation to updated
diagnostic criteria, Clin. Chim. Acta., 449 (2015) 3.
[25] Blennow K., Hampel H., Weiner M. and Zetterberg H., Cerebrospinal fluid and
plasma biomarkers in Alzheimer disease, Nat. Rev. Neurol., 6 (2010) 131.
[26] Strozyk D., Blennow K., White L. R. and Launer L. J., CSF Abeta 42 levels correlate
with amyloidneuropathology in a population-based autopsy study, Neurology, 60 (2003)
652.
[27] Andreasen N. and Blennow K., CSF biomarkers for mild cognitive impairment and
early Alzheimer’s disease, Clin. Neurol. Neurosurg., 107 (2005) 165.
[28] Blennow K., Zetterberg H. and Fagan A. M., Fluid biomarkers in Alzheimer disease,
Cold Spring Harb Perspect Med, 2 (2012) a006221.
[29] Blennow K., Cerebrospinal fluid protein biomarkers for Alzheimer’s disease, NeuroRx, 1
(2004) 213.
[30] Buerger K., Ewers M., Pirttila T. et al., CSF phosphorylated tau protein correlates
with neocortical neurofibrillary pathology in Alzheimer’s disease, Brain, 129 (2006) 3035.
[31] Tapiola T., Alafuzoff I., Herukka S. K. et al., Cerebrospinal fluid beta-amyloid 42
and tau proteins as biomarkers of Alzheimer-type pathologic changes in the brain, Arch.
Neurol., 66 (2009) 382.
[32] Vanmechelen E., Vanderstichele H., Davidsson P. et al., Quantification of tau
phosphorylated at threonine 181 in human cerebrospinal fluid: a sandwich ELISA with
a synthetic phosphopeptide for standardization, Neurosci. Lett., 285 (2000) 49.
[33] Tang W., Huang Q., Yao Y. Y., Wang Y., Wu Y. L. and Wang Z. Y., Does CSF
p-tau181 help to discriminate Alzheimer’s disease from other dementias and mild cognitive
impairment? A meta-analysis of the literature, J. Neural. Transm., 121 (2014) 1541.
[34] Blennow K., Vanmechelen E. and Hampel H., CSF total tau, Abeta42 and
phosphorylated tau protein as biomarkers for Alzheimer’s disease, Mol. Neurobiol., 24
(2001) 87.
[35] Lewczuk P., Esselmann H., Otto M. et al., Neurochemical diagnosis of Alzheimer’s
dementia by CSF Abeta42, Abeta42/Abeta40 ratio and total tau, Neurobiol. Aging, 25
(2004) 273.
[36] McAvoy T., Lassman M. E., Spellman D. S., Ke Z., Howell B. J., Wong O., Zhu
L., Tanen M., Struyk A. and Laterza O. F., Quantification of tau in cerebrospinal
fluid by immunoaffinity enrichment and tandem mass spectrometry, Clin. Chem., 60 (2014)
683.
[37] Bros P., Vialaret J., Barthelemy N., Delatour V., Gabelle A., Lehmann S. and
Hirtz C., Antibody-free quantification of seven tau peptides in human CSF using targeted
mass spectrometry, Front Neurosci., 9 (2015) 302.
[38] Hanger D. P., Byers H. L., Wray S. et al., Novel phosphorylation sites in tau from
Alzheimer brain support a role for casein kinase 1 in disease pathogenesis, J. Biol. Chem.,
282 (2007) 23645.
[39] Mattsson N., Andreasson U., Persson S. et al., The Alzheimer’s Association external
quality control program for cerebrospinal fluid biomarkers, Alzheimers Dement, 7 (2011)
386.
[40] Leinenbach A., Pannee J., Dülffer T., Huber A., Bittner T., Andreasson
U., Gobom J., Zetterberg H., Kobold U., Portelius E. and Blennow K.,
Mass spectrometry-based candidate reference measurement procedure for quantification
of amyloid-beta in cerebrospinal fluid, Clin. Chem., 60 (2014) 987.
[41] Carrillo M. C., Blennow K., Soares H., Lewczuk P., Mattsson N., Oberoi P.,
Umek R., Vandijck M., Salamone S., Bittner T., Shaw L. M., Stephenson D.,
Bain L. and Zetterberg H., Global standardization measurement of cerebral spinal fluid
for Alzheimer’s disease: an update from the Alzheimer’s Association Global Biomarkers
Consortium, Alzheimers Dement, 9 (2013) 137.
[42] https://www.euramet.org/research-innovation/search-research-projects/details
/?eurametCtcp project show%5Bproject%5D=1418&eurametCtcp project listResearch
%5Btitle%5D=neuromet.
[43] Jordan J. B., Poppe L., Haniu M., Arvedson T., Syed R., Li V., Kohno H., Kim
H., Schnier P. D., Harvey T. S., Miranda L. P., Cheetham J. and Sasu B. J.,
Hepcidin revisited, disulfide connectivity, dynamics, and structure, J. Biol. Chem., 284
(2009) 24155.
[44] van der Vorm L. N., Hendriks J. C., Laarakkers C. M., Klaver S., Armitage
A. E., Bamberg A., Geurts-Moespot A. J., Girelli D., Herkert M., Itkonen
O., Konrad R. J., Tomosugi N., Westerman M., Bansal S. S., Campostrini N.,
Drakesmith H., Fillet M., Olbina G., Pasricha S. R., Pitts K. R., Sloan J. H.,
Tagliaro F., Weykamp C. W. and Swinkels D. W. Toward Worldwide Hepcidin
Assay Harmonization: Identification of a Commutable Secondary Reference Material, Clin.
Chem., 62 (2016) 993.
[45] Anderson K. M., Castelli W. P. and Levy D., Cholesterol and Mortality: 30 Years
of Follow-up From the Framingham Study, JAMA, 257 (1987) 2176.
[46] Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in
Adults, Executive Summary of the Third Report of the National Cholesterol Education
Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood
Cholesterol in Adults (Adult Treatment Panel III), JAMA, 285 (2001) 2486.
[47] Lipid Research Clinics Program, The Lipid Research Clinics Coronary Primary
Prevention Trial Results: I. Reduction in Incidence of Coronary Heart Disease, JAMA,
251 (1984) 351.
[48] NCEP, Current status of blood cholesterol measurement in clinical laboratories in the
United States: a report from the Laboratory Standardization Panel of the National
Cholesterol Education Program, Clin. Chem., 34 (1988) 193.
[49] National Institutes of Health, Report of the Expert Panel on Detection, Evaluation,
and Treatment of High Blood Cholesterol in Adults, Bethesda, MD, National Institutes of
Health, 1989, NIH publication no: 89-2925.
[50] National Institutes of Health, Recommendations for improving cholesterol
measurement: a report from the Laboratory Standardization Panel of the National
Cholesterol Education Program, Bethesda, MD, National Institutes of Health, 1990, NIH
publication no: 90-2964.
[51] Cooper G. R., Smith S. J. and Duncan I. W., Interlaboratory testing of the
transferability of a candidate reference method for total cholesterol in serum, Clin. Chem.,
32 (1986) 921.
[52] Edwards S. H., Kimberly M. M., Pyatt S. D., Stribling S. L., Dobbin K. D. and
Myers G. L., Proposed Serum Cholesterol Reference Measurement Procedure by Gas
Chromatography-Isotope Dilution Mass Spectrometry, Clin. Chem., 57 (2011) 614.

[53] Welch M. J., Parris R. M., Sniegoski L. T. and May W. E., CCQM-K6: key
comparison on the determination of cholesterol in serum, Metrologia, 39 (2002) 08001.
[54] Myers G. L., Kimberly M. M., Waymack P. W., Smith S. J., Cooper G. R. and
Sampson E. J., A reference method laboratory network for cholesterol: A model for
standardization and improvement of clinical laboratory measurements, Clin. Chem., 46
(2000) 1762.
[55] National Institutes of Health, Second report of the Expert Panel on Detection,
Evaluation, and Treatment of High Blood Cholesterol in Adults, Bethesda, MD, National
Institutes of Health, 1993, NIH publication no: 93-3095.
[56] National Institutes of Health, Third report of the Expert Panel on Detection,
Evaluation, and Treatment of High Blood Cholesterol in Adults, Bethesda, MD, National
Institutes of Health, 2001. NIH publication no: 01-3670.
[57] Warnick G. R., Nauck M. and Rifai N., Evolution of methods for measurement of
HDL-cholesterol: From ultracentrifugation to homogeneous assays, Clin. Chem., 47 (2001)
1579.
[58] Bachorik P. S. and Ross J. W., National Cholesterol Education Program
recommendations for measurement of low-density lipoprotein cholesterol: executive
summary, Clin. Chem., 41 (1995) 1414.
[59] Warnick G. R. and Wood P. D., National Cholesterol Education Program
recommendations for measurement of high-density lipoprotein cholesterol: executive
summary, Clin. Chem., 41 (1995) 1427.
[60] Lewington S., Whitlock G., Clarke R., Sherliker P., Emberson J., Halsey J.,
Qizilbash N., Peto R. and Collins R., Lancet, 370 (2007) 1829.
[61] Sniderman A. D., Williams K., Contois J. H., Monroe H. M., McQueen M. J.,
De Graaf J. and Furberg C. D., Circ. Cardiovasc. Qual. Outcomes, 4 (2011) 337.
[62] Thompson A. and Danesh J. J., Intern. Med., 259 (2006) 481.
[63] Contois J. H., McConnell J. P., Sethi A., Csako G., Devaraj S., Hoefner D. M.
and Warnick G. R., Clin. Chem., 55 (2009) 407.
[64] Marcovina S. M., Koschinsky M. L., Albers J. J. and Skarlatos S., Clin. Chem.,
49 (2003) 1785.
[65] Danesh J., Collins R. and Peto R., Circulation, 102 (2000) 1082.
[66] Nordestgaard B. G., Chapman M. J., Ray K., Borén J., Andreotti F., Watts
G. F., Ginsberg H., Amarenco P., Catapano A., Descamps O. S., Fisher E.,
Kovanen P. T., Kuivenhoven J. A., Lesnik P., Masana L., Reiner Z., Taskinen
M. R., Tokgözoglu L. and Tybjærg-Hansen A., Eur. Heart J., 31 (2010) 2844.
[67] Otvos J. D., Collins D., Freedman D. S., Shalaurova I., Schaefer E. J.,
McNamara J. R., Bloomfield H. E. and Robins S. J., Circulation, 113 (2006) 1556.
[68] Master S. R. and Rader D. J., Clin. Chem., 59 (2013) 723.
[69] Austin M. A., King M. C., Vranizan K. M. and Krauss R. M., Circulation, 82 (1990)
495.
[70] Gardner C. D., Fortmann S. P. and Krauss R. M., JAMA, 276 (1996) 875.
[71] Hirayama S. and Miida T., Clin. Chim. Acta, 414 (2012) 215.
[72] Mora S., Szklo M., Otvos J. D., Greenland P., Psaty B. M., Goff D. C., O’Leary
D. H., Saad M. F., Tsai M. Y. and Sharrett R., Atherosclerosis, 192 (2007) 211.
[73] Phillips C. M. and Perry I. J., Atherosclerosis, 242 (2015) 399.
[74] Dominiczak M. H., Handbook of Lipoprotein Testing, 2nd Edition (American Association
for Clinical Chemistry) 2000, pp. 1–29.
[75] Mora S., Circulation, 119 (2009) 2396.
[76] Robinson J. G., J. Am. Coll. Cardiol., 60 (2012) 2607.
[77] Grundy S. M., Vega G. L., Tomassini J. E. and Tershakovec A. M., Comparisons
of apolipoprotein B levels estimated by immunoassay, nuclear magnetic resonance,
vertical auto profile, and non-high-density lipoprotein cholesterol in subjects with
hypertriglyceridemia (SAFARI Trial), Am. J. Cardiol., 108 (2011) 40.
[78] Chung M., Lichtenstein A. H., Ip S., Lau J. and Balk E. M., Comparability of
methods for LDL subfraction determination: A systematic review, Atherosclerosis, 205
(2009) 342.
[79] Master S. R. and Rader D. J., Clin. Chem., 59 (2013) 723.
[80] Otvos J. D., Rudel L. L. and McConnell J. P., Concerns regarding lipoprotein particle
measurement by ion mobility analysis, Clin. Chem., 54 (2008) 2086.
DOI 10.3254/978-1-61499-818-1-25
SI traceable measurements of the Earth from space

to monitor and mitigate against climate change
N. P. Fox
National Physical Laboratory - Hampton Rd, Teddington, Middx, TW11 0LW, UK
Summary. — The Earth’s climate is changing, but the rate of change and
scale/timing of its impacts are far from understood and remain the subject of much
debate. Variances of up to 5 ◦ C can be seen between climate forecast models making
it difficult for policy makers to take the necessary measures to mitigate and adapt
to a warming planet. Global observations of the Earth by satellites are the only
way to provide the date necessary to improve the fidelity of the predictions and to
test and constrain the models. These data must be collected over decades to allow
small trends to be sufficiently aggregated so that they can be reliably detected from
a background of natural variability. This places very high demands on the perfor-
mance of satellite sensors since no single satellite will have the lifetime necessary to
attempt to monitor change and thus the accuracy needs to be sufficient to remove
any prospect of bias from instrument drift propagating into the long-time base mea-
surements which will necessarily be derived from a combination of satellites. At
present few satellites are designed specifically for climate and none with the ability
to demonstrate robust SI traceability in orbit at the uncertainty levels necessary
(close to those of primary standards in the laboratory). This paper describes some
of the scientific challenges and culminates in a proposed satellite mission designed
explicitly to address them, and in so doing also has the ability to upgrade the rest
of the Earth Observing system through reference cross-calibration.

25
26 N. P. Fox
Fig. 1. – The uncertainty range in forecast temperatures of the Earth for two extreme emissions
scenarios [1].
1. – Introduction
.
1 1. Climate. – Unequivocal attribution of the causes, consequential impact and ef-
fective mitigation/adaptation of change in the Earth’s climate is arguably the greatest
challenge facing science and society today. The ongoing challenges, to reduce emissions
to prevent the Earth’s global average temperature rising beyond 2 ◦ C (ideally 1.5 ◦ C)
which now stem from the agreement at the Conference of the Parties (COP 21) in Paris
2015, exemplify the intense political, scientific and public debate that is associated with
the issue. This is underpinned by the enormous cost implications of policy decisions that
have to be based on forecasted impacts resulting from global climate models. However,
these forecast models have significant uncertainty with estimates of global temperature
increases ranging from at best, 2 to 5 ◦ C above a year 2000 baseline by 2100, see fig. 1 [1]
for business as usual.
Such a range could be the difference between the need for major new flood defences
to prevent large land losses or maintenance/minor updates of the status quo. These
temperatures are global averages and both the local value of temperature and its impacts
are anticipated to vary widely across the globe (for example some in some regions it may
be as high as 10 ◦ C), with developing nations likely to be in regions most significantly
affected and where even 2 ◦ C rise is considered too high. It is the duty of the science
community to reduce this unacceptably large uncertainty in forecasts by finding and
delivering the necessary information, with the highest possible confidence, in the shortest
possible time. Since it is the forecast models and their predictions that will in practise
be needed to monitor the success or otherwise of the World’s governments efforts to meet
their emissions reduction pledges and more importantly if these have the desired impact
on the Earth’s climate
The Intergovernmental Panel on Climate Change (IPCC) concludes that recent
decades have revealed clear hallmarks of anthropogenic climate change, but the mix
SI traceable measurements of the Earth from space 27
of natural variability and anthropogenic effects on decadal timescales is far from fully
understood or measured, requiring significant improvements in accuracy [1].
Unequivocal attribution and the quantification of subtle fingerprint indicators is fun-
damental to our ability to reliably predict impact and the development of appropriate
mitigation/adaptation strategies. The uncertainty in climate prediction lies in the com-
plexity of the models, our inadequate understanding of the Earth system and its feedback
mechanisms and the relatively poor quality of available data against which to test pre-
dictions on the necessary decadal timescales [2].
The data needed to benchmark and test these models must be global in nature, and
collected over a sufficiently long-time base with appropriate uncertainty to enable the
identification of “change” in a particular measurand above variance that could be caused
by natural or “local” effects. In general, this means that such measurements need to be
considered on decadal timescales and collected from satellite-based instrumentation.
Terrestrial networks, collecting “in situ” data, will always be necessary to provide
local validation of global observations and support detailed studies of Earth system pro-
cesses e.g. air quality, carbon and hydrological cycles, etc. However, it is only remote
and continuous observation of the Earth system by satellite that the key data sets, the
Essential Climate Variables (ECVs) identified by the Global Climate Observing System
(GCOS) of the United Nations [3], can ultimately be adequately addressed. In general,
these ECVs (e.g. albedo, cloud cover, chlorophyll-a, . . . ) are not measured directly but
derived from more fundamental and familiar physical quantities such as: radiances, re-
flectance, transmittance, etc. often now referred to as Fundamental Climate Data Records
(FCDR). The satellite instruments measuring these quantities utilise the full electromag-
netic spectrum and in most cases passively, for example making use of the Sun as a source
for reflectance/radiance/transmittance measurements.
A community workshop held in the USA in 2007, considered the needs of climate [4-6]
and concluded that in the optical domain, the critical underpinning measurands (FCDRs)
and the associated uncertainty requirements were:
– Total Solar Irradiance (TSI) −0.01% (k = 1),
– Solar Spectral Irradiance (SSI) −0.1% (k = 1),

– Earth Reflected solar radiance −0.3% (k = 2),
– Earth Emitted Infrared (IR) radiances −0.1 K (k = 3)

(expressed in terms of resultant temperature).
This, and their priority, was reinforced in the decadal review carried out by the
National Research Council of USA [7].
Sound policy making requires high confidence in climate predictions verified against
decadal change observations with robustly known accuracy. Yet, our ability to monitor
and predict the future of the climate is inadequate as we currently do not possess sufficient
accuracy in our observing capability to confidently observe the small but critical climate
28 N. P. Fox
change signals that are expected to occur over decadal timescales. We currently rely on
data from sensors primarily designed for operational weather forecasting, where short- to
medium-term global consistency and continuity is of greater importance than absolute
accuracy and long-term continuity is achieved largely from overlapping and normalisation
of data sets. It is now increasingly recognised by the community that this is inadequate
and that the robustness and long-term confidence stemming from SI traceability and the
move towards a global climate observing system needs to become a higher priority [8].
In practise this means there is a need to dramatically improve the absolute accuracy
of in-orbit observation (factor of ten) if we wish to maximise our sensitivity to critical
climate signals, mitigate the risks of gaps in the data record and provide the observations
necessary to evaluate the accuracy of climate model predictions. To address this lack in
our observing capability, three critical measurements and their required accuracies were
defined by the US National Research Council Decadal Survey, one of which was Earth
reflected solar spectral radiation [7].
Solar radiation, reflected from the Earth-atmosphere system back to space, constitutes
a powerful and highly variable forcing of the climate system through changes in snow
cover, sea ice, land use, cloud and aerosol properties. Systematic, spatially resolved
observations of the time series of the absolute spectrally resolved flux of near-ultraviolet,
visible, and short wave-IR radiation returned to space by the Earth system, tied to
international standards in perpetuity, underpin a credible climate record of the changing
Earth system.
In combination with establishment of the absolute spectrally resolved solar irradiance
reflected from the Earth-atmosphere system to space, to address the above, it is essential
to continue the long-term, time series measurements of incident total solar irradiance
(TSI). To enhance its climate value and remove the current controversies and uncertain-
ties due to instrumental variations, facilitating a robust linkage into a climate record, the
absolute accuracy and SI traceability needs to be significantly improved. The addition of
a high accuracy spectrally resolved measurement of incoming solar irradiance, provides
the means to link solar variation to climate processes: underpinning studies of atmo-
spheric chemistry and facilitating retrieval of the full range of Earth system products at
both top and bottom of atmosphere.
At present there is no planned (funded) satellite mission capable of meeting these

observation requirements. However, there are two complimentary missions which have
been proposed that can deliver this step change in performance and meet the challenge
which is primarily metrological in nature. One is the NASA CLARREO mission [9] and
the other, which will be discussed in more detail later is the European TRUTHS mission
led by NPL [10,11].
.
1 2. Earth Observation data quality. – Although climate is the primary driver for high
accuracy the number of Earth Observation (EO) satellites has increased considerably in
recent years, reflecting the rapid growth in the importance, application and benefits
of space-borne data to society. The increasing need for more accurate and integrated
observations, as well as opportunities presented by new technology and low cost access
to space, are changing the future landscape of EO. ESA’s new science strategy highlights
this “new era for scientific advances and societal benefits”, where innovative missions
operate holistically in tackling multiple global challenges cost-effectively [12].
Calibration accuracy, stability and sensor-to-sensor consistency are the key on-orbit
performance metrics for EO sensors and the system as a whole. Addressing this challenge
in a consistent, transparent and efficient manner is top of the agenda of space agencies and
the primary topic of the Committee on Earth Observation Satellites (CEOS) Working
Group on Calibration and Validation (WGCV) and World Meteorological Organisation
Global Space Inter-Calibration System (WMO GSICS). The accuracy and consistency
of measurements across multiple space-borne instruments in Low Earth and Geostation-
ary Orbits (LEO and GEO), on all timescales (historic, current, future), are directly
connected to the information and scientific understanding of complex systems such as
the Earth’s climate, as well as the fidelity and confidence of operational data used from
meteorological sensors and the Sentinels of Copernicus. The direct quantitative linkage
of accuracy to climate understanding has been explored in some detail [9] and will be
illustrated in some of the examples below.
A key trend in satellite-based Earth observation is the ever-increasing importance of
combining data from multiple missions, a trend that has led to multi-agency coordination,
for example the establishment by CEOS of virtual constellations to “establish a Global
Earth Observation System of Systems (GEOSS) through the synergistic combination and
harmonisation of data sets to provide knowledge information products to meet the needs
of [society].”
Fully harmonised interoperability of these data sets can only occur through rigorous
quality assurance processes and a detailed understanding of the uncertainties of each
mission. The only robust option to meet the exacting needs of the climate change com-
munity and also the emerging operational needs of the EO community as a whole is to
ensure that all EO data and derived knowledge information products, should be traceable
to SI units and have with it an associated uncertainty estimate and/or “quality indica-
tor”. (Note that many of these information products and indeed ECVs are not relatively
simple physical measurands but often bio-physical in nature and almost always the result
of an algorithm with multiple inputs and non-linear interdependent processes.) Such re-
quirements now pervade the strategy documents and recommendations of organisations

such as CEOS, GEOSS and the WMO and have subsequently been passed down to the
space agencies [4,5,8]. Many of these organisations and agencies recognise that it is not
simply the space element that needs to have exacting QA associated with it but that this
must be embedded within the whole validation and data processing chain. In response to
this demand, CEOS, has led the development of a new internationally endorsed Quality
Assurance Framework for Earth Observation (QA4EO) http://www.QA4EO.org. This
framework or at least its core principle,
“All data and derived products must have associated with them a Quality
Indicator (QI) based on documented quantitative assessment of its traceability
to community agreed (ideally tied to SI) reference standards”
30 N. P. Fox
Fig. 2. – Top level traceability diagram for Albedo, individual boxes drillable further to expand
detail.
is now being implemented throughout the World’s space agencies and the WMO. This
principle is not revolutionary and is widely practised in most other commercial/academic
sectors. In fact there are many good examples of how this has been implemented in the
EO community as well, however, rarely has it been comprehensive or applied in an inter-
nationally harmonious manner, making it difficult to identify and assess any differences
in results. QA4EO therefore provides specific guidance to aid in the implementation and
interpretation of the core “traceability principle” following the best practises that have
been established at National Metrology Institutes (NMI) and the activities needed to
support the Mutual Recognition Arrangement MRA (www.BIPM.org).
Evidence of traceability to SI and the uncertainty to some independent invariant ref-
erence standard is the key-underpinning requirement to enable such a QA infrastructure
to function. The ability to assess the robustness of this traceability process, “end to
end” through the full process chain: data collection to biophysical information prod-
uct, remains challenging and the domain of highly specialised experts. These experts
who are often integrally related to the process and/or its development and thus can-
not be considered independent. To help address this and in particular, improve trans-
parency of the robustness of traceability of all the steps, NPL through a EU project
called QA4ECV (http://www.QA4ECV.org) has helped establish a software tool and an
assessment method to help visualise the status of the process chain in a manner accessible
to the whole community (expert and non-expert). Figure 2 presents an example at top
level for Albedo.
In the Fiduceo (www.Fiduceo.org) project this approach is being further expanded
to drill into the full metrological detail including the mechanics and functions of the
measuring device.
.
1 3. Sensor post launch Calibration and Validation (Cal/Val). – Pre-flight SI traceable
calibration of a satellite sensor is of course paramount if the data from it are to be useable
in the future. However, it should be recognised that for optical sensors in particular, the
radiometric performance is likely to degrade significantly from its pre-flight calibration
and will need some form of post-launch update and regular monitoring of change. With-
out rigorous pre-flight calibration and characterisation it would be virtually impossible
to reliably apply post launch corrections as these are largely based on measurements
which can integrate together a range of potential effects that can only be separated with
pre-flight knowledge.
Post launch Cal/Val can be performed using systems on board the space craft, for
example, diffusely reflecting panels to reflect incident solar irradiance (radiance) to fill
the full aperture of a collecting optic, typically a telescope of some form. In this way
relying on existing knowledge of solar spectral stability and its absolute value together
with the reflectance factor of the diffuse reflector to provide an in-flight calibration. A
variation on this is the use of the moon as a solar illuminated diffuser. In this case there
are no effects of atmosphere and lunar reflectance has been shown to be highly stable
although the absolute value of its reflectance factor is not known with great accuracy.
Thus at present it is largely used to monitor sensor change as opposed to providing a
calibration.
By contrast, vicarious methods where Earth-based targets are used, suffer from un-
certainties due to correction for atmospheric transmittance losses and the difficulty in
assigning an SI traceable value of sufficient accuracy to the relatively large area needed
to fill the field of view of an individual pixel of a satellite sensor. Currently the best ways
to achieve this is:
– To establish dedicated test sites which are well-characterised, by in situ obser-

vations, (ground teams/aircraft) and ideally monitored using autonomous perma-
nently deployed instrumentation.
– Perform comparisons between sensors on the occasions that they simultaneously (or
near) pass over the same part of the globe. For an optical imaging sensor, when this
occurs over spatially uniform surfaces such as deserts and snowfields as a minimum
any biases can be identified. If one sensor is designated as the “reference” then
other satellites can be calibrated to it. However, for this to result in SI traceability
requires measurements made by the reference sensor to be considered SI traceable,
which because of unpredictable ageing effects is rarely the case.
For sensors measuring atmospheric composition, similar approaches are used for the
radiometric calibration of the sensor but in this case test sites are used extensively to
validate the retrievals of gas molecule concentrations. This may be a ground instrument
remotely viewing the atmosphere or one on-board a balloon or aircraft. Other biophysical
variables e.g. vegetation products (land and ocean) similarly have test sites to validate
and or optimise the retrievals
32 N. P. Fox
Fig. 3. – Table listing ECVs of GCOS by theme.
.
1 4. Summary. – This paper will provide an overview of some of the most demanding
requirements for optical based Earth Observation, together with some of the metrology
challenges that need to be resolved to address them and the efforts that are being un-
dertaken. This will culminate in the description of a dedicated “metrology designed”
satellite to allow SI traceable measurements of incoming and outgoing radiation (change
in radiation balance) and associated bio-physical parameters of optimal accuracy to allow
robust attribution and mitigation decisions to be taken in the shortest possible time.
2. – Key climate parameters

.
2 1. Essential Climate Variables (ECV). – GCOS has identified 50 ECV’s as key
indicators of climate change [3], many of which form inputs to climate forecast models,
and have assigned to them specific monitoring criteria e.g. accuracy and stability. They
are listed and grouped thematically in fig. 3. Many of the ECVs are not direct physical
measurands but are instead biophysical in nature, combining physical measurements such
as reflectance with algorithms and other inputs to derive more complex parameters. It
is thus important for the metrology community to understand the final applications and
ensure that any uncertainty estimate can be propagated appropriately through to the
desired parameter.
The following sections will review some of these ECVs, consider their climate impor-
tance and consequence to the underpinning physical measurand.
.
2 2. Earth Radiation Budget (ERB). – The ERB is the overall balance between the
incoming energy from the Sun and the outgoing thermal and reflected solar energy from
the Earth, it is primarily change in this overall balance that causes the Earth’s tempera-
ture to change. Accurate measurements of the ERB are crucial for understanding climate
and climate variability on Earth [13-16]; they have the potential to inform global-scale
climate sensitivity [17], hydrological response [18] and to challenge the ability of climate
models to better represent key processes including inter-hemispheric asymmetries in en-
ergy budget and precipitation [19-21] and correctly predict major climate feedbacks such
as those due to water vapour and cloud, and consequently the magnitude and nature of
climate response [22]. Ultimately, long-term, stable time series of such measurements are
required both to monitor changes of the climate above natural variability, and to validate
our ability to model the evolution of the climate system correctly. The current observing
system is inadequate for this purpose due to substantial uncertainty relating to absolute
calibration [21], sampling [22] and homogeneity [19].
State-of-the-art measurements of the ERB are based on a) outgoing broadband short-
wave (SW, 0.32 μm to 4 μm), b) long-wave (LW, 4 μm to > 200 μm) radiance observa-
tions, and c) knowledge of the incoming solar irradiance. Accurate determination of
the net ERB requires a consistent measurement of all three components, and this can
only be assured by stringent absolute accuracy requirements on each of these variables.
Goals for top of the atmosphere (TOA) flux products are 1 W m−2 absolute accuracy and
0.3 W m−2 or better for decadal stability [4], which places even greater demands on the
measured radiances. The further requirement that the records be unbroken adds an extra
risk to these climate observations as it relies on continuity of support and failure-proof
operations.
The present suite of ERB broadband sensors consists of the CERES instruments [23],
which provide global coverage from LEO, and the GERB instrument [24] which ob-
tain high temporal-resolution observations from the geostationary METEOSAT-8 and 9
satellites. CERES is primarily designed to provide a very accurate high-stability mea-
sure of the global ERB on monthly and longer timescales, launched with 0.5 W m−2 sr−1
(k = 2) absolute accuracy [25], and a decadal stability requirement [26] of better than
0.1 W m−2 sr−1 . They continue the unbroken ERB record which extends for more than
a quarter of a century and includes data obtained from the ERBE (Earth Radiation
Budget Experiment) series of instruments [27]. In contrast, the GERB instruments do

not give a global picture of the ERB, but their high time resolution enables the study of
important climate processes such as the impact of diurnal variability in cloud and aerosol
on the ERB [28,29]. They can be used both for model validation and to facilitate model
improvements [30].
Current observations of ERB do not close the energy balance [31]. Because the pri-
mary aim of these measurements is accurate ERB quantification rather than trend obser-
vation, high absolute accuracy (1%, desired < 0.5%) maintained over the mission life is
the primary concern. To achieve these requirements, ERB instruments undergo highly-
specified ground calibration, and carry on-board calibrators and monitors in order to
transfer this accuracy to orbit and maintain calibration throughout their operation. But
34 N. P. Fox
despite these efforts, in-orbit comparisons demonstrate that, for the SW at least, this
is not sufficient to overcome all the problems of transferring and maintaining in-orbit
calibration [32,34]. For example, the CERES instruments have experienced SW optical
darkening, resulting in changes which are up to nine times greater than their stability
aim. A correction for this has only been made possible through intercomparison of the
four operating CERES sensors and an independent check on this process would be ex-
tremely valuable. As it is possible that the future global ERB record will be maintained
by just a single sensor some alternative way of ensuring calibration stability in orbit is
required if the integrity of this valuable climate record is to be maintained.
Inter-instrument comparisons and measurements of “known” quantities have also led
us to question the ability to transfer ground calibration to orbit. For example, a sanity
check that can be made on the measured ERB is to consider if, when averaged over the
globe and the year, it is zero to within the expected margin of measurement error and
the capacity of the system to store heat on these timescales. CERES determines the
net imbalance in the TOA global annual energy balance to be [33] 6.5 W m−2 , which
is much larger than the best estimate of 0.5–0.85 W m−2 based [19,21,35] on observed
ocean heat content data and model simulations. Uncertainties in instrument calibra-
tion (4.5 W m−2 ), spectral response correction (1.0 W m−2 ) and total solar irradiance
(1 W m−2 ) are the largest contributions to the uncertainty but still fail to explain the
discrepancy. This clearly highlights the value that a more accurate calibration could
provide; the current discrepancy is considerable in comparison to the ocean heat-storage
term itself and even more so when compared to the total anthropogenic forcing expected
over the next 10 years. A further illustration of the difficulties associated with transfer-
ring SW ground calibration to orbit is highlighted by the intercomparison of the GERB
instruments on METEOSAT-8 and 9 which exhibit a 3–6% difference in the SW radiance
depending on the viewed scene.
To summarise, there is a requirement to measure the ERB to very high accuracy
(< 0.5%) and stability over long periods of time. However, despite extensive efforts, we
have clear evidence that the current state-of-the-art instruments struggle to meet these
requirements in the SW. This is compounded by the paucity and reliability of independent
estimates of SW [23] which is less acute for the LW [35]. These problems have important
consequences for our ability to monitor and understand the climate. Essentially the
expected climate trends and significant climate variations in this quantity are so small
that they require stability and absolute accuracy that cannot be achieved by a space-
borne instrument reliant on a ground calibration for its traceability.
This challenge can only be overcome through developing in-flight calibration systems
capable of meeting this accuracy goal in a fully SI traceable manner and then deploying
them on all future satellites. Alternatively, a reference satellite could be utilised which
can measure with sufficient spectral and spatial resolution and an orbit that allows many
cross-overs with other satellites to enable it to transfer its calibration to other sensors via
near simultaneous observation of the same targets and the convolution of its observation
characteristics to match those of the sensor under test. Although for ERB it is only
change in the integrated radiance that is important, high spectral resolution enables an
evaluation of potential changes in albedo that might effect the ERB and also the sensor
under test where degradation is likely to be spectral in nature.
.
2 3. Solar variability
.
2 3.1. Total Solar Irradiance (TSI). Solar radiation, the driving force of the Earth’s
climate, has been observed for many centuries, with quantitative measurements of solar
irradiance being made for over 100 years. It is expected that variations in solar irradiance
influence the terrestrial climate, but we do not yet fully understand how suspected long-
term variations arise or how the climate system actually responds. Monitoring solar
variations is therefore crucial, but the continuous record of space-based measurements
since 1979, which have revealed solar irradiance variations of a few tenths of one per cent
during the 11-year cycle, are as yet inadequate for the detection of long-term variations.
The current extraterrestrial instrument-based record, though key to our ability to
understand the Sun’s contribution, is highly fragile as a climate data set. Figure 4 [36]
shows the relative improvement in disparity between sensors (sensor biases) although
some of the reasons for the previous biases remain unanswered as do instrumental changes
in orbit and of course the discrepancy between integrated SSI and TSI (see below).
Whilst a solar composite can be constructed by normalising the various sensors to a
reference sensor to remove the biases, the results of the composite are subtly different
depending on the assumptions made and choice of reference detector. These differences
have consequences in terms of climate as the current IPCC has assumed very little
variability in the Sun over the last 30 years, and used as evidence of this that exactly the
same baseline is reached during each of the measured solar minimum. It is clear from
fig. 4 that this evidence is open to debate. Although recent studies indicate that the gap
between TIM, VIRGO and others may have closed, the full cause of the instrumental
differences still has not been fully resolved, other than the bias found in the World
Radiometric Reference (WRR) following measurements on the ground in Davos using
the Cryogenic Solar Absolute Radiometer (CSAR-v1) (see sect. 4.2)
The analysis of the observation data [37] shows that the variability of the TSI from the
minimum to the maximum of the solar activity does not exceed 0.1% or ∼ 1.4 W m−2 ,
which translates into a direct radiative forcing of about ∼ 0.25 W m−2 . A long-term
proxy of solar activity, the solar modulation function, shows that in the past 50 years
solar activity was at a 500 year maximum [38,39]; therefore the past 30 years of satellite
observations cannot characterise the full range of possible changes of the solar irradiance
over longer timescales. Substantial long-term variations in solar activity have occurred,
e.g. the Maunder (1645-1715) and the Dalton (1790-1830) minima (see fig. 5).
Solar activity increased in the first half of the 20th century and reached a maxi-
mum (in sunspot numbers) in the year 1957. After five decades of high, but stable,
solar activity, there is now growing evidence for a decline in solar activity in the near
future. Predictions of the strength of the next solar sunspot cycle have been made by
the NOAA, NASA, and ISES Solar Cycle 24 prediction panel and Hathaway and Dikpati
(http://science.nasa.gov). Their recent consensus is that the next 11-year solar cycle
36 N. P. Fox
Fig. 4. – Solar composite illustrating significant biases in the source data and the potential of
very different conclusions depending on composite chosen.
will be below average in intensity (with a predicted maximum sunspot number of 90).
During the last solar minimum (2008/2009) TSI reached the lowest value on record, i.e.
about 0.3 W m−2 lower than during the two previous minima.
While there is general agreement on how TSI has changed over time, there is a dis-
agreement about the magnitude. The TSI relative to the Maunder minimum had been de-
Fig. 5. – Sunspot group numbers (Rg, yellow) and Wolf Sunspot numbers (SSN, dark red).
Predictions for the upcoming solar cycles are shown in light shading.
termined to be −0.7 W m−2 [40], −1.6 W m−2 [41], −3.9 W m−2 [42], −2.7 W m−2 [42,43],
−4 W m−2 [43] and −2.3 W m−2 [44]. Newer reconstructions tend to show smaller am-
plitudes [45,46]. Steinhilber et al. [47] obtained ΔFM = −0.9 ± 0.4 W m−2 , while Schöll
et al. [48], find ΔFM = −3 ± 2 W m−2 on the basis of spectral reconstruction.
Further investigations are needed and these require observations of TSI over a very
long-term time series (more than 30 years) with high precision and absolute uncertainty
better than 0.01%. These uncertainty levels are not available from current TSI experi-
ments. Because it is impossible to have a single stable instrument in orbit for continuous
operation over such a long period, it is necessary to have a series of instruments in orbit,
each traceable to SI standards and each matching the requested accuracy, precision, and
duration in order to be able to finally answer unequivocally, how much the total and
spectral solar irradiance varies on a long timescale. All estimates of the solar influence
on the climate will eventually depend on this experimental determination and can only
be achieved by robust SI-traceable calibrations.
The standard definition [49] of radiative forcing of climate is the change in net down-
ward radiative flux at the tropopause, taking into account the effect of changes in the
stratosphere but without any tropospheric response. Thus an accurate estimate of solar
radiative forcing requires knowledge not only of changes in incoming irradiance at the top
of the atmosphere but also the influence of variations in UV on stratospheric temperature

and composition [50]. Much larger fractional changes take place at shorter wavelengths
than for the integrated visible portion of the spectrum [51]. The solar UV-induced varia-
tions influence the ozone distribution and the heating rates in the stratosphere [52]. The
effect of variation in solar irradiance on the ozone distribution and subsequent effect on
radiation propagation to the troposphere is a non-linear effect, which varies with latitude
and season. This effect has been demonstrated with GCM studies [53] where responses
of the sub-tropical jets and of the tropical overturning circulations had been found, but
the conclusions about solar forcing strengths and mechanisms differ between the AR5
climate models [48].
38 N. P. Fox
Observational requirements. The observational requirements for TSI, are not overly
demanding with the exception of the absolute accuracy needed to meet the climate
benchmark criteria which is 0.02% [4,6]. They are based on best practice from existing
heritage instruments. In summary, they require continuous observations of the Sun for
at least 5 minutes a day to allow averaging of short term local solar surface variations
(although as long and as regular as possible is desired). In observing the Sun, heritage
practice indicates that the defining aperture for irradiance should be a nominal 5 mm in
diameter (although providing it fully underfills the solar disc this is again not critical).
Knowledge of where in the solar disc the aperture is pointing is required to ∼ 2 arc
secs (easily achieved through existing sensors) with pointing accuracy and stability only
required to 0.1◦ .
For SSI the situation is similar, but with at first glance more relaxed uncertainty
requirements ∼ 0.2% although given that they relate to spectrally resolved measurements
are equally stringent.
.
2 3.2. Solar spectral irradiance (SSI). The interaction of solar radiation with the at-
mosphere is fundamental in determining the temperature structure and in controlling
many atmospheric chemical processes. In the middle atmosphere the absorption of solar
UV radiation produces ozone (O3 ) and results in the presence of the warm stratopause.
Here O3 is created through the combination of oxygen molecules (O2 ) with oxygen atoms
(O(3 P)) produced by photodissociation of other O2 molecules at wavelengths less than
242 nm. It is destroyed by radiation at these, and longer, wavelengths, as well as by
chemical reactions with other constituents. In the troposphere a primary source of O3
is through O(3 P) created by photodissociation of NO2 at wavelengths less than 410 nm.
The near-UV spectrum of solar radiation is therefore crucial to the concentration of
ozone, and of many other gases which it interacts with in the stratosphere and tropo-
sphere, and it is important to have precise knowledge of the spectrum of incoming solar
irradiance, and how it varies.
SOLSTICE instruments flown on UARS and SORCE have provided measurements
of the solar UV spectrum since the 1990s. More recently the spectral range has been
extended into the visible and near infrared through the instruments on ERS-2, 1994-
2003 [54], EnviSat, 2002 [55] as well as SORCE, 2003 [56]. But, while variation in
radiation at wavelengths less than about 250 nm is fairly well characterised, the accuracy
of the measurements at longer wavelengths is insufficient to provide reliable indication
of spectral variations.
Figure 6 shows (a) the reference solar spectrum recorded in April 2008 [57] and (b)
the relative spectral solar irradiance variability with negative values indicated by dotted
segments. This comes from measurements observed by UARS/SUSIM (red; [58]) be-
tween the maximum of cycle 23 (March 2000) and the preceding minimum (May 1996),
SORCE/SOLSTICE (light blue; [59]) and SORCE/SIM (dark blue; [60]) between April
2004 and December 2008. Also shown is the variability between 2000 and 1996 predicted
by the semi-empirical model SATIRE-S (green; [61,62]). For each period, averages over
Fig. 6. – (a) Reference solar spectrum recorded in April 2008. (b) Relative spectral solar irra-
diance variability, with negative values indicated by dotted segments (see text).
one month are used [63](1 ). Discrepancies between the datasets are clear, especially at
the longer wavelengths where even the sign of the variation is not well established.
Some examples of how sensitive the response of stratospheric O3 is to details of the UV
spectrum are given in fig. 7 [64]. This shows the percentage change between 2003 and 2008
(i.e. the change, over about 2/3 solar cycle) calculated using 4 different sources for the
UV spectrum. The differences are large, with the maximum change varying from about
0.9% with the SATIRE-S model data to 2.2% using the SORCE/SOLSTICE v.12 satellite
data; In the lower mesosphere values range from around −1.7% to +0.3%; the negative
response here is due to higher values of O(1 D), produced by greater photo-dissociation
of O3 in the 242 nm–310 nm region, reacting with H2 O to produce OH which destroys
O3 . This gives a clear example of how variations in spectra can produce significantly
different responses.
(1 ) Abbreviations: SATIRE-S, Spectral and Total Irradiance Reconstruction for the Satellite
era; SIM, Spectral Irradiance Monitor; SOLSTICE, Solar Stellar Irradiance Comparison Exper-
iment; SORCE, Solar Radiation and Climate Experiment; SUSIM, Solar Ultraviolet Spectral
Irradiance Monitor; UARS, Upper Atmosphere Research Satellite; WHI, Whole Heliosphere
Interval.
40 N. P. Fox
Fig. 7. – Change in stratospheric ozone between 2003 and 2008 using solar spectral irradiance
(SSI) from (a) the NRLSSI model, (b) the SATIRE-S model, and SORCE SOLSTICE (c) version
10 and (d) version 12. All results are for 25 Dec. Contour interval is 0.1% in (a), (b) and 0.2% in
(c), (d) with negative changes (i.e., increases with decreasing solar activity) shown with dotted
contours. Blue and purple reflect negative changes, while green, yellow, and red show positive
changes.
The first general circulation model (GCM) studies of the impact of solar UV-induced
variations in stratospheric temperature and ozone on the dynamical structure of the
troposphere [50] found a particular pattern of response in which the sub-tropical jets
weakened and moved poleward and tropical overturning (Hadley) circulations weakened
and expanded when the Sun is more active. The model runs suggested that the magnitude
of the tropospheric response depended on the specifications of the UV and ozone changes.
The pattern has been confirmed by statistical analyses of observational data [65] and
by experiments with coupled atmosphere-ocean-chemistry GCMs, e.g. [66], providing
additional evidence that coupling between photochemistry and dynamics is important

for surface climate.
More recently evidence has been accumulating to suggest that factors influencing the
strength of the polar stratospheric jet may be able to influence surface climate, at least
at mid-high latitudes. For example, [67] showed that changes in the winter polar vortex
introduced by large (SORCE/SIM-like) decreases in UV can produce a signal in surface
climate like the negative phase of the North Atlantic Oscillation. Analysis of sea level
pressure and temperature data [68] has revealed similar patterns, though with a lagged
response.
These studies provide evidence that changes in solar UV radiation can influence the
troposphere and stratosphere but a lack of long-term precise measurements of the solar
Fig. 8. – Plot showing variances in measurements of solar spectral irradiance from different
corrections to SOLSPEC data (red and green) and those by Weber (blue).
spectrum is hindering our understanding of the issues that surround the response of the
atmosphere to solar variability and, more generally, of chemistry-climate interactions.
Solar spectral irradiance (NIR): The TSI anomaly. Recently there has been and
continues to be, significant controversy about the absolute level of near-IR solar spectral
irradiance. The issue has arisen because of alternative interpretations of data from recent
SOLSPEC measurements on the ISS, where analyses differ by 7% (due mainly to differing
“beliefs” as to when SOLSPEC is at its most reliable: at instrument “first light” or after
“outgassing” (or “degradation” depending on point of view). The main crux of the
controversy stems from two papers in Solar Physics, see fig. 8. Uncertainty is captured
in the difference between the two SOLAR1 and SOLAR2 curves. Thullier et al. [69]
favour a higher value and a reply by Weber [70] favours a lower value. Menang et al. [71]
have provided evidence in support of the lower value using a ground-based calibrated
Sun-pointing FTS.
Significant issues flow from this fundamental debate regarding knowledge of solar out-
put. If TSI is anywhere close to that measured (noting that even including instrumental
biases the uncertainty in TSI has been thought to be 1% and close to or < 0.1%),
then the apparent mismatch between claimed accuracy of TSI and that resulting from
the integral of its spectral resolved components is of major concern. 7% uncertainty in
near-IR SSI, is equivalent to 3% or so of TSI —Thus if the TSI radiation is not in the
near-IR where is it? Is it in the Visible or UV or is TSI in error?
From an atmospheric perspective, the uncertainty in near-IR SSI translates into un-
certainty in heating rates within the atmosphere and surface fluxes, which have been
identified as key uncertain components of the climate system (see e.g. main ERB plot
in [124]) and also clearly a knock-on effect of using ground-based sensing of solar ra-
diation at the surface to understand atmospheric absorption/scattering processes (see
below).
Although not widely publicised as yet, the resolution of this controversy has a very
high priority and could easily be misinterpreted. It is perhaps one reason why the
forthcoming revision of GCOS ECVs has placed a high importance on the need to improve
42 N. P. Fox
Fig. 9. – Recent measurements of near IR water vapour self-continuum compared to models.
measurements of solar spectral irradiance.
Atmospheric absorption anomaly at solar near-IR wavelengths. As well as the gen-

eral issues with SWIR SSI, it is also important to note the uncertainty in knowledge
of atmospheric absorption (primarily due to water vapour), which is one of the largest
uncertainties in modern ERB [72]. In particular, uncertainty in continuum absorption in
near-IR windows, is ill-constrained in recent room-temperature laboratory observations
(e.g. [73-76]). Figure 9 [76] shows recent measurements of the near-IR water vapour
self-continuum compared to models. It should be noted that there is uncertainty in
temperature dependence of this absorption since the lab observations do not access tem-
peratures relevant to atmosphere, and so have to be extrapolated.
Improved understanding is required to understand radiation budget (including current
debate/discussion on interhemispheric differences in planetary albedo [14]), and in remote
sensing (sensors such as MODIS, OCO-2 exploit these near-IR windows; see e.g. [76]).
Thus there is a need for detailed well-calibrated measurements in these windows from
space, to help constrain surface-based estimates of shortwave absorption and increase
our understanding of this poorly understood part of the spectrum.
.
2 4. Climate feedbacks
.
2 4.1. Introduction. Change in ERB is the primary driver of the climate and if we make
the assumption that at present there is little significant change in input solar energy
and that, that is the dominant energy source to the planet, then it is changed in the
interaction of that radiation with the Earth system (including emitted thermal radiation
resulting from its absorption) that are ultimately driving climate change. The principle
warming mechanism is of course thought to be increased anthropogenically produced
greenhouse gases, most notably carbon dioxide, and their absorption of surface-emitted
thermal infrared radiation. However, both as a direct result of this warming and/or other
effects, changes in other aspects of the Earth system can have an impact, both positive
(increased warming) and negative (cooling). Often these can be non-linear and also lead
Fig. 10. – Strengths of individual feedbacks for CMIP3 and CMIP 5 models for Planck (P), water
vapour (WV), clouds (C) albedo (A), lapse rate (LR), combination of water vapour and lapse
rate (WV+LR) and sum of all feedbacks except Planck (ALL), from Soden and Help (2006) [80]
and Vial et al. (2013) [81] following Soden et al. (2008) [82]. CMIPS 5 feedbacks derived from
CMIP5 simulations for abrupt four-fold increases in CO2 concentrations (4 × CO2 ) (taken from
figure 9.43 IPCC AR5) [1].
to a feedback to the climate system and thus the source of much of the variance between
climate forecast models.
Currently there is significant variability between the predictions by different models
of the effects and consequences of climate change, mainly driven by uncertainty in the
climate sensitivity of the equilibrium response of the climate system to a given level
of anthropogenic radiative forcing from greenhouse gases, aerosols, and surface albedo
changes [1]. Uncertainty in climate sensitivity is one of largest sources of uncertainty
in predicting the future economic impacts of the climate change for any given emission
scenario [77]. At 90% confidence equilibrium, climate sensitivity has an uncertainty
factor of 4, which leads to uncertainty in economic impacts of a factor of roughly 16.
Water vapour/lapse rate, albedo and cloud are the principal determinants of equilibrium
climate sensitivity, with cloud feedbacks having the largest uncertainty [1].
Figure 10 shows the intermodal spread in individual feedbacks for the CMIP3 and
CMIP5 models as reported in IPCC AR5 [1]. Differences in cloud feedbacks are by
far the primary source of spread in both equilibrium and transient climate response
simulated by climate models [78,79] and for the CMIP5 models explain around 70%
of their spread in climate sensitivity [81]. Water vapour, lapse rate and albedo (snow,
ice land cover) feedbacks constitute most of the remaining variation. High accuracy
benchmark measurements against which future change can be unequivocally determined
are required for these variables to reduce the uncertainties in the climate feedbacks.
These measurements are also needed to quantify the radiative forcing from anthropogenic
changes in land albedo and quantify aerosol direct radiative forcing.
44 N. P. Fox
.
2 4.2. Cloud feedback on climate. Clouds play a major role in ERB by amplifying or
counteracting global warming [83,72]. Studies have shown a range of amplification values
from close to zero to up to 60% [80], however from CMIP 5 (see fig. 10) there remains
significant disagreement regarding the scale of impact [84]. The global distribution and
radiative properties of clouds is likely to alter in response to global warming, which
could have several different feedbacks on the surface temperature. For instance during
daytimes, reduced coverage of low-level cloud would result in less reflection of solar
radiation, and increased warming [78]. Uncertainty in the scale of CRF remains the
principle cause of variance (2.1 K to 4.7 K [1]) and thus societal impact, in model forecasts
This can only be constrained by detecting a long-term trend against high natural decadal
variability through high accuracy observations of cloud properties and in effect their
change from an unequivocal baseline benchmark.
Reducing the uncertainty associated with climate prediction by a factor of 2 would
require observation stability of 0.5% per decade for visible spectral radiances and 0.3%
per decade for broadband SW radiances [85]. This needs high accuracy spectrally-
resolved observations with high stability over decadal timescales to test the capability
of different climate models in accounting for the various cloud feedbacks (also depen-
dent on cloud type). Spectral resolution of 10–20 nm is needed for cloud-phase and
cloud-type characterization; for observing the transition zone between cloud and clear
sky; and the effect of aerosol on cloud formation. For reflected solar radiation, the
difference between spectral signatures for cloud fraction change and for cloud optical-
depth change is very similar. Therefore, sufficient spatial resolution (< 500 m ideally
250 m) is desirable to allow cloud-masking. A cloud fraction analysis would then allow
determination of cloud radiative forcing (all-sky minus clear-sky) which is needed for
cloud feedback [82]. It would also help in identifying decadal change of all-sky versus
clear-sky properties, when an all-sky spectrum change alone might not be sufficiently
discriminating.
The IPCC AR5 report [1] predicts about 0.2 K per decade change for the next few
decades, or an order of magnitude change of 0.06 W m−2 per decade for SW CRF. Given
the global mean value of SW CRF of 50 W m−2 , this is a relative uncertainty of approx-
imately 0.12% (k = 1) or 0.25% (k = 2).
The same conclusion can be derived by an alternative (order of magnitude) approach.
Expected anthropogenic radiative forcing of the climate system is expected to be ∼

0.6 W m−2 per decade for the next few decades [130]. A 25% cloud feedback would
amplify or dampen this forcing [86] by ∼ 25%, or roughly 0.15 W m−2 . This change
would be 0.3% of the average SW CRF of 50 W m−2 .
Whilst the above analysis can be used to specify the size of the trend to be detected
and thus simplistically the uncertainty requirement needed by a sensor, however, the
critical insight is that even a perfect observing system for measuring long-term forcing and
climate response is fundamentally limited by the noise of natural variability [87]. Such
variability includes a range of timescales: ENSO (3 years to 5 years), solar irradiance and
sunspot cycles (11 years); and Arctic, North Atlantic, and Pacific Decadal Oscillations
(10 years to 30 years). ENSO’s importance is recognized by the IPCC in its use of 5-year
Fig. 11. – The trend in cloud radiative forcing (CRF) with an accuracy in percent per decade
(k = 2) in relation to the length of observation (abscissa) for different calibration accuracies
currently achieved (MODIS, CERES) and (TRUTHS).
running means for comparisons of decadal change, thus a 5-year mission lifetime is an
optimal minimal to ensure that at least this dominant component of natural variability
is accounted for.
While ensemble techniques can reduce noise from natural variability in climate model
predictions or hindcasts, all observed trends are subject to the confounding noise of
natural variability. This means that there is a “floor” for required accuracy in climate
trends: the observations need to have uncertainties smaller than natural variability. The
key, therefore, is to quantify the relationship between natural variability and observing
system accuracy.
Even though climate changes are not simply linear trends, statistical trend analysis is
useful for robust comparison of the impact of different error sources and thus for critical
insights into mission science requirements [31,87-89]. Details of this analysis can be found
in [9].
Plotting the results of the analysis for SW CRF (fig. 11) the dominant source of
uncertainty in climate sensitivity due to albedo change from low cloud, demonstrates
several key points about climate observations: a) Trend accuracy increases with the
length of the climate record, even for a perfect observing system, because of the need to
average out noise in the climate system. b) Absolute calibration uncertainty dominates
the accuracy of and time to detect global average trends and c) That even with a perfect
observing system decadal measurements beyond the life of a single mission are required.
Future observations will be required to greatly reduce uncertainty in climate sensi-
tivity (currently almost a factor of 3) for a clearer understanding of climate change risks
46 N. P. Fox
over the next century, as well as to monitor as rapidly as possible the future effective-
ness of any future carbon emission controls [90]. Studies of the economic impacts of
climate change conclude that a factor of 3 uncertainty in climate sensitivity leads to
roughly a factor of 9 uncertainty in economic impacts, a quadratic relationship [77]. A
decadal change in SW CRF of 1%/decade changes the Earth’s radiation balance as much
as the anthropogenic radiative forcing of 0.5 Wm−2 /decade expected over the next few
decades [1]. The red line in fig. 11 shows that a signal this large would take 12 years to
detect with a perfect observing system. A smaller cloud feedback of half this magnitude
(0.5%/decade, blue line) would require 17 years of observations at 95% confidence for a
perfect observing system, and 20 years with a sensor such as TRUTHS, accuracy of 0.3%
(k = 2) which will be discussed in sect. 4.
For an accuracy change from 0.3% to 1.2%, the time to detect cloud feedback increases
by almost a factor of 2. Society would need to wait an extra decade for accurate climate
sensitivity information.
3. – Establishing SI traceability for the Earth observing system

.
3 1. Introduction. – Although every effort is made to ensure that any pre-flight cali-
bration and characterization of a satellite sensor can be relied upon and is traceable to
SI units, when the instrument is operational in orbit, it is well known that for optical
sensors this is rarely, if ever, the case. The high vibration of launch and sensitivity to
contamination of optical coatings and surfaces make some form of post-launch calibra-
tion, or at best validation, an essential component of any mission. Some sensors have
on-board calibration systems to offer the prospect of direct calibration, but often these
are themselves subject to the same sorts of issues as the instruments under test, partic-
ularly in the solar-reflective spectral domain. For the thermal IR, blackbody radiators,
which for practical (mass/size) reasons may not have quite as good emissivities as the
standards used on the ground, are probably reasonably good at providing some level of
in-flight performance monitoring. With effective design, these might be considered to
offer a sufficient level of “SI traceability” for operational missions, as they are relatively
insensitive to most of the common degradation mechanisms.
However, sensors operating in the solar-reflective domain (< 2500 nm) are a very dif-
ferent issue. Here, various on-board approaches have been and are being used, including
the flight of relatively fragile “standard lamps”. More commonly, diffusers illuminated
by solar irradiance to establish a Lambertian radiance ideally capable of filling the sen-
sor field of view are employed. Knowledge of the diffuser reflectance and solar irradi-
ance allows full in-flight calibration. Even in the absence of reliable values for these,
if they can be relied upon to be stable, then they can at least monitor and provide a
correction for drift. In practice, SSI, although reasonably stable in the short/medium
term, is only known to at best a few per cent in an absolute sense. Similarly, diffusers
degrade with time, particularly when exposed to sunlight. Effort is of course made to
limit exposure, in some cases, e.g. Medium Resolution Imaging Spectrometer (MERIS) of
ESA (http://envisat.esa.int/instruments/meris/), using multiple diffusers. These
methods have never claimed to be more accurate than a few per cent. Thus, although the
best of these are probably adequate for many operational applications, none are really
sufficient for the needs of long-term studies such as climate.
All satellite operators use some alternative vicarious method (or often a collection of
methods) to attempt to establish some level of confidence in the satellite observations
and, if appropriate, on-board calibration systems. These approaches have been recently
reviewed in Chander et al. [91] and included, for example, observation of stable deserts
(some instrumented), snowfields, clouds and also the moon, but none at present can
come close to the uncertainties needed for climate. However, if they could be validated,
and in some cases calibrated, by a sensor in orbit of sufficient accuracy, spectral and
spatial resolution, then the uncertainty of these methods could be reduced significantly.
This would allow the establishment of an operational global calibration system capable of
upgrading the performance of existing sensors (which usually have adequate sensitivity
but not accuracy) and of facilitating a “global climate observing system” rigorously tied
to SI units and now called for by international coordinating bodies of space agencies [8].
The following subsections provide a brief summary of the key methods used for post-
launch radiometric calibration and validation.
.
3 2. Near simultaneous overpass calibrations (SNO). – Simultaneous Nadir Overpass
(SNO) as the name implies is where the satellite under calibration overpasses the same
target (desert, snow, cloud, . . . ) as a designated “reference sensor” simultaneously (or
nearly, typically within around ±5 to 10 minutes). In this way both satellites view the
same target under the same; solar illumination, satellite viewing conditions and through
the same atmosphere, reducing many sources of potential error and uncertainty.
This type of comparison works particularly well when one satellite is geo stationary
and the other an orbiter, as there are regular overpasses. For two orbiting satellites on
a typical Sun-synchronous orbit there are relatively few SNO opportunities. However,
if the reference sensor is placed in a non-Sun-synchronous orbit for example a true 90◦
polar orbit, then there can be many SNO opportunities over a range of global locations,
surface types and solar illumination conditions.
The observational overlap time between sensors is improved (signal-to-noise) by in-
creasing the relative difference in height between the sensors and slowing the effective
ground transit time of the satellite by rotation control. It is of course also important to
ensure that the spatial footprint (pixel size) and spectral bands of the two sensors are
similar and/or ideally the reference sensor is sufficiently higher resolution than the test
sensor so that it can be accurately matched.
.
3 3. Reference standard calibration test sites. – The satellite community has for
many years sought to establish well characterised “test sites” often instrumented, to
provide a means of both calibration and validation of level 1 (top of atmosphere radi-
ances/reflectances) and also in some cases level 2 (e.g. reflectances of vegetated sites) or
higher products for example biophysical variables such as Leaf Area Index (amount of
vegetation). In recent years, significant efforts have been made by the international com-
48 N. P. Fox
Fig. 12. – CEOS endorsed test sites.
munity, particularly through CEOS, to consolidate and focus efforts on a sub-set of some
of these sites to aid interoperability and harmonisation. These have generically been
called “CEOS reference standard test sites”. In addition to the labelling, efforts have
started to establish true “calibration networks” of such sites analogous to the long es-
tablished AERONET sites for aerosols, together with tools to facilitate cross-calibration
between sensors.
The instrumented network on land is called RadCalNet (www.radcalnet.org) and the
non-instrumented are largely deserts and generically called Pseudo-Invariant Calibration
Sites (PICS), on the ocean there are the Buoys: Moby and Bousolle and the Aeronet-
OC network, see http://calvalportal.ceos.org/. The PICS, although seemingly
quite stable (surface reflectance and atmospheric transmittance) with time are generally
not readily accessible by virtue of their geographic locations and cannot have direct
measurements made except remotely from space or potentially by air, which limits their
ability to become SI traceable. They are of course ideal sites to monitor drift of individual
sensors and on which to perform SNO comparisons between sensors. Figure 12 provides a
schematic showing the current CEOS endorsed testsites for radiometric gain calibration
and monitoring.
For complex targets such as vegetation, there are additional challenges related to the
relatively high non-spatial homogeneity of the surface. For these, work is in progress to
develop optimum sampling strategies and range of measurements needed for field-based
measurements to be able to be considered representative, see www.npl.co.uk/trees.
.
3 4. Lunar calibration. – A special test site is the moon. As an extra-terrestrial
site it has been shown to be radiometrically stable, after taking account of libration,
etc., allowing satellite sensor stability to be monitored in orbit and inter-compared. The
dominant work in this field is carried out by the United States Geological Survey (USGS)
and its programme [144] on the Robotic Lunar Observatory (ROLO). This has now been
incorporated into an international programme of the WMO GSICS with the creation of
GIRO as an international reference model. However, until calibrated by some method
or reference sensor the absolute radiometric accuracy of the moon is not sufficient to
allow it to be used as an absolute standard at climate quality levels. However, once
achieved this will allow historic missions to be re-calibrated and FCDRs to be extended
backwards with greater accuracy, since many sensors have been observing the moon for
some decades.
.
3 5. Dominant sources of uncertainty. – For the above techniques and the applications
that the satellite sensor is designed for, the following list highlights the dominant sources
of uncertainty, in the context of remote sensing, that need to be considered. There are of
course many other instrument/measurement sources of uncetainty related to radiometry
in general that are not identified specifically here. The relative importance will of course
vary dependent on the parameter being measured and the final application.
Noise of detecting system: For instantaneous single pixel measurements the satellite
can only view the target for a very short time and so has little opportunity to average
to increase signal to noise ratios.
Knowledge (and control) of positioning: For high resolution (< 50 m) in particular,

geo-location of a specific target can be an issue. For coarse resolution (> 100 m) the size
and homogeneity of the target makes this less of a critical issue.
Drift and ageing of the sensor from pre-flight characterisation: Most optical sensors
incur changes to their pre-flight calibration characteristics on transition to orbit (stor-
age/vibration) and in orbit due to contamination from outgassing and also exposure to
radiation and vacuum.
Out of field stray light: Imperfect optics can lead to stray radiation from outside
the nominal field of view of the sensor (and individual pixels) and if the scene and
surroundings are not very similar in characteristics can lead to spectral dependent biases.
Adjacency (cross-talk) effects: Related to out of field stray light but also from current
leakage between pixels in array detectors, particularly where there are sharp contrasts in
scene reflectances e.g. land/water boundaries.
Atmospheric transmittance: Knowledge of input parameters (aerosols, water vapour,

ozone etc) for each scene and performance/accuracy of the computer algorithm used for
the correction (can be dominant for surface measurement of low reflectance).
50 N. P. Fox
Fig. 13. – Schematic of NPL traceablility chain for radiometric quantities.
Surface topography and angular reflectance characteristics: Knowledge of surface

angular reflectance properties when viewing at other than nadir view and/or effects of
solar illumination angle can have significant impact. This is particularly the case when
the surface is non-homogenous and non-Lambertian.
Spectral characteristics: For most scene types Earth/Sun there is significant spectral
detail, in some cases spectroscopic level e.g. atmospheric composition. In others with
broader smoother features ( 10 nm scales) small variances can still lead to errors in
detection of the bio-geo-physical parameter, for example ability to detect “stress”/disease
in a particular biological observable e.g. vegetation and of course ability to identify
individual species types. Detection of the correct signature will depend on the spectral
accuracy, resolution and coverage of the sensor performing the measurement.
.
3 6. Radiometric accuracy and traceability to SI . – Although all the above factors have
an impact, to varying degrees, on the overall accuracy of the measurements made by the
satellite, most can be constrained to reasonable levels for specific situations. However,
in most cases, it is the ability to determine, and rely upon in orbit, the conversion
factor that allows digital counts of the sensor to translate into physical SI units in a
traceable manner. In a pre-flight mode, the sensor, which can generally be considered as
a filter radiometer/spectrometer of some sort coupled with a light collecting telescope,
can be calibrated by illuminating the entrance aperture with a known source of radiation
(radiance/irradiance). The source often a spectral continuum from an incandescent lamp
illuminated integrating sphere, itself calibrated to SI using a “reference spectrometer”
linked back to SI, see fig. 13.
Traceability for the lamp-based radiance calibration stems from a sequential, quasi-
monochromatic, (using a monochromator and appropriate field of view defining optics)
wavelength by wavelength calibration of the radiance of the integrating sphere source

against a high temperature black body (∼ 3000 K) using for example the NPL SRIPS
facility (see fig. 13). The radiance of the black body being determined from its thermody-
namic temperature and emissivity using Planck’s law. The thermodynamic temperature
of the black body is determined by using a filter radiometer (detector/interference filter)
which itself has been calibrated traceably back to the NPL spectral responsivity scale
using a “trap detector” (essentially a special arrangement of silicon photodiodes) [93]
and ultimately the primary standard detector, the cryogenic radiometer [94,95], wave-
length by wavelength (∼ 0.1 nm intervals) using a tuneable laser illuminated integrating
sphere. The filter radiometer in this chain can be calibrated to measure radiance with
an uncertainty of ∼ 0.03% [96].
Alternatively, a tuneable laser source can illuminate the large integrating sphere,
which is overfilling the satellite sensor and its monochromatic radiance determined using
a calibrated photodiode (trap detector). This latter approach is the most accurate,
removing many steps in the calibration chain.
However, whilst such an approach can provide a pre-flight calibration of the sensor
with uncertainties < 0.1%, the challenge is how to maintain this or even check this in
orbit. All the vicarious calibration approaches described above are limited in terms of
absolute accuracy to at best a few percent, with no obvious way of reducing this signifi-
cantly. The requirement is thus for a method to establish or demonstrate stability of the
radiometric calibration in orbit. Most methods to-date rely on using the Sun as a “refer-
ence source” with its irradiance transformed to radiance by reflection from a lambertian
diffuser on-board the spacecraft that can be deployed in front of the telescope. The prin-
ciple limitations of this method and its variants is that the lambertian diffuser is prone
to drift and the Sun, although reasonably stable in the visible region of the spectrum,
only has a spectral irradiance measured to an uncertainty of ∼ 3 to 5%. Achieving SI
traceability in space with uncertainties needed for climate is a grand challenge directed
to the metrology community [97].
The next section describes a proposed satellite mission with an on-board calibration
system that mimics that used terrestrially pre-flight, allowing SI traceability with un-
precedented uncertainties to be achieved. Together with a hyperspectral imager this
provides the means to address many of the critical climate science questions highlighted
above whilst also serving as a “gold standard” reference sensor to upgrade the calibra-
tion and traceability of other satellite missions, in essence a “climate and calibration
observatory in space”.
4. – Traceable Radiometry Underpinning Terrestrial- and Helio-Studies

(TRUTHS): an NMI in space
.
4 1. Mission requirements and objectives. – TRUTHS [10,11] is a small satellite mission
designed explicitly to address the challenges highlighted above in a manner that facilitates
a paradigm shift in our approach to observing the Earth and Sun. TRUTHS will put
SI-traceable calibration into orbit to establish a fiducial reference data set of sufficient
52 N. P. Fox
accuracy, spectral and spatial resolution that will improve the interoperability of the
global EO system, provide a decadal benchmark for climate change and underpin and,
in many cases, upgrade, downstream applications and services. To achieve these goals
TRUTHS must make:
– Climate benchmark measurements of the Earth: high accuracy spectrally resolved

data, continuously sampled globally, Earth reflectance and radiance
– Climate benchmark measurements of the Sun: high accuracy total and spectrally-
resolved measurements of the solar irradiance
– Reference calibration measurements of surface reference sites: desert and ice PICS,
RadCalNet, AERONET-OC, sampled at nadir when overpassed.
– Reference calibration measurements of the moon: sampled across the lunar libration
cycle
– Simultaneous cross-calibration with other sensors: when opportunities arise, ideally

at Nadir
And, underpinning all of the above
– On-board SI-traceable calibrations of the Hyperspectral Imaging System (HIS) the

observational measurement system.
These objectives are achieved a) through the choice of an orbit that supports climate
benchmark measurements and increases the number of simultaneous cross-calibrations.
b) through the use of a small, agile platform to allow targeted pointing, c) through the
payload instrument: a hyperspectral imaging spectrometer (HIS) which can be used to
measure the Earth, Sun and moon and d) through the radiometric calibration of that
HIS directly, regularly and traceably to an on-board SI primary standard.
These objectives drive the specification of the mission, see table I
These 2σ uncertainties, whilst challenging, are achievable and essential to detect long-
term changes in key climate variables. The spatial resolution is not driven by climate
benchmark measurements themselves, which only require zonal averages, but rather to
ensure a good match to the footprint of other sensors when used as a calibration satellite
The data stemming from these driving requirements can of course have many other ap-
plications e.g. agriculture, forests, resource monitoring, in common with other satellites,
but delivered with significantly higher accuracy and confidence. Although demanding,
the underpinning observation technology and satellite platform are largely developed and
already proven as space instrumentation. The innovation derives from the paradigm shift
in the way it is used and in particular, the disruptive nature of the means to achieve
SI-traceable on-board calibration.
TRUTHS will fly the radiometric capability (including the primary standard) of an
NMI into space. In this way it will be able to act like a calibration or “standards
Table I. – Specification of the TRUTHS satellite to meet climate objectives.
Spectral Spectral Max Uncertainty/

Parameter Swath/km Sampling
range/μm resolution/nm GIFOV/m % (2σ)
Earth
Global nadir +
Spectral 0.32 to 2.3 ∼ 5 to 10 50 50 0.3
multiangle
Radiance
Lunar Weekly
Spectral 0.32 to 2.3 5–10 NA NA (libration 0.3
Irradiance sampled)
Total Solar
Irradiance 0.2 to 35 NA NA NA Daily 0.02
(TSI)
Solar Spectral
Irradiance 0.32 to 2.45 1 to 10 NA NA Daily 0.3
(SSI)
laboratory in space”. TRUTHS will be SI-traceable by design. This paper will not dwell
on the details of the instrumentation, only seek to highlight any key characteristics to
aid the reader understand the concepts.
The key challenge to this mission and thus one where we will concentrate our efforts
in terms of explanation, is the ability to achieve SI traceability at uncertainty levels a
factor of 10 below what has been achieved to-date. We will seek to show how this step
change is not in practice as radical as it might initially appear but rather the result of
taking the terrestrial laboratory calibration chain directly into orbit. In fact, when the
methodology which will be described here was first introduced into NMIs, some 30 years
ago, a step change of > 50× was achieved.
TRUTHS is implemented through a small agile satellite platform with two perpendic-
ular observation axis: Solar and Earth viewing, see fig. 14. The majority of observation
time is allocated to global nadir spectrally and spatially resolved measurements of Earth
reflected solar radiation for decadal climate change benchmarking. A nominal 5 to 10
minutes per day is allocated to solar observations (total and spectrally resolved). In
addition, the agile platform enables:
– near simultaneous angularly co-aligned observations for sensor reference calibra-

tions,
– multi-angular measurements of specific targets for reference inter-calibration of

sensors (surface and moon),
– Multi-angular measurements of specific targets to support “process studies”.
54 N. P. Fox
Fig. 14. – Schematic representation of the TRUTHS payload and its operational configuration.
The satellite has two viewing directions, towards the Sun or Earth, shown here approximately
60 degrees to each other.
The very high radiometric accuracy required by TRUTHS means that it seeks to be
largely self-reliant in most of its key measurands and thus includes the capability required
to provide the information to retrieve the atmospheric parameters (e.g. aerosols, water
vapour) required to account for radiation propagation losses in the atmosphere.
.
4 2. TRUTHS instrumentation. – A full description of the instrumentation employed
on TRUTHS is beyond the scope of this paper but a brief summary of the key instrument,
the primary standard cryogenic radiometer for flight which is called a Cryogenic Solar
Absolute Radiometer (CSAR), is included here to aid in the outline description of the
novel calibration method, more details can be found elsewhere [10,11].
Fig. 15. – Schematic representation of the TRUTHS payload and associated calibration system,
shown here on an Astobus-M platform.
.
4 2.1. Calibration system. Central to the TRUTHS philosophy is the calibration system
which puts SI traceability into orbit and is illustrated in fig. 15. This is achieved by mim-
icking the calibration systems for radiometric measurements pioneered by NPL 30 years
ago, and now widely used at NMIs across the globe. These systems have been simplified
relative to those used at NMIs to minimise the number of components, and in particular,
the number of moving parts compared to measurements in the laboratory and also to
reflect the use, where possible, of heritage space components or heritage constituent tech-
nologies; but they follow the same general principles as NMI calibrations. The individual
components and sub-systems are all at TRL 5 or higher. NPL, together with Airbus, has
recently built and tested under vacuum, a representative, fully operational breadboard
of the whole calibration system, including a new version of the primary standard, CSAR,
which has been integrated with the Airbus Defence and Space, space cooler (see fig. 16).
At the heart of the calibration system is a cryogenic radiometer, the primary standard
which is the reference for almost all radiometric calibrations at NMIs across the world.
The cryogenic radiometer is an instrument which absorbs incoming optical radiation and
compares the heating power of that incoming radiation to that from an electrical heater.
It thus measures optical power directly and this electrical substitution principle and the
basic design is the same as instruments such as PMOD 6 (VIRGO on SOHO) and others
that have been measuring TSI in space for the last 35 yrs. However, by cooling the system
to cryogenic temperatures (in this case < 60 K) makes the instrument more sensitive and
gives it the ability to measure optical power of the relatively low power calibration lasers
diodes (see below) with expanded (k = 2) uncertainties of 0.1%. Terrestrially this
is performed, a factor of ten better with 0.01% (k = 2) uncertainties. The cryogenic
radiometer on TRUTHS is known as CSAR and in addition to serving as the primary
standard in space it can also provide measurements of TSI when pointed to the Sun,
with an uncertainty of 0.02% (k = 2).
A first engineering model of CSAR-v1 was built in 2010 and demonstrated for Earth-
based measurements and is currently providing the World Radiometric Reference for
ground-based TSI from PMOD in Davos, Switzerland. This first prototype has now
been further optimised for space representativeness in conjunction with Airbus.DS. This
new version is coupled to an evolution of the High Power Stirling Cooler (HPSC) flown
on Sentinel 3, see fig. 16. The design of the new CSAR-v2 has been simplified, with 50%
lower mass than the previous version and also operates with similar performance at a
higher temperature and can now be considered as TRL 6.
Primary reference standard, CSAR. CSAR is a cryogenic radiometer that is the pri-
mary standard for radiant power and irradiance measurements on TRUTHS. When flown
it will be the first cryogenic radiometer flown in space and is a key element of the novel
calibration system of TRUTHS. Most NMIs, have adopted the cryogenic radiometer as
the primary standard for radiometric measurements [95,98,99]. The cryogenic radiome-
ter is an extension of the concepts of the Electrical Substitution Radiometer (ESR),
which were independently developed by the meteorologist Ångström (1893) [100] and
the metrologist, Kurlbaum (1894) [101] and are in essence the basis of instruments cur-
56 N. P. Fox
Fig. 16. – Overall layout for the TRUTHS CSAR, schematic diagram.
rently flying in space measuring TSI, e.g. PMO6 on SOHO. A cryogenic radiometer is an
ESR which is cooled down to cryogenic temperatures. Figure 17 illustrates the operating
principle of ESRs. A cavity that is connected to a heat sink via a weak heat link is
first heated up by incoming optical radiation (fig. 17, left hand side); subsequently, the
radiation is blocked and the cavity is electrically heated to the same temperature (fig. 17,
right hand side). This procedure allows the determination of radiant power through the
measurement of electrical power applied to the heater.
Fig. 17. – Operating principle of Electrical Substitution Radiometers.
The main reason for cooling the radiometer is that the cavity can be larger, whilst
keeping the natural time constant the same. This is due to the much reduced heat
capacity of the cavity material (copper) at low temperatures. A physically larger cavity
has much higher absorptivity, with less sensitivity to the degradation of the black surface
coating of the cavity. Being able to make the cavity larger also means that for the TSI
measurements, the larger FOV limiting aperture can be placed in front of the cavity
instead of at the front of the instrument; with significantly reduced bias due to stray light
caused by interaction of the light with the front aperture. Other reasons for operating
the detector at low temperatures are reduced background radiation and elimination of
bias in the electrical substitution by using superconducting leads to the electrical heater.
Cryogenic operation (originally < 30 K) reduces uncertainty levels to < 0.01% (k = 2)
as first shown by NPL in the 1980s [146]. For TRUTHS CSAR-v2 will operate at < 60 K
and provide traceability of Earth spectral radiance measurements to the International
System of Units (SI) to an absolute accuracy of < 0.3% (k = 2). Apart from serving as a
primary standard for radiance and irradiance, CSAR will also provide science measure-
ments in its own right of TSI with an accuracy of 0.02% (k = 2), a factor of ten better
than the currently operational, ambient temperature radiometers.
Because CSAR is both critical to the TRUTHS mission achieving its aims and the
key novel aspect of TRUTHS, the instrument was first prototyped (CSAR-v1) in 2010
and successfully underwent both laboratory and field tests (see below). A second version
of the instrument (CSAR-v2) is an evolved and simplified design making direct use
of an upgraded HPSC. Note that because of the cryogenic components, all cryogenic
radiometers are designed for vacuum operation and the CSAR-v1/2 can be considered
TRL 5/6 commensurate with the cooler.
The CSAR is a relatively simple device consisting of largely metal shields, with only
limited criticality in terms of alignment, with all the major design issues resolved following
more than 30 years of terrestrial heritage. The principal radiometric aspects (other than
being cooled) can be considered to have similar heritage in space with a range of space-
deployed radiometers measuring TSI since the 1970s. The main component of CSAR,
the radiometer detector element is essentially a closed metal cylinder with a thermometer
and a heater. CSAR can be considered as comprising three subsystems, see fig. 17:
– The radiometer head

– Mechanical coolers and drive electronics
– Measurement electronics
CSAR-v2 has been built and undergone extensive testing to demonstrate performance
with it integrated to a space cooler. With pessimistic heat loads, limiting operational
temperatures to ∼ 60 K it fully met its requirements, compromising only on time-constant
(30 s instead of the 10 s target) but this does not impact on the mission at all.
Radiometric head and coolers. The radiometer head contains the field-of-view limiting
apertures (enabling TSI measurements), mounted on radiation-limiting copper shields
and the optically absorbing detector cavities. For redundancy there are three identical
58 N. P. Fox
Fig. 18. – Photographs of the assembly of CSAR-v2 showing, from left to right: (a) two cavities
mounted on the temperature reference block; (b) the radiometer head coupled to HPSC and
mounted inside cryostat and terrestrial vacuum can; (c) close up of cavities showing thermal
link to HPSC.
absorbing cavities, any of which can be used to observe the Sun, for TSI and two of
which (aligned in the calibration axis) can be used to perform absolute calibration of the
HIS. These absorbing cavities are designed to perform optimally for input power levels
of between 5 mW and 20 mW, which is the input power range for both the calibration
laser diodes and for TSI with the defining apertures.
Each cavity consists of a copper cylinder with a sloped end (following the design of
NPL’s primary cryogenic radiometer). The backplate of the detector cavity is coated with
VANTA black paint [102] (a carbon nanotube black), and the front end with 3M Nextel
Black Velvet. This results in a cavity absorptivity of more than 99.998%. The cavities
are very insensitive to degradation of the surface coating and would need to degrade by
more than a factor of ten in order to have an impact on measurement uncertainty for TSI
and a factor of 100 for the on-board calibration. The fact that there are three cavities
provides an additional redundancy and opportunity to cross-compare. 3M black paint
has space heritage in many missions and similarly carbon nano-tube black coatings have
been flown by NASA and the European VANTA coating [102] has undergone testing
for performance change due to cryogenic operation and has recently undergone a full
program of testing to fully qualify it for space funded by UK space agency.
There are two types of aperture in front of each CSAR cavity. The defining apertures
(to define the area for irradiance) are 8 mm in diameter and made of diamond-turned
aluminium with a sharp edge. The FOV limiting apertures are 10 mm in diameter and
are after the defining apertures to reduce the stray light. The FOV limiting apertures
are on the thermal shield, which completely surrounds the radiometer head.
Figure 18 shows the cavities, thermal shields and thermal link of CSAR-v2. Fig-
ure 18(a) also shows how the detector stage has been designed to keep the cold stages to
as low a mass as possible. Three torlon dumbbell links mechanically connect one stage to
the next. These torlon links are at a slight angle to increase the natural eigen-frequencies
of the structure to 130 Hz. Finite element analysis has shown that the structure with-
stands accelerations of 100g in any direction. A cross-section overview of the CSAR-v2
detector head is given in fig. 19.
Fig. 19. – A cross-section through the CSAR-v2 detector head.
The cavity is connected to a copper reference block via a weak thermal link, which
defines the time constant and sensitivity of the cavity. The thermal time constant for
the cavities has been both calculated and measured at a 60 K operating temperature to
be ∼ 30 s.
The absolute power of the radiation incident to the cavity is measured by comparing
the cavity temperature rise due to optical radiation to that caused by electrical heating
via a resistive chip heater. The temperature of the cavity is measured using Lakeshore
CERNOX 1080 thermal sensors, which are suitable for space flight (high resistance to
ionising radiation). The designs for these have been improved between CSAR-v1 and
CSAR-v2 to make them more robust to vibrational shock. CSAR-v2 also uses a more
robust resistive nichrome heater design. The electrical power (and hence the optical
power which created the same heating) is evaluated by measuring the voltage drop across
the cavity heater and across a standard resistor in series with it, thus providing current.
The reference block is connected to the coldtip of the cooler, and is temperature
controlled via PI-control, using a temperature sensor and a cartridge heater. The cold
tip of the cooler is ideally maintained at as low a temperature as possible via a space
cooler and at least 60 K. The baseline design is a slightly upgraded Airbus-DS HPSC.
Pointing accuracy towards the Sun of 0.1◦ is required and easily achievable by the
satellite bus. However, in addition to pointing to the Sun it is important to know where
the instruments are pointing relative to the centre of the solar disc. This can be relatively
easily achieved with a simple quadrant photodiode attached to the front plate of CSAR
and has proven flight heritage from previous PMOD missions.
60 N. P. Fox
Table II. – Uncertainty budget for CSAR used as a reference standard and for TSI measure-
ments.
Source of uncertainty Spectral (calibration) Total Solar Irradiance
LD reference standard Uncertainty (k = 2)/%
Uncertainty (k = 2)/%
Measurement of electrical power 0.010 0.010
Area of defining aperture 0.008
Cavity absorptance 0.010 0.004
Diffraction correction 0.020
Scattered light 0.004 0.004
Random noise 0.06 0.010
Total 0.06 0.026
The CSAR can now be considered to be TRL 5/6. The basic design is well tested
and modelled for space worthiness and is very similar to heritage flown designs of PMOD
with the exception of its cooling.
Uncertainty budget for CSAR-v2 . Table II provides an uncertainty budget for CSAR
to SI as used in space, it differs from that calculated for terrestrial use since there is
no need for a window and its transmittance correction and scatter from sky radiance is
also not applicable. For the spectral calibration power of the LD used for the on-board
calibration system the cavity is under filled and so there are no contributions due to
aperture or diffraction.
Evidence of performance. CSAR-v1 has been tested end-to-end by comparing it to 1)

NPL’s SI standard for radiant power, and 2) the current global standard for TSI. CSAR-
v1 agrees with the SI standard for radiant power within measurement uncertainties (see
fig. 20(a)). CSAR-v1 was also compared with the global standard for Solar Irradiance,
the World Radiometric Reference (WRR) (fig. 20(b)). Figure 20(c) shows that the WRR
is measuring lower than CSAR-v1 by 0.26% ± 0.05% (1σ). The dominant uncertainty
in this comparison relates to the use of a window and its transmittance for terrestrial
applications. This result is consistent with an earlier independent comparison of the
WRR with the SI (fig. 20(d)), which showed that the WRR is measuring too low by
0.34% ± 0.09% [103] and also the recent changes in results of space-based measurements.
It is perhaps notable how valuable a flight of a CSAR in space would be to anchor the
TSI in both space and ground to SI.
.
4 3. On-board calibration methods
.
4 3.1. Overview. The calibration of the hyperspectral imaging spectrometer (HIS) for
Earth radiance and SSI measurements is performed, traceably to the CSAR, using a
system that consists of a Transfer Radiometer (TR) (essentially a photodiode), a small
Fig. 20. – (a) Result of comparison between CSAR-v1 and SI standard for radiant power and
(b) CSAR-v1 installed on solar tracker at World Radiation Center (PMOD/WRC), side-by-side
with the World Radiometric Reference. (c) Result of comparison between CSAR and World
Radiometric Reference. (d) Result of comparison between World Radiometric Reference and SI.
number (∼ 8–10) low power Laser Diodes (LDs), an Integrating Sphere (IS) and a diffuser
wheel with two diffusers (D1, D2). It generally takes place during the Earth shadow part
of the orbit allowing the systems to be well-protected from solar and other radiation
damage. Specific design details for space variants of the various sub-systems are still
to be determined but are readily achievable from heritage systems. Here we outline the
key concepts behind the subsystems to provide the reader with understanding of the
principles.
The following description and schematics provide an outline of the method which
mimics that carried out on the ground but adapted to meet the limitations of a space-
craft. Note that the diagrams here are schematic, and in the actual implementation,
for example, additional turning mirrors may be used, which do not affect the overall
calibration principle.
The primary calibration is performed at a small number of wavelengths corresponding
to the chosen LDs. These LDs can illuminate either of two (identical) cavities of the
CSAR (providing redundancy), the TR and the IS (which provides radiation to illuminate
a diffuser overfilling the input telescope of the HIS). The LDs illuminate each of these
sub-systems in turn using an illumination system based on a rotating pair of rooftop
62 N. P. Fox
Fig. 21. – LD illumination system (schematic).
prisms serving as mirrors, with the central axis (mirror) being fed by laser radiation
from a common optical rod conjoiner linking LDs via optical fibre (fig. 21). The entrance
apertures of each of these sub-systems being on a common plane. The monochromatic
radiation from the LDs illuminates the full aperture of HIS in a Lambertian manner via
a flat plate diffuser in a similar manner to existing on-board systems, the difference being
the use of LD radiation rather than the Sun. The radiance of the diffuser can be measured
directly inflight by an on-board calibrated TR. The responsivity at other wavelengths is
obtained using a white light source e.g. lamp. The following steps scope the calibration
procedure which mimics that carried out terrestrially. It should also be noted that in
addition to radiometric calibration the use of LDs allows wavelength accuracy of HIS
to be checked and also of particular interest, stray-light. The latter whilst not having
enough LDs to provide a full calibration matrix as can be done on ground using tuneable
lasers, does provide sufficient spectral sampling to allows the pre-flight matrix to be at
least monitored in flight.
.
4 3.2. Step 1: Calibration of TR against CSAR using LDs. The optical power of the
temperature-and-current-stabilised LD is measured absolutely by the CSAR and then
used to calibrate the response of the TR to this power level by rotation of the prism
assembly, see fig. 22. The TR is a small integrating sphere with integral Si and InGaAs
semiconductor photodiodes viewing its internal wall. The integrating sphere provides a
convenient means to reduce any alignment or source non-uniformity issues. It has two
field-of-view (FOV) defining apertures, but for this calibration those are underfilled. This
step is repeated for each of the LD wavelengths in turn. The transfer requires the LD
output to be of known wavelength (controlled with an etalon) and stable in intensity
(< 0.1%) during the transfer from the CR to TR (about 1 minute, which is easily
achieved). Individual wavelength calibrations can take place sequentially or on different
orbits. However, it is probably likely to be optimum to perform the full calibration
sequence step 1 and 2 for each wavelength individually.
Fig. 22. – Schematic of step 1. The LDs illuminate (underfill) first (a) the CSAR cavity and
then (b) the TR, by rotating the mirror assembly.
.
4 3.3. Step 2: HIS (Earth Radiance view) calibrated against the TR at each LD wavelength
(radiance mode). The radiance responsivity of the HIS is determined by comparison with
the radiance responsivity of the TR using the LDs. In this step the LD source is allowed
to expand and then collimated to create a uniform illumination on a flat plate lambertian
diffuser, using an IS and mirrors. The IS not only helps to pre-condition the LD radiation
removing effects of non-uniformity but also reduces significantly the effect of speckle, from
both spatial and temporal coherence, The diffuser can move in and out of the FOV of the
HIS by rotation of its mounting plate shown in fig. 23 with a second redundant diffuser
and also mirror. When illuminated, normal to the diffuser as shown (fig. 23), the TR
and HIS simultaneously view the diffuser at reciprocal angles.
In this step the TR measures the absolute radiance of the LD-illuminated diffuser. Its
absolute radiance responsivity having been determined from its power responsivity (step
Fig. 23. – LD illuminates the diffuser, via the IS, and this is viewed by the HIS and IS.
64 N. P. Fox
Fig. 24. – Lamp illuminates the diffuser via the integrating sphere and mirrors to provide
calibration of the HIS at intermediate wavelengths.
1) and knowledge of the areas of the two FOV-defining apertures, which are never directly
exposed to the space environment. Note that, because the TR measures the radiance of
the diffuser directly, the transmittance of the IS and reflectance of any turning mirrors
(optics) do not need to be known. It is advantageous to limit the FOV differences
between the TR and the HIS, but such differences and their effect is small and can be
characterised pre-flight and therefore an exact match is not required.
Thus the HIS is calibrated for radiance responsivity at the LD wavelengths. Note
also that by comparing the TR signals in Step 1 and Step 2 (assuming a stable laser
source between these two measurements) the TR also directly calibrates the throughput
of the IS, mirrors and diffuser at these wavelengths. This provides an irradiance respon-
sivity calibration of the system comprising the IS (underfilling an irradiance defining
aperture on the entrance port), mirrors, diffuser and HIS (this allows it to become the
SSI measuring system).
.
4 3.4. Step 3: HIS (Earth radiance view) calibrated at intermediate wavelengths with
a lamp. An incandescent lamp (with a smooth spectral output) illuminates the IS and
consequently the diffuser and is viewed by the HIS in a similar manner to the LDs, see
fig. 24. At the end of step 2, the HIS has been calibrated at the wavelengths of the
LDs for absolute radiance responsivity. It is now only necessary to interpolate between
these laser wavelengths. This will be achieved using a white light source/lamp with a
broadband, smooth spectral output. Although the lamp will change during operation,
it will do so in a spectrally smooth manner. The absolute level and the basic spectral
shape is determined at the LD wavelengths in step 2, this step is used for interpolation
only. There are many examples of such lamps being used for space calibration, here we
only need functionality and short term (< 10 s) stability.
Fig. 25. – Measurements of the Earth and Sun.
The HIS can then be considered calibrated for making spectral radiance measurements
of the Earth. Similarly, the system comprising the IS, mirrors, diffuser and HIS can be
considered calibrated for making spectral irradiance measurements of the Sun where the
aperture on the IS is the defining aperture.
.
4 3.5. Step 4: Measurements of the Earth and Sun. Measurements of the radiance of
the Earth are made with the HIS directly. The diffuser disc rotates to allow light from
the Earth to enter the HIS telescope, see fig. 25. When the TRUTHS satellite is rotated
to face the Sun, then the CSAR can measure TSI and the IS (illuminating the mirrors,
diffuser and HIS) can measure SSI. For UV measurements of SSI, where the signal levels
are very low, the mirror (M1) rather than diffuser (D1/D2) is used in the path between
the IS and HIS.
These measurements are fully SI-traceable because of this calibration scheme for their
radiometric responsivity. Full traceability to SI also needs traceability for the defining
aperture areas (the two apertures on the TR, the aperture on the IS, and the aperture on
CSAR). These apertures will be calibrated on the ground. In orbit, apertures on CSAR
and IS will both be subject to direct exposure to space and the Sun there is a need
therefore to monitor the stability of those aperture areas. CSAR will have redundant
apertures (different apertures on each cavity, which can have varying degrees of exposure).
For the IS and in principle the TR, a dedicated LD beam (at one wavelength since
geometry has no spectral dependency) is expanded with optics to a size that overfills the
defining apertures of CSAR and the IS. By illuminating each CSAR aperture (redundant
options) in turn, it can compare these apertures and calibrate the IS in irradiance (rather
than power) mode. The same LD irradiance beam can also be used to check the smaller
TR aperture.
.
4 3.6. Summary. Figure 26 shows a schematic representation of the calibration se-
quence for TRUTHS. Each complete iteration through the cycle (single wavelength) is
estimated to take a most 5 minutes including time for any motions and stabilisation of
the system.
66 N. P. Fox
Fig. 26. – Schematic showing overall calibration sequence.
5. – Conclusions
This paper has outlined the need and urgency for significantly improved accuracy
in measurements made of the Earth-Sun system to provide policy makers and society
the unequivocal evidence required to ensure concerted timely action from the worlds
governments.
The necessary improvements in accuracy present a grand challenge to the metrology
community to provide the methods/standards to facilitate a step change in uncertainty,
in space, to that comparable to what is currently achieved using primary standards in
laboratory conditions.
The paper describes how this grand challenge has been embraced by the metrology
community resulting in a small satellite mission proposal, TRUTHS, to establish an NMI
in space. The underpinning metrological concepts of how SI traceability can be estab-
lished on-board the spacecraft have been described. Although the mission is not as yet
funded, it and its sister CLARREO, are receiving significant international attention and
have been through initial engineering studies in preparation for space implementation.
TRUTHS for example could be ready for launch within 3 to 4 years.
Of particular interest is the ability for the missions to upgrade the performance of
other satellite observing systems through reference calibration, in this way enabling the
establishment of a global climate observing system in the relative near term without the
significant cost of many new specifically tailored satellites.
In summary, the urgency of a climate benchmark mission is such that early implemen-
tation is imperative and TRUTHS offers a means to achieve that for the solar spectral
domain and in doing so provides a means to upgrade the performance of the whole Earth
Observing system and provide a showcase example of the benefits of metrology and SI
traceability.
∗ ∗ ∗
I thank UK Space Agency through its Centre for Earth Observation Instrumentation
(CEOI) and the European Metrology Research Program (EMRP) of the European Union
and EURAMET, together with the National Measurement Office of the Business Energy
and Industrial Strategy Department of UK government for their financial support. I also
thank the many contributors from my team at NPL and elsewhere who have carried out
much of the work that his been described in this overview.
REFERENCES
[1] IPCC, 2013: Fifth Assessment Report of the Intergovernmental Panel on Climate
Change, in: Climate Change 2007: The Physical Science Basis, https://www.ipcc.ch/
report/ar5/wg1/ (2013).
[2] Trenberth K. E. and Fasullo J. T., CLIMATE CHANGE Tracking Earth’s Energy,
Science, 328 (2010) 316.
[3] http://www.wmo.int/pages/prog/gcos/index.php?name=EssentialClimateVariables.
[4] Ohring G. (Editor), Achieving Satellite Instrument Calibration for Climate Change
(ASIC3), NPOESS Workshop Report, 144 (2007).
[5] https://www.wmo.int/pages/prog/gcos/Publications/gcos-107.pdf (2006).
[6] Ohring G. B., Wielicki B. A., Spencer R., Emery B. and Datla R., Bull. Am.
Meteorol. Soc., 86 (2005) 1303.
[7] NRC, 2007: Earth Science and Applications from Space: National Imperatives for the
Next Decade and Beyond, 418 (2007).
[8] http://www.wmo.int/pages/prog/sat/documents/ARCH strategy-climate-
architecture-space.pdf.
[9] Wielicki B. et al., Bull. Am. Meteorol. Soc., 95 (2013) 1519.
[10] Fox N. P., Kaiser-Weiss A., Schmutz W., Thome K., Young D., Wielicki B. and
Woolliams E., Philos. Trans. Series A, Math. Phys. Engin. Sci., 369 (2011) 4028.
[11] http://www.npl.co.uk/truths.
[12] ESA, Earth Observation Science Strategy for ESA: A New Era for Scientific Advances
and Societal Benefits (2015).
[13] Loeb N. G. et al., Observational constraints on atmospheric and oceanic cross-equatorial
heat transports: revisiting the precipitation asymmetry problem in climate models, Climate
Dyn., 46 (2016) 3239.

[14] Stephens G. L. et al., The albedo of Earth, Rev. Geophys., 53 (2015) 141.
[15] Global Climate Observing System, GCOS 107: Systematic observation requirements for
satellite based products for climate (WMO/TD 1338) (2006).
[16] Otto A. et al., Energy budget constraints on climate response, Nature Geosci, 6 (2013)
415.
[17] Allan R. P. et al., Physically Consistent Responses of the Global Atmospheric
Hydrological Cycle in Models and Observations, Surveys Geophys., 35 (2014) 533.
[18] Allan R. P. et al., Changes in global net radiative imbalance 1985–2012, Geophys. Res.
Lett., 41 (2014) 5588.
[19] Haywood J. M. et al., The impact of equilibrating hemispheric albedos on tropical
performance in the HadGEM2-ES coupled climate model, Geophys. Res. Lett., 43 (2016)
395.
68 N. P. Fox
[20] Loeb N. G. et al., Observed changes in top-of-the-atmosphere radiation and upper-ocean

heating consistent within uncertainty, Nature Geosci, 5 (2012) 110.
[21] Dessler A. E., Observations of Climate Feedbacks over 2000–10 and Comparisons to
Climate Models, J. Climate, 26 (2013) 333.
[22] Palle E. et al., Earth’s albedo variations 1998–2014 as measured from ground-based
earthshine observations, Geophys. Res. Lett., 43 (2016) 4531.
[23] Wielicki B. A., Clouds and the Earth’s radiant energy system (CERES): An Earth
observing system experiment (vol 77, pg 860, 1996), Bull. Am. Meteorol. Soc., 77 (1996)
1590.
[24] Harries J. E. et al., The Geostationary Earth Radiation Budget Project, Bull. Am.
Meteorol. Soc., 86 (2005) 945.
[25] Priestley K. J. et al., Postlaunch Radiometric Validation of the Clouds and the Earth’s
Radiant Energy System (CERES) Proto-Flight Model on the Tropical Rainfall Measuring
Mission (TRMM) Spacecraft through 1999, J. Appl. Meteorol., 39 (2000) 2249.
[26] Wielicki B. A. et al., Clouds and the Earth’s Radiant Energy System (CERES):
Algorithm overview, IEEE Trans. Geosci. Remote Sensing, 36 (1998) 1127.
[27] Barkstrom B. R., The Earth Radiation Budget Experiment (ERBE), Bull. Am.
Meteorol. Soc., 65 (1984) 1170.
[28] Brindley H. E. and Russell J. E., An assessment of Saharan dust loading and
the corresponding cloud-free longwave direct radiative effect from geostationary satellite
observations, J. Geophys. Res.: Atmospheres., 114 (2009) D23201.
[29] Futyan J. M., Russell J. E. and Harries J. E., Determining cloud forcing by cloud
type from geostationary satellite data, Geophys. Res. Lett., 32 (2005) L08807.
[30] Trenberth K. E. and Fasullo J. T., CLIMATE CHANGE Tracking Earth’s Energy,
Science, 328 (2010) 316.
[31] Leroy S. et al., Testing climate models using thermal infrared spectra, J. Climate, 21
(2008) 1863.
[32] Matthews G. et al., Compensation for spectral darkening of short wave optics occurring
on the cloud’s and the Earth’s radiant energy system (2005).
[33] Loeb N. G. et al., Toward Optimal Closure of the Earth’s Top-of-Atmosphere Radiation
Budget, J. Climate, 22 (2009) 748.
[34] Clerbaux N. et al., Comparison of GERB instantaneous radiance and flux products with
CERES Edition-2 data, Remote Sensing Environ., 113 (2009) 102.
[35] Hansen J. et al., Earth’s energy imbalance: Confirmation and implications, Science, 308
(2005) 1431.
[36] PMOD WRC, Solar Constant: Construction of a composite TSI Time Series from 1978
to present. 2014 [cited 2016 20/06]; available from: https://www.pmodwrc.ch/pmod.

php?topic=tsi/composite/SolarConstant.
[37] Fröhlich C. and Lean J., Solar radiative output and its variability: evidence and
mechanisms, Astron. Astrophys. Rev., 12 (2004) 273.
[38] Muscheler R. et al., Solar activity during the last 1000 yr inferred from radionuclide
records, Q. Sci. Rev., 26 (2007) 82.
[39] Vonmoos M., Beer J. and Muscheler R., Large variations in Holocene solar activity:
Constraints from Be-10 in the Greenland Ice Core Project ice core, J. Geophys. Res.-Space
Phys., 111 (2006) A10105.
[40] Schatten K. H. and Orosz J. A., Solar-Constant Secular Changes, Solar Phys., 125
(1990) 179.
[41] Lean J., Skumanich A. and White O., Estimating the Sun’s Radiative Output During
the Maunder Minimum, Geophys. Res. Lett., 19 (1992) 1591.
[42] Hoyt D. V. and Schatten K. H., A Discussion of Plausible Solar Irradiance Variations,
1700-1992, J. Geophys. Res.-Space Phys., 98 (1993) 18895.
[43] Lean J., Beer J. and Bradley R., Reconstruction of Solar Irradiance since 1610 -
Implications for Climate-Change, Geophys. Res. Lett., 22 (1995) 3195.
[44] Solanki S. K. and Fligge M., A reconstruction of total solar irradiance since 1700,
Geophys. Res. Lett., 26 (1999) 2465.
[45] Wang Y.M., Lean J. L. and Sheeley N. R., Modeling the Sun’s magnetic field and
irradiance since 1713, Astrophys. J., 625 (2005) 522.
[46] Krivova N. A., Balmaceda L. and Solanki S. K., Reconstruction of solar total
irradiance since 1700 from the surface magnetic flux, Astron. Astrophys., 467 (2007)
335.
[47] Steinhilber F., Beer J. and Frohlich C., Total solar irradiance during the Holocene,
Geophys. Res. Lett., 36 (2009) L19704.
[48] Schöll M. et al., Long-term reconstruction of the total solar irradiance based on neutron
monitor and sunspot data, Adv. Space Res., 40 (2007) 996.
[49] Soden B. J., Held I. M., Colman R., Shell K. M., Kiehl J. T. and Shields
C. A., Quantifying climate feedbacks using radiative kernels, J. Clim., 21 (2008) 3504
(doi:10.1175/2007JCLI2110.1).
[50] Haigh J. D., The Role of Stratospheric Ozone in Modulating the Solar Radiative Forcing
of Climate, Nature, 370 (1994) 544.
[51] Lean J., Evolution of the Sun’s spectral irradiance since the Maunder Minimum, Geophys.
Res. Lett., 27 (2000) 2425.
[52] Larkin A., An investigation of the effects of solar variability on climate using atmospheric
models of the troposphere and stratosphere, PhD University of London (2000).
[53] Haigh J. D., The Impact of Solar Variability on Climate, Science, 272 (1996) 981.
[54] Weber M., GOME - The Global Ozone Monitoring Experiment; available from:
http://www.iup.uni-bremen.de/gome/ (2010).
[55] Universität Bremen, SCIAMACHY ON ENVISAT.
[56] University of Colorado B., SORCE: Solar Radiation & Climate Experiment.
[57] Woods T. N. et al., Solar Irradiance Reference Spectra (SIRS) for the 2008 Whole
Heliosphere Interval (WHI), Geophys. Res. Lett., 36 (2009) L01101.
[58] Floyd L. et al., 11 years of solar UV irradiance measurements from UARS, in Solar
Variability as an Input to the Earth’s Environment (2003).
[59] Snow M. et al., The Mg II Index from SORCE, Solar Phys., 230 (2005) 325.
[60] Harder J. W. et al., Trends in solar spectral irradiance variability in the visible and
infrared, Geophys. Res. Lett., 36 (2009) L07801.
[61] Krivova N. et al., Reconstruction of solar UV irradiance since 1974, J. Geophys. Res.:
Atmos., 114 (2009) 6.
[62] Krivova N. A., Solanki S. K. and Unruh Y. C., Towards a long-term record of solar
total and spectral irradiance, J. Atmos. Solar-Terrestrial Phys., 73 (2011) 223.
[63] Solanki S. K., Krivova N. A. and Haigh J. D., Solar irradiance variability and climate,
arXiv preprint arXiv:1306.2770 (2013).
[64] Ball W. T. et al., A New SATIRE-S Spectral Solar Irradiance Reconstruction for Solar
Cycles 21–23 and Its Implications for Stratospheric Ozone, J. Atmos. Sci., 71 (2014)
4086.
[65] Haigh J. D., The effects of solar variability on the Earth’s climate, Philos. Trans. R.
Soc. London A: Math., Phys. Eng. Sci., 361 (2003) 95.
[66] Shindell D. T. et al., Solar and anthropogenic forcing of tropical hydrology, Geophys.
Res. Lett., 33 (2006) 24.
70 N. P. Fox
[67] Ineson S. et al., Solar forcing of winter climate variability in the Northern Hemisphere,
Nature Geosci, 4 (2011) 753.
[68] Gray L. J. et al., A lagged response to the 11 year solar cycle in observed winter
Atlantic/European weather patterns, J. Geophys. Res.: Atmospheres, 118 (2013) 405.
[69] Thuillier G. et al., The infrared solar spectrum measured by the SOLSPEC spectrometer
onboard the International Space Station, Solar Phys., 290 (2015) 1581.
[70] Weber M., Comment on the Article by Thuillier et al. “The Infrared Solar Spectrum
Measured by the SOLSPEC Spectrometer onboard the International Space Station”, Solar
Phys., 290 (2015) 1601.
[71] Menang K. P. et al., A high-resolution near-infrared extraterrestrial solar spectrum
derived from ground-based Fourier transform spectrometer measurements, J. Geophys.
Res.: Atmospheres, 118 (2013) 5319.
[72] Stephens G. L. et al., An update on Earth’s energy balance in light of the latest global
observations, Nature Geosci., 5 (2012) 691.
[73] Ptashnik I. et al., Near-infrared water vapour self-continuum at close to room
temperature, J. Quantitative Spectros. Rad. Transfer, 120 (2013) 23.
[74] Ptashnik I. V. et al., Water vapor self-continuum absorption in near-infrared windows
derived from laboratory measurements, J. Geophys. Res.: Atmospheres, 116 (2011)
D16305.
[75] Mondelain D. et al., Temperature dependence of the water vapor self-continuum by
cavity ring-down spectroscopy in the 1.6 μm transparency window, J. Geophys. Res.:
Atmospheres, 119 (2014) 5625.
[76] Shine K. P., Ptashnik I. V. and Rädel G., The water vapour continuum: brief history
and recent developments, Surveys Geophys., 33 (2012) 535.
[77] Interagency Working Group on Social Cost of Carbon, U.S.G., Technical Support
Document: Social Cost of Carbon for Regulatory Impact Analysis (2010).
[78] Bony S. et al., How well do we understand and evaluate climate change feedback
processes? J. Climate, 19 (2006) 3445.
[79] Dufresne J.-L. and Bony S., An assessment of the primary sources of spread of global
warming estimates from coupled atmosphere-ocean models, J. Climate, 21 (2008) 5135.
[80] Soden B. J. and Held I. M., An assessment of climate feedbacks in coupled ocean–
atmosphere model, J. Climate, 19 (2006) 3354.
[81] Vial J., Dufresne J.-L. and Bony S., On the interpretation of inter-model spread in
CMIP5 climate sensitivity estimates, Climate Dyn., 41 (2013) 3339.
[82] Soden B. J. et al., Quantifying climate feedbacks using radiative kernels, J. Climate, 21
(2008) 3504.
[83] Roberts Y. L., Pilewskie P. and Kindel B. C., Evaluating the observed variability in
hyperspectral Earth-reflected solar radiance, J. Geophys. Res.: Atmospheres, 116 (2011)

D24119.
[84] Flato G. et al., Evaluation of climate models, in Climate change 2013: the physical
science basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change (Cambridge University Press) 2013 p. 741–
866.
[85] Loeb N. G. et al., Angular distribution models for top-of-atmosphere radiative flux
estimation from the Clouds and the Earth’s Radiant Energy System instrument on the
Terra satellite. Part II: Validation, J. Atmos. Oceanic Technol., 24 (2007) 564.
[86] Roe G. H. and Baker M. B., Why is climate change so unpredictable? Science, 318
(2007) 629.
[87] Leroy S. S., Anderson J. G. and Ohring G., Climate signal detection times and
constraints on climate benchmark accuracy requirements, J. Climate, 21 (2008) 841.
[88] Weatherhead E. C. et al., Factors affecting the detection of trends: Statistical

considerations and applications to environmental data, J. Geophys. Res., 103 (1998)
17, 149.
[89] Von Storch H. and Zwiers F., Statistical analysis in climatology (Cambridge
University Press, Cambridge) 1999.
[90] Trenberth K. E. et al., Challenges of a Sustained Climate Observing System, in Climate
Science for Serving Society: Research, Modeling and Prediction Priorities, edited by
Asrar R. G. and Hurrell W. J. (Springer Netherlands: Dordrecht) 2013, p. 13–50.
[91] Chander G., Hewison T. J., Fox N., Wu X., Xiong X. and Blackwell W.J.,
Overview of Intercalibration of Satellite Instruments, IEEE TGRS, 51 (2013) 1056.
[92] http://www.moon-cal.org/.
[93] Fox N. P., trap detectors and their properties, Metrologia, 28 (1991) 197.
[94] Quinn T. J. and Martin J. E., A radiometric determination of the Stefan-Boltzmann
constant and thermodynamic temperatures between −40 ◦ C and +100 ◦ C, Philos. Trans.
R. Soc., 316 (1985) 85.
[95] Fox N. P. et al., A mechanically cooled portable cryogenic radiometer, Metrologia, 32
(1995) 581.
[96] Fox N. P., Martin J. E. and Nettleton D. H., Absolute spectral radiometric
determination of the thermodynamic temperatures of the melting/freezing points of gold,
silver and aluminium, Metrologia, 28 (1991) 357.
[97] http://www.bipm.org/en/conference-centre/bipm-workshops/wmo-bipm workshop/
(2010).
[98] Gentile T. R. et al., National Institute of Standards and Technology high-accuracy
cryogenic radiometer, Appl. Opt., 35 (1996) 1056.
[99] Goebel R., Stock M. and Köhler R., Report on the International Comparison of
Cryogenic Radiometers, in BIPM Report (2000).
[100] Ångström K., Nova Acta Soc. Sci. Upsal. Ser. 3, 16 (1893) 1.
[101] Kurlbaum F., Ann. Phys. (Leipzig), 287 (1894) 591.
[102] Theocharous E. et al., The partial space qualification of a vertically aligned carbon
nanotube coating on aluminium substrates for EO applications, Opt. Express, 22 (2014)
7290.
[103] Fehlmann A. et al., Fourth World Radiometric Reference to SI radiometric scale
comparison and implications for on-orbit measurements of the total solar irradiance,
Metrologia, 49 (2012) S34.
DOI 10.3254/978-1-61499-818-1-73
Amount of substance – the Avogadro constant

and the SI unit “mole”
Bernd Güttler(∗ ), Detlef Schiel, Axel Pramann and Olaf Rienitz
Physikalisch-Technische Bundesanstalt (PTB)
Bundesallee 100, 38116 Braunschweig, Germany
Summary. — The physical quantity “amount of substance” is measured in the unit

“mole” (symbol: mol). In and because of its current definition (via 12 g of 12 C) it is
closely related to another SI base unit, namely the mass unit kilogram. In principle
there is an important difference between mass and amount of substance, because in
case of the amount of substance not only the amount of material is relevant, but also
the unambiguous specification of the entities forming this amount of material. The
practical realization and dissemination of the unit mole is usually based on materials
characterized with regard to their purity. The amount of substance is then derived
from the mass, the molar mass, and said purity. In the wake of the redefinition of
“The International System of Units” (SI) aiming at a set of defining fundamental

constants, also the definition of the “mole” will be rewritten. In its new definition
it will be based on a fundamental constant, namely the Avogadro constant. This
way also the link to the definition of the kilogram will be broken, underpinning the
importance of the “mole” as a base unit in its own right.
(∗ ) E-mail: bernd.guettler@ptb.de

73
74 Bernd Güttler, Detlef Schiel, Axel Pramann and Olaf Rienitz
1. – Introduction
The physical quantity amount of substance is measured in the unit mole (symbol:
mol). It is one out of a total of seven units of the International System of Units (SI).
Although the kilogram and the mole are defined as independent base units, they are
closely related to each other. 1 mol of the carbon isotope 12 C has, by definition, a mass
of exactly 12 g. Usually, the instrument used to determine the mass of an object is
the balance; it is, however, also used to determine the amount of substance. The mass
divided by the molar mass yields the amount of substance expressed in mol.
The molar mass has the unit g/mol. The amount of substance differs from the mass
in particular by the fact that identical amounts of different substances contain the same
number of atoms, molecules or other particles, but do usually not have the same mass.
The Avogadro constant NA expresses the number of atoms, molecules or other parti-
cles in one mole (NA = 6.02214076(12) × 1023 mol−1 ) [1]. When ma (X) is the mass of an
atom, a molecule or of another particle X and M (X) is the molar mass of this particle,
the Avogadro constant NA is obtained as follows:
M (X)
(1.1) NA = .
ma (X)
In 2018, the SI will be completely revised by defining a set of seven fundamental

constants with fixed values and no associated uncertainties at all, among them NA . This
in turn triggers the redefinition of the unit mole. The XRCD (X-ray crystal density)
experiment for the determination of the Avogadro constant with the aid of a silicon
single crystal contributes substantially to the redefinition of the kilogram, but also of the
mole.
2. – History
The concepts “mole” and “Avogadro constant” are based on the interpretation of
chemical processes as connections between atoms and molecules. In 1808, John Dalton
(1766-1844) declared that atoms of an element do not differ from one another and that
they have a defined atomic mass. In 1811, Avogadro (1776-1856) said that the same vol-
umes of all gases contain — at the same temperature and the same pressure — the same
number of molecules. In the first instance, his observation was forgotten until Stanislao
Cannizarro revived it again and published, in 1858, a consistent system of chemical for-
mulas and “atomic weights” (relative atomic masses, see below) of all elements. After
that, the concepts of atomic and molecular weight as well as other concepts borrowed
from atomic theory developed in chemistry [2].
The origin of the concept “mole” is assigned to Wilhelm Ostwald [3]. In his “Manual
and Auxiliary Book for the Performance of Physiochemical Measurements” of 1893, he
wrote: “Let us generally refer to the weight in grams — which is numerically identical to
the molecular weight of a specified substance — as one mole, . . . ” [4]. Similar concepts
such as, for example, “g-molecle” or “g-mole” with a comparable meaning were, however,
Amount of substance – the Avogadro constant and the SI unit “mole” 75
at the same time also used by others [5]. According to this definition, the unit “mole”
was, therefore, closely related to the mass and for a long time it was interpreted as
“chemical mass unit”. Although the atomic perception — which links the mole up with
a particle number and, therefore, requires the introduction of an additional base quan-
tity, the “amount of substance” — had existed since Dalton and Avogadro, experimental
results, which could confirm these models, were first of all missing [5]. The experimen-
tal confirmation of the atomic theory and the determination of the Avogadro constant
(particle number per mole) finally led to two different perceptions of the mole, which
Stille [6] differentiated by the concepts “mole” (as chemical mass unit) and “mole num-
ber” (as a unit related to a particle number which is defined by the Avogadro constant).
The integration of the unit “mole” into the SI system of units solved this contradiction
and made a differentiation of the concepts superfluous. It took place much later, in
October 1971, after the 14th General Conference of the Metre Convention had decided
to introduce the “amount of substance” as the 7th base quantity. The English concept
“amount of substance” was derived from the German concept “Stoffmenge”, according to
Stille [7, 8]. First, there had been a corresponding recommendation of the International
Union of Pure and Applied Physics (IUPAP), of the International Union of Pure and
Applied Chemistry (IUPAC) and of the International Organization for Standardization
(ISO), together with the note that the carbon isotope 12 C had to be selected as the ref-
erence point [9]. To handle questions related to the base quantity “amount of substance”
a committee of the Metre Convention (Comité Consultatif pour la Quantité de Matière
- CCQM) was founded in 1993 [10].
3. – The mole as an SI unit in chemistry
With the aid of the mole, results of measurements in chemistry can be traced back
to SI units and therefore internationally compared. To establish this traceability chain,
a growing number of national standards in chemistry have been developed on the basis
of the mole.
Actual chemical reactions usually involve extremely large numbers of atoms and
molecules. The unit “mole” therefore combines — for practical and historical reasons —
so many particles that reference to other units (e.g. the kilogram) can be made using
reasonably small numbers.

The definition of the SI unit “mole” is [9]:
1. The mole is the amount of substance of a system which contains as many elementary
entities as there are atoms in 0,012 kilogram of carbon 12; its symbol is “mol”.
2. When the mole is used, the elementary entities must be specified and may be atoms,
molecules, ions, electrons, other particles, or specified groups of such particles.
Unlike Ostwald, whose definition related only to the mass, reference is now made to
a particle number. The number of particles corresponds to the numerical value of NA
which is today determined with smallest uncertainty within the scope of the so-called
Avogadro Project.
In practice, traceability of the amount of substance n(X) of a substance X is realized

via its mass m and molar mass M (X):
m
(3.1) n(X) = .
M (X)
The molar mass M (E) of an element E is calculated from its mean, relative atomic mass
Ar and the molar mass constant Mu (from the definition of the mole it follows that Mu is
exactly 1 g/mol). All relative atomic masses use 12 C as the reference point: by definition:
Ar (12 C) = 12 and M (12 C) = 12 g/mol. The mean Ar of an element is calculated from
the Ar,i of its isotopes i E and their respective amount fractions x(i E) [9, 11]:
(3.2) M (E) = Ar (E) × Mu ,

(3.3) Ar (E) = x(i E) × Ar (i E) .
i
Traceability to the mole without reference to the mass via a determination of the
particle number is, however, also imaginable, once the Avogadro constant is fixed as
it is planned in the revision of the SI system. Then, counting elementary particles
may also be based on other properties of a particle than the mass such as its charge
or its optical or magnetic properties. They may be related to observable transitions
between nuclear or electronic energy levels that can be specific for a particular molecular
entity and also correlate with the particle number. Concepts for measurements of the
number of elementary entities have been available for a rather long time, although this
had been limited to special cases. The measurement can, for example, be performed by
determining the amount of substance of a crystalline solid with the aid of its microscopic,
crystallographic lattice parameter and its macroscopic volume. In principle, this is also
done within the scope of the Avogadro Project: a high-purity 28 Si single crystal serves
to perform these measurements. Similar — although much simpler experiments — have
been carried out since the discovery of X-ray diffraction. There are also experiments for
the (direct or indirect) counting of elementary entities, e.g. of ions [12] and electrons
(single-electron-tunnelling - SET circuits) which could become relevant, for example,
within the scope of a quantum-metrological redefinition of the SI base unit ampere [13-15].
Unlike the other base units the mole is, thus, defined by two theorems. The first
defines a number. The second requires the identification of the elementary entities and
establishes the connection to chemistry. The complete description of the determination of
the “amount of substance” of a substance in the unit mole thus requires the identification
and quantification of the elementary entities (analyte). An amount of substance of 1 mol
always contains the same number of entities. This number is identical with the numerical
value of NA .
4. – Realization and dissemination

.
4 1. Primary standards as reference points. – Traceability to an SI unit is achieved
by the practical realization of the unit in the form of primary standards with a specified
quantity value and an associated measurement uncertainty [16]. Primary standards are
standards which have been established with the aid of a primary measurement proce-
dure [16]. Primary measurement procedures are the first link in the dissemination chain
from the SI to the end user [17].
In case of the quantity amount of substance (the unit mole), a primary standard re-
quires a measurement procedure which enables both the identification and quantification
of the measurand. Therefore, the amount of substance related quantities need to identify
explicitly the kind of entities described: for example the substance concentration c(A) of
an analyte A in a matrix
n(A)
(4.1) c(A) = ,
V
or of its amount-of-substance fraction x(A)
n(A)
(4.2) x(A) = .
ntotal
Usually, the quantification of the number of elementary entities is unrealistic. This is

why it is performed by comparison with a pure substance of analyte A, whose amount
of substance n(A) is known from its mass m(A) of the substance and its purity wpur (A):
m(A)
(4.3) n(A) = wpur (A) × .
M (A)
.
4 2. Example: National standards for the determination of element concentrations in
solutions. – Safeguarding the correctness of measurements of the element concentration
in solutions is one of the most frequently performed metrological tasks in chemistry. For
this purpose, PTB provides — in cooperation with the Federal Institute for Materials
Research and Testing (Bundesanstalt für Materialforschung und -prüfung (BAM), Ger-
many) — national standards in the form of monoelemental solutions in aqueous acids
as reference points. Here, the practical link-up to the mole is realized with the aid of
“pure materials” (metals and salts). The purity of these materials specifies the fraction
of the main component (A) of the material, e.g. the fraction of Cu in a Cu material.
Equation (4.3) represents already the link-up to the SI. The uncertainty of the link-up,
i.e. the uncertainty of the amount of substance n(A), is determined by the uncertainties
of the purity, of the mass, and of the molar mass. In contrast to that, the absolute
value of the purity is, in principle, irrelevant to the link-up, although smaller uncertain-
ties associated with the purity may be achieved for high-purity materials. The purity is
calculated from the sum of all possible impurities. Its uncertainty should preferably be
smaller than 5 × 10−5 g/g and depends on the determination of the impurities (metals
as well as non-metals) [18].
The dissemination of these pure substances serving as the national standards is real-
ized via aqueous solutions prepared gravimetrically. The element content in the primary
sample X
sample solution wX(E)
routine measurement procedure
secondary ﬆandard Y
induﬆrial solution, Urel ≤ 0.3% wY(E)
certification measurement procedure
transfer ﬆandard T
transfer solution, Urel ≤ 0.1% wT(E)
precision measurement procedure
primary ﬆandard
primary solution S, Urel ≤ 0.05% wS(E)
gravimetric preparation
pure material, Urel ≤ 0.01% wpur
purity determination
SI
mol NA
Fig. 1. – Dissemination/traceability scheme. Blue downward arrows show the traceability path-
way, while black arrows denote the dissemination. The mass fraction wX of the element E in
a routine sample X is traceable through a routine measurement procedure to the commercially
available secondary standard Y, which in turn, is traceable to the transfer standard T. This
transfer standard is linked via a so-called precision measurement procedure to the primary so-
lution S [19]. The final link to the SI unit “mol” is realized via the purity determination of the
material used to prepare the primary solution S. For more details see text.
solution is determined on the basis of the masses of the pure substance and of the final
solution. In addition, secondary solutions (transfer solutions) are linked up with the
primary solutions with the aid of precision measurement procedures and disseminated
to calibration laboratories. Along the dissemination chain the relative expanded un-
certainty of the element content becomes larger from ≤ 10−4 g/g (primary solution) to
≤ 2 × 10−3 g/g (transfer solution). This dissemination system based on monoelemental
calibration solutions is shown in fig. 1, allows the test laboratories to obtain traceable
and — thus — also comparable measurement results.
When real samples are measured in testing laboratories, matrix effects may occur.
They cause measurement biases although traceability had been established on the basis
of mono-elemental calibration solutions. To overcome this problem, suitable traceable
matrix reference materials must be applied.
.
4 3. Examples of primary measurement procedures. – Different primary measurement
procedures to establish traceability to the mole have been discussed [17, 20].
Traceability is realized with the aid of the so-called “primary ratio measurement
procedures” [16, 17] such as, for example, isotope dilution mass spectrometry (IDMS).
The basic prinicple is the gravimetric addition of the analyte with a non-natural isotopic
pattern (so-called spike) to the sample and the measurement of the resulting isotope
ratio(s) in this blend of sample and spike. From the ratios in this blend and in the
original sample and spike, as well as the spike content and the masses of sample and
spike forming the blend, the analyte content in the original sample is calculated [21].
Since the spike content needs to be calibrated against an amount-of-substance standard,
IDMS is a primary ratio procedure, because it links the result to a quantity of the same
kind.
Primary measurement procedures which work in accordance with this principle allow
relative expanded uncertainties in the range from 0.1% to 3% (k = 2) to be achieved,
e.g. in trace analysis.
As examples for “primary direct measurement procedures”, gravimetry and coulome-
try allow an amount of substance to be directly quantified by weighing or by measuring
the electric charge [17,20] without a link of the result to a quantity of the same kind. The
preparation of a primary solution as described above is an example of gravimetry. In the
case of coulometry, the amount of substance n(A) is determined in an electro-chemical
process via the measurement of a current I over the time t (z: valence, F : Faraday
constant)

1
(4.4) nA = Idt.
z×F
.
4 4. International comparability. – Beyond a realization of the mole, international
comparability of measurement results is required. Comparability over many decades is a
prerequisite e.g. in environmental protection, in climate control, and clinical chemistry.
In the case of the SI unit “mole”, complete traceability would require primary standards
at least for all distinguishable chemical compounds, i.e. ultimately for the whole variety of
natural and synthetic substances. These substances may be concentrations of pollutants
in air or water or clinical markers in the blood serum, but also — as described above
— high-precision solutions of elements. The definition of analyte and matrix is, thus, an
indispensable part of every standard used to establish traceability to the SI unit mole.
The foundation of the “Consultative Committee for Amount of Substance: Metrol-
ogy in Chemistry and Biology” CCQM by the CIPM in 1993 initiated the establishment
of international traceability structures in chemistry [22]. Other organizations, such as
Eurachem (European network of organizations having the objective of establishing trace-
ability in chemical analysis) and CITAC (Cooperation on International Traceability in
Analytical Chemistry) and the regional metrology organizations such as EURAMET (Eu-
ropean Association of National Metrology Institutes) are involved in this lasting process.
With the MRA of 1999 for the mutual recognition of national standards and certifi-
cates (http://www.bipm.org/en/cipm-mra/), a global reference system has been estab-
lished which is, for the first time, also valid for chemical measurements. In chemistry, this
international recognition is based on the international key comparisons of the CCQM,
with the aid of which the degree of comparability of the standards can be quantified. As
in other fields of metrology, the national standards which have been assessed and exam-
ined in this way enter into the CMC database which is made available by the BIPM and
is accessible to the public (http://kcdb.bipm.org/appendixC/). At present, approx.
6150 national standards for chemical measurands are recorded there.
5. – Redefinitions
.
5 1. General . – The kilogram is the only one of the seven base units of the SI that is
defined by a material embodiment as the mass of the international kilogram prototype.
The other base units are defined by reference to a fundamental constant of physics
and/or by an experimental procedure [9]. Some units also depend on other base units.
The metre, which is defined as the path travelled by light in vacuum in a specific fraction
of a second is, for example, defined by a specified value of the speed of light. In that
definition, the second is referred to as a unit of the time. The definition of the ampere
describes an idealized arrangement of two electrical conductors and states for it the values
of measurands in the units kilogram, metre and second. These values in addition define
the magnetic constant. According to the present state of knowledge, it is assumed for
such definitions that the unit is invariable although its practical realization is always
affected by a certain uncertainty. Also, definitions of this kind enable the unit to be
realized at any location and at any time.
It has also been found that the masses of the national prototypes of the kilogram have
significantly drifted during the last hundred years with respect to that of the international
prototype [23, 24]. On average, they have put on about 0.05 mg. As the international
prototype does not basically differ from the other prototypes as regards its physical
properties, it is rather probable that it has lost weight compared, for example, to an
atomic mass. Also, all prototypes might be subject to an additional drift which is still
unknown.
Therefore, for more than 30 years the attempt has been made to define also the
kilogram via an atomic constant or a fundamental constant of physics. The redefinition
of the kilogram is becoming more and more probable and scheduled for 2018. It will be
based on the Planck constant h.
In addition to a redefinition of the kilogram, it is planned to redefine also the ampere,
the kelvin and the mole. According to the current definition of the mole, the mole is
linked directly to the kilogram definition via the mass of 12 C [9]. To break this link is
the most important reason to redefine the mole. Another reason is the general demand
to base all unit definitions solely on fundamental constants [25]. For the mole, the
definition valid so far shall be reworded in such a way that it is based on a fixed value
of the Avogadro constant NA without relating to the unit “kilogram” as it is the case in
the current definition. The most recent drafts reads [26, 27]:
The mole, symbol mol, is the SI unit of amount of substance of a specified elementary
entity, which may be an atom, molecule, ion, electron, any other particle or a specified
group of such particles. It is defined by taking the fixed numerical value of the Avogadro
constant NA to be 6.022 140 857 × 1023 when expressed in the unit mol−1 .
.
5 2. Silicon single crystal and the Avogadro constant. – The Avogadro Project has its
origins in the 1970s, when German researchers started for the first time in determining the
lattice distances in a silicon crystal by X-ray interferometry. This allowed the kilogram
to be linked up with the atomic mass unit. As a connecting link between a macroscopic
mass and the mass of an atom acts the Avogadro constant which indicates the number
of atoms in a mole.
The NA is named after the Italian earl and advocate Amedeo Avogadro who — at the
beginning of the 19th century — dealt with atomism which had been explained by Dalton
and who brought in line all observations connected with it. Dalton had demonstrated
that all gases expand in the same way when they are heated. This, concluded Avogadro,
could be explained only by the fact that the number of gas particles involved is also
always the same: Identical volumes of all gases had to contain — in the case of identical
external conditions — the same number of smallest particles. But it seems that Avogadro
did not really know how to determine this number.
Only in 1865, Josef Loschmidt calculated the particle number per volume with the
aid of the mean free path of gas molecules which was determined by Oskar Meyer and,
later, by James Maxwell. At that time, this value deviated by only 20% from the value
recognized today. Forty years later, Albert Einstein tried to indicate the Avogadro con-
stant more precisely: In his doctoral thesis of 1905, in which he determined the number
of molecules per mole on the basis of a novel diffusion equation from measurements on a
sugar solution, his values deviated — due to a calculation error — by a factor of three.
In 1917, Robert Millikan reduced the error to less than one percent by clearly im-
proving his famous oil droplet test for determination of the elementary charge. Today,
this accuracy is no longer sufficient. Managed by PTB, scientists from several metrology
institutes have joined forces to determine — with the aid of a nearly perfect silicon single
crystal — the Avogadro constant as exactly as never before [1].
The achievement of this objective required, however, many years of development. At
the beginning, silicon crystals of natural isotopic composition were used for the determi-
nation of NA . At several metrology institutes such as, for example, PTB, the National
Institute of Standards and Technology (NIST, USA), the National Physical Laboratory
(NPL, Great Britain), the National Measurement Institute of Japan (NMIJ) and the
Istituto Nazionale di Ricerca (INRIM, Italy), measurements of NA were started at the
end of the 1970s with the establishment of an X-ray interferometer for measuring the
lattice distance in the silicon single crystal. The Institute for Reference Materials and
Measurements (IRMM, Belgium) participated with the measurement of the amount-of-
substance fractions of the three isotopes 28 Si, 29 Si und 30 Si in the silicon. First suc-
cess was achieved at NIST [28] with a relative measurement uncertainty in the range
of 10−6 associated with NA . Despite the increased efforts made at all institutes in-
volved, it has — in the coming years — been impossible to achieve a measurement
uncertainty of less than 3 × 10−7 . Success was reached only when some years ago,
the newly established International Avogadro Coordination (IAC) and other institutions
and companies repeated the measurements with highly enriched 28 Si applying the XRCD
method [29].
Fig. 2. – Principle set-up of the diameter (volume) measurement by the PTB spherical interfer-
ometer. The measurements of d1 and d2 and that of D are performed alternately. By rotating the
sphere, the diameter topography of the entire sphere surface is available. Source: A. Nicolaus,
PTB, Germany.
To determine NA , the volume V and the mass m of the highly enriched silicon single
crystal sphere (fig. 2) are measured:
N × M (Si) × Vsphere
(5.1) NA = .
m × vcell
Additional samples of the same crystal serve to determine the molar mass M and
the volume vcell of the unit cell. N is the number of atoms in the unit cell (in case of
Si: vcell = 8). All quantities must, of course, be measured traceable to the SI units.
The volume of the silicon sphere is determined by interferometric measurements of the
sphere diameter. This provides a complete surface topography of the sphere. The spheres
exhibit a deviation from a perfect spherical shape of only a few 10 nm which in turn is
required to achieve the necessary uncertainty associated with Vsphere . The silicon sphere
is coated with an oxide layer, some nm in thickness, which geometry and mass must be
determined. On top of this layer additional adsorbed layers of water and hydrocarbons
are present and have to be characterized. The mass of the silicon sphere is linked up with
the kilogram prototype by weighing. Due to the large density differences between silicon
and platinum-iridium, the accuracy requirements for the mass determination can, at
present, be achieved only under vacuum conditions, i.e. without air buoyancy correction.
Volume and mass determinations must — in addition — be corrected with respect to
the coating layers. The volume of the unit cell is determined from the lattice distance
of the crystal with the aid of an X-ray interferometer. A sufficient knowledge of the
Fig. 3. – Single crystal consisting of silicon highly enriched in 28 Si of approximately 4.7 kg.
Source: Leibniz-Institut for Crystal Growth (IKZ), Berlin, Germany.
impurities and of the crystal defects is also indispensable. The molar mass is determined
by mass spectrometry from the atomic masses of the three isotopes 28 Si, 29 Si and 30 Si
and from their amount-of-substance fractions. The atomic masses are linked up with the
mass of the carbon isotope 12 C by means of penning traps. Using natural silicon, the
measurement uncertainty associated with the molar mass could not have been improved
any further. This is why the investigations for the determination of NA were stopped
some years ago when a measurement uncertainty of approx. 3 × 10−7 was reached [30].
In addition to insufficient uncertainty for the redefinition of the kilogram, the observed
deviation of the measurement value of approx. 1 × 10−6 relative to other fundamental
constants, also caused quite a headache to the researchers. Has the deviation been due
to miscounting or are there unrecognized inconsistencies in the system of the physical
constants? The possibility of using enriched 28 Si on a large scale for sample preparation
gave new momentum to the Avogadro Project. Estimates showed that the uncertain-
ties associated with the isotopic composition would be considerably reduced, but the
researchers were forced to repeat also all other investigations on this material. In 2003,
in a cooperation with research institutes in Russia, several national and international
Fig. 4. – The principle of the so-called “Virtual Element Isotope Dilution Mass Spectrometry”
(VE-IDMS) is based on ignoring the 28 Si signal completely and mixing the original 28 Si-enriched
silicon x with a 30 Si-enriched material y in a way to adjust a 30 Si/29 Si ratio in the blend bx
close to one. From the measurement of this ratio, the mass fraction w(VE) of the VE can
be calculated. This mass fraction yields the amount-of-substance fractions x(i Si) of all three
isotopes and as the final result the molar mass of silicon in the original sample. Source: A.
Pramann, O. Rienitz, PTB, Germany.
metrology institutes therefore launched the ambitious plan to prepare approx. 5 kg of

silicon highly enriched with respect to 28 Si (x(28 Si) > 0.9999 mol/mol). With this single
crystal material a measurement uncertainty of approx. 2 × 10−8 by the year 2010 was
targeted.
In 2008, the material was successfully prepared with the growing of a perfect single
crystal at the Institute for Crystal Growth (IKZ) in Berlin (fig. 3) and two 1 kg 28 Si
spheres could be polished at the Australian ACPO.
Despite the theoretical prediction the targeted uncertainty associated with NA could
not be achieved, because again the molar mass determination was the limiting quan-
tity due to problems with the chemical blank caused by ubiquitous natural silicon and
the much larger dynamic range necessary to do the mass spectrometric measurements.
Switching from gas mass spetrometry to inductively coupled plasma mass spectrome-
try and the development of the so-called “virtual element isotope dilution mass spec-
trometry” (VE-IDMS) (fig. 4) solved these new problems inherent to the enriched sili-
con [31-34].
The first results using the enriched material have shown that a more exact value of NA
with an uncertainty of 3×10−8 could be obtained and the cause of the observed deviation
with respect to other fundamental constants be detected [29]. Meanwhile (2015), the
uncertainties associated to almost all quantities contributing to the determination of NA
(eq. (5.1)) were substantially improved [35] resulting in a relative uncertainty associated
6.0221420
6.0221418
6.0221416
NA / (1023 mol-1)
6.0221414
6.0221412
6.0221410
6.0221408
6.0221406
10 01
4 14 14 01
6
20 32 20 20 42
C - C C -
IA ST NR IA ST
NI NI
Fig. 5. – Compilation of the most accurate and recent values of the Avogadro constant. Source:
H. Bettin, PTB, Germany.
with NA of less than 2 × 10−8 (fig. 5) [1, 35]. Most notably the relative uncertainty
associated with the molar mass was reduced during the last decade by nearly three
orders of magnitude down to less than 1 × 10−9 [34], shifting its relative contribution to
the uncertainty of NA from 60% (ten years ago) to a mere 6% today.
.
5 3. Realization and dissemination in accordance with the redefinitions. – Also after
the planned redefinition of the mole by fixing the numerical value of NA , the amount of
substance will only in rare cases be determined by the counting of atoms or molecules.
This means that ratios of amounts of substances will also be measured in future with
mass spectrometers. Weighing of the reference material will remain the method for the
determination of the quantitative reference point. Here, the question arises whether —
or how — the molar masses will change after a redefinition.
If the mole will be defined in future by a fixed value of the Avogadro constant NA , and
the kilogram by a value of the Planck constant h — as currently planned — this could
lead to a redundant dimensioning as NA and h depend on each other via other constants
forming the so-called “molar Planck constant” NA h (c speed of light in a vacuum, Ar (e)
relative mass of the electron in u, Mu = 10−3 kg mol−1 molar mass constant, α fine-
structure constant, R∞ Rydberg constant) [36]:
c × Ar (e) × Mu × α2
(5.2) NA × h = .
2R∞
Then, in eq. (5.2), c, h and NA are defined in the definitions of the metre, the kilogram
and the mole, whereas α and R∞ remain independent measurands. This means, only
Ar (e) or Mu have to be considered as variables which depend on these measurands. If
the value of Ar (e) changed as a result of new measurement results of α and R∞ and a
corresponding adjustment, all relative atomic masses would change as well. In particular
Ar (12 C) = 12 would no longer be exactly valid as a reference. As the relative atomic
masses are measured with uncertainties which are comparable to that of α, future changes
would be significant and require the compilation of new tables of the relative atomic
masses. This is why the proposal to introduce Mu as a variable [37] seems to be more
useful and would, in addition, have negligible impacts. As the molar mass of an element
X is calculated according to M (X) = Ar (X)Mu , the molar masses would change with Mu .
From the moment of the redefinition until the availability of new values of α or R∞ ,
Mu = 10−3 kg mol−1 will be valid. After the redefinition, the relative uncertainty of Mu
and, correspondingly, that of the molar mass of 12 C, will depend predominantly on the
measurement uncertainty of α2 which is at present in the order of 10−9 , future relative
changes will probably not exceed this value. In view of the smallest relative uncertainties
which are achieved in the realization of the mole and the molar masses (approx. 10−5 ),
no changes will result in practical applications.
6. – Summary
For the determination of the amount of substance, weighing plays a central role.
Whereas the mass is independent of the kind of the substance, the amount of substance
always relates to a specific, uniform substance. It is proportional to the number of exactly
specified entities (like atoms or molecules), which form the specific amount of substance.
After a short historical survey, the realization of the SI unit “mole” was explained in
accordance with its definition. The mole is realized for a variety of substances in differ-
ent NMIs via primary standards. After the explanations regarding the realization, the
dissemination of the mole via secondary standards in calibration or testing laboratories
to the point of the user of these units was described. Finally, a survey was given of
the planned redefinition of the unit, of the relevant experiments and of the possible con-
sequences for the realization and dissemination. Moreover, problems were highlighted
which remain to be solved.
REFERENCES
[1] Azuma Y., Barat P., Bartl G., Bettin H., Borys M., Busch I., Cibik L.,
D’Agostino G., Fujii K., Fujimoto H., Hioki A., Krumrey M., Kuetgens U.,
Kuramoto N., Mana G., Massa E., Meeß R., Mizushima S., Narukawa T.,
Nicolaus A., Pramann A., Rabb S. A., Rienitz O., Sasso C., Stock M., Vocke
Jr. R. D., Waseda A., Wundrack S. and Zakel S., Metrologia, 52 (2015) 360.
[2] Becker P., Rep. Prog. Phys., 64 (2001) 1945.
[3] Sommerfeld A., Atombau und Spektrallinien (Harri Deutsch Verlag) 1929 (reproduction
1978).
[4] Ostwald W., Hand- und Hilfsbuch zur Ausführung Physiko-Chemischer Messungen (W.
Engelmann Verlag, Leipzig) 1893.
[5] Nernst W., Theoretische Chemie vom Standpunkte der Avogadro’schen Regel und der
Thermodynamik (Enke Verlag, Stuttgart) 1893.
[6] Stille U., Messen und Rechnen in der Physik (Friedrich Vieweg Verlag, Braunschweig)
1955.
[7] Guggenheim E. A., The J. Chem. Ed., 38 (1961) 86.
[8] Milton M. J. T., Philos. Trans. R. Soc. A, 369 (2011) 3993.
[9] Bureau international des poids et mesures (BIPM), Le Système international d’unités (SI)
– The International System of Units (SI) 8th edition, (BIPM, Sèvres/Paris) 2006.
[10] Bureau international des poids et mesures (BIPM), Procès-verbaux des séances du Comité
International des Poids et Mesures, 82nd meeting (1993), (BIPM, Sèvres/Paris) 1993.
[11] De Laeter J. R. et al., Pure Appl. Chem., 76 (2003) 683.
[12] Schlegel C., Gläser M., Scholz F., Mecke M., Kiekenap G., Thomsen-Schmidt
P., Hoffmann K. P. and Ahbe T., Metrologia, 47 (2010) 146.
[13] Bylander J., Duty P. and Delsing P., Nature, 434 (2005) 361.
[14] Zimmermann N. L. and Keller M. W., Meas. Sci. Technol., 13 (2003) 1237.
[15] Ebbecke J., Bastian G., Blöcker M., Pierz K. and Ahlers F. J., Appl. Phys. Lett.,
77 (2000) 2601.
[16] BIPM, JCGM 200 - International Vocabulary of Metrology - Basic and general concepts and
associated terms (VIM), 3rd edition (Joint Committee for Guides in Metrology - JCGM,
Paris) 2012.
[17] Milton M. J. T. and Quinn T. J., Metrologia, 38 (2001) 289.
[18] Matschat R., Kipphardt H., Rienitz O., Schiel D., Gernand W. and Oeter D.,
Accredit. Qual. Assur., 10 (2006) 633.
[19] Kaltenbach A., Noordmann J., Görlitz V., Pape C., Richter S., Kipphardt H.,
Kopp G., Jährling R., Rienitz O. and Güttler B., Anal. Bioanal. Chem., 407 (2015)
3093.
[20] Richter W., Accredit. Qual. Assur., 2 (1997) 354.
[21] Sargent M., Harrington C. and Harte R., Guidelines for achieving high accuracy in
isotope dilution mass spectrometry (IDMS) (RSC, Cambridge) 2002.
[22] Kaarls R., The international infrastructure in metrology in chemistry, PTB-Mitteil, 115
(2005) 283.
[23] Quinn T., IEEE Trans. Instrum. Meas., 40 (1991) 81.
[24] Girard G., Metrologia, 31 (1994) 317.
[25] Fischer J. and Ullrich J., Nat. Phys., 12 (2016) 4.
[26] CCQM/16-04, Draft mise en pratique for the definition of the mole, BIPM,
http://www.bipm.org/en/measurement-units/new-si/.
[27] BIPM, DRAFT 9th edition of the SI Brochure, http://www.bipm.org/en/measurement-
units/new-si/.
[28] Deslattes R. D., Henins A., Bowman H. A., Schoonover R. M., Carroll C. L.,
Barnes I. L., Machlan L. A., Moore L. J. and Shields W. R., Phys. Rev. Lett., 33
(1974) 463.
[29] Andreas B., Azuma Y., Bartl G., Becker P., Bettin H., Borys M., Busch I.,
Gray M., Fuchs P., Fujii K., Fujimoto H., Kessler E., Krumrey M., Kuetgens
U., Kuramoto N., Mana G., Manson P., Massa E., Mizushima S., Nicolaus A.,
Picard A., Pramann A., Rienitz O., Schiel D., Valkiers S. and Waseda A., Phys.
Rev. Lett., 106 (2011) 030801.
[30] Fujii K., Waseda A., Kuramoto N., Mizushima S., Becker P., Bettin H., Nicolaus
A., Kuetgens U., Valkiers S., Taylor P., De Biévre P., Mana G., Massa E., Matyi
R., Kessler E. G. and Hanke M., IEEE Trans. Instrum. Meas., 50 (2005) 854.
[31] Mana G. and Rienitz O., Int. J. Mass Spectrom., 291 (2010) 55.
[32] Rienitz O., Pramann A. and Schiel D., Int. J. Mass Spectrom., 289 (2010) 47.
[33] Pramann A., Rienitz O., Schiel D., Güttler B. and Valkiers S., Int. J. Mass
Spectrom., 305 (2011) 58.
[34] Pramann A. and Rienitz O., Anal. Chem., 88 (2016) 5963.
[35] Fujii K., Bettin H., Becker P., Massa E., Rienitz O., Pramann A., Nicolaus A.,
Kuramoto N., Busch I. and Borys M., Metrologia, 53 (2016) A19.
[36] Mohr P. J., Taylor B. N. and Newell D. B., J. Phys. Chem. Ref. Data, 37 (2008)
1187.
[37] Taylor B. N., Metrologia, 46 (2009) L16.
DOI 10.3254/978-1-61499-818-1-89
Comparisons of gas standards for climate change

and air quality monitoring
R. I. Wielgosz, J. Viallon, E. Flores, P. Moussay, F. Idrees
and T. Choteau
Pavillon de Breteuil, F-92312 Sèvres Cedex, France
Summary. — Reducing air pollution and preventing climate change are major
challenges for the future. The monitoring of atmospheric gas concentration and
emissions that contribute to these effects is vital in assessing the impact of policies to
reduce both air pollutant and greenhouse gas emissions. For this data to be reliable,
calibration of monitoring instrumentation is required with gas standards with values
traceable to Primary Reference Gas Mixtures. Such gas standards are produced by
National Metrology Institutes. The paper describes the level of uncertainty and
compatibility that can be achieved for a number of greenhouse gas and air quality
gas standards based on the methods used to establish their values: CH4 in air and
NO in nitrogen (prepared by static gravimetry); NO2 and HCHO (prepared by
dynamic measurements); O3 and impurities in NOx gases (by spectroscopy); CO2
and O3 cross sections (by manometric methods).
1. – Introduction to air quality and greenhouse gas monitoring
Air pollution damages our health and the environment. Health problems include
serious effects on the cardiovascular and respiratory systems leading to reduced lung
function, asthma, chronic bronchitis and premature deaths. Environmental damage in-
cludes acidification of soils and sources of water; eutrophication — an excess of nutrients,
such as nitrogen oxides and ammonia, in water or soil; and finally physical damage to
89
90 R. I. Wielgosz, J. Viallon, E. Flores, etc.
buildings and monuments, due to corrosion and soiling of their surfaces. This has resulted
in both national and regional regulations being developed around the world to protect
populations and the environment. There are many commonalities to the various legis-
lations many of which contain air-quality standards for ground level ozone, particulate
matter (PM), nitrogen oxides, dangerous heavy metals and a number of other pollutants,
in addition to emissions capping regulations covering sulphur dioxide, nitrogen oxides,
ammonia and volatile organic compounds (VOC). Air pollution ignores national borders
and can be carried very long distances by the wind, so it needs to be tackled through co-
operation at regional, international and global levels. Monitoring of pollutant gases serves
to warn populations about poor air quality and allows them to take action to limit their
exposure, as well as allowing governments to impose actions that will reduce emissions,
and monitor the effectiveness of any remedial actions on both the short and long term.
This in turn requires accurate air quality measurements at the local level throughout
national territories. For these measurements to be useful, they need to be comparable,
which requires calibration to accurate and SI traceable standards. Calibration standards
for this purpose are provided most often by specialty gas companies, which themselves
establish the traceability of the values of their standards by calibration with Primary
Reference Gas Mixtures provided by National Metrology Institutes (NMIs). In this pa-
per, we describe the methods used to compare and in some cases value assign standards
from NMIs, to ensure their equivalence within their stated measurement uncertainties.
Since pre-industrial times (before 1750), atmospheric abundances of Long-Lived
Greenhouse Gases (LLGHGs) have risen considerably due to emissions related to hu-
man activity. Growing population, intensified agricultural practices, increase in land use
and deforestation, industrialization and associated energy use from fossil sources have all
led to the increased growth rate of LLGHG abundances in recent years. It is extremely
likely that these increases have led to the global warming that the world is now experienc-
ing [1]. According to the World Meteorological Organization (WMO), in spring 2015 the
global average mole fraction of carbon dioxide (CO2 ) crossed the 400 μmol mol−1 thresh-
hold [2]. Carbon dioxide is the single most important anthropogenic greenhouse gas in
the atmosphere, contributing ∼ 65% to radiative forcing by LLGHGs. It is responsible
for ∼ 81% of the increase in radiative forcing over the past decade and ∼ 82% over the
past five years. Methane contributes ∼ 17% to radiative forcing by LLGHGs. Approxi-
mately 40% of methane is emitted into the atmosphere by natural sources (e.g. wetlands
and termites), and about 60% comes from anthropogenic sources (e.g. ruminants, rice
agriculture, fossil fuel exploitation, landfills and biomass burning). Atmospheric CH4
reached 256% of the pre-industrial level (∼ 722 ppb) due to increased emissions from
anthropogenic sources. Nitrous oxide contributes ∼ 6% to radiative forcing by LLGHGs.
It is the third most important individual contributor to the combined forcing. It is
emitted into the atmosphere from both natural (about 60%) and anthropogenic sources
(approximately 40%), including oceans, soils, biomass burning, fertilizer use and various
industrial processes.
Nations are setting policies to reduce greenhouse gas (GHG) emissions focused on
Intended Nationally Determined Contributions. Greenhouse gas inventory data provides
Comparisons of gas standards for climate change and air quality monitoring 91
one mechanism for estimating emissions and GHG concentration and flux measurements
are envisioned to provide independent verification of reported emissions. The WMO
Global Atmosphere Watch (WMO-GAW) programme coordinates background level mea-
surements of Greenhouse gases, whilst it is suggested that a much more concentrated net-
work of local monitoring sites in areas with emissions would be needed to verify fluxes of
GHGs. Both types of measurement site will require highly accurate measurements, and
the signature of the WMO of the CIPM-MRA in 2010 has enabled NMIs and WMO Cen-
tral Calibration laboratories to compare their standards for GHGs to ensure worldwide
comparability of these measurements.
2. – Methods for gas standard preparation and verification
The most widely used method for the preparation of primary standard gas mixtures
involves weighing the individual components into a cylinder. Provided the gases being
weighed are stable within the cylinder, the uncertainty in the composition of a standard
gas mixture arises predominantly from three sources: the weighing procedure itself, the
analysis of the purity of the source gases and knowledge of the relative molecular masses
(RMMs) of the components. The ultimate accuracy of gas mixtures prepared by gravime-
try is limited by the smallest mass of gas that can be weighed and added to the mixture
with an acceptable uncertainty. This physical constraint imposes a limit on the most
dilute mixture that can be prepared in a single weighing step. This limit has generally
been overcome by ‘cascading’ a series of dilution steps each carried out gravimetrically.
The general procedure for preparing gas standards by the static gravimetric method is
described in ISO 6142-1:2015 [3] and their verification in ISO 6143:2001 [4]. A review of
static gravimetric methods of gas mixtures has also recently been published [5]. However,
as the reactivity of the component increases, the difficulty of producing accurate stan-
dards by this method increases, and alternative methods are employed, notably dynamic
methods. In certain cases, either because of the reactivity of the gas or the need to have
a reference method for analysis rather than preparation of the standard, either spectro-
scopic or manometric methods are used to provide a reference measurement procedure for
the gas of interest. In this paper, we describe the level of uncertainty and compatibility
that can be achieved for a number of greenhouse gas and air quality gas standards based
on the methods used to establish their values: CH4 in air and NO in nitrogen (prepared
by static gravimetry); NO2 and HCHO (prepared by dynamic measurements); O3 and
impurities in NOx gases (by spectroscopy); CO2 and O3 cross sections (by manometric
methods).
3. – Standards produced by static gravimetric methods (CH4 and NO)
For stable gases, static gravimetric preparation is almost always the preferred choice
for the production of standards in order to achieve the smallest uncertainties (for CO2
standards an alternative analytical manometric procedure is feasible for value assignment
of a gas standard, which will be described in sect. 6). Two examples of the achievable
Table I. – Limits for balance gas composition in standards submitted for the comparison.
Based upon the possible biases that could be introduced into the spectroscopic comparison method
(CRDS) due to variation in the composition of the air matrix in different standards, participat-
ing laboratories were asked to ensure that the composition of their air matrix was within these
limits.
Component in air Minimum amount of substance Maximum amount of substance

fraction permitted within fraction permitted within
submitted cylinder submitted cylinder
Nitrogen 0.77849 mol/mol 0.78317 mol/mol
Oxygen 0.20776 mol/mol 0.21111 mol/mol
Argon 8.865 mmol/mol 9.799 mmol/mol
Carbon Dioxide 360 μmol/mol 400 μmol/mol
uncertainties for such standards are given below (for CH4 a greenhouse gas, and NO, im-
portant for calibration of NOx measurements for air quality), as well as the comparison of
independent realizations of these standards, which demonstrate that the component limit-
ing the uncertainty of the standards arises from purity of either the balance or analyte gas.
.
3 1. Methane in air standards. – In 2013, the CCQM-K82 comparison was designed
to evaluate the level of comparability of NMI preparative capabilities for gravimetric
methane in air primary reference mixtures in the range (1800–2200) nmol mol−1 . The
balance gas for the standards was either scrubbed dry real air or synthetic air. This
study involved a simultaneous comparison of a suite of gas standards with two prepared
by each of the eight participating laboratories. The standards were sent to the BIPM
where the comparison measurements were performed.
The reference value for a given gas standard was calculated from a calibration line
derived from a self-consistent subset of the standards. Measurements at the BIPM were
performed using two independent analytical methods gas chromatograph (GC-FID) and
cavity ring-down spectroscopy (CRDS), and following the advice of the CCQM Gas
Analysis Working Group, results from the CRDS method were used to calculate the key
comparison reference value. The effect of gas matrix composition on the response of gas
analyzers based on optical spectroscopy techniques has been well documented including
biases that can be caused by pressure broadening effects on lines that are measured [6].
In order to ensure that such effects are negligible in the comparison, the tolerances for
air matrix composition were strictly defined as in table I.
Participants in the comparison were required to submit two standards for analysis at
the BIPM, one in the amount of substance fraction range (1800 ± 20 nmol/mol) and the
second in the range (2200 ± 20 nmol/mol). The expanded uncertainties reported by the
participants for the standards they submitted are plotted in fig. 1.
Fig. 1. – Participants’ assigned CH4 expanded uncertainties U (xNMI ).
The CCQM-K82 comparison cylinders were measured by the Cavity Ring Down Spec-
trometer (CRDS) Picarro G1202 model using three different methods. For Method-2,
which was used to report the final results of the comparison, the measurand was the ra-
tio of the CRDS response to a control cylinder. The measurements by this method were
done over one month. Measured ratios were in the range 0.94 to 1.16 with calculated
standard uncertainties of the measured ratio of 0.00026.
The key comparison reference value (KCRV) for CCQM-K82 was calculated using
the measurement results of the CRDS method 2. The analysis of the data from the
comparison was done following the procedures outlined in ISO 6143:2001 [4] (Gas analy-
sis — Comparison methods for determining and checking the composition of calibration
gas mixtures). The regression analysis was performed with XLGenlinev1.1, a computer
programme developed by NPL which implements this methodology by taking into con-
sideration uncertainties in both axes. Standards that were to contribute to the KCRV
were selected by applying the regression analysis first to the entire set of cylinders, so as
to identify possible outliers, which then were not selected in the next set of data to be
analysed. This process led to a self-consistent set of cylinders comprising all cylinders
except one. The goodness-of-fit of the regression performed with this data set is equal
to 1.72, demonstrating consistency of the ensemble.
In order to observe the within-laboratory variability and the between-laboratory vari-
ability the graph of equivalence is represented by a Youden Plot (fig. 2). In a Youden
Plot, degrees of equivalence (D) for one standard of each participant are plotted versus
degrees of equivalence of the other standard. The Youden plot displayed in fig. 2 shows
Method CRDS 2
10
GLS _LCminSet
9
8
7
6
5
4
3
-1
D_2200 / nmol mol
2
1 NIST
VSLNMIJ
0
NIM VNIIM
-1
-2 NPL KRISS
-3
-4
-5
NOAA
-6
-7
-8
-9
-10
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
-1
D_1800/ nmol mol
Fig. 2. – Youden Plot of the CCQM-K82 results. The error bar represents the expanded uncer-
tainty at a 95% level of confidence.
the degrees of equivalence for the standards prepared with the higher CH4 amount of sub-
stance fraction (around 2200 nmol mol−1 ) versus the degree of equivalence for the stan-
dards prepared with the lower amount of substance fraction (around 1800 nmol mol−1 ).
The y = x line of the plot represents the line on which completely correlated pairs of
standards would lie.
For the majority of participants, which were to be positioned uniformly around the
y = x line of the plot, the comparison demonstrated a relatively low within-laboratory
variability, confirming that in most cases the standards within submitted pairs are corre-
lated, and a major reason is that the same balance gas is being used to prepare standards
within pairs. In most cases a major contributor to reported uncertainty of participants’
standards arose from the measurement of trace levels of methane in the balance gas used
to prepare their standards.
The standard deviation of the ensemble of standards about the KCRV value was
1.70 nmol/mol. This standard deviation can be compared to the data compatibility goal
set by the World Meteorological Organization — Global Atmospheric Watch for CH4 in
air which is ±2 nmol/mol (1 SD). Figure 3 compares the results of CCQM-K82 to com-
parisons performed ten years earlier. The level of agreement amongst standards improved
by a factor of ten compared to a similar comparison exercise performed in 2003 (CCQM-
P41) [7, 8]. The comparison of 2003, found the standard deviation of results around
CCQM-P41 part 1, 2003

150 coordinated by VSL
100 CCQM-P41 part 2,

2003, coordinated by VSL
D i / (nmol mol )
-1
50 CCQM-K82, 2013
coordinated by BIPM/NIST
NOAA
NOAA
VNIIM
VNIIM
KRISS
KRISS
NMIJ
NMIJ
-50
NIST
NIST
NIM
NIM
NPL
NPL
VSL
VSL
CSIRO-AR
NOAA
NPL
NPL
KRISS
KRISS
CENAM
NIST
NIST
NMi VSL
NMi VSL
NMIJ
NMIJ
IMGC
CEM
CEM
NMIA
NMIA
BAM
GUM
Fig. 3. – Degree of equivalence between CH4 gravimetric mole fractions during the last ten years.
The error bar represents the expanded uncertainty at a 95% level of confidence.
the reference value to be 30 nmol mol−1 , and 10 nmol mol−1 for a more limited set of
standards. The improvement in performance demonstrated in 2013 follows a number of
advances made in the intermediate years. In 2005, a new scale for CH4 measurements was
established by the National Oceanic and Atmospheric Administration (NOAA), acting
as the Central Calibration Laboratory (CCL) for CH4 in the WMO Global Atmosphere
Watch (GAW) programme [9]. The scale was designated NOAA04 [10]. This scale com-
prises sixteen gravimetric standards covering the CH4 mole fraction range 300 nmol mol−1
to 2600 nmol mol−1 . In 2012 Rhoderick et al. [11] published a series of new CH4 stan-
dards with improved accuracy developed by the National Institute of Standards and
Technology (NIST) achieving comparability against standards from the Korea Research
Institute of Standards and Science (KRISS) of +2.3 nmol mol−1 (KRISS being higher)
and of −2.54 nmol mol−1 against NOAA (NOAA being lower). In 2013, Brewer et al. [12]
described a new series of highly accurate CH4 and CO2 standards made of a whole air
matrix at the National Physical Laboratory (NPL) with an isotopic distribution match-
ing the atmospheric distribution for CO2 and that had excellent agreement against NIST
standards for CH4 mole fraction values.
In order for a primary standard, or rather a change in primary standard, to have
negligible influence on measurement comparability at a level defined by the WMO DQO,
the standard uncertainty of the primary standard and the standard deviation of a set
of compared primary standards should be arguably less than one quarter the value of
the compatibility goal, which would be 0.5 nmol/mol. This level of agreement is the
goal to set for a future repeat of the CCQM-K82 comparison, and will require further
improvements in uncertainties for the measurement of trace level methane in balance
gases.
.
3 2. Nitrogen monoxide in nitrogen standards. – The aim of the CCQM-P73 [13] com-
parison was to evaluate the level of comparability of laboratories’ preparative capabilities
for gravimetric nitrogen monoxide/nitrogen primary reference mixtures in the range (30–
70) μmol/mol. Such primary gas reference mixtures are prepared and maintained by a
number of NMIs in order to provide NOx calibration services in the areas of environmen-
tal and emissions analysis at the national level. The comparison was designed so that
measurements would be performed at a central laboratory (the BIPM) and measurement
results compared to values assigned by each National Metrology Institute (NMI) based
on gravimetry. The study was organised as a comparison of a suite of 2n primary gas
standards, two standards prepared by each of the n participating laboratories. The refer-
ence value for a given gas standard was to be determined from a regression line calculated
from all standards, or from a self-consistent subset of the standards. Measurement results
were based on two independent analytical methods, notably UV spectrophotometry and
chemiluminescence. Each participating laboratory provided: one high pressure cylinder
containing a NO/N2 mixture in the range (30–50) μmol/mol; one high pressure cylinder
containing a NO/N2 mixture in the range (50–70) μmol/mol.
Two analysers, the ABB Advance Optima LIMAS 11-UV analyser and the Thermo
Environmental Instruments Model 42C Chemiluminescence NO-NO2 -NOx analyser, were
used for the comparison of nitrogen monoxide gas standards. The two analysers were
connected in series, with the 42C downstream of the LIMAS. A sample flow of 500 ml/min
was sufficient to supply both analysers when connected in series.
To avoid post measurement correction for changes in pressure within the measure-
ment cells of the 42C and LIMAS analysers, arising from changes in ambient pressure, the
facility was designed so that it was isolated from the effects of ambient pressure changes.
In the case of the LIMAS, a software-controlled electronic pressure controller (EPC) im-
mediately downstream of the measurement cell, in combination with the MFC-controlled
sample flow, actively regulated the pressure within the LIMAS measurement cell at ap-
proximately 1050 hPa with a typical variation (1σ) over several hours of 0.01 hPa (0.001%
relative). In the case of the 42C analyser a capillary tube in the sample stream imme-
diately upstream of the measurement cell and a diaphragm vacuum pump immediately
downstream of the reaction chamber, ensured pressure stability within the measurement
cell. With this configuration, in combination with the MFC- and EPC-controlled sample
flow, the pressure within the 42C reaction chamber was stable at approximately 230 hPa
with a typical variation (1σ) over several hours of 0.04 hPa (0.02% relative). A 30 port
auto-sampler was used to allow automatic sampling of gas cylinders.
Analysis of all gas standards by FTIR was undertaken to quantify impurities within
the gas standards, and to compare these with the impurities and their uncertainties
0.5
0.4
0.3
Δ NO 2 μmol/mol
0.2
0.1
0.0
-0.1
-0.2
NMIA (M10)
NMIA (M23)
SMU (M6)
VNIIM (M8)
IPQ (M12)
IPQ (M25)
LNE (M5)
SMU (M18)
VNIIM (M22)
LNE (M19)
NIST (M9)
NIST (M24)
NMi-VSL (M13)
NMi-VSL (M21)
KRISS (M1)
CSIR-NML (M2)
KRISS (M15)
CSIR-NML (M16)
Laboratory (cylinder)
Fig. 4. – Graph of the difference between NMI assigned values and BIPM FTIR measurements
of NO2 mole fractions within gas standards analysed in CCQM-P73.
reported by participating laboratories. Significant differences between impurity mole

fractions reported by laboratories and the values measured by FTIR at the BIPM would
be used as confirmation that the data should be considered as an outlier and not included
in the calculation of the regression line. Nitrous oxide (N2 O) and nitrogen dioxide (NO2 )
impurity concentrations were of particular importance, as these impurities would be
expected in pure nitrogen monoxide gas, and would increase in concentration in the pure
gas with time and at elevated pressures. The results of the FTIR measurements are
plotted in fig. 4 and fig. 5. There was a strong correlation between deviations from the
zero line found in these graphs and the difference from the reference value calculated
from the comparison, indicating that deviations could have been corrected by further
purity analysis of the standards submitted for the comparison.
The differences between predicted and assigned gravimetric values are plotted in fig. 6.
By considering the predicted value as the reference value for each gas standard, the cal-
culated difference can be compared to degrees of equivalence (D) calculated in previous
key comparisons for nitrogen monoxide. In order for biases to appear in the same sense,
the values of –D from CCQM-K1.c and EUROMET.K1.c were plotted in fig. 6 together
with the values of D determined in CCQM-P73. The results of CCQM-P73 can be com-
pared to previous key comparisons for nitrogen monoxide in nitrogen at 100 μmol/mol,
and illustrates the reductions in uncertainties that can be achieved through a comparison
with measurements performed at a central facility.
The advantages in organizing a comparison with measurements performed at a
central laboratory have been clearly demonstrated, notably: analytical measurement
1.0
0.8
0.6
Δ N 2O μmol/mol
0.4
0.2
0.0
-0.2
NMIA (M10)
NMIA (M23)
SMU (M6)
VNIIM (M8)
IPQ (M12)
IPQ (M25)
VNIIM (M22)
SMU (M18)
LNE (M5)
LNE (M19)
NIST (M9)
NIST (M24)
NMi-VSL (M13)
NMi-VSL (M21)
KRISS (M1)
CSIR-NML (M2)
KRISS (M15)
CSIR-NML (M16)
Fig. 5. – Graph of the difference between NMI assigned values and BIPM FTIR measurements
of N2 O mole fractions within gas standards analysed in CCQM-P73.
3.000
2.000
1.000
0.000
mol/mol
-1.000
-2.000
-3.000
D/
-4.000
-5.000
CCQM-P73 (LIMAS 11UV analysis) CCQM-K1.c EUROMET-K1.c

-6.000
-7.000
NMIA (M10)
NMIA (M23)
SMU (M6)
VNIIM (M8)
OMH
FMI
IPQ (M12)
IPQ (M25)
LNE (M5)
SMU (M18)
VNIIM (M22)
GUM
LNE (M19)
VNIIM
VNIIM
IPQ
CHMI
NPL
NPL
NIST (M9)
NRLM*
CEM
NIST (M24)
CENAM (M7)
NIST
CENAM (M20)
NMi-VSL (M13)
NMi-VSL (M21)
CERI (M3)
CERI (M14)
NMi-VSL
NMi-VSL
VALCRM1 (M4)
METAS
BNM-LNE
BNM-LNE
KRISS (M1)
VALCRM2 (M11)
VALCRM3 (M17)
KRISS (M15)
CSIR-NML (M2)
KRISS
CSIR-NML (M16)
NRCCRM
Fig. 6. – Comparison of results from CCQM-P73 with degrees of equivalence determined in

previous key comparisons for nitrogen monoxide in nitrogen at 100 μmol/mol. The quantity –D
has been plotted for the results of CCQM-K1.c and EUROMET-K1.c, so that deviations from
the reference value in these key comparisons and CCQM-P73 appear in the same sense.
uncertainties can be reduced; a degree of equivalence parameter and its uncertainty can
be calculated. Following FTIR analysis, the standards of three laboratories (six gas
standards in total) were omitted from the regression analysis data set, as a significant
difference between reported and measured values of impurity contents was observed. An
additional standard was removed from the regression analysis set as its deviation from
the regression line was of the same order of magnitude as the standards already omit-
ted. A regression line consistent with the remaining calibration data (15 gas standards)
could be obtained, using analytical standard uncertainties of 0.12% (at 70 μmol/mol)
and 0.27% (at 30 μmol/mol), which resulted in predicted standard uncertainties of the
nitrogen monoxide mole fractions of 0.09 μmol/mol. This can be compared to the gravi-
metric uncertainties which were submitted by participating laboratories and were in the
range (0.004 to 0.104) μmol/mol.
4. – Dynamic methods (NO2 and HCHO)
Nitrogen oxides (NOx =NO+NO2 ) emissions into the atmosphere constitute a major
global air pollution issue. Currently, ambient air quality monitoring regulations [14] have
as objectives the limiting of nitrogen dioxide (NO2 ) concentrations to below (40 μg m−3
to 200 μg m−3 ) depending on the monitoring time, and lower values are expected in the
future with the development of new NOx reduction technologies in transport and indus-
try [15]. In order to support these monitoring requirements highly accurate and traceable
measuring systems are indispensable. A wide diversity of measuring systems are avail-
able and documented in the literature [16, 17]. Chemiluminescence is the most common
method used to monitor NOx and NO2 air pollution, where NO2 is not measured directly
but calculated from the difference of nitric oxide (NO) and NOx concentrations. Recent
studies [18] have shown that whereas total NOx emissions have been dropping or sta-
bilised in recent years, the proportion of NO2 to total NOx in the exhausts from vehicles
has been increasing and the principal factor for this increase has been considered to be
the increased use of oxidation catalysts and particle filters on light duty diesel vehicles.
This together with the foreseen development of sensitive optically based methods [19,20]
for the direct measurement of NO2 concentrations has led to the requirement for accurate
standards of NO2 in N2 in addition to NO in N2 standards.
The reactivity of NO2 , means that static gravimetric preparation techniques are dif-
ficult to employ for the preparation of these standards, and dynamic facilities for the
generation of NO2 in N2 concentrations are favoured for value assigning secondary stan-
dards of NO2 in N2 . A full description of these types of dynamic facilities can be found
elsewhere [21-25]. More recent developments are based on a magnetic suspension bal-
ance system for mass loss measurements of a permeation tube [24, 25]. The system at
the BIPM combines this with quantitative FT-IR measurements of both NO2 and other
nitrogen oxide impurities as well as measurements with a continuous gas analyzer ABB
Limas 11 (AO2020), resulting in the production of highly accurate concentrations with a
standard uncertainty of ∼ 0.04 μmol mol−1 (k = 1) at a mole fraction of 10 μmol mol−1 ,
a standard relative uncertainty of 0.4%. The performance of the BIPM facility has been
Fig. 7. – Schematic of the NO2 facility permeation tube chamber and magnetic suspension
balance.
demonstrated in the international comparison CCQM-K74 [24] of NO2 standards in-

volving National Metrology Institutes (NMIs). In this comparison, the BIPM facility
was used to value assign NO2 /N2 gas standards contained in high pressure cylinders
where, through careful quantitative purity analysis and as reported by previous similar
studies [26, 27], the presence of significant amounts of HNO3 in the gas mixtures was
confirmed.
Within the BIPM facility (fig. 7), the magnetic suspension balance (MSB; Ruboth-
erm, Germany) is central to the system. An electromagnet is suspended from the base
of the weighing pan. Below this electromagnet there is a long vertical glass vessel, the
measurement cell of the MSB. At the top of the glass vessel there is a permanent magnet
which is held suspended by the electromagnet attached to the balance. The position of
the permanent magnet is detected electronically and maintained by a servo-control of

the current of the electromagnet. An NO2 permeation tube is suspended from the per-
manent magnet (permeation rates in the range (5000–10000) ng min−1 are used for this
purpose). Thus, the balance measures the mass of the permeation tube without being
mechanically in contact with it, since the balance and the weighing load are separated
by a layer of glass. The coupling between the permeation tube and the balance is purely
magnetic and the sensitive balance is protected from the highly corrosive NO2 gas and the
sometimes elevated temperatures and gas flows surrounding the permeation tube. This
facilitates continuous monitoring of mass loss of the permeation tube, which is located in
a temperature controlled environment by means of a double glass wall jacket containing
water circulating at a constant temperature controlled by a remote cryothermostat. At
constant temperature, as a first approximation, the tube emits NO2 through its per-
meable fluoropolymer membrane at a constant rate. The balance is a high resolution
comparator (model AT20, Mettler, USA) with range of (0 to 22 g) and 2 μg resolution.
The balance is configured with two mass pieces used to perform an external calibration
of the balance. The two mass pieces have nominally the same volume but different mass,
as one is made of titanium (Ti) and the other of stainless steel (SS). They are used to
correct for the effect on the mass measurements arising from changes in the density of
the ambient atmosphere surrounding the balance itself.
To generate primary mixtures using the MSB, a well characterized flow of NO2 -free gas
(nitrogen) is required. The total gas flow is measured by means of a molbloc /molbox
facility. An electronic digital pressure controller is used to maintain the pressure of the
incoming gas entering the molbloc at about 2700 hPa that is the optimal pressure to
minimize the uncertainty of the molbloc flow measurement (∼ 0.1%). The gas flow is
then introduced to a gas purifier that removes the remaining water and oxygen that may
leak into the gas. The gas flow is divided in two streams, a carrier and a diluent, both
regulated by two mass flow controllers (MFCs). The flow of the carrier stream is set at a
constant value, 100 mL min−1 , mixing with the NO2 emerging at constant rate from the
permeation tube. The pressure conditions of the permeation chamber are controlled by
an electronic digital pressure to avoid any buoyancy variation. The gas mixture of the
carrier line is diluted by a larger flow, the diluent stream, varied within the range (0.3
to 5) L min−1 in order to, dynamically, generate primary NO2 mixtures in nitrogen (or
air) at various concentrations in the range (1-15) μmol mol−1 .
The final module of the system enables comparison with or calibration of standards
of NO2 in nitrogen contained in high-pressure cylinders with the dynamically generated
gas mixtures. This is achieved via the response of the NO2 analyser, either the ABB
Limas11 or FTIR. The system enables the sequential sampling of up to 15 standards
contained in cylinders by means of a 16 position valve (MPV-16). In addition, a 4-
port 2-position valve is used to select which sample stream, from either the MSB or
from a cylinder, is directed to the analysers, the other stream being directed to waste,
without perturbing the flow of either stream. The determination of the nitrogen dioxide
mole fraction of the stream emerging from the facility is calculated from the following
equation:
Mimp × ximp
P × Vm MHNO3 × xHNO3
(1) xNO2 = − − ,
qv × MNO2 MNO2 MNO2
where:
xNO2 is the NO2 mole fraction in μmol mol−1 ;
P is the NO2 permeation rate in ng min−1 ;
Vm = 22.4038 L mol−1 , is the molar volume of air/N2 at standard conditions (273.15 K,

101.3 kPa);
MNO2 = 46.0055 g mol−1 , is the molar mass of NO2 ;
qv is the total flow of N2 given by the molbloc/molbox facility;
xHNO3 is the HNO3 mole fraction in μmol/mol measured by FT-IR spectroscopy;
MHNO3 = 60.005 g mol−1 is the molar mass of HNO3 ;
ximp are the mole fractions in μmol/mol of other impurities measured by FT-IR Spec-
troscopy; and
Mimp are the molar mass of the impurities.
The second part of this equation considers the nitric acid mole fraction observed as
the main gas impurity in the gas mixtures produced in the facility and permeated from
the permeation tube. The third part comprises impurities that were below the limit
of detection of the FTIR spectrometer. They were considered negligible but with an
uncertainty estimated by FTIR analysis of the gas mixture emerging from the permeation
chamber without further dilution by nitrogen. The uncertainty budget for the operation
is described in table II.
In the CCQM-K74 comparison, the Key Comparison Reference Value for each cylinder
was the NO2 mole fraction assigned by the BIPM. Following the CCQM GAWG guidance,
it was decided that the standard uncertainty of the Key Comparison Reference Value
(KCRV) can be quantified by the following equation:

(2) uKCRV (xNO2 ) = (u(xBIPM ))2 + (u(xNO2 Losses ))2 + (u(xNO2 Drift ))2 ,
where u(xBIPM ) is the uncertainty associated with the value assigned by the BIPM
following the procedure described above, u(xNO2 Lossess ) the uncertainty contribution due
to NO2 losses equivalent to 5.7 nmol/mol and u(xNO2 Drift ) the uncertainty contribution
due to observed drift in NO2 estimated in 21 nmol/mol, resulting in a KCRV standard
uncertainty of about 0.041 μmol/mol.
The CCQM-K74 results indicated consistency of the majority of measurement results

within limits of ±3%. This can be compared to a relative standard uncertainty in the key
comparison reference value of 0.4%. It was concluded that the additional variance was
due to the additional challenge of measuring and differentiating nitric acid from NO2 ,
and current research in this area is directed at improving nitric acid measurements as
well as reducing the content of nitric acid in standards.
A similar system can be used for the dynamic generation of formaldehyde mixtures
in nitrogen, by use of either a paraformaldehyde permeation source or a diffusion tube
containing trioxane and a thermal convertor to produce HCHO [28, 29]. Two systems
implementing these principles were used by the BIPM to pilot the Key Comparison
CCQM-K90 in 2016 [30].
Table II. – Uncertainty budget for a NO2 /N2 primary mixture generated with the BIPM facility.
Note: the molar masses MN2 O4 , MN2 O3 , MN2 O5 , MHONO , MHO2 NO2 were not included in this
budget as they represent negligible uncertainty contributions.
Quantity Estimate Assumed Standard Sensitivity Uncertainty Index

distribution uncertainty coefficient
contribution %
u(xi ) ci = ∂xNO2 /∂x ui (y)
xi mol mol−1
P 8.35 × 10−6 g min−1 Normal 4.18 × 10−9 g min−1 1.1 4.5 × 10−9 2.2
Vm 22.4038 L mol−1 Normal 340 × 10−6 L mol−1 400 × 10−9 140 × 10−12 0.0
−3 −1 −6 −1 −6 −9
qv molbloc1 452 × 10 L min Normal 455 × 10 L min −20 × 10 −9.1 × 10 8.8
−1 −3 −1 −9 −12
MNO2 46.0055 g mol Normal 1.40 × 10 g mol −190 × 10 −270 × 10 0.0
xHNO3 0.10 × 10−6 mol mol−1 Normal 0.021 × 10−6 mol mol−1 −1.4 −29 × 10−9 88.5
xN2 O4 0 mol mol−1 Normal 866 × 10−12 mol mol−1 −2.0 −1.7 × 10−9 0.3
−1 −12 −1 −12
xN2 O3 0 mol mol Normal 307 × 10 mol mol −1.7 −510 × 10 0.0
xN2 O5 0 mol mol−1 Normal 361 × 10−12 mol mol−1 −2.3 −850 × 10−12 0.0
−1 −12 −1 −12
xHONO 0 mol mol Normal 520 × 10 mol mol −1.0 −530 × 10 0.0
xHO2 NO2 0 mol mol−1 Normal 572 × 10−12 mol mol−1 1.7 −980 × 10−12 0.1
−1 −3 −1 −9 −12
MHNO3 63.013 g mol Normal 1.17 × 10 g mol −2.3 × 10 −2.6 × 10 0.0
Quantity Value Standard

uncertainty
xNO2 8.86 μmol mol−1 30 nmol mol−1
5. – Spectroscopic methods (O3 and FTIR)

.
5 1. Ozone standards. – Ground-level ozone concentration is an important air quality
parameter which is monitored and reported world-wide. The reference method for ozone
measurements is based on UV Photometry, with the NIST SRP ozone reference standard
acting as the primary standard for numerous national and international ozone monitoring
networks. The NIST SRP was developed in the 1980s as a joint collaboration between
the US Environmental Protection Agency (EPA) and NIST to ensure the traceability in
the US ground-level ozone network. Two of these instruments are maintained by NIST
(SRP0 and SRP2) [31] and are regularly used to ensure proper comparability of ten other
SRPs maintained by EPA regional centres. Similar requirements for other networks exist
globally, for example, the European Directive 2002/3/CE related to ozone in ambient
air sets information and alert thresholds, target values and long-term objectives, as well
as rules and guidelines on the implementation of ozone monitoring networks. The UV
photometry method is described as the reference method for ozone analysers as well as
calibration instruments, following the ISO standard 13964 [32]. The Directive also gives
a data quality goal by specifying a maximum uncertainty of 15% attached to individual

measurements.
In 2000, the BIPM, in collaboration with NIST, started a programme to establish
an international network for ozone reference standard comparisons and calibrations. In
the framework of the CCQM comparison programme, the BIPM has coordinated the
comparison CCQM-P28 of the ozone reference standards of 23 laboratories, which main-
tain ambient ozone analyser/standard calibration services within their roles as national
metrology institutes or national/international air quality reference laboratories [33]. The
comparison was followed on by BIPM.QM-K1, starting in 2007, as an on-going bilateral
comparison of standards at the BIPM, with 42 comparisons of standards having been
performed so far. The SRP (fig. 8) has been designed for the measurement of ozone
concentrations in air up to 1000 nmol mol−1 . The measurement is based on the absorp-
tion of radiation at 253.65 nm by ozone/air mixtures in the gas cells of the instrument.
One particularity of the instrument design is the use of two gas cells to overcome the
instability of the light source.
The measurement equation is derived from the Beer-Lambert and ideal gas laws. The
number concentration (C) of ozone is calculated from
−1 Tmes Pstd
(3) C= ln(D),
2σL Tstd Pmes
where
σ is the absorption cross-section of ozone at 253.7 nm in standard conditions of temper-
ature and pressure. The value used is: 1.1476 × 10−17 cm2 /molecule [34]. In (3):
L is the optical path length of one of the cells,
Tmes is the temperature measured in the cells,
Tstd is the standard temperature (273.15 K),
Pmes is the pressure measured in the cells,
Pstd is the standard pressure (101.325 kPa),
D is the product of transmittances of two cells, with the transmittance (T ) of one cell
defined as
Iozone
(4) T = ,
Iair
where
Iozone is the UV radiation intensity measured in the cell when containing ozonized air,
and
Iair is the UV radiation intensity measured in the cell when containing pure air (also
called reference or zero air).
Fig. 8. – A schematic of the Ozone SRP system.
Using the ideal gas law eq. (3) can be recast in order to express the measurement
results as a mole fraction (x) of ozone in air:
−1 R Tmes
(5) x= ln(D),
2σL NA Pmes
where
NA is the Avogadro constant, 6.022142 × 1023 mol−1 , and
R is the gas constant, 8.314472 J mol−1 K−1
The uncertainty budget for the operation of the SRP is given in table III:
The results of the BIPM.QM-K1 comparison demonstrate excellent agreement be-
tween ozone standards based on UV photometry. It should be noted however that for
the comparison the same conventional value of the ozone cross section is used in these.
Recent measurements of the ozone cross section based on UV absorption and manometry
(described in sect. 6) [35]and more recently by gas phase titration [36] appear to indicate
that the conventionally used value may be biased about 2% high, and discussions on
revising this value are currently on-going.
Table III. – Uncertainty budget for ozone mole fraction measurements with an SRP.
Uncertainty u(y) Sensitivity contribution to
coefficient u(x)
Combined ci = ∂x
∂y |ci | · u(y)
Component (y) Source Distribution Standard standard
Uncertainty uncertainty
u(y) nmol/mol
Measurement
Rectangular 0.0006 cm
Scale
Optical Path Lopt Repeatability Normal 0.01 cm 0.052 cm − Lx 2.89 × 10−3 x

opt
Correction factor Rect 0.052 cm
Pressure gauge Rectangular 0.029 kPa

x
Pressure P 0.034 kPa −P 3.37 × 10−4 x
Difference
Rectangular 0.017 kPa
between cells
Temperature
Rectangular 0.03 K
probe x
Temperature T 0.07 K T 2.29 × 10−4 x
Temperature
Rectangular 0.058 K
gradient
Ratio of Scaler resolution Rectangular 8 × 10−6 x

1.4 × 10−5 D ln(D) 0.28
intensities D Repeatability Triangular 1.1 × 10−5
Absorption cross 1.22 × 10−19 1.22 × 10−19 x

Hearn value −σ 1.06 × 10−2 x
section σ cm2 /molecule cm2 /molecule
.
5 2. FTIR for the comparison and analysis of gas standards. – FT-IR spectroscopy
provides a powerful analytical technique for gas concentration measurements as it can
deliver high sensitivity and selectivity and quantitative measurements [37]. FT-IR in-
strument response and stability have been well studied, and a number of calibration
strategies can be employed to deliver accurate measurement results. In the metrology
laboratory setting, FT-IR instruments can be readily used to compare gas standards with
small uncertainties [38-40]. Alternatively, the instruments can be used to make measure-
ments with the traceability of results to very high-resolution line strength data, and
algorithms that produce synthetic calibration spectra and enable quantitative measure-
ments at 1 cm−1 resolution, an approach that was originally developed in the atmospheric
remote-sensing community. In order to ensure the SI traceability of measurement results,
the uncertainties in the reference line strength values must be included in the calculation
of the uncertainty of the measurement result. Within the laboratory, the use of synthetic
spectra for calibration will be particularly useful to quantify impurities in gas matrices,
for which no gas standards are available, but for which spectroscopic high resolution
line strength data exist. This has been successfully demonstrated for measurements of
nitric acid (HNO3 ) present as an impurity in the nitrogen dioxide (NO2 ) in nitrogen
(N2 ) primary gas mixtures [25]. Since no reference standards of HNO3 were available,
synthetic reference spectra created by the software MALT (Multiple Atmospheric Layer
Transmission) [37, 41, 42] and the molecular database HITRAN version 2004 [43] (more
recently the 2012 version has become available) were used, together with a fitting algo-
rithm of the experimental spectra by Classical Least Square (CLS) [44], to determine
the mole fraction of this component. The methodology for determining the uncertainty
of measurements resulting from the application of this calibration technique has been
described in detail elsewhere [45], together with values of realistically achievable stan-
dard measurement uncertainties, which, for example, were not below 3.4% relative for
the measurement of NO2 mole fractions at nominal values of 10 μmol mol−1 .
The FT-IR facility used in the NO2 /HNO3 study consisted of a ThemoNicolet Nexus
FT-IR spectrometer configured with a Mercury Cadmium Telluride (MCT) high D* liq-
uid N2 -cooled mid-infrared detector, and set to 1 cm−1 resolution. To keep the internal
optical path of the spectrometer free of any interference species this ensemble was placed
in a stainless-steel enclosure which was constantly purged with high-purity N2 (dewpoint
∼ −95 ◦ C, i.e. ∼ 200 nmol mol−1 H2 O) flowing at ∼ 15 L min−1 . The spectrometer was
configured with a 6.43 m multipass White-type cell (Gemini Scientific Instruments, USA).
The gas cell had wetted surfaces of electro-polished stainless steel and gold (mirror coat-
ings) to minimize surface interactions with reactive gas phase species. The path length of
the gas cell (Ls ) was measured experimentally by reference to a single pass gas cell. The
precise knowledge of the optical path length is of major importance [37, 39, 46-49] when
using FT-IR spectroscopy to determine the mole fraction of a compound. It is strongly
recommended that the optical path length is measured rather than relying on the value
given by the cell provider, especially for multi pass cells. A common method to achieve
this path length measurement is to use a stable gas standard of known concentration and
a single pass reference gas cell in which the path length is easier to measure [50]. The
path length of the 6.43 m gas cell (Ls ) used for NO2 quantification was measured exper-
imentally using a single pass gas cell (cell length of 100.097 mm) and two NO2 primary
reference gas standards (PRSs) following equation (6):

xr ∗ Lr ∗ Pr Ts
(6) Ls = K ∗ ∗ ,
Tr xs ∗ Ps
where x was the NO2 mole fraction in μmol mol−1 , L the path length in m, P the
gas pressure in kPa, T the absolute gas temperature in K, and K the ratio between
NO2 absorbances in both spectra. Care was taken to ensure that the mole fractions
of NO2 used in both gas cells resulted in approximately the same absorbance in both
spectra (less than 0.1) to avoid non-linearities. The gas standard used in the reference
gas cell was a high concentration NO2 mixture in N2 prepared by the National Physical
Laboratory (NPL), UK, with a certified mole fraction of (99.99 ± 0.8) μmol mol−1 . In
the case of the 6.43 m gas cell, a second gas standard of NO2 in N2 with a certified mole
fraction of (11.00 ± 0.22) μmol mol−1 prepared by the National Metrological Institute in
the Netherlands (VSL) was used.
A variety of line-by-line atmospheric radiation codes for the generation of synthetic
infrared reference spectra have been produced since the first version of the atmospheric
Fig. 9. – Fit and residual for NO2 matched run by MALT at 10.95 μmol mol−1 .
transmittance and radiance model FASCODE was created [51]. In the NO2 /HNO3 study
MALT version 4.4 was used to generate the synthetic infrared spectra. MALT4.4 calcu-
lates atmospheric transmission or absorbance spectra by a radiative transfer model and
then modifies the ideal spectra to take into account instrumental effects that occur in a
real spectrometer. This code has been compared with other codes such as SFIT [52, 53],
a commonly used code for ground base FT-IR, and agreement to better than 0.7% has
been observed [54]. Figure 9 illustrates the difference between a measured (cross dots)
and simulated (solid line) of a NO2 spectrum, with the measured spectrum obtained on
a gas mixture generated by the BIPM-NO2 primary gas facility. A full description of
how uncertainties associated with the use of MALT4.4 have been evaluated and described
elsewhere [45].
The relative standard uncertainty of laboratory-based FT-IR measurements of NO2
mole fractions at 10 μmol mol−1 using the v3 band and HNO3 at 200 nmol mol−1 using the
v2 band calibrated with synthetic spectra traceable to HITRAN reference line strength
data have been determined, and were found to be 3.4% and 11%, respectively. These
uncertainties include contributions from the uncertainties of the values of molecular line
strengths in HITRAN, which is a requirement for the results to be considered SI traceable.
The standard uncertainty assigned to the molecular line parameters (3% for NO2 and
5% for HNO3 ) appeared to be the most important uncertainty contributor followed by
the uncertainty of the optical path-length of the gas cell used for the measurements
and the uncertainty of the gas temperature in the gas cell uncertainty assigned to the
instrument. In the case of NO2 this uncertainty was one order of magnitude higher
than the uncertainty that can be achieved by the calibration of FT-IR instruments with
SI traceable gas standards generated by a dynamic weighing system (relative standard
uncertainty of 0.4%) [25]. Nevertheless, the technique remains useful for minor impurities,
where the contribution to the uncertainty of the major component will remain small.
.
5 3. FTIR measurements of isotope ratios in CO2 . – FTIR has the additional advan-
tage of being sensitive to individual isotopologues, which opens the possibility of mea-
suring isotope ratios within gases. The accurate measurement of isotope ratios in CO2 in
air has recently been demonstrated [55]. For these measurements, a vacuum Bruker Ver-
tex 70v FTIR Spectrometer equipped with a RockSolid interferometer (vacuum better
than 0.2 hPa), with 1 cm−1 resolution (0.16 optional), a 40 mm beam diameter, a globar
source and CaF2 beam splitter was used. The spectrometer was configured with a liq-
uid N2 -cooled mid-infrared Indium Antimonide (InSb) detector and a 10.01 m multi-pass
White-type gas cell of volume 0.75 L (Gemini Scientific Instruments, USA). The wetted
surfaces of the gas cell were electro-polished stainless steel treated with silconert 2000
(Silcotek) and gold (mirror coatings) to minimize surface adsorption and desorption ef-
fects for CO2 . The interferometer was scanned at 64 scans min−1 and spectra co-added
for five minutes to obtain an acceptable signal-to-noise ratio. The transmission spectra
of gas reference standards obtained following this procedure had a very high signal to
noise ratio of typically ∼ 1 × 104 peak-peak from (2240–3380) cm−1 . In order to prevent
nonlinear responses produced by excess photon flux reaching the InSb detector special
care was put into adjusting the instrument parameters of the software to ensure that
the apparent intensity from the InSb detector was zero at 1850 cm−1 . The spectrometer
user interface was controlled using a BIPM developed software named B-FOS, that al-
lowed the automatic setting of all instrument parameters into Bruker’s proprietary OPUS
software for control, spectral acquisition and on-line analysis through the use of MALT
spectrum analysis software, version 5.56. The spectra were constructed by co-adding up
to 320 scans recorded in about 5 minutes to provide a single spectrum of a sample.
This single spectrum was ratioed with a similar spectrum of ultra-pure N2 collected
under similar conditions to provide an absorbance or transmission spectrum of the gas

sample (relative to ultra-pure N2 ) in the gas cell. The sample flow rates were kept at
∼ 400 mL min−1 In order to ensure the complete exchange of samples, only the measured
spectra obtained after flowing the sample through the White-type cell for 35 min were
used for mole fraction determinations. The sample pressure was measured by means of
the calibrated barometer Mensor Series 6000. The relative uncertainty of this pressure
sensor was u(P ) = 0.02%. The sample temperature was measured with a calibrated 100
Ω RTD temperature probe introduced into the outlet gas of the White-type gas cell with
a standard uncertainty u(T ) = 0.02 K.
The recorded spectra were analysed online by non-linear least squares fitting in two
sections of the spectra, 3500–3800 cm−1 (plotted in fig. 10) and 2200–2310 cm−1 (plotted
1.00
0.95 Measured
Fitted
0.90
Transmitance
0.85
0.80
0.75
0.70
3500 3550 3600 3650 3700 3750 3800

0.002 Residual
0.000
-0.002
3500 3550 3600 3650 3700 3750 3800

-1
Wavenumber/cm
Fig. 10. – Typical measured and fitted spectra (upper part), and residual signal (lower part)
in the region 3500–3800 cm−1 obtained on a standard of CO2 in air at 384.33 μmol mol−1 . The
absorption feature is due to the major isotopologue 626.
1.00
0.95
0.90 Measured
0.85 Fitted
Transmitance
0.80
0.75
0.70
0.65
0.60
0.55
0.50
220 0 225 0 230 0
0.002 Residual
0.000
-0.002
2200 2250 2300

-1
Wavenumber/cm
Fig. 11. – Typical measured and fitted spectra (upper part), and residual signal (lower part)
in the region 2200–2310 cm−1 obtained on a standard of CO2 in air at 395 μmol mol−1 . The
absorption feature is due to the isotopologues 626, 636 and 628, and to N2 O when present.
in fig. 11), with modelled spectra calculated by MALT using absorption line parameters
from the HITRAN database version 2012 [56]. As previously described [57], the first re-
gion shows peaks that are due to the major isotopologue 12 C16 O16 O, whereas the second
region can be used to analyse the two other isotopologues 13 C16 O16 O and 12 C16 O18 O. In
order to correctly account for interferences, N2 O mole fractions were also calculated from
measurements in the second spectral region. Since all standards used during this work
were ultra-dry CO2 /air gas mixtures the cross sensitivity with H2 O was not considered.
With appropriate calibration standards it has been demonstrated [55] that δ 13 C and
18
δ O can be measured in CO2 at nominally 400 μmol/mol in air with standards uncer-
tainties of 0.09 and 1, respectively, with traceability to the VPDB scale.
6. – Manometric methods (CO2 and O3 )

.
6 1. Manometric reference method for CO2 mole fraction value assignment. – Changes
in carbon dioxide concentration in the atmosphere are of high environmental and political
importance, with increasing levels of emissions and background monitoring. Performance
criteria for CO2 monitoring networks are being set, with the most stringent of these being
for background ambient levels where data compatibility from different monitoring sites
should be at the 0.1 μmol/mol level for mole fractions at the 400 μmol/mol. This in turn
requires accurate calibration standards, which historically have been traceable to one
ensemble of primary gas standards, which establish a scale for atmospheric background
CO2 measurements, and were value assigned by a manometric method [58, 59]. The
consistency of manometric and gravimetric methods for gas standard value assignment
is being investigated at the BIPM, which has developed its CO2 -PVT measurement
system. The name reflects that the CO2 amount fraction in air is deduced from accurate
measurements of pressure, volume and temperature of whole air and then of the CO2
extracted from it by cryogenic trapping. The whole air and pure CO2 samples parameters
are measured in two distinct volumes, the ratio of which is predetermined using a well-
known method based on series of expansions of air, introducing three auxiliary volumes
to reduce the uncertainty.
A diagram of the system is displayed in fig. 12. This system was constructed in Pyrex
and installed inside an oven, so as to maintain it at a constant temperature of nominally
40 ◦ C. The facility includes a number of glass vessels indicated by their volume (V a, V c,

V d, V e, V f ), linked by tubbing either in glass or stainless steel. A number of valves
in glass (GVi ) or stainless steel (SVi ) allow the transfer of gas through the system. All
valves are air actuated. A Silcosteel cryotrap (CRYO) is installed on the outside of the
oven and used to extract CO2 from air. A second, smaller trap, is situated inside the
oven to transfer the extracted CO2 in a small volume V a. The flow rate of the gas can be
controlled in two points with mass flow controllers (MFCi ) of nominal range 0 L min−1 to
1 L min−1 . A Non Dispersive InfraRed (NDIR) CO2 analyzer is used to check the value
and stability of the CO2 mole fraction in the samples. The entire system can be vacuumed
with a scroll pump (Edwards nXDS6i) and/or a turbopump (Edwards STP-301C).
A measurement is made by first collecting a sample of air containing CO2 in the
Fig. 12. – Scheme of the PVT-CO2 system: Pi is the pressure gauge; CRYO the cold trap;
NDIR the Non Dispersive Infrared CO2 analyzer; Vi the glass vessels; GVi the Glass valve;
SVi the Stainless-steel valve; MFCi the Mass Flow Controler. The square represents the oven
maintained at 40 ◦ C.
“large volume” (V e + V a + V b) and taking a precise measurement of its temperature and

pressure. The gas is then flowed via a mass flow controller (MFC2 ) at a controlled rate
through the cold trap CRYO, which is held at a temperature below the condensation
point of CO2 . The CO2 , N2 O (of which a sample usually contains ∼ 350 nmol mol−1 ), and
any residual water vapor in the sample are condensed on the walls of the cold trap, while
the N2 , O2 , Ar, and CH4 pass through the trap and are pumped away and exhausted.
After all the untrapped gases have been removed from the system, it is sealed off while
the “small volume” V a is cooled through direct contact with liquid nitrogen (LN2 ).
The next step is distillation, wherein the cold trap is heated until the CO2 (along
with the N2 O) sublimates and is cryopumped into the small volume. The final step is to
allow the small volume to return to the ambient temperature in the oven and stabilize,
at which point the temperature and pressure are again measured. Combining the ratio
of pressure and temperature measurements in the small and large volumes, along with a
separately performed gas expansion measurement of the ratio of the small to the large
volume, and correcting for deviations from ideal gas behavior and for the small amount
of N2 O gives an absolute measure of the amount fraction of CO2 in the initial air sample.
As developed by Zhao et al. [59], the measurement equation is derived from the
equation of state of a real gas applied in the volume V at pressure P and temperature
T , taking into account the second virial coefficient BV :

nBV (T )
(7) P V = nRT 1 + .
V
This formulation being quadratic in the number of moles n, it has two solutions, of which
only one has a physical meaning. Solving this equation leads to expressions for nCO2 and
nair , hence for the CO2 mole fraction xCO2 which is the ratio of them:

BV 2 (T ) γ1
(8) xCO2 = RV ,
BV 1 (T ) γ2
where the function γi is defined by
4Pi Bi (Ti )
(9) γi = 1 − 1+
RTi
and the index 1 is for air, 2 for CO2 .

The volume ratio RV = Va +V b +Ve
Va is best measured by the technique of static expan-
sions in a total of five different vessels, applying a method which is commonly used in
pressure metrology to calibrate pressure sensors [60]. For each expansion, dry air near
to atmospheric is introduced in a first volume Vi . After its pressure and temperature
has reached equilibrium, the gas is expanded into a second (larger) volume Vi , where its
pressure Pi is measured. Each ratio ri = Vi /Vi is calculated, so as to compute the final
ratio RV . Therefore the accuracy of the volume ratio measurement strongly relies on the
pressure measurements.
The pressure of the air and pure CO2 samples has to be measured accurately, and
this is ensured by an absolute digiquartz pressure transducer (Paroscientific 215A-101,
measurement range from 0 kPa to 103.4 kPa) (P1). In the system described above, a
resonant silicon pressure gauge (P2) (General Electric DPS-80HA) was also included for
tests. The pressure gauge is calibrated versus the BIPM pressure balance [61] via a
Quartz-Reference Pressure Transducer (DHI RPM4 equipped with one A100KS) used as
transfer standard.
Temperature measurements are also critical. It is important to first assess the residual
instability of the air and CO2 samples temperature after the stabilization period. In
addition despites the presence of fans inside the oven to help the air circulation, small
temperature gradients are observed through the volumes in which samples are kept and
need to be taken into account. Therefore the temperature is measured at various points
of the oven by negative temperature coefficient (NTC) resistive thermometers.
The main uncertainty contributors are the pressure, temperature, and volume ratio,
with similar relative uncertainties for all of them. Zhao et al. have demonstrated typ-
ical standard uncertainty of 0.07 μmol mol−1 with a volume ratio equals to 643. The
BIPM system has a volume ratio of about 250, resulting in typical uncertainties of
0.2 μmol mol−1 . A second version of the system is being developed with a volume ratio of
1000 to reach lower uncertainties. In addition this second version is constructed in stain-
less steel with a surface treatment (SilcoNert), as this material has been successfully used
in the cold trap, demonstrating interesting properties to avoid absorption/desorption
of CO2 .
.
6 2. Manometric method for ozone cross-section measurements. – The property of
ozone to strongly absorb UV radiation, notably in the Hartley Band, and the relative
ease of reproducing a mercury line at 253.65 nm (in air) has led to the value of the ozone
absorption cross-section at this wavelength becoming particularly important for global
ozone atmospheric monitoring. Efforts to improve its accuracy continue [62], because
it directly impacts on all results from instruments based on UV absorption. The NIST
SRP, operates on this principal, and also acts as the primary standard for numerous
national and international ozone monitoring networks, such as the WMO Global At-
mosphere Watch (GAW) Programme [63, 64]. Several replicas of this instrument are
maintained by the BIPM, one of which is the reference for international comparisons of
national ozone standards coordinated by the BIPM. The ozone absorption cross-section
value at the 253.65 nm wavelength, [65] is the measured value used in the SRP, and was
adopted as a conventional value during the Quadriennal Ozone Symposium in 1984 to-
gether with measurements by Bass and Paur for other wavelengths [66]. Measurements of
ozone absorption at other wavelengths have often been scaled to the value at 253.65 nm,
since they were performed in ozone in air mixtures, of which the ozone concentration is
determined using its absorption at this wavelength.
During the first international comparison of ozone standards for ground level ozone
(CCQM-P28) conducted by the BIPM, twenty three laboratories reported results based
on UV absorption compared to two laboratories reporting ozone concentration measure-
ments based on gas phase titration systems: the National Institute for Environmental
Studies of Japan [67] and the BIPM [33]. Gas phase titration involves reacting ozone
in air with nitrogen monoxide, and measuring either the loss of nitrogen monoxide or
the gain of the reaction product, nitrogen dioxide, to deduce the ozone concentration
in the sample. The 2% to 3% bias observed between the methods (gas phase titration
reporting higher ozone concentrations) could be explained by a biased ozone absorption
cross-section value, which represents the major uncertainty component in measurements
based on UV photometry.
In 2007, the BIPM started a laboratory programme to perform new measurements of
the ozone absorption cross-section with improved accuracy. Efforts focused on two major
sources of uncertainty in the measurements: ozone purity and knowledge of the light
path length. The BIPM first developed a laser ozone photometer, capable of measuring
ozone concentrations in the same range as the SRP but with improved accuracy, as
described earlier [68]. This instrument was used to deduce new values of the ozone
cross-section in the Hartley band, relative to the reference value obtained by Hearn at
253.65 nm. Subsequent studies focused on determining absolute measurements of the
ozone cross-section at the same wavelengths with the smallest uncertainties published
to date. This required a cryogenic ozone generator to be developed, together with a
Fig. 13. – Measurement setup. AOM: Accousto-Optic Modulator; BS: Beam Splitter; CDG:
Capacitive Diaphragm Gauge (Baratron); CRYO: cryogenic ozone generator; LT: Light Trap; M:
Mirror; RGA: Residual Gas Analyser (Quadrupole mass spectrometer); TMP: Turbo Molecular
Pump; V: Valve; WM: Wavemeter.
method based on cycles of evaporation-condensation of ozone so as to better evaluate the

purity of gaseous samples on which UV absorption measurements were performed. The
setup, shown in fig. 13, consisted of four major parts: a silent discharge cryogenic ozone
generator (CRYO), an absorption cell (AC), a pressure gauge (CDG) and a quadrupole
mass spectrometer (RGA). The discharge cryogenic ozone generator was where ozone was
generated from an electric discharge in pure oxygen (Linde, grade 6.0), using a method
similar to that described by Janssen et al. [69].
The absorption cell was made of quartz and was where gaseous ozone at room temper-
ature was introduced to allow absorption measurements to be performed with an intensity
stabilized UV laser beam. The pressure gauge was a capacitive diaphragm gauge with
its heater turned off (1 Torr MKS 690A Baratron coupled with a MKS 670B signal con-
ditioner) and was used to monitor the pressure of ozone samples at room temperature
as well as the pressure of residual gases when ozone was condensed in the cryostat. The
quadrupole mass spectrometer (RGA model MKS Vision 1000 C) was used to check the
composition of residual gases. An additional pressure gauge (Pirani, Pfeiffer vacuum
MPT100 / PTR35-130) was used to monitor the oxygen pressure during ozone genera-
tion and the vacuum pressure during evacuation of gases with a turbo-molecular pump
(TMP - Edwards STP 301C backed up with a primary pump Edwards XDS 10). Except
for the quartz absorption cell, the measurement volume, i.e. the volume closed by valves
V1, V2, V3 and V4 was made of 316L stainless steel, including the 4 valves (VAT 26524-
KE01-BLV1) in which the gas is in contact with stainless steel 316, except when they
are open, allowing contact with the gate seals in Viton. Vacuum flanges were connected
together and to the valves using aluminium seals, as it was found to minimize ozone
losses compared to the original Viton seals, and avoid the formation of CO2 and CO.
Ozone was produced in a silent discharge cryogenic ozone generator especially designed
for this purpose. Using the liquid ozone trapped in the bottom of the ozone generator, a
series of evaporation-condensation cycles were performed so as to obtain about 9 different
ozone pressures in the gas cell to cover the range 0.2 mbar to 1 mbar, whilst measuring
the absorbance Ae at a fixed laser wavelength to determine the ozone cross-section.
The purity of the ozone sample was of primary importance when measuring accurate
ozone cross-sections. An analysis of the uncertainties shows that any impurity that
would be above 1% in the sample should be measured and taken into account if the target
uncertainty of the ozone cross-section is to be below 1%. It is well known that, even when
starting measurements with pure ozone samples, dissociation of ozone following collisions
and interactions with surfaces will lead to the formation of oxygen. In addition, leaks
within the system and reaction of ozone with contaminated surfaces could potentially
lead to other trace contaminants being present. The amount of oxygen was quantified by
comparing the total pressure measured while the Baratron gauge was in contact with the
ozone sample with the variation of the ozone partial pressure, determined from optical
absorption measurements. The presence of other minor impurities, notably CO2 and
CO, was evaluated and quantified by FTIR and Mass Spectrometry (with the RGA).
The optical setup was the same as described earlier [68] for the laser ozone photometer,
except that the measurement cell was a 5 cm long absorption cell (AC in fig. 13) in quartz.
The UV light was produced with an argon-ion laser doubled in frequency emitting light at
either 244.06 nm, 248.32 nm or 257.34 nm (wavelengths in vacuum). The method used for
measuring the absorption cell length was based on counting interference fringes produced
in a Michelson interferometer with the absorption cell placed in one of its arms, when
evacuating the cell from air at ambient pressure to vacuum, due to the varying index
of refraction in the arm with the cell. This method follows the process described by
Castrillo et al. [70].
Ozone absorption cross-sections could then be derived from the Beer-Lambert law:
RAe Tcell Ae
(10) σ= = ,
NA Lopt x(O3 )PT CLopt
where Ae = − ln(I/I0 ) is the absorbance measured in real time in the absorption cell, PT
the total pressure, x(O3 ) is the ozone mole fraction in the sample, and C = NART cell
x(O3 )PT
−3
is the ozone concentration in molecule cm .
This permitted new measurements of absolute values of ozone absorption cross-
sections of 9.48 × 10−18 , 10.44 × 10−18 , and 11.07 × 10−18 cm2 molecule−1 , with relative
expanded uncertainties better than 0.7%, for the wavelengths (in vacuum) of 244.06,
248.32, and 257.34 nm, respectively, to be published. The cross-section at the 253.65 nm
line of mercury was determined by comparisons using a SRP equipped with a mercury
lamp as the light source, and a value of 11.27 × 10−18 cm2 molecule−1 with an expanded
relative uncertainty of 0.86% (coverage factor k = 2) was reported. This is lower than
the conventional value currently in use and measured by Hearn in 1961 with a relative
difference of 1.8%.
REFERENCES
[1] Stocker T. et al., Technical summary, in Climate Change 2013: The Physical
Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change (Cambridge University Press) 2013, p. 33-
115.
[2] WMO, WMO Greenhouse Gas Bulletin, Geneva, 2016.
[3] ISO, Gas analysis – Preparation of calibration gas mixtures – Gravimetric method, in 6142-
1:2015 2001.
[4] ISO, Gas analysis- Comparison methods for determining and checking the composition of
calibration gas mixtures, in 6143:2001. 2001.
[5] Milton M. J. T., Vargha G. M. and Brown A. S., “Gravimetric methods for the
preparation of standard gas mixtures” Metrologia, 48 (2011) R1.
[6] Nara H. et al., “Evaluation of factors affecting accurate measurements of atmospheric
CO2 and CH4 by wavelength-scanned cavity ring-down spectroscopy” Atmos. Meas. Tech.
Discuss., 5 (2012) 5009.
[7] Veen A. M. H. v. d. et al., “International comparison CCQM-P41 Greenhouse gases. 1.
Measurement capability” Metrologia, 44 (2007) 08002.
[8] Veen A. M. H. v. d. et al., “International comparison CCQM-P41 Greenhouse gases. 2.
Direct comparison of primary standard gas mixtures” Metrologia, 44 (2007) 08003.
[9] GAW Research and Monitoring Reports. Report No. 206. 16th WMO/IAEA Meeting on
Carbon Dioxide, Other Greenhouse Gases, and Related Measurement Techniques (GGMT-
2011) (Wellington, New Zealand) 2011. 67.
[10] Dlugokencky E. J. et al., “Conversion of NOAA atmospheric dry air CH4 mole fractions
to a gravimetrically prepared standard scale” J. Geophys. Res.: Atmos., 110 (2005)
D18306.
[11] Rhoderick G. C., Carney J. and Guenther F. R., “NIST Gravimetrically Prepared
Atmospheric Level Methane in Dry Air Standards Suite” Anal. Chem., 84 (2012) 3802.
[12] Brewer P. J. et al., “Preparation and Validation of Fully Synthetic Standard Gas
Mixtures with Atmospheric Isotopic Composition for Global CO2 and CH4 Monitoring”
Anal. Chem., 86 (2014) 1887.

[13] Wielgosz R. I. et al., “Final report on CCQM-P73: International comparison of nitrogen
monoxide in nitrogen gas standards (30-70) μmol/mol” Metrologia, 45 (2008) Tech. Supl.
08002.
[14] Baldasano J. M., Valera E. and Jimenez P., “Air quality data from large cities” Sci.
Total Environm., 307 (2003) 141.
[15] Skalska K., Miller J. S. and Ledakowicz S., “Trends in NOx abatement: A review”
Sci. Total Environm., 408 (2010) 3976.
[16] Spicer C. W. et al., “Evaluation of NO2 Measurement Methods for Indoor Air Quality
Applications” J. Air & Waste Manag. Assoc., 44 (1994) 163.
[17] McClenny W. A. et al., “Preparing to Measure the Effects of the NOx SIP Call– Methods
for Ambient Air Monitoring of NO, NO2 , NOy , and Individual NOz Species” J. Air &
Waste Manag. Assoc., 52 (2002) 542.
[18] Carslaw D. C., “Evidence of an increasing NO2 /NOX emissions ratio from road traffic
emissions” Atmosph. Environ., 39 (2005) 4793.
[19] Wada R. and Orr-Ewing A. J., “Continuous wave cavity ring-down spectroscopy
measurement of NO2 mixing ratios in ambient air” Analyst, 130 (2005) 1595.
[20] Fuchs H. et al., “Intercomparison of measurements of NO2 concentrations in the
atmosphere simulation chamber SAPHIR during the NO3Comp campaign” Atmos. Meas.
Tech., 3 (2010) 21.
[21] Frank P., Scaringelli A. E. O. K., Rosenberg Ethan and Bell John Paul,
“Preparation of known concentrations of gases and vapors with permeation devices
calibrated gravimetrically” Anal. Chem., 42 (1970) 871.
[22] Hughes E. E. et al., “Performance of a nitrogen dioxide permeation device” Anal. Chem.,
49 (1977) 1823.
[23] ISO, Gas analysis – Preparation of calibration gas mixtures using dynamic volumetric
methods – Part 10: Permeation method, in ISO 6145-10:2002. 2002.
[24] Flores E. et al., “Final report on international comparison CCQM-K74: Nitrogen dioxide,
10 μmol/mol” Metrologia, 49 (2012) Tech. Suppl. 08005.
[25] Flores E. et al., “Highly accurate nitrogen dioxide (NO2 ) in nitrogen standards based on
permeation” Anal. Chem., 84 (2012) 10290.
[26] Fried A. et al., “Determination of nitrogen dioxide in air compressed gas mixtures
by quantitative tunable diode laser absorption spectrometry and chemiluminescence
detection” Anal. Chem., 60 (1988) 394.
[27] Rafiu A. et al., FTIR Spectroscopy of HNO3 and NO2 Relevant to Stratospheric Wake
Analysis in The First National Student Conference: NASA University Research Centers
at Minority Institutions; 295-298; (NASA-CR-205049) 1997.
[28] Brewer P. J. et al., “Gaseous reference standards of formaldehyde from trioxane”
Talanta, 108 (2013) 83.
[29] Aoki N. et al., “Evaluation of the permeability of formaldehyde and water through a
permeation tube for the preparation of an accurate formaldehyde reference gas mixture”
Analyst, 138 (2013) 6930.
[30] CCQM-K90. [cited 2017; Available from: http://kcdb.bipm.org/appendixB/KCDB ApB
info.asp?cmp idy=1095&cmp cod=CCQM-K90&prov=exalead.
[31] Paur R. J. et al., Standard Reference Photometer for the Assay of Ozone in calibration
Atmospheres, in NISTIR (NIST, 2003) p. 65.
[32] ISO, Ambient air - Determination of ozone - Ultraviolet photometric method, 1996,
International Organization for Standardization.
[33] Viallon J. et al., “PILOT STUDY: International Comparison CCQM-P28: Ozone at
ambient level” Metrologia, 43 (2006) Tech. Suppl. 08010.

[34] Viallon J. et al., “A study of systematic biases and measurement uncertainties in ozone
mole fraction measurements with the NIST Standard Reference Photometer” Metrologia,
43 (2006) 441.
[35] Viallon J. et al., “Accurate measurements of ozone absorption cross-sections in the
Hartley band” Atmos. Meas. Tech., 8 (2015) 1245.
[36] Viallon J. et al., “Ozone cross-section measurement by gas phase titration” Anal. Chem.,
88 (2016) 10720.
[37] Griffith D. W. T., “Synthetic Calibration and Quantitative Analysis of Gas-Phase FT-
IR Spectra” Appl. Spectrosc., 50 (1996) 59.
[38] Flores E., V. J. a. W. R., Accurate measurements of nitrogen dioxide and nitric acid
mole fraction by FT-IR spectroscopy calibrated by gas standards and synthetic spectra. in
GAS 2011 (WTC Rotterdam, the Netherlands) 2011.
[39] Flores E. et al., “Final report on Pilot Study CCQM-P110: Study on the accuracy and
uncertainty of FT-IR methods calibrated with synthetic spectra for NO2 concentration
measurements” Metrologia, 50 (2013) 08011.
[40] Flores E. et al., “Final report of the pilot study CCQM-P110-B1: A comparison of
nitrogen dioxide (NO2 ) in nitrogen standards at 10 μmol/mol by Fourier transform infrared
spectroscopy (FT-IR)” Metrologia, 49 (2012) Tech. Suppl. (Issue 1A).
[41] Esler M. B. et al., “Precision Trace Gas Analysis by FT-IR Spectroscopy. 1. Simultaneous
Analysis of CO2 , CH4 , N2 O, and CO in Air” Anal. Chem., 72 (1999) 215.
[42] Esler M. B. et al., “Precision Trace Gas Analysis by FT-IR Spectroscopy. 2. The 13 C/12 C
Isotope Ratio of CO2 ” Anal. Chem., 72 (1999) 216.
[43] Rothman L. S. et al., “The HITRAN 2004 molecular spectroscopic database” J. Quant.
Spectrosc. Radiat. Transfer, 96 (2005) 204.
[44] Haaland D. M., Easterling R. G. and Vopicka D. A., “Multivariate Least-Squares
Methods Applied to the Quantitative Spectral Analysis of Multicomponent Samples” Appl.
Spectrosc., 39 (1985) 84.
[45] Flores E. et al., “Accurate measurements of nitrogen dioxide (NO2 ) and nitric acid
(HNO3 ) mole fraction by FT-IR spectroscopy calibrated by gas standards and synthetic
spectra” Appl. Spectrosc., 67 (2012) 1178.
[46] Birk M. et al., “Determination of line strengths by Fourier-transform spectroscopy” Appl.
Opt., 35 (1996) 2985.
[47] Chu P. M. et al., “The NIST Quantitative Infrared Database” J. Res. Nat. Inst. Stand.
Technol., 104 (1999) 81.
[48] Sharpe S. W. et al., “Gas-Phase Databases for Quantitative Infrared Spectroscopy” Appl.
Spectrosc., 58 (2004) 1461.
[49] Smith T. E. L. et al., Absolute accuracy and sensitivity analysis of OP-FTIR retrievals of
CO2 , CH4 and CO over concentrations representative of “clean air” and “polluted plumes”
Atmos. Meas. Tech., 4 (2011) 97.
[50] EPA, METHOD 320 Measurement of Vapor Phase Organic and Inorganic Emissions
by Extractive Fourier Transform Infrared (FTIR) Spectroscopy (U.S. Environmental
Protection Agency) 1998.
[51] Smith H. J. P. et al., FASCODE: Fast Atmospheric Signature Code (Spectral
Transmittance and Radiance), AFGL-TR-78-0081, 1978 ADA057506.
[52] Rinsland C. P. et al., “Stratospheric N2 O mixing ratio profile from high-resolution
balloon-borne solar absorption spectra and laboratory spectra near 1880 cm−1 ” Appl. Opt.,
21 (1982) 4355.
[53] Rinsland C. P. et al., “Northern and southern hemisphere ground-based infrared
spectroscopic measurements of tropospheric carbon monoxide and ethane” J. Geophys.

Res., 103 (1998) 28217.
[54] Griffith D. W. T. et al., “Intercomparison of NDSC Ground-Based Solar FTIR
Measurements of Atmospheric Gases at Lauder, New Zealand” J. Atmos. Oceanic Technol.,
20 (2003) 1153.
[55] Flores E. et al., “Calibration strategies for FTIR and other IRIS instruments for accurate
δ 13 C and δ 18 O measurements of CO2 in air” Anal. Chem., 89 (2017) 3648.
[56] Rothman L. S. et al., “The HITRAN2012 molecular spectroscopic database” J. Quantit.
Spectrosc. Radiat. Transfer, 130 (2013) 4.
[57] Griffith D. W. T. et al., “A Fourier transform infrared trace gas and isotope analyser
for atmospheric applications” Atmos. Meas. Tech., 5 (2012) 2481.
[58] Zhao C. L., Tans P. P. and Thoning K. W., “A high precision manometric system for
absolute calibrations of CO2 in dry air” J. Geophys. Res.: Atmos., 102 (1997) 5885.
[59] Zhao C. L. and Tans P. P., “Estimating uncertainty of the WMO mole fraction scale for
carbon dioxide in air” J. Geophys. Res.: Atmos., 111 (2006) D08S09.
[60] Bergoglio M. and Calcatelli A., “Uncertainty evaluation of the IMGC-CNR static
expansion system” Metrologia, 41 (2004) 278.
[61] Pierre O. et al., “Final report on supplementary comparison EUROMET.M.P-S3 of
absolute pressure standards in the barometric range from 80 kPa to 110 kPa” Metrologia,
45 (2008) 07004.
[62] Orphal J., A critical review of the absorption cross-sections of O3 and NO2 in the 240-
790 nm region (European Space Agency, Noordwijk) 2002.
[63] Galbally I. E. et al., Guidelines for Continuous Measurements of Ozone in the
Troposphere (World Meteorological Organization, Geneva) 2013.
[64] Klausen J. et al., “Uncertainty and bias of surface ozone measurements at selected Global
Atmosphere Watch sites” J. Geophys. Res.: Atmos., 108 (2003) 4622.
[65] Hearn A. G., “The absorption of ozone in the ultra-violet and visible regions of the
spectrum” Proc. Phys. Soc., 78 (1961) 932.
[66] Bass A. M. and Paur R. J., The ultraviolet cross-section of ozone I: Measurements, in
Quadriennial Ozone Symposium. Halkidiki, Greece (Reidel, Dordrecht) 1984.
[67] Tanimoto H. et al., “Intercomparison of ultraviolet photometry and gas-phase titration
techniques for ozone reference standards at ambient levels” J. Geophys. Res.: Atmos., 111
(2006) D16313.
[68] Petersen M. et al., “Relative measurements of ozone absorption cross-sections at three
wavelengths in the Hartley band using a well-defined UV laser beam” J. Geophys. Res.,
117 (2012) D05301.
[69] Janssen C., Simone D. and Guinet M., “Preparation and accurate measurement of pure
ozone” Rev. Sci. Instrum., 82 (2011) 034102.
[70] Castrillo A. et al., “Combined interferometric and absorption-spectroscopic technique
for determining molecular line strengths: Applications to CO2 ” Phys. Rev. A, 67 (2003)
062503.
DOI 10.3254/978-1-61499-818-1-121
Chemical primary reference materials:

From valine to C-peptide
R. I. Wielgosz, S. Westwood, R. D. Josephs, N. Stoppacher,
A. Daireaux and T. Choteau
Bureau International des Poids et Mesures, Département de Chimie
Pavillon de Breteuil, 92312 Sèvres Cedex, France
Summary. — The accurate measurement of organic compounds in a wide variety of

matrices is important to a great number of sectors including health, the environment,
forensics, food safety, pharma. The instruments used to make such measurements
require calibration, and for well-defined chemical entities the calibration hierarchy
leads to a pure material that has been value-assigned for the mass fraction of the
analyte within the material. Such organic primary reference materials are provided
as Certified Reference Materials (CRMs) by National Metrology Institutes. This
paper reviews the various methods that can be used to value-assign such reference
materials, both for small organic molecules such as valine and folic acid, as well as
larger molecules such as peptides.
1. – Introduction: Organic primary reference materials for analytical

chemistry
The accurate measurement of organic compounds in a wide variety of matrices is

important to a great number of sectors including health, the environment, forensics, food
safety, pharma. The instruments used to make such measurements require calibration,
and for well-defined chemical entities the calibration hierarchy leads to a pure material
121
122 R. I. Wielgosz, S. Westwood, R. D. Josephs, etc.
that has been value-assigned for the mass fraction of the analyte within the material.
Such organic primary reference materials are provided as Certified Reference Materials
(CRMs) by National Metrology Institutes. This paper reviews the various methods that
can be used to value assign such reference materials, both for small organic molecules
such as valine, and larger molecules such as peptides.
The simple mathematical description of the mass fraction of the main component of
a primary reference material is shown in the following equation:
mA
(1) wA = ,
mA + m x
where wA is the mass fraction of the main component A in the material, mA is the mass
of A in a gravimetrically defined aliquot of the material, Σmx is the combined mass of x
minor components (impurities) present in the same aliquot.
It can be expressed in the SI units of kg/kg but for numerical convenience, wA is
reported in units of milligram per gram (mg/g).
Purity may also be reported as the amount of substance fraction (mol/mol) of the
main component in a material, although this is less frequently encountered in quantitative
organic analysis.
Measurements of the mass fraction content of the main component A in an organic
material which contains x independent minor components (impurities) can be obtained
by a number of approaches. Two of these will be described here in detail:
i) The mass balance approach, which proceeds with determination of the mass fraction
(w1 , w2 , . . . , wx ) for each of the x independent impurities in the material followed
by subtraction of the summation of impurities from the limit value;
ii) Quantitative NMR (qNMR) which relies on the direct determination of the mass
fraction of the main component.
In addition for peptides, considered as large organics in this paper, methods which quan-
tify smaller chemical entities within the peptide (amino acids or elemental analysis) can
be used to determine the mass fraction content of the peptide in a material. These
methods will also be described.

Valine is used as an example of how the mass balance approach can be applied to
small molecules; the application of qNMR is described for folic acid; Angiotensin and
c-peptide are used as examples to describe the application of methods to value assign
the purity of pure peptide materials.
2. – The mass balance purity assignment method applied to valine
For the mass balance technique, the individual classes of impurity that generally
require assessment and quantification are:
1) related structure impurities (wRS )
Chemical primary reference materials: From valine to C-peptide 123
2) residual organic solvent (total volatile organic compounds) (wOS )
3) water (wW )
4) non-volatiles (wN V. ), with potential contributions from either or both:
a) inorganic impurities;
b) non-volatile oligomeric organics.
When the mass fraction assignments of impurities are orthogonal and adequately cover
the total impurities present in the material, the equation for assignment of the mass
fraction content of the main component A (wA ) becomes, where all components are in
units of mg/g:
(2) wA = 1000 − (wRS + wW + wOS + wN V ).
The combined standard uncertainty associated with wA is given by simple quadratic

combination of the uncertainties associated with each contributing impurity assignment:

(3) u(wA ) = u(wRS )2 + u(wW )2 + u(wOS )2 + u(wN V. )2 .
In the following sections the approaches used to quantify each class of impurity in valine
are summarized, with the validation of the approach described elsewhere [1-7].
.
2 1. Measurement of mass fraction of related structure impurities (wRS ) in valine
.
2 1.1. Experimental and method description. The qualitative identification and quantifi-
cation of related structure components present in a sample of L-valine was undertaken by
tandem high performance liquid chromatography using a charged aerosol detector (LC-
CAD) or a linear Q-Trap MS/MS detector (LC-MS/MS). The structural assignment of
the main component and the presence of the major impurities identified by LC methods
(alanine, leucine and isoleucine) were separately confirmed by 1 H NMR spectroscopic
analysis of a D2 O solution of the valine material.
Related structure impurities present in the comparison material at levels in excess

of 0.1 mg/g, as summarized in table I, were alanine, leucine, isoleucine, α-aminobutyric
acid and methionine. The impurities were quantified by LC-MS/MS using external cal-
ibration against a reference standard for each compound and the assigned value was
confirmed by a separate LC-CAD analysis. A method for the detection of D-valine as
a potential impurity in L-valine was developed using derivatization of the amino acid
as the chloroformate and subsequent quantification by GC-FID using a chiral GC col-
umn. In addition to demonstrating that D-Valine was not present in the material above
the method limit of detection (< 0.05 mg/g) it was also able to confirm the presence
of L-alanine, L-leucine and L-isoleucine at levels consistent with the values assigned by
LC-MS/MS.
Fig. 1. – Chromatogram of the amino acid standard solution mix with CAD detection. Notice
the baseline separation of valine from all other components.
The characterization of the valine material commenced with the development of HPLC
methods capable of resolving a range of potential amino acid impurity components.
Gravimetrically prepared test solutions consisting of mixtures of nine amino acids (Ile,
Leu, α-AB, γ-AB, Met, Phe, Pro, Tyr, Val) were used for the method development
studies and to establish the performance characteristics of an LC-CAD method. The
limit of detection (LOD), limit of quantification (LOQ) and linearity of the response of
each component to the CAD at mass fraction levels in the range (400–1500) ng/g were
established. A representative example of the chromatographic separation achieved with
the optimized HPLC-CAD method is shown for the test mixture in fig. 1.
Due to its relatively low sensitivity the CAD method was only suitable for the charac-
terization of a potential impurity present in valine at levels in excess of 0.5 mg/g. However
Table I. – Related structure impurity assignments in valine.
Component Mass fraction in valine (mg/g) u (mg/g) urel (%)

Alanine 2.68 0.12 4.5
Isoleucine 1.76 0.08 4.5
Leucine 2.02 0.06 3.0
α-Aminobutyric acid 0.32 0.025 7.8
Methionine 0.02 0.0013 8.1
Combined related 6.80 0.16 2.4
structure impurities (wRS )
the CAD provided a “universal detector” capable of responding to each non-volatile or-
ganic component present in the column effluent above the method detection limit. When
the same samples were analysed using a diode-array UV detector (DAD) connected in
series with the CAD, as expected, only the components containing a UV-chromophore
(Phe and Tyr) were detected.
A separate column type and chromatographic method were used to achieve the base-
line separation from valine of nine out of ten selected amino acids (Ala, Ile, Leu, α-AB,
γ-AB, Phe, Pro, Tyr and HPro) and their quantification by LC-MS/MS using a Q-Trap
detector. The tenth amino acid (Met) co-eluted with valine under the conditions used,
but quantification could still be achieved in this case based on Met –specific MRM tran-
sitions. Full scan (total ion current) mode was used for the initial detection of each
component with subsequent quantification based on unique MRM transitions identified
for each component. As for the LC-CAD method, the LOD, LOQ and linearity of
the response of each component using the LC-MS/MS method was suitable and could
be achieved at lower at mass fraction levels (30 ng/g–250 ng/g). The higher sensitivity
of detection by MS/MS compared to that achievable using the CAD meant the LC-
MS/MS method could detect potential impurities present in valine at levels in the range
0.05 mg/g–0.2 mg/g, up to an order of magnitude better than the CAD. Both methods
were applied directly to solutions of the native material and have the important advan-
tage for accurate quantification of not requiring an initial chemical derivatization step.
As shown in table I the main related substance impurities present in the material
were alanine, leucine and isoleucine. In addition smaller levels of α-aminobutyric acid
and methionine could also be detected. The identity of each impurity was established by
their co-elution and identity of response to the CAD and MS/MS when compared with
authentic standards of each compound.
The value for the combined mass fraction of related structure impurities (wRS ) in
the valine material is 6.80 mg/g with an associated standard uncertainty in this value of
0.16 mg/g. The mass fraction of each individual related structure impurity was assigned
from the response obtained by the sample using LC-MS/MS quantification against the
response from external calibration curves generated using authentic standards.
Figure 2 shows overlaid MRM chromatograms, specific for each component identified
and quantified in the valine material. In addition to the main peak corresponding to
valine, diagnostic MRM ions are observed for Ala, α-ABA, Ile and Leu. A small MRM
diagnostic for Met can be observed co-eluting with the large Val peak.
The relative response using the CAD detector, where the peak area of each resolved
component was normalized against the total peak area and adjusted for individual re-
sponse factors, provided an independent check of the values assigned through LC-MS/MS.
In addition high-field NMR spectroscopy showed the presence of the alanine, leucine and
isoleucine impurities and allowed for a semi-quantitative determination of the amount of
each. These values were also consistent with the values assigned by LC-MS/MS quan-
tification.
The potential for contributions of related structure impurities invisible to the detector,
or that co-elute with the main peak, or that are present below the limit of detection
Fig. 2. – MRM chromatogram of the L-valine sample at a mass fraction of main component of
about 100 μg/g.
but cumulatively are significant, was also assessed. The use of two independent sets of
chromatographic conditions, each capable of resolving the main component from longer
and shorter retention time potential impurities, provided initial confidence that the co-
elution of related impurities with the valine component was unlikely to be a significant
issue. This assumption was independently confirmed by NMR and by high-resolution
MS/MS studies using direct infusion of a solution of the valine material into an Orbitrap
system. In addition the results of C/H/N elemental analysis of the material did not
deviate significantly from the theoretical values based on the molecular formula of valine.
Taken together these results were the basis for the assignment of the level of “non-
detectable” related structure impurities in this case as zero with negligible associated
uncertainty.
.
2 1.2. Metrological uncertainty and SI-traceability of mass fraction of related structure
impurities. The external calibration approach to quantify related structure impurities us-
ing LC-MS/MS quantification allows for sensitive and selective quantification of minor
impurities in the presence of a structurally related major component. The relative stan-
dard uncertainty in the mass fraction assignment of minor components are comparable to
those that can be obtained for the main component under the same circumstances — in
the range 2–5% for the main impurities. Larger relative uncertainties are associated with
the quantification of impurities present at levels below 1 mg/g but these do not make
a significant contribution to the overall absolute standard uncertainty for the assigned
value for the main component.
These levels of measurement uncertainty in the individual impurity quantifications
lead to a smaller absolute uncertainty in the contribution to the uncertainty due to the
mass fraction of the related structure impurity component than would be associated
with direct assignment of the main component by external calibration against a refer-
ence standard. To establish SI-traceability of the result for the total related structure
impurities, it is necessary to demonstrate that the result can be related, within asso-
ciated uncertainty, to stated references through an unbroken chain of calibrations. In
this context the stated references are the individual reference materials used to generate
the external calibration curve for each impurity quantification. In this case a reference
material was available for each identified impurity component. The traceability of the
mass fraction for each impurity was to the reference material used for the external cali-
bration. Traceability requires a fit-for-purpose, SI-traceable mass fraction assignment for
each reference material used as the basis for each external calibration. For the related
structure impurity quantifications described in this paper, the mass fraction of the main
component in the individual reference materials in the analysis certificates provided by
each producer was not reported explicitly as an SI-traceable mass fraction with an as-
sociated uncertainty. Additional characterization was required by the BIPM to confirm
the purity of each reference material and the uncertainty associated with the assignment.
The extent of additional characterization required was a function of the absolute amount
of the impurity in the material being characterized.
For the impurities present at levels below 1 mg/g in a material, even a large (10%)
relative uncertainty in the assigned value for the impurity does not make a significant
contribution to the uncertainty in the assigned value of the combined related structure
impurity content. Traceable results can be assumed to be obtained from external cali-
brations based on these materials provided chromatographic analysis of the material is
consistent with the provider’s assigned value.
For the impurities present at levels above 1 mg/g, in-house checks by qNMR and Karl
Fischer titration in addition to a confirmatory chromatographic analysis consistent with
the provider’s assigned purity value provided sufficient confidence.
.
2 2. Measurement of mass fraction of water (wW ) in valine. – The Karl Fischer (KF)
titration method is the principle source of information on the water content of all or-
ganic compounds characterized by the BIPM. Direct addition to the titration cell or
transfer of the water content from a solid sample via a heated sample oven were used as
appropriate. The difference between sample and blank values provided the water con-
tent estimate for the material under examination. Where heated sample oven transfer
is used the influence of oven temperature must be investigated to establish the mini-
mum temperature for complete release of water from the sample but before generation
of water from chemical breakdown of the compound. In the case of valine, heated oven
transfer up to an oven temperature of 220 ◦ C indicated no significant release of water
relative to the background levels seen when using a blank vial. As the temperature was
increased above 220 ◦ C water release began to occur but this was consistent with chem-
ical breakdown of valine by condensation of amine and acid functional groups forming
amide bonds, releasing a molecule of water. The water content estimate obtained by KF
titration was consistent with results obtained by thermogravimetric analysis (TGA) and
elemental analysis. The TGA data was also consistent with irreversible breakdown of
valine commencing at temperatures above 220 ◦ C rather than the release of bound water
from the material.
The general approach for the calculation of water content in a sample of an organic
solid by Karl Fischer titration using the oven transfer approach is
mV al(H2 O)
(4) wW = ,
mV al
where wW = mass fraction of water content in valine, mV al(H2 O) = mass of water by

sample oven KFT for a given mass of valine, mV al = mass of valine, and
(5) mV al(H2 O) = ms(H2 O) − mblk ,

yKF T
(6) ms(H2 O) = R ∗ T ∗ .
a
ms(H2 O) , the observed water content for a given mass of valine by the sample oven transfer
KFT determination, includes a contribution from the water content of the sample vial.
Correction by mblk , the assigned value for water content of a blank vial, is assumed
equal to the contribution to ms(H2 O) from the sample vial. R is a recovery factor for
oven transfer of water content from valine and the sample vial and is assumed to be
as determined for a suitable reference material of assigned water content. T is a factor
for the uncertainty of the assigned value of the NIST SRM 1290 which was the primary
traceability source for the water content estimation. yKF T is the total water estimate
obtained directly by KFT and a is the slope of a linear regression determined separately
using the NIST SRM 1290 reference material for the relation between gravimetric amount
of water added and the value observed by Karl Fischer titration.
In the specific case of valine, there was no significant difference found between ms(H2 O)
and mblk . That is, the values obtained with samples containing aliquots of the valine
were not significantly different from those obtained using blank vials.
In this simple case, the water content of valine was assigned a value of 0 mg/g with a
potential one-way bias of 0.16 mg/g, equivalent to a triangular distribution from zero to
the estimated limit of detection of 0.28 mg/g.
.
2 3. Measurement of the mass fraction of residual organic solvent (wOS ) for valine.
– Residual solvent content (wOS ) was determined by direct injection of a solution of the
material onto a capillary GC column suitable for analysis of volatile organic solvents
and fitted with a guard column, the column outlet being connected to a mass selective
detector.
Two independent solutions were analysed, one in methanol and the other in DMSO.
The large difference in retention time for the solvent peaks between the two solutions
allows for control for the possibility of residual solvent peaks being obscured by one
or other of the sample solvent peaks. No residual solvent was detected by the mass
analyser. Analysis by 1 H NMR independently verified the absence of any significant
organic solvent residue in the material. In this case the wOS was assigned a value of
zero with an associated positive uncertainty of 0.03 mg/g, equivalent to a conservative

estimate of the limit of detection for the GC-MS analysis.
.
2 4. Measurements of the mass fraction of non-volatile materials (wN V ) in valine.
– The presence of non-volatile materials were assessed by combination of data from
complementary methods. The inorganic content of the compound was determined by
ICP-MS, allowing for individual determination of levels of the common metallic elements
(Na, Mg, Al, Si, Fe, K, Ca). None were found to be present at levels above their
individual limits of detection which are in the range 20 μg/g to 50 μg/g for the equipment
and method used. In addition the results of microanalysis for C, H and N content were
consistent with the empirical formula of the main component and there was no significant
mass residue remaining after oxidative combustion of the material at high temperature
in the TGA. The results for non-volatile residue determinations by TGA for valine were
indistinguishable from blank analyses. In this case the wN V was assigned a value of
zero with an associated positive uncertainty of 0.28 mg/g, equivalent to a conservative
estimate of the combined limits of detection for the ICP-MS and TGA analysis.
.
2 5. Assignment of the mass fraction content of valine. – Applying the mass assign-
ments obtained as described above into eqs. (2) and (3) gives the assigned value for the
mass fraction of valine in the material (wV al ) and its expanded uncertainty (k = 2) of
993.2 ± 0.72 mg/g.
3. – qNMR applied to pure material standards of folic acid
.
3 1. Introduction to qNMR. – Nuclear magnetic resonance (NMR) is a physical phe-
nomenon resulting from the ability of any atomic nucleus containing an odd number of
protons and neutrons (i.e. atoms having a spin quantum number I > 0) to absorb and re-
emit electromagnetic radiation when placed in a magnetic field. 1 H (proton) NMR is by
far the most critical and widely used application. When placed in an external magnetic
field, a nucleus with I = 1/2 can take up one of two quantum energy states — either in a
lower-energy state aligned in the same direction as the applied field or in a higher-energy
state aligned against it. An NMR experiment uses a radio frequency pulse to excite all
nuclei in the sample, followed by a time domain free induction decay (FID) relaxation
of the bulk radiofrequency emission response as the system returns to its equilibrium
state. This time domain bulk response is amplified and detected. The acquired FID
is then subjected to Fourier transformation (FT) to convert it into a frequency domain
spectrum.
1
H NMR spectroscopy provides an enormous amount of information about the two-
and three-dimensional structure of organic compounds. This arises because nuclei in non-
equivalent magnetic environments experience a local magnetic field that differs to a small
but detectable level from the applied field. In the case of hydrogen atoms attached to the
carbon skeleton of an organic compound, this local difference leads to distinct, separate
NMR absorption frequencies for each set of non-equivalent atoms. The information
available from the set of signals that comprise the 1 H NMR spectrum for an organic
compound includes the:
– chemical shift (δ, ppm relative scale) of each signal, indicating the number and
nature of different chemical environments within a molecule.
– coupling constants (J, Hz) for each signal due to adjacent non-equivalent 1 H nuclei.
This leads to fine structure (doublets, triplets, etc) for a given signal and shows
the number and arrangement of neighboring atoms within the molecule.
– peak shape (line width at half height), indicating the nature of dynamic motion
and/or chemical exchange of nuclei within the molecule.
– peak intensity (integral), from which the number of equivalent nuclei giving rise to
each signal can be assigned.
The area of a peak in an NMR spectrum of a pure compound is directly proportional to

the number of nuclei giving rise to that peak. Because the number of nuclei is in turn
related to the molar concentration of the compound, the amount of an organic analyte
in solution can be assigned from the relation:
Sa Na × ca
(7) = ,
Sb Nb × cb
where Sa , Sb is the peak area integrals of resolved 1 H NMR signals a and b, Na , Nb is

the number of nuclei giving rise to 1 H NMR signals a and b, ca , cb is the concentration
in solution of the compound giving rise to 1 H NMR signals a and b.
In the case of signals arising from the same molecule, where ca and cb are equal and
thus cancel out, the signal ratio provides the relative ratio of 1 H atoms giving rise to
each signal in a molecule.
If the two NMR signals arise from a target analyte of unknown purity and from a
separate reference compound of known purity, both present in the solution at known
concentrations, the relation can be rearranged and a quantitative NMR (qNMR) mea-
surement can be used to assign the absolute purity of the target analyte according to the
measurement equation:
SA NIS MA mIS
(8) PA = · · · · PIS .
SIS NA MIS mA
In this equation MA and MIS are the molar masses of the target analyte and the internal
standard, mA and mIS the mass of the target analyte and the internal standard used
for sample preparation, PIS is the purity of the internal standard, NA and NStd are the
number of nuclei giving rise to the signals due to analyte and internal standard and SA
and SIS are the integrated areas of the signals they give rise to.
Accurate qNMR measurements by the internal standard approach require the follow-
ing:
– purity of the reference compound is independently assigned,
– gravimetric operations are undertaken with suitable levels of trueness and accuracy,
– NMR signals do not overlap each other or signals from impurities or the solvent,
– full potential peak area response of both signals are achieved by appropriate choice
of the NMR spectrum acquisition and processing parameters.
The rapid increase in qNMR applications for organic purity assignment is reflected in the
recent publication of several reviews of the technique and the published literature [8-12].
qNMR has important advantages over conventional mass balance methods or external
calibration using high-resolution chromatographic methods for the purity assignment of
both chemical standards and technical grade material. These include a significant re-
duction in the time, resources and amount of material required to perform the purity
assignment along with the ability to potentially quantify both the main component and
organic impurities simultaneously. In the case of valuable materials available in limited
amounts, it is possible to recover the analyte if desired. In addition a single reference
material can be used for the quantification of a range of compounds, avoiding the need
for value assigned CRMs corresponding to each individual compound to be quantified,
as is required for chromatographic approaches using external calibration. As the under-
lying physicochemical mechanism is completely different from other quantitative organic
analytical techniques, it also provides a powerful independent method for confirmation
of a purity assignment obtained by traditional methods or to flag an otherwise unde-
tected bias.
Apart from the high capital cost of the equipment, the main technical disadvantage
of the techniques is the relative lack of sensitivity and limited resolution of NMR sig-
nals compared to high-resolution chromatographic methods using mass selective detection
techniques. For complex organic structures containing related structure (particularly iso-
meric) impurities, it may not be possible to obtain a NMR quantification signal unique
to the main component. Key factors in obtaining an appropriate qNMR spectrum are
the nature of the bulk excitation pulse, which must induce an excitation response that
is uniform relative to the number of nuclei contributing to each signal over a suitable
window of the spectral region, the acquisition time for collection of the time domain data
to maximize signal without adding excessive noise and the length of the relaxation delay
between pulses. The relaxation delay process follows an exponential decay curve and
must be sufficient to allow for complete reestablishment of the initial equilibrium state.
The digital signal amplification process can also introduce a small but systemic bias.
Finally the signal to noise ratio must be sufficiently high to make variation in peak area
due to contribution from the baseline noise negligible. After the Fourier transformation
is applied to the raw acquired FID signal, further critical parameters for accurate quan-
tification include the establishment of the signal baseline and selection of an appropriate
integration range for each quantified signal.
Fig. 3. – Chemical structure of folic acid.
Fig. 4. – NMR spectrum of folic acid (FA) and maleic acid (MA), recorded with optimized
instrument parameters.
.
3 2. Application of qNMR for purity measurements of folic acid. – Folic acid provides
a challenging example for the application of qNMR. The structure of folic acid is shown in
fig. 3, with the hydrogen atom attached to the heterocyclic ring system of folic acid (H-7)
used in the qNMR measurement indicated. Samples containing a nominal concentration
of 5 mg folic acid (FA) and 2 mg of maleic acid (MA) as an internal standard, were
dissolved in 1 mL of DMSO-d6 . Complete dissolution of the chemical powder without
resorting to ultrasonication, required up to 48 h at room temperature. Fifty microlitres
of D2 O was added to each sample in order to reduce the size of exchangeable hydroxyl
peaks in the NMR spectra, then aliquots of 800 μL of the sample solution were transferred
into NMR tubes for measurement. NMR measurements were carried out with a JEOL
ECS-400 NMR spectrometer. Figure 4 shows a NMR spectrum of FA and MA, recorded
with the instrument under optimized conditions.
Table II. – Example of an uncertainty budget for qNMR measurement on one sample of folic
acid material.
Term Abbreviation Unit Values Uncertainties rel u

Purity of Internal Standard (MA) PIS [mg/g] 999.9 0.145 0.00014501
Number of 1H nuclei of Internal NIS 1 2 0
Standard (MA)
Number of nuclei of analyte (FA) NA 1 1 0
Peak area ratio Analyte/Internal SA /SIS 1 0.299034 0.000825 0.00275938
Standard
Molecular weight of Internal MIS [g/mol] 116.0719 0.002150 1.8519E-05
Standard (MA)
Molecular weight of Analyte (FA) MA [g/mol] 441.3973 0.004629 1.0488E-05
Weighed amount of Analyte (FA) mA [mg] 5.28177 0.001240 0.00023477
Weighed amount of Internal mIS [mg] 2.12357 0.001240 0.00058392
Standard (MA)
Purity of Analyte (FA) PA [mg/g] 914.3155

Combined uncertainty for the purity u(PA ) [mg/g] 2.591202 0.283%
of analyte
The NMR spectrum of folic acid is considerably more complex than valine, but from
the various NMR signals arising from the non-equivalent hydrogen atoms present in the
molecule, the resonance due to the hydrogen atom attached to the heterocyclic ring
system of folic acid (H-7), at chemical shift δ = 8.6 ppm, was selected as suitable for
qNMR measurement. The unique proton signal from maleic acid (MA) at δ = 6.2 ppm
was used as the internal standard.
The qNMR estimate is assumed to include contributions from two isomeric forms
of folic acid, identified and quantified in the material by LC-HRMS and LC-UV and
estimated to be present at 3 mg/g total in the material. The qNMR values reported for
“FA” are thus the combined value for native FA and its isomeric impurities. Subtraction
of 3 mg/g from the values for “FA” was used to correct for the contribution due to
isomeric impurities. In this case the folic acid content assigned by qNMR measurements
was 912.9 ± 4.4 mg/g, in good agreement with the result obtained by a separate mass
balance approach of 911.1 ± 4.3 mg/g. An example of an uncertainty budget for qNMR
measurement of folic acid is given in table II.
4. – Methods for large organics: peptides
In general, the purity of peptides can be assessed by using the full mass balance
approach. However, a full mass balance approach requires relatively large quantities of
peptide material, and therefore other approaches, as summarized in fig. 5, may be used).
Fig. 5. – SI traceable peptide purity value assignment approaches.
A common theme to all approaches is the identification and quantification of peptide

impurities. The methods for doing this will be described first, followed by a description
of the Peptide Impurity Corrected Amino Acid (PICAA) method for determination of
the mass fraction of the main peptide component in the material. The PICAA method
generally requires only a few milligrams of peptide material. It involves quantification
of constituent amino acids (AAs) following hydrolysis of the material and correction
for AAs originating from inherent structurally-related peptide impurities. Identification
and quantification of structurally-related peptide impurities by liquid chromatography
coupled to (high-resolution or tandem) mass spectrometry (LC-hrMS or LC-MS/MS) is
required to obtain the most accurate results. Exact matching double liquid chromatog-
raphy — isotope dilution tandem mass spectrometry (LC-IDMS/MS) can be used for
amino acid analysis (AAA). Traceability of the AAA results is to pure amino acid certified
reference materials (CRMs) or directly to NMI purity capabilities for amino acids. NMI
capabilities to determine the purity of L-valine were recently assessed in an international
inter-comparison study [13].
.
4 1. High-resolution mass spectrometry, identification and quantification of impurities
in peptides
.
4 1.1. Typical configuration of a mass spectrometer. Mass spectrometers generate a
beam of gaseous ions from a sample that are separated by mass and charge ratio (m/z),
and finally display a mass spectrum [14,15]. All these instruments share basic structural
components, listed in their technical order: a system for sample introduction (interface),
an ion source (e.g. ESI, APCI) for generation of ions from sample molecules, one or more
mass analysers (e.g. quadrupole, Orbitrap) for the separation of ions according to their
m/z ratio, an ion detector/collector (e.g. electron multiplier) and finally a computer for
controlling the instrument and processing of analytical data. Each mass spectrometer
Fig. 6. – The process of electrospray ionization (ESI).
also requires an efficient pumping system for the generation of a high vacuum inside the
instrument (ranging from approx. 10−2 to 10−6 Pa), in order to reduce as far as possible
unwanted collisions of analyte ions with other ions and neutral gas atoms, so that the an-
alyte ions can reach the detector [16]. Mass spectrometers can be operated as standalone
instruments with direct introduction of the sample into the interface (e.g. syringe infu-
sion) or be used in hyphenated analysis systems [17,18] (e.g. liquid chromatography-mass
spectrometry (LC-MS), gas chromatography-mass spectrometry (GC-MS)).
In single-stage mass spectrometry (MS) either an MS full scan or a Single Ion Moni-
toring (SIM) experiment is carried out in a single mass analyser. Provided that analyte
molecules undergo a soft ionization (e.g. by ESI) without fragmentation in the ion source,
MS delivers information of the masses and charges of intact ions. In contrast, tandem
mass spectrometry [19] includes two (MS/MS) or more stages of mass spectrometric
analysis (MSn ).
Between each of these MS stages, the ions are subjected to activation by input of
energy, which leads to their fragmentation. Ion activation can occur by applying various
techniques such as collision induced dissociation (CID) where analyte ions are colliding
with molecules of an inert gas such as nitrogen or helium. Mass spectrometers with
two or more physical mass analysers are referred to as tandem mass spectrometers (MSn
separated in space). However, single-stage mass spectrometers containing just one mass
analyser of a particular type, the ion trap, are also capable of doing MSn experiments
sequentially (MSn separated in time).

.
4 1.2. Electrospray ionization (ESI). This ionization technique was developed in the
1980s for the mass spectrometric analysis of large proteins [20, 21], but due to its soft
mode of action (largely non-destructive ionization), ESI has become very popular, es-
pecially for the MS analysis of many polar, non-volatile and thermolabile organic com-
pounds. Consequently, this technique is nowadays widely used for the mass spectrometric
investigation of e.g. biomolecules (such as peptides, proteins, nucleotides, carbohydrates),
pharmaceutic drugs and small polar molecules [20]. The ESI process comprises several
sequential steps (fig. 6): A liquid (often the LC effluent), including the dissolved ana-
lyte molecules is pumped through a metal capillary inside an ion source at atmospheric
pressure [22]. The flowrate usually ranges between 1 μL/min and 1 mL/min for conven-
Fig. 7. – MS data recorded at low (left), medium (middle) and high (right) mass resolution.
tional ESI. The capillary is held at a high voltage, e.g. 4000 volts, and the desired polarity
for generating either positive or negative ions. A spray of microscopically small charged
droplets is generated, consisting of ions which undergo iteratively a series of solvent
evaporation, increase of the charge density at the droplets’ surface, spontaneous decay
of droplets (Coulomb explosions) into microdroplets carrying much fewer ions, until the
droplets have got a size of few nanometers, and finally enter the mass analyser of the
mass spectrometer [23, 24].
.
4 1.3. Mass analysers and mass resolution. This component of a mass spectrometer
separates ions coming from the ion source by their mass (m)-to-charge (z) ratios (m/z).
Another important parameter in MS is the mass resolving power or mass resolution
(R), i.e. the ability of the mass analyser to separate signals of two ions with closely
related m/z values. R is expressed by the formula R = m/Δm, where m stands for the
ion mass and Δm for the full width of a mass signal (“peak”) at half of its maximum
height (FWHM). If an ion peak at m/z 800 has a width of 0.4 at 50% of its height,
then its RFWHM is 2000 [25]. The magnitude of mass resolving power needed in MS
work depends on the analytical task: For quantitative purposes in MS/MS a lower R
value (broader peaks) may be tolerated, whereas for the qualitative discrimination of two
ions with closely related m/z values higher R values (peaks as narrow as possible) are
advantageous.
Currently there are various types of mass analysers available, which provide analysis
capabilities at different mass resolutions: Ion traps (IT) [26] and quadrupoles [27-29] op-
erate at low-to-medium mass resolution (1000–10 000), the time-of-flight (TOF) [30] and
the Orbitrap [31] at high mass resolution (10 000–240 000), and the Fourier-Transform
Ion-Cylotron-Resonance (FT-(ICR) at ultrahigh mass resolution (1 000 000) [32]. To vi-
sualise the concept of mass resolving power, mass spectra of a QTrap4000 and an LTQ-
Orbitrap XL have been compared for the same compound: the triply positive charged
molecular ion ([M+3H]3+ of the peptide Angiotensin I (fig. 7)).
Fig. 8. – The ion path of an LTQ-Orbitrap XL tandem mass spectrometer.
Narrower peaks in the spectrum allow for easier distinction of two very closely related
masses of e.g. an analyte and its impurity. Also, a mass can be assigned to a possible
molecular formula with high mass precision/small deviation.
Currently the BIPM runs two hybrid tandem mass spectrometers: A QTrap4000TM
from AB-Sciex, which includes two quadrupoles separated by a central collision cell,
and an LTQ-Orbitrap XLTM from Thermo Fisher Scientific, whose ion path is com-
posed amongst other elements of a linear trap quadrupole (LTQ) and of an Orbitrap
analyser (fig. 8). Inside the instrument, the ions are first stored in the LTQ, axially
ejected, trapped in the C-trap, temporarily transmitted to the HCD collision cell for
fragmentation if required or directly injected into the Orbitrap mass analyser for hrMS
analysis. Filtering of precursor masses, MS fullscan at low resolution and MS/MS or
MSn fragmentation can also take place in the LTQ.
.
4 1.4. The Orbitrap mass analyser. The Orbitrap represents a relatively new electro-
static mass analyser, which has been invented by A. Makarov in the year 2000 [31]. It
consists of a spindle-shaped central electrode and a barrel-shaped outer electrode. The
ions are injected tangentially into the hollow space between the spindle and the barrel
electrode, where they rotate in the electric field in stable orbits around the spindle elec-
trode due to a potential of this inner electrode, which is of opposite sign as the charge of
the ions. Additionally, they move back and forth (harmonically) along the z-axis. The
frequency of these movements along the spindle is indirectly proportional to the square
root of the m/z values of the oscillating ions. Comparably to an FT-ICR-MS, an image
current is detected at the barrel electrode and the resulting signal converted to a mass
spectrum by Fourier transformation by a computer.
Depending on the time span required per mass scan, the mass resolving power of a
most recent Orbitrap model amounts up to RFWHM 240 000, which lies between the R
values of TOF- and FT-ICR-MS analysers. Due to its wide mass range (up to 4000 Da)
it is equally suitable for the analysis of small molecules, as well as for the investigation
of peptides and proteins in proteomics [31].
.
4 2. High-resolution mass spectrometry of Angiotensin I . – The linear oligopeptide An-
giotensin I is composed of 10 amino acids, with the sequence (from amino to carboxy ter-
minus): aspartic acid-arginine-valine-tyrosine-isoleucine-histidine-proline-phenylalanine-
$VS $UJ 9DO 7\U ,OH +LV 3UR 3KH +LV /HX
Fig. 9. – Amino acid sequence of the peptide Angiotensin I.
histidine-leucine, (or shorter: Asp-Arg-Val-Tyr-Ile-His-Pro-Phe-His-Leu, or very short:

D-R-V-Y-I-H-P-F-H-L; fig. 9). Its elemental formula is C62 H89 N17 O14 and its molecular
weight is 1295.68 g/mol.
Being an inactive prohormone in the renin-Angiotensin-aldosterone system in the
body, Angiotensin I is converted by enzymes to shortened and biologically active an-
giotensin peptides ANG II, III, IV, . . .. Among these peptides, ANG II exerts several
biological effects to control the water and salt balance in the body and to increase the
blood pressure.
A mass spectrum was recorded from Angiotensin I at high mass resolution (60 000
FWHM) in the MS fullscan mode (scanning from low to high mass) at an LTQ-Orbitrap
XL and it shows some charge states of the positively charged molecular ion of this peptide
(fig. 10). Typical for ESI, often more than one charge state is visible for a large molecule
(peptide, protein) in the same spectrum. The insert shows the isotope pattern of the
[M + H]+ ion.
.
4 2.1. Sequencing of Angiotensin I using MS/MS. Tandem mass spectrometers are used
to obtain information from the structure of molecules. Particular ions of a compound
Fig. 10. – hrMS spectrum of Angiotensin I.
Fig. 11. – MS/MS fragmentation behaviour of a peptide.
(e.g., a peptide) are selected by the first mass analyser (and all other ions are removed).
The selected ions, also called precursor ions, are transferred to a collision cell, inside
which the vacuum is slightly reduced due to the presence of molecules of an inert gas
such as helium. The precursor ions collide with the gas molecules, take up a lot of energy,
become instable and fragment. Subsequently, the resulting product ions are scanned by
the second mass analyser according to their m/z ratios. Finally the ions create a signal
at the detector, which is strongly amplified.
Peptides and proteins show a characteristic behaviour in this fragmentation, which
leads to the generation of defined product ions [33, 34] containing the amino acids, of
which they have been composed (fig. 11). These ions usually are members of the series
a, b, c, x, y and z, and they have different chain lengths (anything between a single
amino acid and the intact peptide). The mass difference between two fragments of the
same series, e.g. between y2 and y3 , indicates the mass difference of an amino acid that
is incorporated inside the peptide.
In an experiment, the full amino acid sequence of the peptide Angiotensin I was de-
termined by MS/MS using the Orbitrap to scan the fragment ions. For better readability
the m/z values have been limited to 2 decimal places (fig. 12).
Fig. 12. – MS/MS sequencing of the peptide Angiotensin I.
.
Fig. 13. – Characterization of the impurity Angiotensin (2-10) by LC-hrMS (a), with hrMS
spectrum of the molecular ions recorded at RT 30.21 min (b) and MS/MS sequencing of the
peptide (c).
.
4 2.2. Identification of impurities in ANG I peptide material. LC-hrMS and LC-hrMS/MS
analysis of a candidate reference material of the peptide Angiotensin I have revealed the
presence of a structurally related impurity, a peptide of nine amino acids length and
missing the first amino acid at the amino terminus and hence referred to as ANG (2-
10). The extracted ion current chromatogram in fig. 13a) specifically displays only the
signal generated from all ions of the m/z 394.55746, corresponding to the diagnostic
triply charged molecular ion of ANG (2-10). As in ESI some degradation of the analyte
molecule may also occur in the ionization process (leading to the artificial formation of so-
called in-source fragments), care has to be taken during the experimental work: Signals of
smaller m/z, which elute from the HPLC column at the same chromatographic retention
time (RT) as the analyte compound of interest, should not be misinterpreted as real
impurities. Only if they appear at a different RT in the chromatogram, they can be
considered as impurities, which are physically present in the sample material. This was
the case for ANG (2-10), that elutes at a different RT (30.21 min) than ANG I (35.0 min).
If a high concentration of the peptide is analysed for the possible presence of low-
abundant impurities, the mass spectrometer has to be protected from a possible contam-
ination with the main compound. This can be achieved by preventing the LC effluent
Fig. 14. – Impurities detected in ANG I peptide material using the Orbitrap.
from temporarily entering the MS instrument during the elution of the chromatographic
peak of the main compound (peak “cut-off”). Then, the diagnostic molecular ions of
ANG (2-10), shown in fig. 13b were detected with small mass deviation (−0.1 until
−2.6 ppm) with regard to their theoretical masses. Finally, an MS/MS spectrum has
verified the presence of the impurity peptide (fig. 13c).
ANG II, ANG (2-10) and three ANG I isomers were found to be the major
structurally-related impurities present in the ANG I material., using the same LC-hrMS/
MS methodology (fig. 14). The occurrence of isobaric impurities of ANG I is caused by
incorporation of the isobaric amino acids isoleucine (Ile) and leucine (Leu) at the 5th
and 10th position in the amino acid sequence of ANG I. The findings were published
in the first Angiotensin paper of the BIPM [35], with focus on a qualitative analysis
and semi-quantitative estimation of the mass fraction of the impurities. In a follow-up
paper [36], the same ANG I peptide material has been investigated and quantified metro-
logically in a comparative way with both Qtrap4000 and LTQ-Orbitrap XL systems and
the resulting mass fractions were stated together with their uncertainties.
.
4 2.3. PICAA analysis of ANG I. The ANG I material was subjected to microwave-
assisted vapour phase hydrolysis in order to release its constituent AAs, which are then
analyzed by liquid chromatography coupled to charged aerosol detection (LC-CAD) or to
isotopic dilution mass spectrometry (LC-IDMS). The amino acids are analyzed directly
without derivatization in order to avoid any carry-over of additional impurities. An
LC-IDMS/MS method which included 30 MRM transitions for five non-labelled and
five isotopically labelled AAs, was employed to quantify the amino acids. Figure 15
shows a representative chromatogram with SRM transitions overlays from LC-IDMS/MS
measurements of a hydrolyzed ANG I standard spiked with isotopically labelled AAs,
wherein the chromatographic co-elution of corresponding 12 C and 13 C AA for Pro, Val,
Fig. 15. – Representative LC-IDMS/MS chromatogram with SRM transition overlays of a hy-
drolyzed ANG I sample spiked with 13 C-AA standards. Elution order of amino acids: Pro, Tyr,
Val, Ile, Leu and Phe.
Ile, Leu and Phe isotopes and their presence in the hydrolyzed ANG I material are well
observed.
AA mass fractions obtained by the microwave-assisted vapour phase hydrolysis and
exact matching double LC-IDMS/MS approach were calculated according to eq. (9).

mz my R B
(9) Wx = Wz · · · ,
myc mx RBc
where Wx is the mass fraction of AA in sample, Wz is the mass fraction of original AA

in calibration blend, mz is the mass of original AA solution in calibration blend, myc is
the mass of the labelled AA solution in calibration blend, my is the mass of the labelled

AA solution in sample blend, mx is the mass of sample used, RB is the sample ratio,

RBc is the calibration ratio.
The presence of peptide impurities was corrected using the following equation:

Mr (P) nAA xIMPi
(10) xp = − YIMPi ,
Z1 mm Mr (IMPi )
where xP is the mass fraction of peptide in the material, mm is the mass of the material
analyzed, Mr (P) is the relative molecular mass of peptide, Z1 is the number of molecules
of the amino acid of interest per peptide molecule, nAA ia the amount of substance of
the amino acid of interest measured in the material, Yi is the number of molecules of the
amino acid of interest per peptide impurity molecule (IMPi ), xIMPi is the mass fraction
of the peptide impurity IMPi , Mr (IMPi )is the relative molecular mass of the peptide
impurity IMPi .
The resulting ANG I purity mass fractions values obtained by PICAA of Leu, Phe,
Pro and Val were combined to achieve a final purity mass fraction of 789 mg/g ANG I
in the material, with a standard uncertainty of 29 mg/g.
.
4 3. Characterization of C-peptide primary reference material. – Human C-peptide
(hCP), defined as human proinsulin [57–87] fragment with the amino acid sequence
EAEDLQVGQVELGGGPGAGSLQPLALEGSLQ [relative molecular mass (Mr) 3020.3],
is an important biomarker in laboratory medicine and forensic science. The hCP level in
serum provides an accurate assessment of residual β-cell function and has become an im-
portant marker of insulin secretion in patients with diabetes [37,38]. Equimolar amounts
of hCP and insulin are then stored in the pancreatic β-cells, and both are eventually
released to the portal circulation. In laboratory medicine, the hCP level is determined to
distinguish type 1 and type 2 diabetes and to control the status of insulin secretion. In
addition, hCP measurements are playing a key role in forensic science. A higher than ex-
pected mass concentration of insulin without an accompanying increase in the hCP mass
concentration in a serum sample of a victim in a hypoglycemic state provides evidence
of exogenous insulin administration.
Standardization studies of hCP measurements concluded that hCP results gener-
ated by different methods and laboratories were not always in agreement with each
other [39, 40]. Reference methods based on LC-IDMS have been developed and the fea-
sibility for their usages in the standardization of hCP has been demonstrated [41-43]. In
order to implement these methods, a pure peptide reference material is required.
The assignment of the mass fraction content of high-purity materials was the subject
of the CCQM-K115/P55.2 comparison [44]. hCP was proposed as the most appropriate
choice for a study material for a first CCQM key comparison and parallel pilot study
looking at competencies to perform peptide purity mass fraction assignment as it repre-
sents short (1 kDa to 5 kDa), non-cross-linked synthetic peptides. The variance in results
was found to be due to the measurement challenge of identiftying and quantifying all
peptide impurities within the material.
.
4 3.1. Peptide impurity identification and quantification. Liquid chromatography cou-
pled to tandem mass spectrometry was operated with a linear ion trap as a first mass
analyzer and a high-resolution Orbitrap as a second mass analyser for the identification
and quantification of structurally-related peptide impurities [45]. Numerous classes of
structurally-related impurities such as amino acid deletion/insertion, alkylated fragment,
deamidation, and pyroglutamylation have been identified in the hCP sample. In general,
the longer the primary structure of a peptide the greater the number and possibility
of occurrences of structurally-related peptide impurities at higher mass fraction levels.
Amino acid deletion/insertion, missing/insertion of one or more amino acid residues, are
usually generated during the process of peptide synthesis. Truncated peptide impuri-
ties, missing one or more desired amino acid residues at either N- or C-terminal end
most likely produced by degradation during transportation or storage are also classi-
fied as amino acid deletion impurities. Acetylated fragments and formylated fragments
(e.g. formyl-hCP(11-31), formyl-hCP(8-31)) are other major classes of hCP impurities.
Deamidation is another prominent impurity. It occurs on the side chain by replacing
an amide functional group with a carboxylated group. Three possible deamidated hCPs
(deahCP) were identified in the hCP material at 6th, 9th, and 22th glutamine positions
Fig. 16. – hCP impurity identification and quantification: overview.
in the sequence of hCP. Pyroglutamylation [46] happens on N-terminal glutamic acid by

loss of one molecule of water creating pyroglutamylated hCP (phCP).
Six hCP isomers were identified based on the information of m/z with high mass
accuracy, and tandem mass spectra. Also, some minor impurities with a mass shift
of 126.1 Da to some hCP fragments were observed. This is attributed to the forma-
tion of a substituted guanidine by the reaction with the amino acid activator N, N -
Diisopropylcarbodiimide (DIC-hCP) at N- terminal [47]. Two minor impurities with the
same m/z of 833.43 (z = 2) were not exactly identified. They are truncated hCPs with
unknown modification at N-terminal. One minor impurity with serine replaced with
glycine at 29th position in the sequence of hCP.
All the impurities identified in hCP material by the BIPM are listed in fig. 16 to-
gether with impurities found by other participants in the CCQM-K115/55.2 study on
hCP purity.
Table III. – Total hydrolysis efficiency means and standard deviations for leucine, proline and
valine in hCP for both gas/liquid phase hydrolysis and microwave-assisted hydrolysis methods
Total hydrolysis efficiency

Gas/liquid phase hydrolysis Microwave-assisted hydrolysis
Mean (%) SD (%) Mean (%) SD (%)
Leucine 97.1 0.8 97.5 3.0
Proline 97.9 1.4 98.5 4.2
Valine 96.8 1.8 96.2 4.0
.
4 3.2. PICCA analysis of C-peptide material. The CCQM-K115/P55.2 study also allowed
the impact of the hydrolysis methods employed for PICAA on the quantification of the
hCP mass fraction to be studied. Previous pilot studies on peptide hydrolysis have shown
discrepancies in hydrolysis efficiencies for a certain peptide [48, 49]. Seven participants
used the PICAA approach in CCQM-K115/P55.2. Two participants used microwave-
assisted hydrolysis while five employed gas/liquid phase hydrolysis. All participants that
e used PICAA performed an efficiency correction for the hydrolysis methods.
In general, the hydrolysis efficiencies are very high (> 95%) (see table III). No
significant differences are observed for hydrolysis efficiencies between both hydrolysis
methods (microwave-assisted hydrolysis or classical gas/liquid phase hydrolysis) em-
ployed or amino acids (leucine, proline and valine) selected for analysis. The variance
of the microwave-assisted hydrolysis is slightly larger than the variance of the classi-
cal gas/liquid phase hydrolysis which can probably be explained by the slightly larger
number of data sets for the classical gas/liquid phase hydrolysis.
It can be summarized that the hydrolysis even of a large, 31 amino acids containing,
peptides is very efficient and independent of the method used or amino acid analysed.
Nevertheless, small biases need to be corrected and/or need to be considered in the
calculation of the uncertainties as was done by the participants in CCQM-K115/P55.2.
In general, an excellent comparability of hydrolysis efficiencies with small variances was
obtained with variances of the microwave-assisted method being slightly larger. The
accurate identification and quantification of peptide related impurities has a larger impact
on the individual results of the hCP purity.
REFERENCES
[1] Westwood S., Josephs R. D., Daireaux A., Wielgosz R. et al., Metrologia, 46 (2009)
08019.
[2] Westwood S., Josephs R. D., Choteau T., Daireaux A., Wielgosz R. et al.,
Metrologia, 48 (2011) 08013.
08009.
08014.
[5] Duewer D. et al., NIST Special Publication, 1012 (2004).
[6] Josephs R. D., Daireaux A., Westwood S. and Wielgosz R. I., J. Chromatogr. A,
1217 (2010) 4535.
[7] Westwood S., Choteau T., Daireaux A., Josephs R. D. and Wielgosz R. I.,
Analytical Chemistry, 85 (2013) 3118.
[8] Malz F. and Jancke H., J. Pharmaceut. Biomed. Anal., 38 (2005) 813.
[9] Pauli G., Jaki B. and Lankin D., J. Nat. Prod., 68 (2005) 133.
[10] Holzgrabe U., Prog. Nucl. Magn. Reson. Spectrosc., 57 (2010) 229.
[11] Pauli G., J. Nat. Prod., 75 (2012) 834.
[12] Giradeau P., Magn. Reson. Chem., 55 (2017) 61.
[13] Westwood S. et al., “CCQM-K55. c (L-(+)-Valine) Final Report: June 2014; CCQM-
K55. c Key Comparison on the Characterization of Organic Substances for Chemical
Purity” (2014).
[14] Gross J. H., Mass Spectrometry: A Textbook (Springer Edition) 2004.
[15] Budzikiewicz H., Schäfer M., Massenspektrometrie – Eine Einführung (Wiley-VCH
Verlag. 5. Auflage) 2005.
[16] Finehout E. J. and Lee K. H., Biochem. Mol. Biol. Educ., 32 (2003) 93.
[17] The Coupling of Gas and Liquid Chromatography with Mass Spectrometry, J. Mass
Spectrom., 34 (1999) 157.
[18] Klampfl C. W., J. Chrom. A, 1044 (2004) 131.
[19] De Hoffmann E., J. Mass Spectrom., 31 (1996) 129.
[20] Fenn J. B., Mann M., Meng C. K., Wong S. F. and Whithouse C. M., Science, 246
(1989) 64.
[21] Gaskell S. J., J. Mass Spectrom., 32 (1997) 677.
[22] Bruins A. P., Mass Spectrom. Rev., 10 (1991) 53.
[23] Manisali I., Chen D. D. Y. and Schneider B. B., Trends Anal. Chem., 25 (2006) 243.
[24] Nemes P., Marginean I. and Vertes A., Anal. Chem., 79 (2007) 3105.
[25] Murray K. K., Boyd R. K., Eberlin M. N., Langley G. J., Li L. L., Naito Y. and
Tabet C. J., Abstr. Pap. Am. Chem. Soc., 229 (2005) U123.
[26] Wong P. S. H. and Cooks R. G., Curr. Separ. Com Drug Develop., 16 (1997) 1.
[27] March R. E., J. Mass Spectrom., 32 (1997) 351.
[28] Le Blanc J. C. Y., Hager J. W., Ilisiu A. M. P., Hunter C., Zhong F. and Chu
I., Proteomics, 3 (2003) 859.
[29] Hopfgartner G., Varesio E., Tschäppät V., Grivet C., Bourgogne E. and
Leuthold L. A., J. Mass Spectrom., 39 (2004) 845.
[30] Sparbier K., Wenzel T., Dihazi H., Blaschke S., Müller G.-A., Deelder A., Flad
T. and Kostrzewa M., Proteomics, 9 (2009) 1442.
[31] Perry R. H., Cooks R. G. and Noll R. J., Mass Spectrom. Rev., 27 (2008) 661.
[32] Barrow M. P., Burkitt W. I. and Derrick P. J., Analyst, 130 (2005) 18.
[33] Roepstorff P. and Fohlman J., Biomed. Mass. Spectrom., 11 (1984) 601.
[34] Biemann K., Annu. Rev. Biochem., 61 (1992) 977.
[35] Stoppacher N., Josephs R. D., Daireaux A., Choteau T., Westwood S. W. and
Wielgosz R. I., Anal. Bioanal. Chem., 405 (2013) 8039.
[36] Stoppacher N., Josephs R. D., Daireaux A., Choteau T., Westwood S. and
Wielgosz R. I., Rapid Commun. Mass Spectrom., 29 (2015) 1651.
[37] Kjems L. L., Vølund A. and Madsbad S., Diabetologia, 44 (2001) 1339.
[38] Polonsky K. S., Licinio-Paixao J., Given B. D., Pugh W., Rue P., Galloway J.,
Karrison T. and Frank B., J. Clin. Invest., 77 (1986) 98.
[39] Little R. R., Rohlfing C. L., Tennill A. L., Madsen R. W., Polonsky K. S.,
Myers G. L., Greenbaum C. J., Palmer J. P. and Rogatsky E., Clin. Chem., 54
(2008) 1023.
[40] Little R. R., Wielgosz R. I., Josephs R. D., Kinumi T., Takatsu A., Li H., Stein
D. and Burns C., Clin. Chem. 2017. accepted.
[41] Fierens C., Stöckl D., Baetens D., De Leenheer A. P. and Thienpont L. M., J.
Chromatogr. B, 792 (2003) 249.
[42] Rodrı́guez-Cabaleiro D., Stöckl D. and Thienpont L. M., Rapid Commun Mass
Spectrom, 19 (2005) 3600.
[43] Rodrı́guez Cabaleiro D., Stöckl D., Kaufman J. M., Fiers T. and Thienpont
L. M., Clin. Chem., 52 (2006) 1193.
[44] Josephs R. D., Li M., Song D., Daireaux A., Choteau T., Stoppacher N.,
Westwood S., Wielgosz R. I., Li H., Flatschart R., Borges Oliveira R.,
Melanson J. E., Ohlendorf R., Henrion A., Kinumi T., Wong L., Liu Q., Akgöz
M., Gören A. C., Un ¨ I., Quaglia M. and Warren J., Metrologia, 54 (2017) Tech.
Suppl., 08007.
[45] Li M., Josephs R. D., Stoppacher N., Wielgosz R. I. and Li H., Impurity
determination for human C-peptide by liquid chromatography-high resolution tandem mass
spectrometry, in preparation.
[46] Kinumi T., Goto M., Eyama S., Kato M., Kasama T. and Takatsu A., Anal. Bioanal.
Chem., 404 (2012) 13.
[47] Sanz-Nebot V., Benavente F. and Barbosa J., J. Chromatogr. A, 870 (2000) 315.
[48] Bunk D., P55.1 Scope of applicability BAWG/12-54.
[49] Hydrolysis investigation study BAWG/12-59 (2012) BIPM webpage, CCQM BAWG
working documents.
MODULE II. FUNDAMENTALS OF METROLOGY
DOI 10.3254/978-1-61499-818-1-149
Uncertainty of measurement
W. Bich
INRIM - Torino, Italia
Summary. — We live in a world of uncertainty. We measure phenomena and

quantities, and the outcomes are uncertain. Our predictions, be they about future
climate or stock trend, are based on uncertain data. We continuously take decisions
in conditions of incomplete knowledge. The purpose of this paper is to make the
reader aware of the uncertainty that surrounds our lives, and to give him some tools,
both conceptual and technical, to cope with uncertainty, specifically in the domain
of measurement. It is shown that probability theory is the appropriate framework
in which uncertainty can be treated in a convenient and rigorous way. The author is
convener of the Joint Committee for Guides in Metrology (JCGM) Working Group
1 (GUM). The opinion expressed in this paper does not necessarily represent the
view of this Working Group.
In this world nothing can be said to be certain, except death and taxes.
Benjamin Franklin
Disclaimer
Most of the contents of this paper are taken or adapted from [1]. New parts are
sects. 1, 2 and 10. Some of the remaining original material has been considerably
revamped.
149
150 W. Bich
The author is convener of the Joint Committee for Guides in Metrology (JCGM)
Working Group 1 (GUM). The opinion expressed in this paper does not necessarily
represent the view of this Working Group. However, he is grateful to his colleagues of
both Working Groups for the many fruitful discussions that contributed to form the ideas
expressed in this paper.
1. – Introduction
Uncertainty means doubt, or lack of certainty, or incomplete knowledge, and, accord-

ing to Franklin, is much more pervasive than one might imagine at first thought. Take
for example the number π. We define it exactly as the ratio of the circumference to
its diameter. We know to this date (11 November 2016), 22 459 157 718 361 (i.e., in a
more picturesque description, ≈ π e trillion) decimal digits of π [2]. But π is an irrational
(actually, a transcendental) number, so that its next unknown decimal digit(s) cannot
be inferred from those we know. So, our knowledge of π is not complete and never will,
although the uncertainty in its value is so low that it can be neglected in all practical
applications. However, this pale shade on something believed absolutely exact carries a
high conceptual significance. Nobody doubts that π is exact, in the sense that it pos-
sesses a well-defined, unique value made of an infinite numbers of digits, whichever is the
representation chosen, decimal, binary, exadecimal, etc. Yet, this unique value, that we
can define exactly, is unknown.
That of π is by no means an exception, as the vast majority of numbers are incom-
pletely known. The only set of completely known numbers is that of rational numbers,
Q, that includes integers, Z. By “a completely known number” I mean that all its digits
are known, regardless of the representation chosen. However, Q is an infinite, countable
subset of the infinite, uncountable set of real numbers R. And (some) irrational numbers
may be calculable in principle to within any stipulated numerical accuracy, but the in-
finite, non-repeating sequence of their digits by definition cannot be completely known.
Hence, the conclusion holds that most numbers are incompletely known. So, π is just an
outstanding example of a common situation.
2. – Physical quantities and their values
What about physics? In short, since mathematics is the language of physics, the
situation is pretty similar, in that most of our knowledge is incomplete, thus uncertain.
In addition, to learn something about (the value of) a physical quantity, one has to
measure it, which introduces further uncertainty. Before having a better look at what is
certain and what uncertain in physics, it is appropriate to spend some words about the
concepts of quantity and value of a quantity.
.
2 1. Quantities. – In his Treatise on Electricity and Magnetism of 1873 [3], Maxwell
wrote:
Uncertainty of measurement 151
EVERY expression of a Quantity consists of two factors or components. One

of these is the name of a certain known quantity of the same kind as the
quantity to be expressed, which is taken as a standard of reference. The other
component is the number of times the standard is to be taken in order to make
up the required quantity. The standard quantity is technically called the Unit,
and the number is called the Numerical Value of the quantity.
This approach is also adopted in the currently authoritative document for terminology,
the “International vocabulary of metrology — Basic and general concepts and associated
terms (VIM)”, JGCM 200:2012 [4], usually abbreviated VIM3. In it,
quantity “property of a phenomenon, body, or substance, where the property

has a magnitude that can be expressed as a number and a reference” ([4],
definition 1.1).
The concepts “property” and “reference” are explicitly considered as “primitive”, there-
fore undefined ([4], page xii). A NOTE to the definition of quantity reads:
“A reference can be a measurement unit, a measurement procedure, a refer-

ence material, or a combination of such”.
Here (following Maxwell) only those quantities will be considered whose magnitude
can be expressed in terms of a unit, i.e.
“real scalar quantity, defined and adopted by convention, with which any
other quantity of the same kind can be compared to express the ratio of the
two quantities as a number” ([4], definition 1.9).
In the traditional notation of quantity calculus ([4], definition 1.21, and [5]), quantities
are denoted by uppercase italics fonts, say, Q, whereas numerical values and units are
denoted by the same letter within curly, {Q}, and square, [Q], brackets, respectively. In
this notation Maxwell’s statement would be Q = {Q}[Q], where Q is the quantity, [Q]
the unit and {Q} the numerical value of the quantity. In the case of the speed of light
in vacuum c0 , the quantity (or its value?, see below) is invariant, whereas its numerical
value depends upon the unit chosen to express it. For example, c0 = 1 c0 in a natural
system of units, in which the unit of speed is c0 itself, whereas c0 = 299 792 458 m s−1
in the international system of units, SI. Or, c0 = 1 802 613 915 461 fur/ftn (furlongs per
fortnight) in a more exotic system of units [6].
The definition of quantity in the VIM is appropriate in measurement, but too narrow
for physics. Mass, current, spin, momentum are all generic quantities we use to model
our understanding of nature by means of equations in which the quantities are treated
as mathematical variables. These variables can take any value in their domain, mostly
continuous in classical physics, mostly discrete in quantum physics. The expression
F = ma (we just consider vector magnitudes for simplicity) implies that there is a (in
152 W. Bich
this case, uncountable) infinity of sets of three values (real numbers) for F , m and a that
satisfy the expression. For example, the mass mA of body A to which a specific force Fs
is applied experiences a specific acceleration as . As a further example, speed is a generic
quantity, the speed of light in vacuum c0 , which happens to be one of the fundamental
physical constants, i.e., universal in nature and invariant in time, is a specific, well-
defined quantity. The definition given in the VIM applies only to specific quantities such
as mA or c0 , possessing a unique value. One might say that the VIM definition includes
only quantities well identified, for example by a subscript. Accordingly, only labelled
quantities can be measured.
.
2 2. Quantity values. – Despite the narrow definition of quantity, the VIM further
distinguishes between a quantity and its value, (quantity value, or preferably value of
a quantity or simply value)
quantity value “number and reference together expressing magnitude of a

quantity” ([4], definition 1.19),
where the number mentioned in the definition of quantity value is the numerical value
([4], definition 1.20).
Needless to say, the distinction between value and numerical value of a quantity has
originated endless misunderstandings, as people tend to identify the two concepts and
confound value with numerical value.
In addition to the misleading, but conceptually clear distinction between value and
numerical value of a quantity, there is a further ambiguity in the distinction between
a quantity and its value(s). Whereas in a broad definition of quantity the distinction
between a generic quantity and some of its specific values can be understood (although
its usefulness remains questionable), in the VIM definition of quantity it is hard to see
any conceptual difference between c0 and its value. After all, in quantity calculus there
is an equality sign between them, which to me means that they are mathematically
interchangeable. Everybody understands the statement “the speed of light in vacuum is
constant and independent of the unit chosen”, whereas “the value of the speed of light in
vacuum is constant and independent of the unit chosen” would be questioned and would
need explanation. For a discussion on quantity and quantity value, see [7]. For further
information about quantities, properties and their classification, see also [1, 8-12].
.
2 3. Physical constants. – Fundamental physical constants constitute in many respects
a special category among physical quantities. Without entering in the debate on which
and how many constants are really fundamental, we will consider as authoritative the list
given in [13]. The constants are indeed cornerstones in our understanding of nature. Fur-
thermore, they are Lorentz-invariant, i.e., their value is the same in any inertial reference
frame and, apart from some theories suggesting a change in their values according to the
age of the universe [14], they are also considered time-invariant. From the viewpoint of
measurement, each of them constitutes a neat example of ideal measurand. As for π, we
do not doubt that they have a precise and unique value, represented by a real number.
Yet, most of them are imperfectly known.
One exception is the magnetic constant or permeability of free space μ0 . This constant
was fixed in 1948 by virtue of the definition of the ampere in the International System
of Units, SI [15], and has the value μ0 = 4π × 10−7 NA−2 . This value is exact, although,
at least in principle, affected by the infinitesimal uncertainty associated with π.
There is a second exception, that is the speed of light in vacuum c0 . This was fixed in
1983 by the 17th General Conference on Weights and Measures, CGPM, with Resolution
1 ([15] Appendix 1), by which the metre was re-defined in terms of the speed of light
and the unit of time, the second. The effect was that c0 was assigned the conventionally
exact value c0 = 299 792 458 ms−1 . The reason was that it had been measured to an
accuracy exceeding that of the realization of the metre at that time [16]. Conventionally
exact means that an infinite sequence of zeros has to be imagined after the last significant
decimal digit. It is worth noting that, contrary to μ0 , c0 is known exactly both in practice
and in principle.
These two definitions impacted on another constant which is a monomial combination
of μ0 and c0 , i.e. the electric constant, or permittivity of free space 0 = 1/μ0 c20 , which
became exact as well. As a further consequence, also the characteristic impedance of
vacuum Z0 , defined as Z0 = μ00 , or Z0 = μ0 c0 , became exactly defined.
The SI is undergoing an epochal revision that will hopefully be made effective at the
26th CGPM in 2018. The revision includes re-definition of four units in terms of as many
physical constants. The units are the kilogram, the ampere, the kelvin and the mole; the
constants are the Planck constant h, the elementary charge e, the Boltzmann constant k
and the Avogadro constant NA , respectively [17]. The re-definitions will imply, similarly
to what happened for c0 , that the values internationally recommended at that time for
the constants involved will be made conventionally exact, thus removing any uncertainty
from them.
Uncertainty is a measure of state of knowledge. As such, it cannot magically disap-
pear (or increase) unless the state of knowledge has changed accordingly. However, the
(overall) state of knowledge about physical constants and units can only change if new
information is provided by, say, new measurements or discoveries, which is not the case
in the envisaged revision of the SI. Yet, there is no trick in the revision, as well as there
was no trick when the speed of light c0 was promoted from uncertain to exact. In both
cases, uncertainty does not disappear, rather it is distributed differently from before. We
trust (with the caveat expressed above) that each physical constants has a unique value,
and we are allowed to fix its numerical value in the SI by choosing a suitable real number
exactly known at the expenses, so to say, of the corresponding (SI) unit. So, in a sense,
uncertainty is transferred from physical constants to the artefacts traditionally used to
realize the unit, from h, k, e and NA to the international prototype of the kilogram, the
triple point of water and so on. In addition, some of the previously exact constants ac-
quire an uncertainty. This is the case of the magnetic constant μ0 (and 0 and Z0 , which
are related to it), and of the molar mass constant Mu . This aspect has been variously
154 W. Bich
questioned [18,19] but overall the re-distribution of uncertainty determined by the forth-
coming revision of the SI is more rational and corresponds better to our understanding
of nature. See also [20].
.
2 4. Other quantities. – As with numbers, besides the restricted set of conventionally
exact constants the vast majority of physical quantities (in the VIM sense) is incompletely
known, thus uncertain. As already mentioned, to learn about a quantity it is necessary
first to specify (or define) it, then to measure it. Let us consider measurement in the
next section.
3. – Measurement
.
3 1. Measurement and measurand. – The VIM definition for measurement is
measurement “process of experimentally obtaining one or more quantity
values that can reasonably be attributed to a quantity” ([4], definition 2.1).
The quantity mentioned in the definition of measurement is known as
measurand “quantity intended to be measured” ([4], definition 2.3)
and the “quantity value that can reasonably be attributed to a quantity” is the
measured quantity value “quantity value representing a measurement re-
sult” ([4], definition 2.10)
There is no doubt that the aim of any measurement is to improve the state of knowl-
edge about the measurand, so that a definition to be considered in a future edition of
VIM might sound as
“process of experimentally updating the state of knowledge about a quantity”.
This definition emphasizes the role of measurement as a process by which fresh data
are collected by experiment and combined with existing relevant information to update,
and hopefully improve, the state of knowledge about the measurand. The situation of no
prior knowledge (which seems unlikely in practice) can anyway be encompassed in this
scheme as a limit case. Note that the update might worsen the state of knowledge, when
a new measurement of a quantity provides a datum in disagreement with previous values.
.
3 2. Measurement result and measurand estimate. – According to its current definition,
measurement is the process of measuring, not its outcome. The latter is the
measurement result (or result of measurement) “set of quantity values
being attributed to a measurand together with any other available relevant
information” ([4], definition 2.9).
This definition changed considerably with respect to that given in the previous edi-
tion (the second), which was “value attributed to a measurand, obtained by measure-
ment”. The meaning of the current definition is that a measurand estimate alone is not
a measurement result. In order to be considered as such, the outcome of a measure-

ment must comprise a whole range of plausible estimates, perhaps with a further piece
of information about their respective plausibility (or probability). A common way is
to specify the measurand estimate and an associated uncertainty (see sect. 4). It will
still take a long time before the idea —that an estimate with an associated uncertainty
is by far better than another without it— is accepted worldwide by all the concerned
environments, beside those of science and technoloy, such as for example forensic science
and laboratory medicine. In this respect, only some countries, mostly of English descent,
show a consolidated tradition [21-23].
At the same time, the change in the definition engendered (at least) two problems. The
first, inherent in any change, is that people still use the term with the old meaning. The
second is that in the VIM a term for the old meaning is missing. The VIM-recommended
term is measured quantity value, but the meaning is not the same. A non-VIM term
commonly used is estimate, or measurand estimate.
As a further terminological remark, in laboratory language, and unfortunately also
in the scientific literature, the distinction between measurement and measurement result
is often overlooked, which, combined with the problems previously described, causes
considerable confusion.
4. – Measurement uncertainty
.
4 1. General . – In its broad sense, un-certainty is a term that does not need a defi-
nition, meaning simply lack of certainty, or doubt. Therefore, measurement uncertainty,
or uncertainty of measurement would indicate doubt in the result of a measurement.
However, the term has also a technical, specific meaning, that is, that of measure of the
amount of doubt.
The word “uncertainty” means doubt, and thus in its broadest sense “uncer-
tainty of measurement” means doubt about the validity of the outcome of a
measurement. Because of the lack of different words for this general concept
of uncertainty and the specific quantities that provide quantitative measures
of the concept, for example, the standard deviation, it is necessary to use the
word “uncertainty” in these two different senses. ([24], 2.12.)

In the context of measurement, the word “uncertainty” without adjectives
refers both to the general concept of uncertainty and to any or all quantitative
measures of that concept. When a specific measure is intended, appropriate
adjectives are used. ([24], 2.2.)
These two quotation are taken from the
“Guide to the expression of uncertainty in measurement”, JCGM 100:2008 [24],
usually abbreviated GUM, hereafter the Guide. This is the reference document as con-
cerns specific measures of uncertainty. It is supported by a number of ancillary docu-
ments, including Supplements. For a historical outline and more about the whole suite
156 W. Bich
of documents, see [1]. For an update and an overview of the envisaged developments,
see [25-29].
The measurement uncertainty is defined as
measurement uncertainty “non-negative parameter characterizing the dis-

persion of the quantity values being attributed to a measurand, based on the
information used” ([4], definition 2.26).
NOTE 2 reads
“The parameter may be, for example, a standard deviation called standard
measurement uncertainty (or a specified multiple of it), or the half-width of
an interval, having a stated coverage probability.”
.
4 2. Definitional uncertainty. – Before treating specific measures of uncertainty, it is
necessary to mention that, according to a line of thought, also adopted in VIM3, the
set of “values being attributed to a measurand” seldom can reduce to one single value,
not because of dispersion introduced by the measurement of the quantity, but due to an
intrinsic impossibility to completely define the measurand. This is reflected in the VIM
definition of true value of a quantity:
true value “quantity value consistent with the definition of a quantity” ([4],
definition 2.11).
The intended meaning of this definition is that, since a quantity cannot be completely
defined without an infinite amount of information, there would be a whole range of (true)
values consistent with its incomplete definition. A classic example among philosophers
is “the length of a given table” (metrologists preferring perhaps a gauge block), which,
as such, is ill-defined in many respects. The short sides of the table are neither perfectly
parallel nor perfectly rectilinear. Even if both these requisites were respected, the edge
surfaces would not be ideally polished, et cetera. Hence, the length of the table, defined in
terms of distance between the short sides of the table, would be a multi-valued quantity,
and a whole set of true values would be consistent with its definition. So, the features of
the table would be such that, even before starting the measurement, there is a vagueness
in what really the length of the table is. This vagueness is called in the VIM
definitional uncertainty “component of measurement uncertainty result-

ing from the finite amount of detail in the definition of a measurand” ([4],
definition 2.27).
NOTE 1 Definitional uncertainty is the practical minimum measurement un-

certainty achievable in any measurement of a given measurand.
The concept of definitional uncertainty is better understood in the framework of oper-

ationalism, a theory developed in the early decades of the past century by Percy William
Bridgman [30], an American physicist and Nobel laureate known for his contributions
in the field of ultra-high pressure. Loosely speaking, in his view any item or concept,
thus including measurands, should be defined operationally, i.e., through the operations
performed to measure it. As also these operations need to be defined, there would re-
sult an unavoidable limit in the completeness of any operational definition. Hence, the
definitional uncertainty follows.
Bridgman’s operationalism has its merits, but is currently more influential outside
physics, for example in psychology and specifically in psychometrics, where the measur-
ands do need an operational definition, being things such as anger, or stress, or comfort,
whereas in physics measurands are or, perhaps better, can be defined, at least ideally,
sufficiently well.
Indeed, it is a well established belief (at least in classical physics) that quantities in-
volved in a physical system have at any time unique values. Even in quantum mechanics,
many observable quantities can have simultaneously exact values, although this does not
hold for any quantity. In any case, at least in principle, and at least as concerns the
classical quantities that constitute the measurand of most of our measurements, we do
believe that they have a value, and that this value is unique. This is certainly true for
.
physical constants, as discussed in sect. 2 3. It is the task of the experimenter to define
sufficiently well the measurand. This task is accomplished through a thorough modelling
of the measurement, which, in a sense, is an operational definition. In this respect, in
the Guide, subclause 1.2 reads:
“This Guide is primarily concerned with the expression of uncertainty in

the measurement of a well-defined physical quantity —the measurand— that
can be characterized by an essentially unique value. If the phenomenon of
interest can be represented only as a distribution of values or is dependent
on one or more parameters, such as time, then the measurands required for
its description are the set of quantities describing that distribution or that
dependence.” ([24], 1.2.)
The second part of this quotation considers an aspect that is often overlooked. It
may happen, and indeed it happens frequently, that the experimenter is interested in the
value of a quantity but, for some reason, does not want to (or can not) actually measure
it, preferring (or needing) to use some prior knowledge. For example, the measurement
of the mass of a body requires in principle the knowledge of the air density during the
weighing. The measurement of air density implies a somewhat expensive instrumentation,
skill and time. As a counterpart, a remarkable accuracy can be achieved which, however,
might not be required for the application [31].
As another example, the construction of a standard gas mixture requires knowledge
of the atomic weights of the individual gas components used in the construction, but the
actual measurement of atomic weights in the available samples is not easily feasible.
In both cases, it is known that the quantity of interest varies over a range of values.
The air density at a given location varies with time on a daily and seasonal basis, following
158 W. Bich
changes of temperature, pressure, humidity et cetera. Similarly, the atomic weight of an

element in a specific sample may depend on the isotopic composition of the sample
which, in turn, depends on the origin of the sample. As a result, the quantity of interest
has not a unique value, but many values distributed over a population. Therefore,
the situation is as depicted in the quotation above from the Guide. The population
of values of the quantity is described by a distribution. In most cases, and certainly
in the examples given, there exists a large set of data gathered from measurements on
samples or individuals belonging to the population. The main parameters characterizing
the population, e.g., the mean and the variance (and the associated uncertainties), can
be inferred from these data. Of course, tasks of the kind of that concerning the atomic
weights are not affair of individual experimenters. Actually, the Commission on Isotopic
Abundances and Atomic Weights (CIAAW), a specific body of the International Unions
for Pure and Applied Chemistry (IUPAC), is mandated with publishing periodic updates
of the recommended values of the atomic weights of the elements [32].
A user needing to know the atomic weight of an element contained in his sample
can therefore use the internationally recommended value and the associated uncertainty,
which, for some elements, is dominated by the dispersion of the corresponding population.
Should this uncertainty be too large for his purposes, he would have no other choice than
that of measuring the atomic weight of his sample. Similar considerations hold for the
air density (at a given location) and for many other quantities and situations.
Situations such as those described above are relevant, in my opinion, to definitional
uncertainty. After all, a measurand like “the atomic weight of carbon” is in principle in
a similar category as “the length of a given table”. Both measurands are ill-defined, but
the conundrum can easily be solved by adopting a better-defined measurand, such as the
mean atomic weight (and, in this case, a suitable measure of the population size) and the
mean length of the table. In the latter case, the standard deviation of (a sample taken
from) the population made of the individual length values “compatible with the definition
of the measurand”, taken as a measure of the population dispersion, would serve as the
definitional uncertainty arising from the incomplete definition of the measurand “length
of the table”.
5. – Modelling a measurement
It is a well established fact that a measurand Y is rarely measured directly, but

its value is determined through the measurement of a number of other quantities Xi
(i = 1, . . . , n) on which the measurand depends,
(1) Y = f (X),
where X = (X1 , X2 , . . . , Xn ) and denotes transpose.

To be honest, a direct method of measurement is defined in the International Elec-
trotechnical Vocabulary published by the IEC [33], but the definition is questionable and
does not seem to hold up under a close scrutiny [34].
The paradigm of indirect measurement applies even in the case of measurement by

counting, in wich it seems hard to write a model [35].
Modelling a measurement plays a role in the issue of definitional uncertainty. In
the approach of the Guide, the measurand is defined by model (1). So, the model
itself constitutes the operational definition of the measurand. The task of writing a
model for a given measurement is to the experimenter, who typically embeds in the
model all the input quantities on which the measurand (or output quantity) is supposed
to depend, to the best of his knowledge. Of course, there is a possibility that some
contributions are unrecognized, and thus neglected, with the consequence that the model
is incomplete and the estimates it produces are biased, and the associated uncertainties
typically underestimated. There is no way to quantify such incompleteness unless an
external help is sought, such as a reference standard or a comparison with other results
obtained in different measurements of the same measurand.
There is a further remark concerning the modelling of a measurement. Let us con-
sider mass measurements as a case example. As far as mass is measured by means
of a balance in atmosphere, the corresponding measurement model is more or less the
same, be the measurement carried out in a National Metrology Institute or in the floor
shop. There might be minor differences due to the particular type of balance used,
electromagnetically-compensated or mechanical with equal arms, or analytical, but the
input quantities to the model are basically the same: reference mass (perhaps hidden
within the balance calibration), balance sensitivity and indications, air density, volume
et cetera. For a discussion, see [36]. Of course, some of the input quantities, contributing
in a marginal way to the estimate (marginal with respect to the accuracy required by the
application) will be neglected in the lowest-level measurements, but their contribution to
the uncertainty cannot be neglected, rather, this contribution will be considerably larger.
The fact that the measurement model for a given type of measurement is basically
the same, and that it is worldwide accepted and adopted, ensures the good worldwide
harmonization of measurement results, as shown by the typically good results of inter-
comparisons.
6. – Propagating uncertainty
In the Guide, the approach of estimation theory [37-39] is adopted. In this framework,
a parameter Y is estimated by a suitable estimate Ŷ . This estimate in most cases is not
perfect, as it is affected by an unknown error
(2) Ŷ = Ŷ − Y.
The estimate Ŷ for the measurand is obtained from model (1) by evaluating it at the
estimates X̂ obtained for the input quantities X:

(3) Ŷ = f X̂ .
160 W. Bich
The estimates X̂ are themselves affected by errors
(4) X̂ = X̂ − X.
If the errors are sufficiently small, the corresponding error Ŷ = Ŷ − Y is the total
differential

∂ Ŷ
(5) Ŷ ≈ X̂ ,
∂ X̂i X̂ =ξ i
i i
where the partial derivatives are evaluated at the specific estimate values ξi .
One has always an idea of the probable magnitude of the unknown errors, this idea
being supported by some experimental evidence or other knowledge. Therefore, errors
will be considered as zero-mean random variables, RVs [40-42], or

(6) E X̂ = 0,
∞
where E stands for expectation or mean: E(X) = −∞ XgX (ξ)dξ, with gX (ξ), being the
probability density function, PDF, of X.
Condition (6) implies E(Ŷ ) ≈ 0 and, from the error definitions (4) and (2), also
E(X̂) = X and E(Ŷ ) ≈ Y , respectively.
In this way, by squaring formula (5) and by taking the expectations of both sides, one
has

∂ Ŷ ∂ Ŷ

(7) V Ŷ = V Ŷ ≈ cov X̂i , X̂j ,

i j ∂ X̂i ∂ X̂j X̂ =ξ i,j i,j
where V (Ŷ ) = E[(Ŷ − E(Ŷ ))2 ] = E[2Ŷ ] = V (Ŷ ) is the variance and cov(X̂i , X̂j ) is
the covariance between X̂i and X̂j . Since cov(X̂i , X̂i ) = V (X̂i ) = V (X̂i ), if the input
quantities X̂i are uncorrelated, eq. (7) reduces to the perhaps more familiar form
2

∂ Ŷ
(8) V Ŷ ≈ V X̂i .
i ∂ X̂i X̂
i =ξi
The positive square roots of the variances V (X̂i ) are taken as the standard uncertain-
ties of the corresponding estimates, [V (X̂i )]1/2 = u(X̂i ), and similarly for the covariances,
cov(X̂i , X̂j ) = u(X̂i , X̂j ).
An elegant formulation in matrix notation, useful for extension to any number of
measurands (the multivariate case), is

(9) V Ŷ = u2 Ŷ ≈ CY U X̂ CY ,
where
⎡ ⎤
u2 X̂1 ··· u X̂1 , X̂N
⎢ ⎥
⎢ .. .. .. ⎥
(10) U X̂ = ⎢ . ⎥,
⎣ . . ⎦
u X̂N , X̂1 ··· u2 X̂N
is the covariance matrix of X̂.

When Y = (Y1 , . . . , Ym ) , eqs. (1) and (10) become
(11) Y = f (X),
and

(12) U Ŷ ≈ CY U X̂ CY ,
respectively.
In both eqs. (9) and (12)
⎡ ⎤

∂ Ŷ1 ∂ Ŷ1
⎢ ··· ⎥
⎢ ∂ X̂1 ∂ X̂n ⎥
⎢ ⎥
⎢ .. .. .. ⎥
(13) CY =⎢ . ⎥
⎢ . . ⎥
⎢ ⎥
⎣ ∂ Ŷm ∂ Ŷm ⎦
···
∂ X̂1 ∂ X̂n
X̂=ξ
is the Jacobian or sensitivity matrix, where the partial derivatives, evaluated at X̂ = ξ,

take the name of sensitivity coefficients.
Equations (7) and (8) represent the (univariate) law of propagation of uncertainty
in the general case and when the input variables are uncorrelated, respectively. Equa-
tion (12) is a generalization to the multivariate case (see also [43]).
Further generalizations are possible, to the case of implicit multivariate models, i.e.,
f (Y , X) = 0, with real or complex variables. For an exhaustive account, see [44].
Formula (5), and therefore also (7), (8) and (12) are linear, first-order approximations.
Therefore, for a model that is linear in the input quantities, they are exact. For a generic
non-linear model, it is necessary that, for each estimate, model (1) can be considered
linear in the relevant variable in a neighbourhood the size of the error. These formulæ,
which deal with the second moments of PDFs, are valid independently of the specific
shapes of the PDFs themselves, provided that condition (6) is fulfilled.
162 W. Bich
7. – Evaluating uncertainties
.
7 1. General . – In sect. 6 we learned how to propagate through models (1) and (11)
the input variances V (i ) and covariances cov(i , j ). We still need to evaluate them.
The covariances, although per se interesting as a topic and indispensable for a correct
evaluation of uncertainty, will be reluctantly neglected here as beyond the scope of this
paper. For a discussion of an archetypal situation in which covariances arise, see [45].
As to variances, they can be evaluated in essentially two ways. Either by applying
statistics to a data sample, originated from repeated indications, or by using probability
theory, when no data sample is available. The two procedures stem from two distinct
views of probability, viewing it as a limit of relative frequency or as a degree of belief,
respectively, in the occurrence of events. The former is usually called frequentist, the
latter Bayesian, or, perhaps better, subjective.
.
7 2. Random and systematic effects. – The formulæ given in the previous sect. 6 have
been in use for a long time. However, they were used only with random errors. Indeed,
definition (6) is perfectly acceptable in their case, in which the actual estimates X̂ and Ŷ
are realizations of random variables, and the whole apparatus of statistics can be used.
In particular, when a sample X̂ij , (j = 1, . . . , q), is available, the sample mean
1
q
X̂ i = X̂ij
q j=1
is taken as the best estimate for Xi , and the sample standard deviation of the mean
⎡ ⎤1/2
1 2
s X̂ i =⎣ X̂ij − X̂ i ⎦
q(q − 1) j
is taken as an estimate of the true standard deviation σ(X̂i ) of the population supposed
to exist behind the sample, i.e., of the PDF of the RV X̂i . The degrees of freedom
ν = q − 1 give a measure of the reliability of the estimate (X̂ i ) and of its uncertainty
u(X̂ i ) = s(X̂ i ). As a consequence, u(X̂ i ) has itself an uncertainty.
Things are different with systematic errors. With them, there is no sample taken
from a population, therefore statistics cannot be used and, above all, condition (6) has
no easy interpretation. As a consequence, the propagation formulæ given in sect. 6 seem
to be not applicable.
Traditionally, systematic and random errors have been given separate treatments.
Still nowadays, in many fields of physics the situation in this respect has not changed.
A good example can be found, in which the two types of errors are declared separately.
In the two papers announcing the first evidence of a Higgs-like boson [46, 47], the mass
of the new-found neutral boson is given as 126.0 ± 0.4 (stat) ± 0.4 (sys) GeV and 125.3 ±
0.4 (stat) ± 0.5 (syst) GeV, respectively, where “stat” and “sys” (or “syst”) stand for
“statistical”, i.e. random, and “systematic”.
It is desirable that uncertainties arising from both random and systematic effects be
declared in a unique way. This has been the main motivation for publishing the Guide.
The underpinning idea is that a systematic error is a contribution to the measurand
estimate, rather than to its uncertainty [48]. Therefore, it must be estimated, even
roughly, and suitably included in the measurement model as a further input quantity
Xi . The latter correction is equal to the error changed of sign if the error is additive, or
equal to the reciprocal of the error if it is multiplicative.
An estimate of a systematic error is called “bias” in the VIM ([4], definition 2.18).
Experimenters are reluctant to include bias correction in the measurement model, espe-
cially when bias is estimated without supporting experimental data but simply on the
experimenter’s educated judgement, as this procedure is perceived as introducing a sub-
jective element in a scheme otherwise completely objective. In this respect, there are
two remarks, the first being that complete objectivity is an illusion, the second being
that both kinds of estimates, objective and subjective, are equally respectable, the only
measure of their quality being their uncertainty.
Once the bias has been corrected, there remains an uncertainty about hod good
the correction is. This uncertainty goes legitimately in the budget, as it is sometimes
called, of the input uncertainties contributing to the measurand uncertainty u(Ŷ ). The
evaluation of bias uncertainty is the subject of the next subsection.
.
7 3. Frequentist and Bayesian (or subjective) approaches. – A frequentist or, perhaps
better, orthodox statistician is somebody who tends to consider nonsense a question such
as: “What is the probability that a nuclear war will explode in the next five years?”.
Questions of this kind are nonsense in a frequentist context, the reason being that there
is not a population of past nuclear wars from which it could be possible, by repeatedly
sampling from the population, to infer the probability sought. Loosely speaking, in
the frequentist framework probability is viewed as the limit of the relative frequency of
occurrence of events.
Imagine now an experimenter needing a value for, say, the thermal expansion co-
efficient of copper αCu . He goes to a handbook of materials properties, and finds a
value with no further pieces of information, no references, no uncertainty. Of course,
he knows that the value is not exact, therefore some evaluation of its uncertainty is
needed. Also in this case there is no population of data from which a dispersion can
be inferred, so that no frequentist technique can be applied. But even in the happy
case in which some further piece of information is provided, say, a maximum estimated
error for the given value, frequentist procedures are not applicable. Indeed, the exper-
imenter’s task is conceptually very similar to that of the politician needing to estimate
the probability of a nuclear war, or of the geologist wondering about the probability
of a major earthquake. Actually, the task od the experimenter can be formulated as
“given a value αCu = x for the thermal expansion coefficient of copper, what is the
probability that the true value of αCu lies within a given neighbourhood of x?” Or,
“determine a neighbourhood of x in which the true value of αCu lies with a stipulated
probability”.
164 W. Bich
To fulfill this task it is necessary to adopt a different, subjective view of probability,

in which the latter is interpreted as a measure of the degree of belief in the occurrence of
a given event. In terms of quantities, degree of belief translates into state of knowledge.
Specifically, the state of knowledge on a quantity, whatever the source, be it experiment or
subjective judgement, is modelled by a probability density function, or PDF. Accordingly,
the quantity itself is treated as a random variable, which implies an extension in the
meaning of the term “random” to simply include what is incompletely known. According
to this view, everything is random. Even an exact value, such as those discussed in
sects. 1 and 2 can be modelled by a Dirac delta function centred on that value. As
another example, the incomplete knowledge of π is modelled by a (very narrow) PDF, so
that even π is (provocatively) considered a random variable [49]. Whether π is really a
random variable or rather this is only a useful way to convey our state of knowledge on
it, is a philosophical question. The common belief is that π has a unique, (true) value,
regardless of its description in terms of RVs.
The advantage of the subjective approach over the frequentist is that, with it, uncer-
tainties, be they originated by random or systematic effects, are treated on a common
footing. Each of the input quantities is assigned a PDF, or, more generally, a joint PDF
is assigned to the input vector X when its components are not independent. About the
assignment of the appropriate PDF according to the available information, two are the
main tools. The Bayes-Laplace’s rule when a data sample is available (see, e.g. [50, 51]),
and the Principle of Maximum Entropy, PME, in other cases [52]. Extensive guidance is
given in [53].
.
7 4. Change of paradigm. – In the previous paragraph I made a subtle trick, when
stating that a PDF is assigned to X, rather than to X̂. This seemingly innocent change
contains the essence of the Bayesian approach. In it, as already mentioned, the quantities
X, rather than their estimates X̂, are viewed as RVs. If we come back to the definition (4)
of errors, X̂ = X̂ − X, and to their re-interpretation (6) as zero-mean RVs, E(X̂ ) = 0,
we can conclude that in the frequentist view the estimates are viewed as (realizations
of) random variables, and the quantities as fixed, whereas in the Bayesian or subjective
approach the estimates are nothing more than known, or certain, pieces of information
used to assign a PDF to the quantities, which are represented by random variables.
Typically, the PDFs are chosen in such a way that the estimates X̂ are taken as the
expectations of the PDFs themselves.
Once the joint PDF is assigned, its first and second moments, i.e., the expectation
E(X) = (E(X1 ), . . . , E(Xn )) and the uncertainty matrix UX , respectively, can be
obtained and propagated using model (11), as concerns the expectation, and the propa-
gation law (12) (which embeds also the special cases (9) (univariate correlated) and (8)
(univariate uncorrelated)), as concerns the uncertainty matrix. The propagation law is
now rewritten as
(14) U (Y ) ≈ CY U (X)CY ,
in which the hats denoting estimates have been dropped.
At the time of writing this paper the Guide still adopts a fully frequentist approach.
This untenable attitude is obtained by artificially attaching (subjective?) degrees of
freedom to non-statistical uncertainties. The consequencies are unpleasant. Since in this
approach the input uncertainties are estimates, as demonstrated by their degrees of free-
dom, also the uncertainty associated with the measurand has an effective degrees of free-
dom, obtained from those of the input uncertainties by means of a Welch-Satterthwaite
formula [54-56] that has severe limitations and has been questioned [57,58]. This degrees
of freedom indicates that the measurand uncertainty has itself an uncertainty, being an
estimate of a “true” uncertainty, an outdated concept with respect to the now domi-
nating view that uncertainty is a subjective measure of the state of knowledge on the
measurand. In the Bayesian, or subjective approach, on the contrary, exact first and
second moments of PDFs are propagated, which, incidentally, is much more satisfactory
from a theoretical viewpoint. The next Guide, currently under preparation, will adopt
this view, already adopted in its Supplements [44,53]. For a discussion of the anticipated
changes, see [26, 27, 59, 60].
8. – Coverage intervals – propagation of PDFs

Once the input joint PDF gX (ξ) is available (in most cases a product of independent
PDFs, or at most a multivariate Gaussian), the temptation is to directly propagate it
through the model to obtain the PDF gY (η) for the measurand Y (or the joint PDF
gY (η) for the multivariate case). An analytical solution exists in the mathematical
theory of random variables [40, 42], but calculations, involving multiple integrations,
are typically impossible. Even a numerical integration is not viable. A more crude, but
effective way is offered by an implementation of Monte Carlo methods [61-63]. Exhaustive
treatments are given in [53] and [44], as concerns the univariate and multivariate cases,
respectively. With this implementation, coverage intervals (regions) can be obtained for
the measurand(s) at any stipulated coverage probability. These are genuine Bayesian, or
credible, intervals, which completes the transition from point to interval estimation [64].
Of course, also measurand estimates and standard uncertainties can be easily obtained.
Coverage intervals are not necessarily symmetric about the estimate. This is partic-
ularly advantageous when the estimate is close to the lower limit (typically zero) or to
the upper limit of its domain. In these cases, using the first-order approximate GUM
treatment, based on the propagation of first and second moments of the PDF only, ex-
poses to the risk that standard uncertainty encompasses non-physical regions. This risk
is greatly reduced using PDF propagation.
9. – Bayesian inference
The approach described in the previous sect. 7 is certainly subjective, in that it uses
PDFs to encapsulate knowledge, but only weakly Bayesian. Bayes-Laplace rule is actually
used to build a posterior PDF for an input quantity for which a data sample is available,
and the Principle of Maximum Entropy is intrinsically Bayesian. The main outcome
of the procedure are credible intervals, another typically Bayesian feature. However,
166 W. Bich
a full implementation of Bayesian inference implies use of Bayes-Laplace rule taking into
account all the available pieces of prior information, including that concerning the mea-
surand. This knowledge is encapsulated in a prior PDF, which is updated by combining
it with experimental data by using Bayes-Laplace rule. The outcome is a posterior PDF,
from which the PDF for the measurand can be obtained by marginalization.
Prior information on the measurand is not considered in the approach described in
sect. 7, although it can be incorporated in it [65] in a way perhaps less elegant and natural
than with Bayes-Laplace rule [66]. Intuitively, prior information on the measurand can
safely be disregarded when it is meagre compared to that provided by fresh experimental
data, which is often the case in metrology.
Bayesian inference, i.e., full use of Bayes-Laplace rule was traditionally frustrated
by computational difficulties, which were magically solved by Markov Chain Monte
Carlo, MCMC, a beautiful technique of numerical simulation able to provide a mean-
ingful posterior PDF in virtually any application, from cryptography to geology and
meteorology [67]. A huge literature exists on Bayesian inference in various fields, in-
cluding cosmology [68] and astrophysics [69], inverse problems [70] and many others.
For a good overview of Bayesian inference in physics, with a number of case examples,
see [71].
First-order examples of Bayesian inference are some estimators based on least-squares,
such as the minimum-variance and the Kalman estimators [72]. These estimators, espe-
cially the latter, have been in use for a long time, in mass metrology [73] and especially
in time metrology (see, e.g., [74] and related bibliography).
In recent years, Bayesian inference has gained popularity among metrologists, and
measurement results given as posterior PDFs for the measurand or as moments of pos-
terior PDFs are becoming more frequent. For a case example, see [75]. Applications are
reported in [76-78].
There is no doubt that Bayesian inference will be more and more popular in metrology,
in parallel with the its broader interaction with different fields, such as biology, meteo-
rology, laboratory medicine and psychometrics, in which in general it may be difficult, if
not impossible, to define the measurand in terms of a rigorous model.
10. – Conclusion
At the end of this long journey, my hope is that the reader may have reached a clear
view of the intimate connection between uncertainty and probability. Probability theory
is the appropriate framework currently adopted and universally accepted when there is a
need for reasoning under incomplete state of knowledge. So, like probability, uncertainty
can be interpreted in (at least) two ways, as a property of an event (objective view) or
as a state of mind about that event (subjective view). In the former interpretation, the
uncertainty in the occurrence of an event, or in the value of a quantity, would be inherent
in the event or the quantity. In the latter, events can occur or not, and quantities have a
value, with no uncertainty or probability attached to them. We assign an uncertainty to
these events or quantities based on our feelings about them. These feelings can be totally
intuitive or based on educated guess, inspired by objective data, but they are ultimately
subjective.
Uncertainty is a personal matter; it is not the uncertainty but your uncertainty.

Dennis Lindley [79]
∗ ∗ ∗
The author is grateful to his colleagues of the Joint Committee for Guides in Metrology
(JCGM) Working Group 1 (GUM) for the many fruitful discussions that contributed to
form the ideas expressed in this paper. However, they remain personal views.
REFERENCES
[1] Bich W., Error, uncertainty and probability, in Proceedings of the International School of
Physics “Enrico Fermi”, Course 185, “Metrology and Physical Constants”, edited by E.
Bava, M. Kuehne and A. M. Rossi (IOS Press, Amsterdam and Società italiana di fisica,
Bologna) 2013, pp. 47–73.
[2] Trueb P., π e trillion digits for π, (2016), available at http://www.pi2e.ch/.
[3] Maxwell J. C., A Treatise on Electricity and Magnetism (Oxford University Press) 1873.
[4] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML, International vocabulary
of metrology — Basic and general concepts and associated terms (VIM), JCGM 200:2012,
http://www.bipm.org/utils/common/documents/jcgm/JCGM 200 2012.pdf.
[5] De Boer J., Metrologia, 31 (1995) 405.
[6] Wikipedia entry: http://en.wikipedia.org/wiki/FFF system.
[7] Mari L. and Giordani A., Metrologia, 49 (2012) 756.
[8] Stevens S. S., Science, 103 (1946) 677.
[9] Dybkaer R., Accred. Qual. Assur., 16 (2011) 479.
[10] Level of measurement. Available: http://en.wikipedia.org/wiki/Level of measurement.
[11] White R., Accred. Qual. Assur., 16 (2011) 31.
[12] Dybkaer R., Accred. Qual. Assur., 16 (2011) 649.
[13] Mohr P. J., Newell D. B. and Taylor B. N., Rev. Mod. Phys., 88 (2016) 035009.
[14] Dirac P. A. M., Proc. R. Soc. A, 165 (1938) 199.
[15] Bureau international des poids et mesures, BIPM, Le Système international
d’unités, 8e édition, 2006, http://www.bipm.org/utils/common/pdf/si brochure 8.pdf.
[16] Evenson K. M. et al., Phys. Rev. Lett., 29 (1972) 1346.
[17] BIPM, http://www.bipm.org/en/measurement-units/rev-si/.

[18] Leonard B. P., Metrologia, 47 (2010) L5.
[19] Price G., Accr. Qual. Assur., 16 (2011) 121.
[20] Cabiati F. and Bich W., Metrologia, 46 (2009) 457.
[21] Aitken C., Roberts P. and Jackson G., Communicating and Interpreting Statistical
Evidence in the Administration of Criminal Justice – Practitioner Guide N. 1,
Fundamentals of Probability and Statistical Evidence in Criminal Proceedings (The Royal
Statistical Society, 2010) http://www.rss.org.uk/site/cms/contentviewarticle.asp?
article=1132.
[22] Puch-Solis R., Roberts P., Pope S. and Aitken C., Communicating and Interpreting
Statistical Evidence in the Administration of Criminal Justice – Practitioner Guide N.
2, Assessing the probative value of DNA evidence (The Royal Statistical Society, 2012)
http://www.rss.org.uk/site/cms/contentviewarticle.asp?article=1132.
168 W. Bich
[23] Wallace J., Science & Justice, 50 (2010) 182.

[24] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML, Evaluation of measurement
data — Guide to the Expression of Uncertainty in Measurement, JCGM 100:2008, GUM
1995 with minor corrections. http://www.bipm.org/utils/common/documents/jcgm/
JCGM 100 2008 E.pdf.
[25] The GUM and GUM-related documents. Available: http://www.bipm.org/en/
publications/guides/gum.html.
[26] Bich W., Cox M. G. and Harris P. M., Metrologia, 43 (2006) S161.
[27] Bich W., Cox M. G, Dybkaer R., Elster C., Estler W. T., Hibbert B., Imai H.,
Kool W., Michotte C., Nielsen L., Pendrill L., Sidney S., van der Veen A. M. H.
and Wöger W., Metrologia, 49 (2012) 702.
[28] Bich W., Metrologia, 51 (2014) S158.
[29] Bich W., Cox M. G. and Michotte C., Metrologia, 53 (2016) S149.
[30] Bridgman P. W., The Logic of Modern Physics (Mc Millan) 1927, reprinted in 1960,
Arno Press.
[31] Organisation Internationale de Métrologie Légale, International Recommen-
dation R 111, OIML, Paris 2004, http://www.oiml.org/publications/R/
R111-1-e04.pdf.
[32] Meija J. et al., Pure Appl. Chem., 88 (2016) 265.
[33] IEC, 60050-300:2001, International Electrotechnical Vocabulary - Electrical and electronic
measurements and measuring instruments - Part 311: General terms relating to
measurements - Part 312: General terms relating to electrical measurements - Part 313:
Types of electrical measuring instruments - Part 314: Specific terms according to the type
of instrument. On line at: http://www.electropedia.org/iev/iev.nsf/display?
openform&ievref=311-02-01.
[34] Bich W., Accred. Qual. Assur., 14 (2009) 581.
[35] Bich W. and Pennecchi F., Metrologia, 49 (2012) 15.
[36] Bich W., Cox M. G. and Harris P. M., Metrologia, 30 (1993/94) 495.
[37] Lewis F. L., Optimal Estimation (John Wiley & Sons, New York, Chichester, Brisbane,
Toronto, Singapore) 1986.
[38] Beck J. V. and Arnold K. J. Parameter estimation in engineering and science (John
Wiley & Sons, New York, London, Sydney, Toronto) 1977.
[39] Bich W., Estimation and uncertainty in metrology, in Proceedings of the International
School of Physics “Enrico Fermi”, Course CXLVI, “Recent Advances in Metrology and
Fundamental Constants,” edited by T. J. Quinn, S. Leschiutta and P. Tavella (IOS
Press, Amsterdam and SIF, Bologna) 2001, pp. 653–688.
[40] Bickel P. J. and Docksum K. A., Mathematical Statistics: Basic Ideas and Selected
Topics (Prentice Hall, Englewood Cliffs, New Jersey) 1977.

[41] ISO, ISO 3534-1. Statistics – Vocabulary and symbols – Part 1: Probability and general
statistical terms, International Standards Organization, Geneva, 2006.
[42] Gibbons J. D. and Chakraborti S. Nonparametric Statistical Inference, 3rd edition
(Marcel Dekker, New York, Basel, Hong Kong) 1992.
[43] Brandt S., Data Analysis, third edition (Springer, New York) 1999.
[44] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML, Evaluation of
measurement data — Supplement 2 to the “Guide to the Expression of Uncertainty
in Measurement” — Extension to any number of output quantities, JCGM 102:2011,
http://www.bipm.org/utils/common/documents/jcgm/JCGM 102 2011 E.pdf.
[46] The ATLAS Collaboration, Phys. Lett. B, 716 (2012) 1.
[47] The CSM Collaboration, Phys. Lett. B, 716 (2012) 30.
[48] Müller J. W. “The Assignment of Uncertainties to the Results of Experimental

Measurements”, in Precision Measurements and Fundamental Constants II, edited by
B. N. Taylor and W. D. Phillips (Natl. Bur. Stand., US) Spec. Publ. 617 1984, pp. 375–
381.
[49] Press S. J., Bayesian Statistics: Principles, Models, and Applications (John Wiley &
Sons, New York, Chichester, Brisbane, Toronto, Singapore) 1989; Meas. Sci. Technol., 4
(1993) 1.
[50] Lira I. and Kyriazis G., Metrologia, 36 (1999) 163.
[51] Kacker R. and Jones A., Metrologia, 40 (2003) 235.
[52] Weise K. and Woeger W., Meas. Sci. Technol., 4 (1993) 1.
[53] BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML, Evaluation of measurement
data — Supplement 1 to the “Guide to the Expression of Uncertainty in Measurement” –
Propagation of distributions using a Monte Carlo method, JCGM 101:2008, http://www.
bipm.org/utils/common/documents/jcgm/JCGM 101 2008 E.pdf.
[54] Satterthwaite F. E., Psychometrika, 6 (1941) 309.
[55] Welch B. L., Biometrika, 34 (1947) 28.
[56] Welch B. L., Biometrika, 29 (1938) 350.
[57] Ballico M., Metrologia, 37 (2000) 61.
[58] Hall B. D. and Willink R., Metrologia, 38 (2001) 9.
[59] Bich W., IEEE Trans. Instrum. Meas., 61 (2012) 2153.
[61] Fishman G. S., Monte Carlo, Concepts, Algorithms, and Applications (Springer, New
York) 1996.
[62] Manly B., Randomization and Monte Carlo methods in Biology (Chapman and Hall,
London) 1991.
[63] Noreen E., Computer Intensive Methods for Testing Hypotheses (John Wiley & Sons,
New York) 1989.
[64] Sivia D. and Skilling J., Data Analysis – A Bayesian Tutorial, second edition (Oxford
University Press) 2006.
[65] Elster C., Metrologia, 44 (2007) 111.
[66] Possolo A. and Toman B., Metrologia, 44 (2007) 464.
[67] Diaconis P., Bull. Am. Math. Soc., 46 (2009) 179. Available at http://www.ams.org/
journals/bull/2009-46-02/S0273-0979-08-01238-X/S0273-0979-08-01238-X.pdf.
[68] Christensen N., Meyer R., Knox L. and Luey B. B., Astrophys. J., 826 (2016) 41.
[69] Stenning D. C. et al., Class. Quantum Grav., 18 (2001) 2677.
[70] Wang J. and Zabaras N., Inverse Problems, 21 (2005) 183.
[71] Dose V., Rep. Prog. Phys., 66 (2003) 1421.
[72] Luenberger D. G., Optimization by Vector Space Methods (John Wiley & Sons, New
York, Chichester, Brisbane, Toronto, Singapore) 1969.
[73] Bich W., Metrologia, 29 (1992) 15.
[74] Arias E. F., Panfilo G. and Petit G., Metrologia, 48 (2011) S145.
[75] Klauenberg K. and Elster C., Metrologia, 53 (2016) S32.
[76] Elster C. and Wübbeler G., Metrologia, 53 (2016) S10.
[77] Allard A. et al., Metrologia, 53 (2016) S1.
[78] Bodnar O. et al., Metrologia, 53 (2016) S46.
[79] Lindlay D. V., Understanding Uncertainty, revised edition (Wiley, New York) 2014.
DOI 10.3254/978-1-61499-818-1-171
International recognition of NMI calibration

and measurement capabilities: The CIPM MRA
Andy Henson
International Liaison and Communication Department,
Pavillon de Breteuil, F-92312 Sèvres Cedex, France
Summary. — The CIPM Mutual Recognition Arrangement (CIPM MRA) has now
been signed by the representatives of 101 institutes and covers a further 153 institutes
designated by these signatories. These come from 57 Member States, 40 Associate
States and Economies of the CGPM, and 4 international organizations. Through
it the national metrology institutes can demonstrate their measurement abilities
and publish internationally recognized statements of their so-called calibration and
measurement capabilities (CMCs). All the data are openly available in the CIPM
MRA database (the KCDB), which has become an essential reference for the NMIs
themselves, accredited laboratory community as well as a small number of high

end industrial and other organisations. In this paper we review the situation that
led to the development of the CIPM MRA, identifying the three main drivers: the
challenges of regulators wanting traceability to the national NMI in an increasingly
globalised world; the emergence of a laboratory accreditation and with it the need
for laboratories to demonstrate metrological competence; and finally the emergence
and strengthening of the Regional Metrology Organizations (RMOs). The paper
also addresses the CIPM MRA structure, its operational mechanisms and impact,
and concludes with some speculative remarks as to how it might evolve in the future.

171
172 Andy Henson
1. – Introduction
The CIPM Mutual Recognition of national measurement standards and of calibration

and measurement certificates issued by national metrology institutes (known as the CIPM
Mutual Recognition Arrangement, or CIPM MRA) [1] was signed in Paris on 14 October
1999 by the Directors of 38 National Metrology Institutes (NMIs) and two international
organizations. Today, it provides a primary source to identify internationally recognized
national capabilities within the NMI and wider metrology community. The underpinning
science and the outcomes are openly available to all interested parties.
The CIPM MRA is built on demonstrated and peer-reviewed capabilities provid-
ing a high level of confidence and trust. Participation in the CIPM MRA has grown
substantially since its launch. At the end of September 2015 it had been signed by
the representatives of 98 institutes — from 54 Member States, 40 Associate States and
Economies of the CGPM, and four international organizations — and covers a further
152 institutes designated by the signatory bodies [2].
The requirement to participate in scientific comparisons with international counter-
parts, together with the peer-review process covering both the quality management sys-
tem and the individually declared Calibration and Measurement Capabilities (CMCs),
ensures a high degree of rigour within the CIPM MRA. This in turn provides the basis
for and underpins the international mutual recognition. The names of the participants,
the comparison reports and results, and the CMCs are all publicly and freely available in
the BIPM key comparison database, the KCDB [3], which is maintained by the BIPM.
By March 2016, 933 Key Comparisons, 454 Supplementary Comparisons and 24 561
peer-reviewed CMCs were listed in the KCDB [4].
Through the Joint Committee of the Regional Metrology Organizations and the BIPM
(JCRB) [5], the BIPM operates the CMC inter-regional review website; the BIPM also
chairs the JCRB and provides the Executive Secretary for it. The Executive Secretary
position has always been a secondment position from one of the Member State NMIs to
the BIPM, the secondment typically being for a two year period.
After more than 15 years of operation the CIPM MRA has matured into a well-
recognized pillar of the international quality infrastructure. The CIPM MRA is signed
by the Directors of NMIs and as such it is not binding on Governments. Nevertheless,
over time the CIPM MRA database has become the essential reference on the interna-
tionally accepted calibration and measurement capabilities of the NMIs and Designated
Institutes (DIs). One unforeseen aspect of this success has become evident in recent
years. The Directors of the major participating laboratories, whose staff bear the brunt
of the workload of piloting comparisons and reviewing CMCs, had understood that in
the early phase there would be a significant workload. What they had not anticipated
was just how the CIPM MRA would continue to expand in both scope and participation,
with the consequent continued and ongoing high workload which remains today. After a
decade and a half it was time to review the implementation and operation of the CIPM
MRA and ensure its sustainability for the coming years. That review is currently well
advanced.
The CIPM MRA 173
This paper will look at some of the original motivations which led to the CIPM
MRA being drawn up; review its structures, mechanisms and operation; and consider its
impact, as well as its evolution over time. The article will conclude with the author’s
speculations as to what the future might hold for the Arrangement.
2. – The origins of the CIPM MRA
The backdrop to the CIPM MRA was the major increase in world trade triggered by
the General Agreement on Tariffs and Trade [6]. GATT provided fair trade rules and
led to the gradual reduction of tariffs, duties and other trade barriers, for goods, services
and intellectual property. Although GATT began in 1947, the early decades focused on
preventing increased tariff barriers rather than on any reduction in tariffs. However, in
its later years, tariff barriers began to tumble and world trade expanded accordingly. The
final round — the 1986-1994 Uruguay Round [6] — led to the creation of the World Trade
Organization (WTO). As fiscal barriers to trade were reduced, non-tariff barriers, and
the need to address them, were brought into far sharper focus, leading to the Technical
Barriers to Trade (TBT) Agreement which first came into force alongside the WTO in
1995. The need to measure consistently when performing conformity assessment, and
to have those measurements accepted across trading partners, was fundamental to an
increasingly globalized world.
In addition to these changes on the world stage, three different developments collec-
tively reinforced the need for a comprehensive and coherent solution to establishing and
demonstrating the degree of international equivalence at the NMI level. It is worth look-
ing a little more closely at each of these drivers, and at the response of the international
community.
The first driver related to countries trading with the largest economy in the world, the
USA. Many US regulators required instruments to be specifically calibrated by the US
NMI, the National Institute of Standards and Technology (NIST). Whilst comparability
related to the SI base units, and particularly the metre and kilogram, were established
under the auspices of the CIPM, the need for confidence in industrially relevant derived
units such as pressure or force was becoming a major issue. In particular, regulators
in the USA, and consequently the companies supplying equipment with their sphere of
regulation, were demanding that calibrations must be “traceable to NIST” (rather than
“traceable to the SI”). That is to say the only acceptable traceability route for instrument
calibration for some regulators in the USA was via the USA’s NMI, irrespective of whether
or not the manufacturer was in the USA. This represented a major cost and delay hurdle
for non-USA manufacturers and at the same time placed an unwelcome additional burden
on NIST, which was increasingly being requested to calibrate instruments for non-USA
customers. From the early 1980s onwards the NMIs of the major trading nations were
able to ease the burden for their manufacturers by concluding bilateral agreements with
NIST, often supported by bilateral comparisons, demonstrating the equivalence of their
measurement standards with those held at NIST. NIST in turn educated USA regulators
as best they could, by providing the regulators with advice and a technical understanding
174 Andy Henson
of the issue, often on a case-by-case basis. This enabled the regulator, if they so chose,
to accept the foreign NMI’s calibration of the instrument manufacturer’s equipment.
The CIPM recognized this trend as early as 1983 [7], although it was initially reluctant
to intervene in what was essentially an issue between pairs of sovereign nations. It
recognized, however, that if the same approach were widely adopted by other nations
it could result in an unmanageable web of bilateral relationships, and also that the
NMIs and industries of smaller nations were at risk of being disadvantaged, without the
“trade clout” to warrant the priority of concluding multiple bilateral agreements and
conducting the supporting comparisons for their NMI. By 1986 the first tentative and
modest concepts for some sort of centrally coordinated approach were being discussed
and considered within the CIPM. It took more than ten years, and the two other key
drivers, before the CIPM reached consensus to move forward.
The second driver related to the emergence of a laboratory accreditation system in
the European Community. By the early 1990s certification of manufacturers, and con-
sequently accreditation of the calibration and testing laboratories supporting them, was
becoming increasingly prevalent. Laboratory accreditation in Europe expanded partic-
ularly rapidly, as a way of ensuring industrial confidence in calibrations throughout the
European Community (now the European Union), with its many different languages,
cultures and legal systems. Before long, the accreditation system was adopted, with
some variations in implementation, worldwide. Accreditation of laboratories systemi-
cally formalized the need to have and to demonstrate metrological competence under-
pinned by credible metrological traceability. The International Laboratory Accreditation
Cooperation (ILAC) began to appraise the international metrology community of the
needs of their national accreditation bodies, accredited calibration and test laboratories
(of which there are now some 50000 worldwide). These — largely commercial — cal-
ibration labs were required to demonstrate that their equipment had been calibrated
by a competent body and that they had appropriate metrological traceability. The
accreditation community needed some way to be sure that the measurement capabil-
ities claimed by the NMIs were justified, and ILAC called for a database to provide
information and assurance on sources of reliable, internationally recognized metrological
traceability.
The third driver arose in parallel with the emergence and strengthening of regional
metrology cooperation through what are now known as Regional Metrology Organiza-
tions (RMOs). Emerging from the 1977 Commonwealth Science Council Initiative, the
Asia Pacific Metrology Programme (APMP) was established in 1980. This was followed
by EUROMET, the European Association of National Metrology Institutes (now EU-
RAMET) in 1987; the Inter-American Metrology System (SIM) in 1988; the Euro-Asian
Cooperation of National Metrological Institutions (COOMET) in 1991; and the Inter-
Africa Metrology System (AFRIMETS) in 2007. As scientific comparisons of measure-
ment standards are the key tool enabling NMIs to benchmark their scientific progress, as
well as to demonstrate their service delivery capabilities, the RMOs began to organize
regional comparisons to provide evidence of comparability for their members whose ca-
pabilities were not sufficiently advanced for them to participate at the global level. There
The CIPM MRA 175
was an increasingly urgent need to somehow tie together these regional comparisons with
those of the CIPM and the BIPM.
By 1994, with growing acceptance of the need for a coordinated worldwide activity,
the CIPM outlined a draft Resolution to be put before the CGPM at its meeting in
October 1995. This resolution, which was adopted as Resolution 2 (1995) [8], set out the
principles for what would become the CIPM MRA four years later.
Thus, towards the end of 1995 the foundation for the CIPM MRA had been laid.
What lay ahead was a protracted consultation and drafting exercise undertaken by the
BIPM and involving the CIPM and the Directors of the NMIs (who would be the future
signatories). The main elements were outlined by the time of the CIPM meeting in
1996, but fine tuning of the details, and winning support across the breadth of NMIs
took a considerable amount of work and diplomatic skill from all involved. A meeting
of Directors of NMIs took place at the BIPM in February 1997 at which a draft of the
CIPM MRA was discussed line by line with the Directors. Shortly after a meeting of
representatives of the Regional Metrology Organisations, that would take on a formal
responsibility within the CIPM MRA, ensured that they understood and were happy
with their proposed role. A second meeting of NMI Directors took place in 1998 where
again a detailed discussion of the text took place. Thus the CIPM MRA was, in effect,
drawn up by the ensemble of NMI Directors. Nevertheless a number of challenges had to
be addressed. In particular, many NMIs were not used to the concept of routinely having
their capabilities internationally peer reviewed, a cornerstone of the proposed process,
and it took time for some of them to become comfortable with this approach. Some NMIs
did not operate formal quality management systems and would have to develop them.
Further, at that time ISO Guide 25 (now ISO/IEC 17025) was not universally accepted
for use by all NMIs, so the requirements for laboratory quality management systems
had to be resolved to the satisfaction of the expected participants. The approach to
establishing reference values and the meaning of equivalence had to be looked at afresh.
Resolving how best to address different perspectives on a wide variety of detailed issues
took time.
3. – Launch of the CIPM MRA

In October 1999, during the 21st meeting of the CGPM, Resolution 2 was adopted,
formally paving the way for signature of the CIPM MRA. In Resolution 2 (1999) the
CGPM invited [8]:
– all Member States of the Metre Convention to participate in the arrangement by

giving authority to the director of the designated national metrology institute in
their country to sign the arrangement,
– all Member States to make every effort to implement the arrangement and to
encourage other authorities in their country to recognize the equivalence of na-
tional measurement standards and calibration and measurement certificates thereby
demonstrated,
176 Andy Henson
– all States to use this arrangement as the basis for recognizing the national measure-
ment standards and calibration and measurement certificates of signatory national
metrology institutes.
On 14 October 1999 the Directors of the NMIs from 38 Member States and two
international organizations signed the document, and the CIPM MRA was finally under
way.
In parallel, following consultation with the WTO, consideration was given to ensure
that the CIPM MRA did not itself become a technical barrier to trade. The CIPM
created a new status of “Associates of the Conférence Générale des Poids et Mesures”
to allow States that were not yet ready to become Member States, (and in special cases
Economies) the opportunity to participate in the CIPM MRA. Thus alongside Resolution
2 (1999) addressing the CIPM MRA, the 21st CGPM also adopted Resolution 3 (1999) [8]
to lay out the basis for participation in the CIPM MRA by NMIs from Associate States
or Economies. Depending on the size of a state’s economy, Associates were allowed to
participate in the CIPM MRA with a subscription as low as one tenth of the minimum
contribution that would be paid if they were a Member State(1 ).
4. – Structure and mechanisms of the CIPM MRA
The objectives of the CIPM MRA are [1] to establish the degree of equivalence of
national measurement standards maintained by NMIs; to provide for the mutual recog-
nition of calibration and measurement certificates issued by NMIs; and thereby to provide
governments and other parties with a secure technical foundation for wider agreements
related to international trade, commerce and regulatory affairs.
NMI directors sign the CIPM MRA with the approval of the appropriate authorities
in their own country and thereby accept the process specified in the CIPM MRA for
establishing the CIPM MRA database. They agree to recognize the results of key and
supplementary comparisons as stated in the CIPM MRA database and to recognize
the published Calibration and Measurement Capabilities (CMCs) of other participating
NMIs and DIs(2 ).
A limited number of international organizations also participate. There are cur-

rently four such organizations [2]: the European Space Agency (ESA); the International
Atomic Energy Agency (IAEA); the Institute for Reference Materials and Measurements
(IRMM); and the World Meteorological Organization (WMO).
(1 ) This was revisited by the 24 th CGPM in 2011, with the adoption of Resolution 4 [7],
which increased the minimum subscription level for an Associate State to one fifth of that for a
Member State, and also imposed increases in subscriptions for those Associate States that have
been Associates for five years, and which have reached a certain level of engagement with the
CIPM MRA, yet choose to remain Associates rather than accede and become a Member State.
(2 ) It is important to understand that signature of the CIPM MRA engages NMIs but not
necessarily any other agency in their country.
The CIPM MRA 177
The three fundamental elements leading to approval of an institute’s CMCs are

1. participation by the institute in reviewed and approved scientific comparisons;
2. operation by the institute of an appropriate and approved quality management
system;
3. international peer-review (regional and inter-regional) of claimed calibration and
measurement capabilities.
A generalized overview of the process is given below, but for full details please refer
to the text of the CIPM MRA, available on the BIPM website [1].
The outcomes of the CIPM MRA are published, internationally recognized statements
of the CMCs of the participants [3]. The technical basis relies on demonstrated compe-
tence through international key and supplementary comparisons, and the operation of
peer-reviewed quality systems at the NMIs.
Participating institutes are required to operate an appropriate quality system (essen-
tially this currently means ISO/IEC 17025, and for those providing reference materials,
ISO Guide 34, soon to be ISO 17034) which must cover the calibration and measure-
ment capabilities that are to be declared through the CIPM MRA. Due to geographic,
technical and organizational differences between the RMOs, each has tailored its quality
management system review process to be optimal for its own region, whilst remaining
within the JCRB guidelines. For example in APMP, where most laboratories are also ac-
credited, the review process is closely integrated with the regional accreditation system.
Assessors are jointly chosen and the assessment evidence used for both accreditation and
the CIPM MRA, avoiding duplication of effort.
Following satisfactory participation in appropriate comparisons the participating in-
stitutes declare their CMCs, which are subject to two rounds of peer review. The CMCs
are firstly reviewed within the RMO of which the declaring institute is a member. After
any comments have been resolved at the RMO level, the CMCs are subject to a second
round of inter-regional review by the other RMOs. This second interregional review is
carried out in parallel by the various RMOs. The outcome consists of internationally
recognized statements of the measurement capabilities of the participating institutes.
All the data are openly available in the CIPM MRA database [3], which is maintained
by the BIPM and publicly available on the internet. The database, widely known as the
KCDB (the BIPM key comparison database, which goes far beyond just details of the
comparisons), comprises four parts:
– Appendix A listing the signatory NMIs together with any designated institutes;
– Appendix B with full details of the registered comparisons;
– Appendix C listing the internationally approved Calibration and Measurement Ca-
pabilities (CMCs);
– Appendix D listing Key Comparisons (although this is somewhat redundant in
practice given the information in Appendix B).
178 Andy Henson
The overall coordination is by the BIPM under the authority of the CIPM. The
Consultative Committees of the CIPM, the RMOs and the BIPM are responsible for
carrying out the key and supplementary comparisons. The Joint Committee of the
Regional Metrology Organizations and the BIPM (JCRB) [5] is charged with coordinating
the activities among the RMOs, particularly with regard to the inter-regional CMC
review.
.
4 1. Key and supplementary comparisons. – The scientific comparisons are the basic
building block that enables NMIs to show they are ‘getting the right answer’ and appro-
priately estimating the uncertainties of their results. The subjects of key comparisons
are decided by the CCs, and these same subjects are often also adopted as key compar-
isons by the RMOs, which may in addition also undertake supplementary comparisons
to address specific measurement requirements.
At the launch of the CIPM MRA a formal transition period was defined [1] as running
until the first wave of key and supplementary comparisons had been completed. In this
period some flexibility was exercised, recognizing that it took some time for the processes
to catch up with a backlog of reviews, of, for example, the quality systems. The transition
period was deemed to have been concluded at the end of 2003 (and at the end of 2005
for Chemistry).
In these early years the NMIs, RMOs, and the CIPM Consultative Committees were
hard at work organizing and conducting the comparisons and analysing the results. At
the same time, whilst some NMIs had formal quality systems, many did not, and had
to begin developing a suitable system from scratch. All the quality systems then had
to be taken through the RMO peer-review system. The NMIs and DIs also worked on
developing their own CMCs and reviewing the CMCs of other NMIs and DIs.
By May 2004, just after the end of the transition period, 470 key comparisons had
been registered in the KCDB [4], among which 324 were conducted by the CCs and
BIPM, and 146 were conducted by one of the five RMOs participating in the JCRB.
Of these, about one fifth were conducted before the CIPM MRA was signed and so did
not necessarily fully meet all procedural aspects; however their results were considered
to provide “Provisional equivalence” which allowed them to be used to support CMC
declarations.
Figure 1 shows the cumulative listing of comparisons registered in the KCDB. Since
2003, the average rate of registration of new key comparisons has been roughly constant at
about 40 new key comparisons per year (fig. 2). A slight reduction of this rate can be seen
in recent years, probably resulting from the strategic planning exercises carried out since
2013, which have led to some rationalization. It is clear that new comparisons continue
to be needed, both because new capabilities need to be underpinned, and because the
original comparisons become old and need to be repeated.
The number of RMO supplementary comparisons, on the other hand, shows a modest
but steady increase, perhaps driven by RMO members who are not able to operate at
the highest levels of metrology and who are developing capability and needing to partic-
ipate in comparisons. Supplementary comparisons are typically conducted for two main
The CIPM MRA 179
Fig. 1. – Total number of key comparisons and supplementary comparisons registered in the
KCDB.
Fig. 2. – Number of new comparisons registered in the KCDB over the one-year period ending
at the date indicated on the x-axis [4].
reasons. Firstly, an NMI may miss a comparison cycle and need to demonstrate its capa-
bilities. Secondly, the RMO may have specific regional needs that are not covered by the
key comparisons. Key comparisons only address the key techniques or “pinning points”,
an RMO may wish to undertake a comparison related to a more specific technique that
is not considered key. This may be because they have member NMIs who are not able
to participate in the high-level metrology addressed in key comparisons, but nonetheless
need to be able to support CMC claims.
180 Andy Henson
Fig. 3. – Evolution of the number of CMCs listed in the KCDB [4].
.
4 2. Calibration and Measurement Capabilities (CMCs). – CMCs are initially reviewed
by the NMI creating them and then by experts drawn from the relevant technical com-
mittee of the supporting RMO. Once the initiating RMO is satisfied, the CMCs are
submitted into an inter-regional review process. The CMCs, usually in batches, are
posted on a dedicated page of the JCRB website. The JCRB has instituted deadline
requirements in the inter-regional review to prevent it from stalling due to inaction by
the reviewing RMOs. RMOs must indicate their intention to review within three weeks
of the CMC file being posted, using a standardized online process, otherwise their review
rights are lost. There is no fixed deadline for reviews because the size of a batch and the
complexity of the CMCs within any given batch vary enormously. RMOs set and post
their own review deadlines, but having done so they must respect them. Historically, all
batches have been reviewed by all RMOs, but more recently some CCs have organized
themselves by dividing up the review work to reduce the amount of redundancy in the
review process. However, all CMCs must be reviewed by experts from at least one addi-
tional RMO. In the vast majority of cases, even today, CMCs are reviewed by more than
one region at the inter-regional stage.
Figure 3 shows the total number of published CMCs as a function of time. By May
2004, just after the end of the transition period, the KCDB contained more than 17 350
CMCs, around two thirds of the number in the database today. By the time of the ten-
year anniversary in late 2009 there were just over 21 339 CMCs published in the KCDB
and the number of registered comparisons had doubled to 664 key comparisons and 218
supplementary comparisons.
The number of CMCs continued on an upward trend until March 2013, when the curve
has flattened off. However, interpretation of the numbers is complicated by changes that
have been made to the way some of the CMCs are formulated. In late 2004 EUROMET
The CIPM MRA 181
Fig. 4. – Who visits the KCDB? [4].
(now EURAMET) introduced the concept of using uncertainty tables for current and
voltage transfer allowing uncertainty information to be displayed in a more succinct way
and reducing the need for the NMIs to make multiple line entries. This resulted in a drop
in the number of CMCs, but not of course in the amount of information in the KCDB.
In 2013 EURAMET decided to adopt the uncertainty tables (sometimes referred to as
the uncertainty matrix) across its entire portfolio of Electricity and Magnetism CMCs,
and in 2015 APMP followed suit.
By 2005 it was realized that a mechanism was needed to allow for temporary suspen-
sion of CMCs. This became known, somewhat misleadingly, as “greying out” of CMCs.
For the KCDB users these greyed out CMCs are invisible and inaccessible, but they
remain in the database ready to be reinstated when appropriate evidence of addressing
the reason for suspension has been provided. Much later it was realized that some CMCs
sat in this greyed out status for long periods, and formal procedures were introduced to
handle both the greying out and the reinstatement or deletion of CMCs.
In 2004 at the suggestion of a number of NMIs a CIPM MRA logo [9] was adopted.
This allows those NMIs which have been granted permission to include the CIPM MRA
logo on calibration certificates covered by CMCs published in the KCDB. Much more
recently this has been extended to verification certificates (particularly important in
COOMET), and to Certified Reference Material (CRM) documentation, again provided
they are covered by CMC entries in the KCDB.
182 Andy Henson
5. – The CIPM MRA today
At the time of writing, the CIPM MRA has been signed by the representatives of
101 institutes — from 57 Member States, 40 Associates of the CGPM, and 4 international
organizations — and covers a further 153 institutes designated by the signatory bodies [2].
The KCDB website receives approximately 11 000 unique visits per month [4] (dis-
counting minor visits where little is examined in the database). Although not surprisingly
the NMI community is the single largest community visiting the KCDB (see fig. 4), there
are substantial numbers of external visitors too, most notably from calibration and test
laboratories.
NMIs access the database for many reasons, for example to check and benchmark their
own capability, to assess the state of the art during the CMC review process, or perhaps to
source traceability for national standards that are not primary. Also, in many countries
the NMI acts as a “portal” for regulators and other users. That is to say the third party
enquiry is addressed to the NMI, which in turn uses its knowledge of and familiarity with
the database and its contents, together with its expertise in understanding measurement
challenges, to provide advice to the client or customer. Consequently, many of the NMI
database visits may be to service external enquiries.
As of March 2016, the KCDB included a total of 24 561 CMCs [4]:
– 14 433 in General Physics,
– 4 050 in Ionizing Radiation, and
– 6078 in Chemistry.
The distribution of CMCs among the RMOs is very uneven, as can be seen below:
– AFRIMETS: 451,
– APMP: 5 612,
– COOMET: 2 344,
– EURAMET: 10 949,
– SIM: 4 974.
The balance not included in the RMO distribution comes from the international
organizations.
As of March 2016, the KCDB covered 933 key comparisons, with the distribution of [4]:
– 91 from the BIPM,
– 457 from the CCs,
– 4 from AFRIMETS,
– 130 from APMP,
– 44 from COOMET,
– 151 from EURAMET, and
– 56 from SIM.
The CIPM MRA 183
One of the wider objectives quoted in the CIPM MRA is to provide governments and
other parties with a secure technical foundation for wider agreements related to interna-
tional trade, commerce and regulatory affairs. The most obvious expression of this, and
the widespread recognition and acceptance of the KCDB, was reflected in its inclusion
as a reliable and convenient source of internationally accepted traceable measurements
in the 2013 “ILAC-P10:01/2013. ILAC Policy on the Traceability of Measurement Re-
sults” [10]. In this way the 50 000 accredited calibration and testing and medical labo-
ratories worldwide are all linked back into the international system, helping ensure the
unbroken chain of measurements used by industry and wider society to the SI.
There are many examples of where the CIPM MRA has had a practical impact and a
number of these have been summarized by KRISS, the Korean NMI. One example relates
to a Korean manufacturer contracted to develop, manufacture and delivers two special
oil offshore platforms to an oil consortium operating in the Russian Federation. Such
platforms contain thousands of instrumentation loops, and in this case some 600 loops
subject to state metrological control, comprising some 10000 measuring instruments of
approximately 60 different types. The requirement was for traceability to the national
measurement standards of the Russian Federation, verified by Russian Federation Ver-
ification Officers, or by some body accredited and authorized by them. VNIIMS, the
Russian NMI, relied on the joint participation of Russian and Korean NMIs in the CIPM
MRA as justification for calibration to take place in Korea, saving 16 million $US. A
similar building project, this time for an oil major operating in the USA, relied on the
participation of KRISS and NIST in the CIPM MRA, allowing in country calibration
leading to a saving of some 11 million $US. A third involved a Mexican automobile
parts manufacturer, maintaining SI traceability through the Korean NMI, able to call on
the CIPM MRA when supplying an Indian client leading to a saving of some 5 million
$US [11].
In addition the CIPM MRA has undoubtable helped provide a basis for comparing the
performance of NMIs and hence raise the general standard of metrological performance
in many participating states.
6. – The CIPM MRA review and the way forward

In 2009 a Symposium was held to celebrate the ten-year anniversary of the CIPM
MRA. A wide range of presentations were given by representatives from the organiza-
tions that rely on sound and widely accepted measurements. These included, amongst
others, the WTO, the International Organization for Standardization (ISO), ILAC, the
International Organization of Legal Metrology (OIML), the United Nations Industrial
Development Organization (UNIDO) and Boeing. The symposium was largely a cele-
bration of the success of the CIPM MRA, but not surprisingly thoughts also turned to
the future. In a session entitled “The CIPM MRA today and tomorrow: How the CIPM
MRA might evolve to support metrology needs in other sectors of society” representatives
from the stakeholder and metrology community outlined the challenges for the future,
including some speculation around the way the CIPM MRA itself might evolve. Whilst
184 Andy Henson
there was recognition of the need for and value of the CIPM MRA in facilitating eas-
ily accessible internationally accepted traceability, the NMI community questioned the
sustainability of the CIPM MRA workload in the longer term. This 2009 discussion
encapsulated what was to become a growing concern of the Directors from the major
NMIs over the coming years.
It is probably true to say that when first conceived the success and take-up of the
CIPM MRA had not been fully envisaged. Whilst the initial workload had been properly
anticipated, the ongoing workload had not. New areas of metrology, such as chemistry,
together with emerging NMIs wishing to demonstrate their capability, as well as es-
tablished NMIs expanding their scopes, all add to the leadership burden which falls
disproportionately on a limited number of leading NMIs. This burden not only relates to
the running of comparisons, but also to the peer review of quality management systems
and the examination of CMCs.
In March 2013 the JCRB held a workshop on CMC review, which brought some
useful incremental improvements in efficiency but no substantive changes in scope or
implementation. A discussion at the October 2013 meeting of NMI Directors and Member
State Representatives, echoed in the CIPM, concluded that it was time for a deeper look
and formally concluded “There is a need to review the effectiveness and efficiency of the
CIPM MRA” [7].
Around the same time and in parallel, a strategy exercise was undertaken (and the
consequent strategies published on the BIPM webpages) by each of the CIPM Consulta-
tive Committees. This exercise has already helped manage and rationalize the number
of planned CC comparisons.
The CIPM began planning a workshop for the CIPM MRA stakeholders to address the
sustainability of the CIPM MRA and, to ensure all Member States were clear regarding
the objectives, drafted a resolution which in November 2014 was adopted as Resolution
5 at the 25th CGPM meeting [8] which invited:
the Consultative Committees and the JCRB to continue their ongoing efforts to
streamline operations within the existing framework, and to prepare for and con-
tribute to the wider review in 2015,
the CIPM to establish a working group under the chairmanship of its President,
with membership to be determined at the 2015 workshop, to conduct a review of the
implementation and operation of the CIPM MRA,
With consultations to the CCs, JCRB/RMOs and the NMI Directors and other stake-
holders undertaken and back ground preparations in place the October 2015 workshop
framed the task and established a representative Working Group. The discussions during
the review period were wide-ranging.
Certainly NMI Directors were pressing for their concerns over the workload to be
addressed, particularly in piloting comparisons, but also in the CMC review process.
The CMC review process currently has considerable redundancy built into it (as the
review is first done in the region of the initiating NMI, and then in a second step, carried
The CIPM MRA 185
out in parallel by selected NMI experts from the other regions). There is a general feeling
that those laboratories that have completed the learning curve and participated in many
comparisons, publishing many CMCs, should now be prepared to volunteer and carry a
greater share of the workload, particularly in the piloting of comparisons.
It is equally clear that the KCDB will be redesigned when the new needs are better
known. At this stage we can predict that a minimum requirement will be for better
data input tools and an improved KCDB search capability. The search facilities for the
physics and chemistry areas of the KCDB are already treated separately, but could of
course be separated further if required. The BIPM, as the operator of the system, has
input its ideas, under the guidance of the CIPM, but the main input was from the NMI
Directors, for they are the signatories of the CIPM MRA and they provide most of the
resources necessary to make it work.
The Working group then developed and brought the key ideas to streamline imple-
mentation and operation to a second workshop held in March 2016. This workshop
allowed the Working Group to identify and agree key points and issues for discussion,
and following a further consultation with the CIPM and NMI Directors 9 Recommen-
dations were adopted under which 28 action points are identified. The Working Group
recommendations were:
1. on managing the level of participation in KCs more effectively,
2. on providing better visibility of the services supported by the CMCs in the KCDB,
3. on constraining the proliferation of CMCs,
4. on improving the efficiency of the CMC review processes,
5. on encouraging and enabling states with developing metrology systems to become

signatories and fully participate in the MRA,
6. on the governance of the MRA by the JCRB and the CIPM,
7. on the effective and timely implementation of improvements from this review

through the JCRB and the CIPM,
8. on the scopes and processes used for developing CMCs in Chemistry,
9. on the development of a new strategy for KCs and CMCs in Ionizing Radiation.
The detailed recommendation points, amongst other suggestions, include strengthen-

ing strategic planning of comparisons, sharing the coordination roles more widely, mea-
sures to prevent unnecessary proliferation of CMCs, adopting a risk-based approach to
allow a reduction in the (currently significant) redundancy in the review system, greater
consistency in the evidential requirements supporting CMC claims, training to promote
“right first time” approaches, a web-based CMC submission system, and an updated
KCDB and search and display capability.
186 Andy Henson
7. – In conclusion
In conclusion it is clear that the CIPM MRA has been a huge success, helping to
underpin free trade and improving the comparability of measurement worldwide. After
15 years of operation it is time for the implementation and operational aspects were
reviewed at the workshop held in March 2016. At this workshop a formal review group
identified and agreed key recommendations for change to ensure the sustainability of the
CIPM MRA over the coming decades.
REFERENCES
[1] CIPM MRA: Mutual recognition of nation al measurement standards and of calibration
and measurement certificates issued by national metrology institutes, BIPM, October
1999 (updated October 2003); available online at http://www.bipm.org/en/cipm-mra/
cipm-mra-text/.
[2] Laboratories participating in the CIPM MRA: available online at http://www.bipm.org/
en/cipm-mra/participation/signatories.html.
[3] BIPM key comparison database (KCDB): http://kcdb.bipm.org/.
[4] Reports on the KCDB, available online at
http://www.bipm.org/jsp/en/ViewKCDBReport.jsp.
[5] Joint Committee of the Regional Metrology Organizations and the BIPM (JCRB):
http://www.bipm.org/en/committees/jc/jcrb/.
[6] GATT: General Agreement on Tariffs and Trade, see http://onlinebookshop.wto.org/
shop/article details.asp?Id Article=236&lang=EN.
[7] CIPM, Procès-Verbaux des Séances du Comité International des Poids et Mesures,
available online at http://www.bipm.org/en/committees/cipm/
publications-cipm.html.
[8] Resolutions adopted by the CGPM: available online at http://www.bipm.org/en/
worldwide-metrology/cgpm/resolutions.html.
[9] Recommendations from the Working Group on the Implementation and Operation of the
CIPM MRA http://www.bipm.org/utils/common/documents/CIPM-MRA-review/
Recommendations-from-the-WG.pdf.
[10] Authorized users of the CIPM MRA logo: available online at http://www.bipm.org/en/
cipm-mra/logo/; see also CIPM MRA-D-02: Use of the CIPM MRA logo and certificates
statement, http://www.bipm.org/utils/common/documents/CIPM-MRA/
CIPM-MRA-D-02.pdf.
[11] ILAC-P10:01/2013. ILAC Policy on the Traceability of Measurement Results’.
[12] KRISS website: http://krisswebnew.kriss.re.kr/eng/sub03/sub03 03.html.
DOI 10.3254/978-1-61499-818-1-187
On the proposed re-definition of the SI:

“for all people for all time”
Martin J. T. Milton
Bureau Internationale des Poids et Mesures (BIPM)
Pavillon de Breteuil, Sèvres, 92312, France
Summary. — The proposals to re-define the base units of the SI are expected
to be implemented in 2018. This paper describes the development of the proposals
from their first publication in 2006 through to 2016. They will provide greater
opportunities to perform primary realizations of the base units and thereby address
one of the longest-standing goals of measurement science.
1. – Metrology and the role of the SI
The aim of metrology is articulated in the vision statement of the Bureau International
des Poids et Mesures (BIPM) as being “to provide science and engineering with a basis
for measurements that are stable in the long term and comparable worldwide in order to
support innovation and trade and to enhance the quality of life”. One of the principal
ways this has been achieved is by the agreement of a system of units for use in all
applications of measurement around the world. The foundation of the system is a set
of base units each of which has a definition and an agreed protocol as to how it should
be put into practice (known as a mise en pratique). These base units, together with a
set of derived units defined from them, and a set of prefixes, form what is called the
International System (or SI).
187
188 Martin J. T. Milton
Although the aim of metrology is to provide the basis for measurements that are un-
changing over long periods of time, it has always been a practical and dynamic enterprise
that has exploited scientific advances in order to improve measurement results and to
increase the effectiveness with which accurate measurements can be disseminated [1, 2].
A recent manifestation of the desire to improve the global basis for stable and comparable
measurements is the proposal to revise the definitions of some of the base units of the
SI. These proposals are often referred to as being the proposals for a “new SI” and are
the subject of this paper.
2. – The development of the SI
The name Système International d’Unités (SI) was officially given to the system of
units developed by the BIPM in 1960. It is a system that has been modified when
there has been consensus about the need for improvements. Many of the elements of the
system had been established before 1960 and since that date the system has continued
to be refined and extended. In particular, there have been three significant changes to
the set of base units agreed in 1960 that are relevant to the discussion presented here:
– In 1971, the mole was added to the system as the base unit for the quantity amount
of substance. It thus became the seventh base unit.
– In 1983, the definition of the metre was changed to be the distance travelled by
light in a specified fraction of a second thus fixing the speed of light. This was the
first example of a definition of a base unit that made reference to a fixed numerical
value for a specified fundamental constant.
– In 1990, a set of units was adopted for practical electrical measurements based
on agreed (“conventional”) values for the Josephson constant (KJ ) and the von
Klitzing constant (RK ). These superseded the ampere for practical electrical mea-
surements.
Whilst the first two of these changes had no significant drawbacks, the change made to
the electrical units by introducing the 1990 “convention” was a great deal more significant
and is described below.
3. – Establishing units for electrical quantities

A starting point for considering how units for electrical quantities are defined is
through their link to the units used for mechanical quantities. For example through
Coulomb’s law and Ampere’s law which show the force generated by static and dynamic
charges. They are expressed by two equations each linking mechanical quantities to
electrical quantities:
qq
F = k1
r2
and
dF II
= 2k 2 .
dl r
On the proposed re-definition of the SI: “for all people for all time” 189
They are written here in a form with proportionality constants that depend on the
choice of unit system. Maxwell’s equations enable us to show that the proportionality
constants must meet the condition that
k1
= c2 .
k2
Hence either k1 or k2 can be fixed and the other will be determined by this relationship [3].
Since 1948, the definition of the SI ampere has set the choice of k2
“The ampere is that constant current which, if maintained in two straight parallel
conductors of infinite length, of negligible circular cross-section, and placed 1 metre
apart in vacuum, would produce between these conductors a force equal to 2×10−7
newton per metre of length.”
Hence, k2 = μ0 /4π and consequently k1 = μ0 /4πc2 , where μ0 is the permeability

of free space (sometimes referred to as the magnetic constant). The factor 4π appears
because the SI equations are “rationalized”, meaning that by placing a 4π here, it does
not appear in the formulation of Maxwell’s equations.
4. – The 1990 convention for the electrical units
A central principle of the SI has always been that it is a coherent system of units.
This means that measurement results expressed in terms of the seven base units (and the
derived units) can be combined consistently using the equations of physics and chemistry
without the need for any additional conversion or scaling factors.
However, in 1990 the CIPM decided to adopt a convention for the electrical units in
order to take advantage of the great precision of the quantized voltage and impedance
observed in the Josephson and quantized Hall effects. The values of these effects are
given by
n
U =f
KJ
and
R = RK /p,
where n is a step number and p a plateau number. The Josephson constant (KJ ) and
the von Klitzing constant (RK ) are given by
2e
KJ =
h
and
h
RK = .
e2
In order to take advantage of the great precision of the two quantum effects, it was
decided in 1990 to adopt a convention by which the values of KJ and RK would be fixed
and thus not incur the uncertainties associated with the CODATA values of e and h in
use at that time. The fixed values were based on recommended experimental values.
The 1990 convention is the basis for measurements that are not coherent with the
other SI units because the best experimental values for KJ and RK have departed from
those chosen in 1990, which are usually denoted KJ-90 and RK-90 . This is a significant
drawback with the 1990 convention and proves to be a major argument in favour of finding
a new definition for the ampere that would have the precision and stability brought by
the 1990 convention, but without the implicit loss of coherence.
5. – The Kibble balance and the silicon Avogadro experiment
Whilst the 1990 convention was the basis for a great improvement in the precision
of electrical measurements, it had no impact on measurements traceable to the other
six base units. It was its application to the operation of what is now called the Kibble
balance that was instrumental to the proposal to re-define the kilogram that is discussed
here.
The starting point was an insight by Bryan Kibble published in 1976 into the use of
the current balance in a mode that he referred to as a “moving-coil current balance”. This
subsequently became known as a watt balance and is now known as a Kibble balance.
His insight was that a current balance could be used in both a static and a dynamic
mode. In the static mode a mass (m) is supported against the force of gravity (g) by a
current (I) flowing in a coil in a magnetic field (written here as a flux gradient dϕ/dz)
according to
dϕ
mg = −I .
dz
If in the second mode, the coil is moved through the same magnetic field with a linear
velocity vz , then a voltage is generated given by
dϕ dϕ
U =− = −vz .
dt dz
The flux gradient can be eliminated by combining the equations for the static and dy-
namic modes to give:
U I = mgvz .
Hence the “moving-coil current balance” is governed by an equation that sets electrical
power (on the left-hand side) equal to mechanical power (on the right-hand side). This
is despite the fact that the apparatus itself does not realize a direct conversion between
electrical and mechanical power during its operation.
The experiment assumed greater significance following the adoption of the 1990 con-
vention because, when the voltage (U ) is measured using the Josephson effect, and the
current is measured using (I = U 2 /R) with the Josephson and the quantum Hall effects,
then we can write
2 2
h h h
UI = f n1 n2 = f1 f2 n1 n2 p,
2e e2 4
where the formulae for KJ and RK have been assumed to be exact and n1 and n2 are
integer values for the number of quantized steps in the Josephson junction devices and
p the plateau number in the quantum Hall device respectively. Then,
h
mgvz = f1 f2 n1 n2 p.
4
This equation has no dependence on μ0 nor the present definition of the ampere. It
links three mechanical quantities (on the left-hand side) with the Planck constant and
hence can be used to link mass measured in kg to the value of the Planck constant.
It is this equation that was proposed by Taylor in 1999 [4] as a possible basis for a
definition of the kg as discussed below.
6. – A method for establishing a mass unit from the atomic mass unit
Another innovation made in the 1970’s that was quite unrelated to the Kibble balance,
also plays an important role in the current work to re-define the kilogram. It was the
development by Deslattes of an experiment to measure the Avogadro constant (NA ) by
high-accuracy X-ray crystallography of silicon. He observed
“It would appear that a value of NA is obtainable with an uncertainty of the order
0.01 ppm. Should this be achieved, one might assert that the artefact kilogram is
redundant and that masses can be obtained via an algorithm of realization from
the atomic mass unit” [5].
That the result reported for NA had an unexpected systematic error indicated the
magnitude of the effort that would be required to reach the goal. Significant investments
have since been made in this experiment leading to very many important developments
in the measurements of density, isotopic abundance and lattice constants of highly pure
silicon crystals [6]. These have finally led to results that approach the target accuracy of
0.01 ppm foreseen by Deslattes.
7. – The 2005 proposal to redefine the kilogram
The origin of the set of proposals considered here to re-define the base units of the
SI is found in a paper published in 2005 with the title Redefinition of the kilogram: a
decision whose time has come [7]. A title that makes a specific reference to the historic
publication on metrication, A Metric America: A Decision Whose Time Has Come [8],
thus suggesting that the authors considered that they were proposing a matter of some
historical significance.
The search for a definition of the kilogram that was not based on an artefact has been
going on since the artefact itself was agreed in 1889. For example, the CIPM President
remarked at the 11th CGPM that, Il y a là une faiblesse du Système Métrique à laquelle
les métrologistes devront tôt ou tard porter remède. We start this discussion by reference
to a paper published in 1999. As described above, this paper [4] recommended that
the kilogram be re-defined in terms of a fixed value of the Planck constant (h). It was
justified at that time because of “the superiority of the watt balance to other methods
of linking an invariant of nature to the kilogram”.
Subsequently, the 2005 publication broadened the 1999 proposal by reference to the
development of X-ray diffraction experiments of the type pioneered by Deslattes. It
argued that the kilogram could be re-defined in terms of a fixed value of either the
Planck constant (h) or, alternatively, the Avogadro constant (NA ). Recognizing that the
best values available at that time for these constants were inconsistent and insufficiently
accurate to form the basis of a new definition, the authors proposed that such a definition
might be facilitated by
“adopting a conventional value for the mass of the international prototype,

m(K)07 = 1 kg exactly, it could remain the basis for the worldwide system of
practical mass measurement” [7].
The 2005 proposal is not the same as the one subsequently made in 2006. It was
motivated primarily by the desire to avoid the influence of any possibly unsatisfactory
properties of the International Prototype of the Kilogram (IPK) that might limit the
accuracy with which the definition of the kilogram could be realized. These were widely
thought to be the cause of changes in the differences between the mass of the IPK and
the masses of the copies maintained with it (the temoins). When measured in 1989,
these had changed by as much as 50 μg from the values of the same differences measured
100 years previously. By assuming these differences had accumulated linearly over 100
years, this has been equated to a possible drift in the mass of the IPK with an upper
limit of 0.5 μg/year. Since these estimates presented a possible level of instability in the
IPK that was less than the inconsistency between the best values of h and NA , the 2005
publication proposed that a conventional value be used for the IPK. This would allow
its value to be defined to be constant whilst the unit of mass was realized in practice
through experiments capable of measuring either h or NA .
The authors of the 2005 publication also argued that the adoption of their proposals
would have the benefit of reducing the uncertainty of some of the fundamental constants.
Even in 2005, the argument for progressing with the change was made strongly:
“The advantages of redefining the kilogram immediately outweigh any apparent

disadvantages” [7].
It was suggested that it could be implemented at the next meeting of the CGPM in
2007. The proposals were endorsed enthusiastically by the Consultative Committee on
Units (CCU) in 2005 which recalled in a recommendation that:
“consensus now exists on the desirability of finding ways of defining all of the base
units of the SI in terms of fundamental physical constants so that they are universal,
permanent and invariant in time” [9].
Later that year the proposals were also endorsed by the Comite Internationale des
Poids et Mesures (CIPM), which, in its Recommendation 1 (CI-2005) approved “in prin-
ciple” the preparation of new definitions for the kilogram, ampere, kelvin and mole,
although it gave no indication of the motivation for such a change, except indirectly
through its references to the previously published recommendations of the CCU.
Despite the enthusiasm of the CCU and the CIPM, the consultative committees re-
sponsible for measurements of mass and temperature (the CCM and the CCT) feared a
break in the continuity of measurements in their areas if the proposals were adopted too
soon. Hence they developed criteria that experimental data for h and k (the Boltzmann
constant) must meet before they would be willing to accept that the proposals could be
adopted [10].
8. – The 2006 proposals
The following year the proposals were presented in full in a publication [11] that also
included the authors’ responses to criticisms made of the 2005 proposal. It is this 2006
publication that is the subject of the discussion here. It updated the 2005 proposal by
confirming the preference of the authors for the kilogram to be defined by fixing the nu-
merical value of h (rather than NA ) and it added a proposal to re-introduce the electrical
units into the SI by removing any need for the continued use of the 1990 conventional
values. Of particular importance to the discussion here is that the authors stated that
their detailed proposals were structured according to three “guiding assumptions”. These
were that:
i) The overall structure of the SI should be preserved, because “these quantities and
units are deemed to meet current and future needs”,
ii) “it is not always necessary that a new definition of an SI base unit should allow
the unit to be realized with a reduced uncertainty”,
iii) “the units to be redefined and the constants to which they are to be linked should
be chosen in such a way as to maximize the benefits to both metrology and science”.
These guiding assumptions provide some insight into how the authors balanced the
motivating factors for their proposed changes. Whilst the first and third of these as-
sumptions are articulated in a way that supports the interests of a broad range of users
of the SI, the second is not as easy to rationalise. It was justified by the authors because
in their view:
“the benefits to both metrology and science of replacing the current definition of
the kilogram by one that links it to an exact value of the Planck constant h, and
the current definition of the kelvin by one that links it to an exact value of the
Boltzmann constant k, are viewed as far outweighing any marginal increase in the
uncertainty of the realization of the SI unit of mass or thermodynamic temperature
that might result” [11].
Although not stated, it might be presumed that the authors were anticipating an
increased uncertainty in the short term, but that the route to a reduced uncertainty in
the longer term could be foreseen, perhaps through technical advances.
At this point it is useful to summarize the technical content of the 2006 proposals.
They comprise two major points that are inter-linked. The first was to make the system
of electrical measurements coherent with the SI by abrogating the use of the conventional
values for the Josephson constant (KJ ) and the von Klitzing constant (RK ) adopted in
1990 and to replace them with new values based on fixed values of h and e. The values
chosen for h and e would be the best available values from a least-squares fitting of the
fundamental constants carried out immediately prior to their adoption.
The second point is a consequence of the first. It is that experiments that were
previously used to determine h in terms of standards of mass (such as the watt balance
and the silicon XRCD method) would be constrained to work in the opposite mode. This
could be achieved by allowing them to determine standards of mass, and even to realise
the kilogram, in terms of an agreed fixed value of h. A possibility that would only be
acceptable if it could be done with a practically useful uncertainty. (It would also be
possible to realize atomic-scale masses with respect to a fixed value of h using atomic
recoil methods, but these could not be used to realize macroscopic masses.)
These two major points relating to the kilogram and the ampere were combined with
proposals relating to the five other base units; they were:
– To re-define the kelvin such that primary thermometric methods can be used to
realise the kelvin by reference to a fixed value of the Boltzmann constant.
– To re-define the mole as the amount of substance of a fixed number of entities in

place of the present definition which refers to the amount of substance of a specified
mass of a pure isotope of carbon [12, 13].
– To re-word the definitions of the second, metre and candela. There is no change in
the nature of these definitions, they would simply be re-worded.
However, the proposals would bring some possible technical disadvantages, of which
the most prominent would be:
– An extremely small uncertainty component would be added to the mass of the IPK
when expressed in terms of the re-defined kilogram. This would be propagated
through to the uncertainty of all mass values that continued to be traceable to the
IPK.
– There would be a small shift in the ohm and a larger shift in the volt when moving
from the 1990 convention to the proposed new values [14].
– An extremely small additional uncertainty would be added to the molar mass con-
stant (when expressed in kg/mol) and to the temperature of the triple point of
water (when expressed in K).
– A change to the definition of the mole [13,15] might be considered by some authors
to change the rationale for retaining it as a base unit.
The proposed changes would bring immediate benefits for those concerned with prac-
tical electrical measurements, but the impact on practical mass measurements would not
be as straightforward. Following the publication of the proposals in 2006, the consulta-
tive committee concerned with measurements of mass (the CCM) again considered the
impact of the proposed changes by reviewing the progress against the criteria it had
developed in 2005. Their approach was later summarized as being that:
“Although these conditions will lead to an increase in the best possible uncertainty
attributed to a national platinum iridium standard [. . . ] the CCM considers this
acceptable as an upper limit in view of current practical uncertainty requirements
and the possible instability of the IPK” [10].
9. – Articulating the definitions
The 2006 proposals were very detailed. In addition to advocating the principles
motivating the changes, the publication proposed how new definitions for the base units
might be worded. This is not a simple matter since a definition based on a fixed value of
a constant is most easily articulated using wording that explicitly states the value of the
constant from which the unit may be realized. This has become known as an “explicit
constant” formulation for a definition. For example an explicit-constant definition for

the kilogram might be:
“The kilogram, unit of mass, is such that the Planck constant is equal to exactly
6.626 068 96 ×10−34 joule second”.
By contrast, the same definition can be expressed using an “explicit-unit” formula-

tion as:
“The kilogram, unit of mass, is the mass of a body whose de Broglie-Compton

frequency is (299 792 458)2 /(6.626 068 96 ×10−34 ) hertz”.
This formulation evokes a frequency of the order 1050 Hz which is vastly larger than
any frequency that is encountered routinely. It is therefore not preferred.
The 2006 publication suggests that the proposals could be “formally adopted by the
24th CGPM in 2011”. Subsequently, at its meeting in 2011, the CGPM did not adopt
new definitions but did go as far as recognising the possibility of adopting the changes.
It used similar wording to that used in 2007:
“Considering [. . . ] that the NMIs and the BIPM have rightfully expended signifi-
cant effort [. . . ] by extending the frontiers of metrology so that the SI base units
can be defined in terms of the invariants of nature —the fundamental physical
constants” [16, 17].
10. – The motivation for re-definition
The technical advantages and disadvantages of the proposed changes to the SI de-
scribed in the previous section are subtle. They cannot be explained simply and would
only concern those laboratories working with the SI units at the highest levels of accu-
racy. In this section, we consider some of the more general arguments that have been
developed to motivate and justify the adoption of the changes.
.
10 1. Eliminating the dependence on artefacts. – The most frequently used of these is
the case for eliminating any dependence of the definitions of the base units on physical
artefacts. This is often phrased very broadly in terms of an objective
“To define the units in such a way that they would be based on natural physical
phenomena, thereby providing a stable and universally reproducible system” [18].
As is illustrated below, this argument is often presented in a way that goes beyond
specific technical issues about the performance of the IPK and becomes articulated at
the level of a principle justified axiomatically. For example:
“Although the international prototype has served well as the unit of mass since it
was so designated by the CGPM in 1889, it has one important limitation: it is not
linked to an invariant of nature. Thus, the possibility of redefining the kilogram in
terms of a true natural invariant —the mass of an atom or a fundamental physical
constant— has been discussed during at least the last quarter century” [11].
In this statement, we read that the limitation of the IPK is that it is not linked to
a natural physical phenomenon or an invariant of nature, not that its performance is
simply inadequate in some respect.
The use of artefacts to act as standards is a recurring theme in the history of measure-
ment. The signature of the Metre Convention by seventeen nations in 1875 established
an international agreement that unique artefacts should be carefully prepared to form
the basis for the definitions of the metre and the kilogram. These artefacts were formally
endorsed by the CGPM in 1889. Whilst the use of these artefacts to define measure-
ment units provided the practical basis for the world’s measurement system, even at
that time it had been observed that a more stable and enduring basis might be found.
This is referred to directly in the widely referenced quotations:
“If then, we wish to obtain standards of length, time and mass which shall be abso-
lutely permanent, we must seek them not in the dimensions, or the motion, or the
mass of our planet, but in the wavelength, the period of vibration, and the absolute
mass of these imperishable and unalterable and perfectly similar molecules” [19].
and
“with the help of fundamental constants we have the possibility of establishing
units of length, time, mass and temperature, which necessarily retain their validity
for all times and cultures, even extraterrestrial and nonhuman . . . ” [20]
Great progress was made during the 20th century towards exploiting the stable and
enduring properties of atoms to act as measurement standards. The most prominent
examples were:
– The metre which was defined in 1960 as a specified number of wavelengths in
vacuum of a certain transition of the krypton-86 atom.
– The second which was defined in 1967 as the duration of a specified number of
periods of the transition between two hyperfine levels in the ground state of the
caesium-133 atom, and
Subsequently, in 1983 a definition was agreed for the metre that introduced a funda-
mental constant into the SI when the metre was defined in terms of the speed of light.
The use of the speed of light for the definition of the metre is highly effective. There is a
link between distance and velocity that makes it intuitive and straightforward to explain.
Also, the speed of light plays a fundamental role in the theories of electromagnetism and
special relativity. Hence the 1984 definition of the metre is both straightforward to
articulate and takes advantage of an important principle from modern physics.
The argument for the elimination of artefacts from the definitions of the base units
focusses on the one remaining unique artefact, the IPK which has been used as the
definition of the kilogram since 1889. Drawbacks that can be rationally associated with
it include its vulnerability to accidental damage and the possibility that the alloy itself
might be intrinsically unstable, perhaps because of effects at its surface. Such arguments
should not be accepted without scrutiny:
“There is clear experimental evidence that the masses of 1 kg prototypes are not
stable when compared among themselves over many dozens of years. To date, how-
ever, there is no experimental evidence to support a hypothesis that the mass of
the international prototype is changing with respect to a mass derived from a fun-
damental constant of physics. The current lack of such evidence is presumably due
to the present level of precision of watt balance and Avogadro experiments” [21].
More evidence on the stability of the IPK has become available recently. The results
of the extraordinary campaign of calibration of the prototype mass standards held at the
BIPM in 2014 show that the difference between the mass of the temoins and the IPK
has only changed by an average of 1 μg since 1989 [22]. This sets a significantly different
quantitative limit on the practical stability of the IPK as an artefact than the upper
limit proposed for its drift that has been discussed above.
The alternative to the use of artefacts is to base measurement standards on “con-
stants of nature” or “fundamental constants”. This view is usually referenced to the
above quotation from Maxwell, although at least one contemporary author proposes
that fundamental constants are more suitable than atomic properties:
“The question of whether the mass of an atom or an elementary particle is more

fundamental than the Planck constant has been raised. Since there is a spectrum
of particles, each one having a different mass and none more fundamental than
the other, giving a fixed value to the Planck constant is surely less arbitrary than
giving a fixed value to the mass of an atom” [23].
.
10 2. Widening access to the realization of the base units. – There is another group
of arguments in favor of the 2006 proposals that was also referred to in [10]. This is that
they provide greater opportunities for access to primary realizations of base units.
“the most significant benefit in our view is that it liberates mass metrology from
an artefact-based unit. This means that different laboratories can realize the unit
at will —the long-sought goal of the kilogram being realizable at any time at any
place by anyone with the requisite uncertainty will now be limited only by the
financial and/or human resources available at a given laboratory” [11].
This view has been repeated more recently
“we have but one realization of the unit of mass in the entire world (one mass to
rule them all!). If redefinition is successful, we will immediately have at least three
{laboratory} groups capable of realizing the unit and calibrating artifacts for use as
primary standards. [. . . ] it will be possible to derive a unit of mass using whatever
means is convenient and at whatever scale point desired” [24].
And also with some caution
“Whilst “quantum” phenomena are, in principle, accessible “for all time to all
people”, in practice the experimentation required to realize them is not. In the
most extreme cases, they depend on experimentation that can only be carried out
by highly specialized laboratories with the necessary experience and funding to do
so” [1].
It is important to recognize that such universality of access can also be achieved

through classical experiments (for example, the present definition of the kelvin provides
universality of access) as much as it can be by a quantum definition. Indeed a new
definition may not be required, for example a watt balance could be used under the
present definition of the ampere to realize a measurement of the kilogram. It would give
a result in terms of the 1990 convention for the electrical units that is readily convertible
to the SI.
Since the first proposal for a metric system advocated at the time of the French
revolution, we note that its motivation was, at least in part, based on the notion of
facilitating freedom and equality of access to measurement standards. However, such
ambitions have proved difficult to achieve. Attempts to adopt a definition based on the
size of the earth that was originally proposed in order to allow universal access did not
achieve their objective. It has been observed that
“The great virtue of the meridian experiment [. . . ] was that it could not easily
be repeated —as a simple pendulum experiment might have been. The meridian
experiment, by its very grandeur, difficulty and expense, had fixed the meter —
permanently” [25].
The challenge of achieving the goal of universality of access can be limited by the
practical.
Returning to the motivation for metrology outlined in the introduction to this paper
it is a pragmatic enterprise that succeeds when it is able to provide measurement results
that are stable, comparable and coherent. It is improvements in these three respects that
rightly motivate improvements to the system.
Increased access to the system is an objective that is motivated pragmatically. As
one of the earliest advocates of quantum metrology wrote
“The most important property of a system of measurement is that it should be

widely and easily available” [26].
11. – Conclusions
Considering again the three most significant consequences of the proposed changes
– That practical electrical measurements will be brought into the SI (but at the cost
of step changes when practical electrical measurements move from the conventional
values to the SI values of RK and KJ at the time of implementation).
– The opening of the possibility of using any method that can relate a mass mea-
surement to the defined value of h to realize the kilogram.
– The opening of the possibility of using primary thermometric methods to make

measurements of temperature in the SI unit kelvin.
The proposed changes are steps forward that have been justified pragmatically with
respect to the benefits of making electrical measurements coherent within the SI. Addi-
tionally, we see that a common feature of them all is that any laboratory will, in principle,
be able to realize the definitions and the scope of methods allowed for their realization
will increase. In this respect, the proposed changes may bring the SI closer to one of
the greatest ambitions of measurement science —that it should be “for all people for
all time”. A statement that goes back to the law of 1799 that initiated the first metric
system in France.
We should not expect the proposed changes to be the end of the process of improving
the international system of measurement units. It is to be expected that further progress
will be made in metrology eventually leading to proposals to renew the “new SI”, not least
to incorporate a definition of the second that uses an optical rather than a microwave
transition. This would be a change within the framework of a system in which the
realization all of the base units is already open “for all people for all time”. This aspect
of the present change will be one of historical importance. There should be no reticence
about this. It was highlighted by the authors of the 2006 proposals as being “the most
significant benefit”. It is a point that may have been inadvertently diminished by the
emphasis on “defining constants” and “invariants of nature” to the point where the
CGPM has not touched upon it in any of its three recommendations on the subject.
REFERENCES
[1] Milton Martin J. T., Williams Jonathan M. and Bennett Seton J., “Modernizing
the SI: towards an improved, accessible and enduring system” Metrologia, 44 (2007) 356,
doi: 10.1088/0026-1394/44/5/012.
[2] Milton Martin J. T., Davis Richard and Fletcher Nick, “Towards a new SI:
a review of progress made since 2011” Metrologia, 51 (2014) R21, doi: 10.1088/0026-
1394/51/3/R21.
[3] Jackson J. D., Classical Electrodynamics (John Wiley, New York) 1975.
[4] Taylor B. N. and Mohr P. J., “On the redefinition of the kilogram” Metrologia, 36
(1999) 63, doi.org/10.1088/0026-1394/36/1/11.
[5] Deslattes R. D. et al., “Determination of the Avogadro Constant”, Phys. Rev. Letts.,
33 (1974) 463.
[6] Bettin H. et al., “Accurate measurements of the Avogadro and Planck constants by
counting silicon atoms” Ann. Phys. (Berlin), 525 (2013) 680.
[7] Mills I. M., Mohr P. J., Quinn T. J., Taylor B. N. and Williams E. R., “Redefinition
of the kilogram: a decision whose time has come” Metrologia, 42 (2005) 71.
[8] NBS (1971) A Metric America: A Decision Whose Time Has Come, Special Publication
345, National Bureau of Standards, Washington, DC, July 1971.
[9] CCU RECOMMENDATION U 1 (2005), On possible changes to the definitions of the
kilogram, the ampere, the kelvin and the mole, www.bipm.org/utils/common/pdf/CC/
CCU/CCU17.pdf.
[10] Glaser M., Borys M., Ratschko D. and Schwartz R., “Redefinition of the kilogram
and the impact on its future dissemination” Metrologia, 47 (2010) 419, doi: 10.1088/0026-
1394/47/4/007
[11] Mills I. M., Mohr P. J., Quinn T. J., Taylor B. N. and Williams E. R., “Redefinition
of the kilogram, ampere, kelvin and mole: a proposed approach to implementing CIPM
recommendation 1 (CI-2005)” Metrologia, 43 (2006) 227.
[12] Taylor Barry N., “Molar mass and related quantities in the New SI” Metrologia, 46
(2009) L16, doi:10.1088/0026-1394/46/3/L01.
[13] Milton M. J. T. and Mills I. M., “Amount of substance and the proposed redefinition
of the mole” Metrologia, 46 (2009) 332, doi: 10.1088/0026-1394/46/3/022.
[14] Fletcher Nick, Rietveld Gert, Olthoff James, Budovsky Ilya and Milton
Martin, “Electrical Units in the New SI: Saying Goodbye to the 1990 Values” NCLSI
Measure J. Meas. Sci., 9 (2014) 30.
[15] Milton Martin J. T., “The mole, amount of substance and primary methods” Metrologia,
50 (2013) 158, doi: 10.1088/0026-1394/50/2/158.
[16] Resolution 12 of the 23rd CGPM (2007), On the possible redefinition of certain base units
of the International System of Units (SI), http://www.bipm.org/en/CGPM/db/23/12/.
[17] Resolution 1 of the 24th CGPM (2011), On the possible future revision of the International
System of Units, the SI, http://www.bipm.org/en/CGPM/db/24/1/.
[18] Mohr Peter J., “Defining units in the quantum based SI” Metrologia, 45 (2008) 129.
[19] Maxwell J. C., (1870) “Address to the Mathematical and Physical Sections of the British
Association, Liverpool, 15 September, British Association Report, Vol. XL”, reproduced in:
The Scientific Papers of James Clerk Maxwell, Vol. 2, edited by Niven W. D. (Cambridge
University Press, Cambridge) 1890, p. 225.
[20] Max Planck (1900) reproduced in Huebner R. P. and Luebbig H. A Focus for Discoveries
(World Scientific Publishing Ltd, Singapore) 2008, p. 150.
[21] Davis R. S., “Possible new definitions of the kilogram” Phil. Trans. R. Soc. A, 363 (2005)
2249, doi: 10.1098/rsta.2005.1637.
[22] Stock Michael, Barat Pauline, Davis Richard S., Picard Alain and Milton
Martin J. T., “Calibration campaign against the international prototype of the kilogram
in anticipation of the redefinition of the kilogram part I: comparison of the international
prototype with its official copies” Metrologia, 52 (2015) 310, doi: 10.1088/0026-
1394/52/2/310.
[23] Becker P., de Bièvre P., Fujii K., Glaeser M., Inglis B., Luebbig H. and Mana
G., “Considerations on future redefinitions of the kilogram, the mole and of other units”
Metrologia, 44 (2007) S.1.
[24] Pratt Jon R., “How to Weigh Everything from Atoms to Apples Using the Revised SI
NCLSI Measure” J. Meas. Sci., 9 (2014) 26.
[25] Alder Ken, The Measure of all Things (Abacus, London) 2002.
[26] Cook A. H., “Quantum metrology - standards of measurement based on atomic and
quantum phenomena” Rep. Prog. Phys., 35 (1972) 463.
DOI 10.3254/978-1-61499-818-1-203
The Metre Convention and the creation of the BIPM

T. J. Quinn(∗ )
92 rue Brancas, 92310 Sèvres, France
Summary. — The Metre Convention was signed in Paris on 20 May 1875. Its stated
purpose was to assure the international unification and perfection of the metric
system. At its origin was a set of recommendations made at a Conference on geodesy
held in Berlin in 1867 which called for the manufacture of a new European prototype
of the metre and the creation of a European international bureau of weights and
measures. The response of the French Government was to create an International
Metre Commission to consider the question. The Commission duly recommended
what the Berlin Conference had proposed and a Diplomatic Conference on the Metre
took place in Paris. It opened on 1 March 1875 and culminated on 20 May in the
signing of the Metre Convention by the representatives of 17 States. Included in the
Convention was the creation of an International Bureau of Weights and Measures, a
General Conference on Weights and Measures and an International Committee for
Weights and Measures. The international organization comprising these three organs
still exists and is the means through which Governments today arrange for and
support the International System of Units, SI, the successor to the metric system.
This article gives a brief description of the discussions and events surrounding all
this.
(∗ ) Emeritus Director of BIPM.

203
204 T. J. Quinn
1. – Introduction
This is the first of two articles related to the Metre Convention, the International
Bureau of Weights and Measures (BIPM) and the development of the metric system
from its origins at the time of the French Revolution to the present day. The main
points in this first article will be:
– the call by geodesists in 1867 for a better standard of the metre and for a European
international Bureau of weights and measures,
– the response of the French Government by creating an International Metre Com-
mission in 1869,
– the meetings of the International Metre Commission from 1870 to 1873 and proposal
for a Diplomatic Conference,
– the Diplomatic Conference of the Metre in 1875 which ended with the signature
of the Metre Convention on 20 May which included a decision to create the first
international scientific institute, the BIPM,
– the modifications to the text of the Convention first in 1907 and then in 1921 when
a new Convention was signed broadening the mission of the Convention of 1875,
– the creation of the BIPM, its laboratories and staff and early programme,
– the Centenary of the BIPM in 1975 and its development since then.
The Metre Convention is an intergovernmental Treaty whose aims were explicitly
stated in the preamble to the Treaty: “assurer l’unification internationale et le perfec-
tionnement du système métrique” (to assure the international unification and perfection
of the metric system). These have been its guiding principles ever since. In 1960 the
Metric System was made international and renamed the International System of Units,
SI. The broader aims of the Convention in terms of international unification now extend
much wider than those envisaged by those who set up the Convention in 1875 or even
the Convention of 1921. This is notably through the CIPM MRA signed in 1999 which
now includes formal participation of some one hundred States. The “perfection of the
metric system” is now on the verge of a major step forward with the proposed revision
of the SI in 2018 when it will be based on a set of seven defining constants of nature. As
I show in the second of these two articles, this will finally meet the aspirations of those
who set up the metric system at the time of the French Revolution “for all time for all
nations”.
We begin with events that led up to the Metre Convention of 1875.
2. – The call by geodesists in 1867 for a better standard of the metre and for
a European international Bureau of weights and measures
Although the decimal metric system had been formally adopted at the time of the
French Revolution, by the law of 7 April 1795 (18 germinal an III in the Republican
The Metre Convention and the creation of the BIPM 205
calendar), its implementation in France was slow. From 1812 pre-metric units began
to be allowed again not to be finally forbidden until 1837, but it began to spread out-
side France [1]. Stimulated by France exhibiting a set of metric standards from the
Conservatoire des Arts et Métiers at the Great Exhibition of 1851 in London, many
countries began to use metric measures based on standards provided by France. These
were not standards calibrated directly against the Metre and Kilogram of the Archives
but against copies at the Conservatoire made at the same time as the standards of the
Archives and deposited not only at the Conservatoire but also at the Paris Observatory
and the Académie des Sciences. In 1848 the Conservatoire had taken on the role of inter-
mediary for foreign links to the official French standards. While for most purposes this
was perfectly satisfactory, it was not so for geodesy where the former French standard of
length, the toise, continued to be used.
Among the many things carried out by Napoleon at the beginning of the nineteenth
century had been military surveys in those countries in Europe then part of, or allied to,
the French Empire. The Napoleonic surveys at the beginning of the nineteenth century
were all linked by triangulation to the French system that was based on the measurements
of the arc of the meridian by Delambre and Méchain [2]. Later on, during the first half
of the century, similar surveys were carried out in the many independent German States.
New instruments and methods of calculation were developed, as a result of which the
accuracy of the resulting maps was significantly improved. During the years 1821 to 1824
the great Carl Friedrich Gauss had carried out a survey of the Kingdom of Hannover.
During the calculation of the results of the triangulation data he became interested in
the shape of the Earth. While Gauss’ survey was linked to some others it did not extend
all over Central Europe.
The idea of Gauss was for a comprehensive central European triangulation from the far
North to the South of Italy. This was not pursued at the time but advances then began to
be made in understanding the figure of the Earth and the importance of including gravity
measurements to account for variations in the vertical in evaluating astronomical data
for finding latitude. In the thirty years following Gauss’s measurements, two leading
figures in European practical geodesy appeared, Friedrich Wilhelm Bessel and Johan
Jacob Baeyer. Bessel was an astronomer and Director of the observatory of Koenigsberg
where he kept a copy of the toise used as reference for much geodesy in that region..
Johan Jacob Baeyer was a Prussian army officer who began his geodesy studies as a
major in the Prussian general staff detached to work with Bessel. He subsequently rose
rapidly in the Prussian general staff to oversee many new triangulations in Prussia. In
1861 he made a proposal to the Prussian Minister for War for a Central European Arc
Measurement. The aim was to make a new determination of the anomalies in the shape
of the Earth by means of accurate triangulation combined with gravity measurements
and including all modern ideas developed by Bessel and others. This was a real scientific
problem but of course with wider implications.
In 1862 this proposal was approved by the Prussian government and the first negoti-
ations then took place in Berlin between representatives of Prussia, Austria and Saxony.
These were then widened. In 1864, the first General Conference of the Representatives to
206 T. J. Quinn
the Central European Arc Measurement took place in Berlin. The main accomplishments
of the Conference were to set up a formal organization for Central European collabo-
ration in geodesy and set out a list of scientific recommendations for future geodesy in
Europe. This was the birth of what is today the International Association for Geodesy.
Included in the scientific recommendations were ones to use the toise of Bessel for all
geodetic calculations, to insist that all measurements should be compared with Bessel’s
toise, that its length should be compared with the metre and that in the future, results
should be given in both toise and metres. In addition, it was recommended that a
campaign of height measurements above sea level be carried out in Europe. The prime
importance for geodesy of agreement on length standards thus became very clear. From
30 September to 7 October 1867 there took place in Berlin what was called the “Second
International Conference on Geodesy for the Measurement of Degrees in Europe”, the
first having taken place also in Berlin in 1864. It adopted recommendations that marked
the beginning of formal moves that led to the signing of the Metre Convention in Paris
in 1875.
The Recommendations adopted by the Conference were the following [3]:
1. The Conference confirms the resolution taken three years ago recognizing the need to
compare the standards and rules that are used for the measurement of base lines in
the different countries. To carry out these comparisons we request the construction
of two comparators, one for the reference standards and the other for the working
standards for the base equipment; these comparators should be made such that they
can be used for both line and end standards and the determination of expansion
coefficients.
2. A special commission is charged with establishing the principles to be followed in

these comparisons and for the construction of the comparators. The Commission
set up with this in mind three years ago should be enlarged by at least six additional
members from countries participating in the Geodesic Association; the permanent
Commission is charged with designating these new members.
3. The Commission recommends to the Conference to instigate research into the time
variation of the expansion coefficients of scales and length standards; it also wishes
that this Commission be ready to concern itself with the construction of new proto-
type standards.
4. It is in the interest of the sciences in general and of geodesy in particular, that

a single system of weights and measures with decimal submultiples be adopted in
Europe.
5. Because of all the units of measurement that might be considered the metre has the
greatest chance of being accepted, the Conference favours the adoption of the metric
system.
6. The Conference recommends that where the metric system is introduced, it be ac-
cepted without changes and that the decimal subdivisions be retained in all countries.
In particular the Conference opposes the introduction of the metric foot.
7. The Conference recommends that a new European prototype be constructed in order

to define the common unit of measurement for all the countries of Europe and for
all time as precisely and stable as possible. The length of this metre should differ
as little as possible from that of the metre at the archives in Paris and should in
any case be compared with it as accurately as possible. In the construction of the
new prototype standard, ease and accuracy of the necessary comparisons should be
the primary objective.
8. The construction of the new prototype metre and the manufacture and comparison
of copies for the various countries should be entrusted to an international commis-
sion on which these countries should be represented.
9. The Conference favours the establishment of a European international bureau of

weights and measures.
10. The Conference recommends that the delegates bring these resolutions to the atten-
tion of their governments and asks the permanent Commission (of the Association)
to do everything possible to see that they are carried out.
During the discussion preceding the adoption of these recommendations, it is stated

in the Report of the Conference, that they were drawn up by an ad hoc committee headed
by Otto Wilhelm von Struve, Director of the Pulkova Observatory (Saint Petersburg),
but that von Struve invited Adolph Hirsch, the Director of the astronomical observatory
at Neuchatel in Switzerland, to present them as it was he who had in fact drawn up
the recommendations. It was Hirsch who later translated the German Reports of the
Conference into French from which the above English version has been translated.
The full extent of the work that would be required to establish a new metric standard
had become apparent at the Conference. It was this that probably prompted Hirsch to
propose the creation of a European international bureau of weights and measures. Such
a bureau is, of course, much broader than an institute devoted solely to length standards
and as we shall see this broad proposal did not go down well in France. Before the
vote was taken to adopt these recommendations, at the sixth session on 5 October, some
members were of the opinion that some of them went beyond the competence of the
Conference and one of the Prussian delegates even refused to take part in the discussion
for this reason. In the end, however, they were adopted by a large majority. A final
recommendation was adopted before the end of that day’s session instructing the central
bureau of the Association to take the necessary steps to construct a new geodetic base
for the common use of those countries associated for the measurement of degrees in
Europe.
Taken together, these recommendations represent a clear and precise plan for estab-
lishing a new European system of measurement. While it would be based on the metric
208 T. J. Quinn
system, the recommendations could easily be interpreted to mean that all of this would
be done independently of France. In view of the composition of the Conference this is
perhaps not surprising but it stimulated quick reactions from Paris.
3. – The International Metre Commission
In France, the Recommendations of the Berlin Conference led to considerable anxiety

as, quite apart from other considerations, it was not clear that there was any intention
that the proposed international European bureau of weights and measures would be
located in France. The three important scientific bodies in France at the time: the
Bureau des Longitudes, the Conservatoire Imperial des Arts et Métiers and the Académie
des Sciences, quickly reacted to the Recommendation from the Berlin Conference.
First of all the Bureau des Longitude commented. The interest of the Bureau des
Longitudes in the possible creation of a laboratory stemmed from one of its original
missions which was to look after the instruments of the Paris Observatory. The members
of the Bureau had clearly understood the 1867 Recommendation to refer to the creation
of a scientific institute or laboratory because they wrote in December 1867 [4]:
“Your Commission is of the opinion that it would be right to put the Berlin declaration
under the eyes of the competent authority, to ask it to see if sufficient measures have
been taken to give satisfaction to the needs now coming to light and if it would not
be convenient to complete these measures by the creation, in Paris, of a special and
permanent establishment for the rigorous study of metric standards and geodesic scales.
One would install in this sort of laboratory, the most precise apparatus for comparison
that today’s science can achieve, such as the great microscope-comparator for geodesic
scales of our colleague M. Brunner. A wide range of other work related to the propagation
of the metric system could be entrusted to this laboratory under the direction of the
Bureau des Longitudes and with the cooperation of special Commissioners delegated by
the French Government and by other interested Governments. Included was the following
footnote: This establishment would in no way duplicate the system of verification of
weights and measures destined for internal French trade and commerce, the fully complete
system that exists today at the Conservatoire des Arts et Mètiers”.
The decision by the French Government in 1869 to take the initiative and call a
meeting to discuss these proposals was based on a Report it had commissioned from
the Académie des Sciences [5]. The details were largely drawn up by the Conservatoire
Imperial des Arts et Métiers. In 1869 an invitation was thus sent to all the States with
which France had diplomatic relations to send official delegates to participate in an In-
ternational Metre Commission to prepare plans for the construction of a new prototype
of the metre to be based on the original Metre of the Archives, but one that would be
a line standard rather than an end standard. It was proposed that the work of manu-
facturing the new standard and its copies for participating States would be undertaken
by a permanent French Section of the new Commission but with the help and advice
of international members of the Commission. No mention was made of the proposal to
create the international European bureau of weights and measures nor was there any
mention of the kilogram.
Twenty-four governments accepted the invitation to send delegates to take part in the
new Metre Commission. The first meeting was called for 8th August 1870, at which it
created a committee for Preparatory Research. It did little else because in August 1870
France was in the middle of its war with Prussia and many delegates were unable to be
present. The committee for Preparatory Research however included members from the
French Section and a number of foreign delegates. The delegates were almost all high
level men of science, occupying high positions in their own countries and were in most
cases members of their national Academies of Science and, where they existed, directors
of institutes responsible for weights and measures.
The Committee for Preparatory Research met for the first and only time in April
1872 and in its Report [6] set out what was essentially to become the structure of the
Metre Convention: It proposed:
“[. . .] that the International Metre Commission is declared as the central scientific
body for all metrological matters of the countries that adopt the metric system;
formed of the delegates of the governments of these countries it meets periodically to
consider all important matters related to weights and measures which by their nature
call for an international solution”. [This was to become the General Conference on
Weights and Measures],
“[. . .] the Metre Commission nominates a permanent Committee of five members
each of different nationality charged with overseeing the execution of its decisions,
to undertake research and studies demanded by the development of the metric sys-
tem and to prepare solutions for all those questions that call for an international
solution”. [This was to become the International Committee for Weights and Mea-
sures],
“[. . .] as executive organ of the Commission and under the surveillance of the its per-
manent Committee an International Metrological Institute will be created in Paris
where the international prototypes will be deposited as well as instruments and ap-
paratus necessary for their conservation and comparison. This establishment will
carry out all the work and metrological research that the Commission or its per-
manent Committee decides in the interest of the metric system, in particular the
comparisons and verifications of international standards. In order to guarantee the
international character and independence of the institution, its director and staff
will be nominated by the International Commission and the running costs will be
covered by contributions from all the interested countries in proportion to the num-
bers of their inhabitants”. [This was to become the International Bureau of Weights
and Measures].
The only meeting of the full International Metre Commission took place in Septem-
ber 1872 at which it approved the outline proposals of the Committee for Preparatory
Research and created a Permanent Committee of 12 members charged with supervising
210 T. J. Quinn
the actual construction of the new metric prototypes and recommended the creation of
an international Institute(1 ) for weights and measures to be located in Paris [7]. The
Metre Commission set up a number of Commissions from among its members one of
these, Commission 7, addressed the question of the creation of an international bureau.
On page 132 of ref [7] we find “We now propose the creation of an international institute
(French institut) for weights and measures. This institute will require the construction of
a special building, it will have its own budget a Director and staff”. There then followed
a broad discussion at the end of which the Commission made a series of recommenda-
tions in which we find the essential of what later became the Metre Convention. As
regards the creation of an international bureau we find: “The International Commis-
sion informs governments of the utility of creating in Paris an international Institute for
weights and measures [italics and capital I for Institute in the original] [8]. The estab-
lishment will be international and declared neutral, its seat will be in Paris, it will be
supported by contributions from all the governments that sign the treaty that creates
the Institute”.
Following a list of activities related to the construction and calibration of the new
metric standards, it ends by saying
“The Institute will undertake all the work that the Commission and its executive Com-
mittee decide to carry out in common in the interests of metrology and the propagation
of the metric system”.
The meaning of all this is perfectly clear and explicit. The proposed international
establishment will be a scientific institute with its seat in Paris having short and long
term objectives. The Report of the International Metre Commission, formalised by its
Permanent Committee in 1873, was sent to Governments, as a result of which the French
Government called the Diplomatic Conference of the Metre to draw up the necessary
Treaty to implement the Commission’s recommendations.
4. – The Metre Convention of 1875
The Diplomatic Conference that drew up the Convention took place in Paris from
1 March to 20 May 1875, the date on which the Metre Convention was signed by the
representatives of seventeen States. Plenipotentiary delegates from the twenty two States
(1 ) Close to the end of the meeting of the Metre Commission the French delegate suggested
changing the word Institute to Bureau. This was accepted without comment although no reason
was given for the change. In fact, the two words are essentially synonymous and in the future
discussions at the Diplomatic Conference in 1875 three words were frequently used to describe
the international body being created, Institute, Bureau and Establishment, all being qualified
by the words permanent and scientific. The meaning of the word Bureau in the 19th Century
French Dictionary (Littré, 1889 edition) is the same as it is now, either a piece of furniture, a
place where operations take place or a small group of people charged with the running of an
organization or association.
that had sent delegates to meetings of the International Metre Commission were present,
accompanied by scientific delegates who had actually taken part in the earlier discus-
sions [9](2 ).
After the opening plenary session of Plenipotentiary delegates, a series of meetings of
the scientific delegates took place to try and reach a unanimous proposal to put to the full
Conference. The discussions that took place during the meetings of scientific delegates
were at times difficult and reflected the different opinions that existed regarding public
support for science and the future role of an international scientific institute. In the end
no unanimity could be reached and the decision was finally made by the full Conference
of Plenipotentiary delegates. All agreed from the outset that an international scientific
institute should be created to carry out the work needed to establish and disseminate
the new standards of the metre and kilogram as had been proposed by the Metre Com-
mission. However, three delegations, those of Great Britain, Netherlands and Portugal
were strongly against it being permanent. Each had a different reason. A number of
other delegations reserved their position waiting to see how opinion developed.
The British Government had decided in advance that it was against any permanent
scientific institute because a) although the costs would be small initially they would
undoubtedly increase in the future b) such an international institute would create what
might later be an inconvenient precedent and c) Great Britain had no interest in adopting
the metric system anyway [10, 11]. This was a somewhat surprising decision because in
the years since the 1851 Great Exhibition considerable enthusiasm had grown in Britain
for the possible adoption of the metric system. Committees had been created following
other international Exhibitions and Congresses, the Statistical Congress in Brussels in
1853, the Paris Exhibition of 1855 and statistical Congress in London in 1860. Trade with
France increased markedly following the Cobden Chevalier Treaty of 1860 and in 1863
the British Association for the Advancement of Science weighed in with strong support.
Parliament had shown strong support for possible legislation to make the metric system
compulsory. With this in view a Royal Commission on Standards was created to report
on the matter. Surprisingly, it reported that Great Britain was not ready for such a step
and that it had not been shown the metric system was superior to the English system.
Such was the authority of a Royal Commission that moves towards metrication
stopped and hence the response of Great Britain to the proposed International Bureau
of Weights and Measures. However, Great Britain signed the Convention in 1884 when
it found that it was unable to obtain copies of the new metric standards unless it signed
the 1875 Metre Convention. This should be seen in the context of the debate during
the last quarter of the 19th Century in Britain as to the role of Government in scientific
research in the face of strong laissez-faire attitudes prevalent at the time. In 1870 the
(2 ) A full account of the Conference appears in the official publication by the French Foreign
Ministry of the Proceedings of the Diplomatic Conference, namely, Documents Diplomatique de
La Conférence du Mètre, Paris, Imprimerie Nationale 1875, a document of 149 pages, available
on the BIPM website www.bipm.org.
212 T. J. Quinn
Government created another Royal Commission, the Cavendish Commission, to report

on the state of scientific teaching and research in Britain. In its eighth Report, in 1874
the Commission recommended the creation of public laboratories. Despite this, it was
not until 1900 that the National Physical Laboratory was actually created.
For the Netherlands, the delegate himself was of the view that a permanent scientific
institute would be bad for science as it would on the one hand have too much power
and on the other, that the staff would inevitably be second rate because once the new
standards had been completed there would be little of interest to do. For Portugal the
delegate was in fact the Director of the French Conservatoire National des Arts et Métiers
and it is clear that he wished the work to be handed to his own institute. He had been
excluded from the French delegation because of his views and difficult personality and
had been able to persuade Portugal to allow him to be the Portuguese delegate [12].
Such are the human and personality difficulties in international relations!
In the face of these strong objections from three delegations to the idea of a perma-
nent institute two proposals were developed at the meeting of scientific delegates. The
first, immediately supported by a majority of delegates was for a permanent interna-
tional scientific institute and the other for an institute which after the work of preparing
the new metric standards had been completed would become simply a depository of the
standards. The Title of the first proposal was: Draft International Organization for
metrological work for the fabrication and verification of new metric prototypes, the de-
pository of these prototypes and for their subsequent use. It included two Articles on the
proposed international institute(3 ), the first was:
To execute the work decided by the Commission, an international Institute for weights
and measures (Institut international des poids et measures) would be founded on the
following basis:
The establishment would be international and declared neutral
Its seat would be in Paris
Its creation and maintenance would be by funds from all the countries that adhere to the
treaty concluded to create it
The institute would be placed under the direction and surveillance of the international
Commission for weights and measures . . . .
The next article set out its scientific duties. These were very close to those in the
final text of the Convention and appear in it as Article 6 (1875). Thus the status and
tasks of the proposed international bureau was set out right at the beginning and were
not modified later in the Conference.
The second proposal was simply for a place in an existing building in Paris where the
work of manufacturing the new standards would be done after which it would become a
(3 ) See ref. [9], page 56.
depository of the standards available to participating States to send scientists to use when
necessary. It included an article giving the possibility of those States wishing to make
it a permanent scientific institute making whatever arrangements they so wish. Various
attempts were made to find a compromise but without success, France it this stage
reserving its position. Thus it was at the Conference of Plenipotentiary that delegates
had to decide which one to adopt.
After statements of position by delegations that largely reflected the opinions of their
scientific delegates, the delegate from France, Jean Baptiste Dumas, stated that France
was firmly in favour of the first proposal. His key statement was the following: “The
French Government accepts that the institution now being proposed should be scientific
of a high order, that it be permanent to allow not only the completion of the work now
underway under the terms already agreed but in addition all the work related to the
propagation of the metric system and required by the continual progress of the science
of metrology” [13]. This statement persuaded most of the few delegations who were
undecided and the proposal for a permanent scientific institute was in due course adopted
with only the Netherlands voting for the second proposal, Great Britain and Portugal not
voting. A last minute attempt to delay a final decision was countered by a clear statement
from the German plenipotentiary that the German Government accepted the invitation
of the French Government with the idea that the principal object of the Conference was
the creation of a permanent neutral scientific institute exactly as expressed in the draft
No 1 and it is for this he has authorized to give his signature.
A few days later, on 15 April 1875 at the penultimate session of the Diplomatic
Conference on the Metre, it was decided to add a preamble in which the purpose of the
Convention would be stated to precede the Articles of the Convention. The preamble
leads directly into the Articles of the Convention and appears as follows [14]:
Their Majesties . . . etc., etc., (list of names of Heads of State beginning

with His Majesty Emperor of Germany, Allemagne in French),
wishing to assure the international unification and perfection of the met-

ric system, have decided to conclude a Convention to this effect and have
designated as their Plenipotentiaries, namely:(4 )
(list of names of Plenipotentiaries,)
who, after having communicated their full powers, found to be in good

and due form, have decided upon the following arrangements:
(4 ) Désirant assurer l’unification internationale et le perfectionnement du système mètrique,

ont résolu de conclure une Convention à cet effet et ont nommé pour leurs Plénipotentiairies,
savoire :
214 T. J. Quinn
First Article
The High Contracting Parties undertake to create and maintain, at their

common expense, a scientific and permanent International Bureau of
Weights and Measures with its seat in Paris.
Article 2 (1875)
The French Government will take the necessary steps to facilitate the
acquisition or, if the case arises, the construction of a building specif-
ically dedicated to this purpose, in accordance with the conditions set
out in the Regulations annexed to the present Convention.
Article 3 (1875)
The International Bureau shall operate under the exclusive direction and
supervision of an International Committee for Weights and Measures,
itself placed under the authority of a General Conference on Weights
and Measures, consisting of the delegates of all the contracting Govern-
ments. [Italics as in the original]
There then followed the remaining 13 Articles of the Convention and 22 Articles of the
Regulations which, following Article 22 of the Regulations have the same force as those
in the Convention. In addition, there was a supplementary Annex of 6 Articles setting
out the transition arrangements for States that participated in the International Metre
Commission but had not signed the Convention plus the tasks of the French Section of
the Commission and the costs of making the new prototypes.
The stated mission of the Convention, “to assure the international unification and
perfection of the metric system” is clearly a much more general mission for the orga-
nization than was that implied by the title of the first draft of the Convention given
earlier in this article and reflects the broad long-term aims set out by Dumas in his key
intervention during the Conference.
Unfortunately, this preamble to the Convention has mostly been omitted in published
texts of the Convention, which begin simply with Article 1. The last time it was used
was in the Convention signed during the 6th General Conference on Weights and
Measures held at the International Bureau of Weights and Measures on 6 October 1921
by the Plenipotentiaries of the then 27 States Party to the Convention of 1875 [15].
This new Convention modified many Articles of the 1875 Convention and Annexed
Regulations, its title was:
International Convention bringing modifications:
1. to the Convention signed in Paris on 20 May 1875 to assure the inter-
national unification and perfection of the metric system;
2. to the Annexed Regulations of this Convention.
The omission of the preamble, which gives a name and mission to the Convention,
has been that the whole international organization created by the Convention seems to
be missing a name. This has recently led to misunderstanding as to the structure of
the organization created by the Convention. It seems evident that since the Convention
was specified as the “Convention to assure the international unification and prfection
of the metric system” there was no need to give a specific name to the international
organization created by the Convention because it would simply repeat the same words.
The Convention of 1875 created the first international scientific institute, the Inter-
national Bureau of Weights and Measures, always referred to by its acronym BIPM from
its French name Bureau international des poids et mesures. The Convention placed the
BIPM under the exclusive authority and supervision of an International Committee for
Weights and Measures(5 ), which itself would be under the authority of a General Con-
ference on Weights and Measures(6 ) made up of delegates of all the States Parties to the
Convention.
The mission of each of the three organs of the Convention, set out in Articles of the
Convention and Regulations in 1875 modified slightly as shown in 1921, very clearly
provide for the accomplishment of this mission:
The mission of the General Conference on Weights and Measures: Article 7 (1875)
of the Regulations states:
The General Conference, mentioned in Article 3 of the Convention, shall meet in Paris,
on the convocation of the International Committee at least once every six years.
Its mission is to discuss and instigate measures necessary for the propagation and
perfection of the metric system as well as to approve new fundamental metrological
determinations that might have been made in the interval between its meetings. It
receives the Report of the International Committee on work accomplished and proceeds
by secret ballot to the renewal of half of the International Committee.
Votes etc., etc. . . .
The mission of the International Committee for Weights and Measures is set out in a
number of Articles, in particular Article 3 of the Convention, given above, in respect of
its full responsibility for the International Bureau, then in Articles of the Regulations in
respect of the acquisition of a building and instruments, then in 1921 its responsibilities
were broadened in Article 10 (1921) of the Regulations:
The International Committee directs all metrological work the High Contracting parties
shall decide to have carried out in common.
It is also charged with supervising the conservation of the prototypes and international
standards.
(5 ) Now always referred to as the CIPM, standing for Comité international des poids et mesures.
(6 ) Now always referred to as the CGPM, standing for Conférence général des poids et mesures.
216 T. J. Quinn
Finally, it may initiate the cooperation of specialists in metrological matters and

coordinate the results of their work.
The mission of the International Bureau of Weights and Measures was set out in
Article 6 (1875) of the Convention:
The International Bureau of Weights and Measures is charged with:
1◦ all comparisons and verifications of the new prototypes of the metre and the kilo-
gram;
2◦ the conservation of the international prototypes;
3◦ the periodic comparisons of national standards with the international prototypes

and their official copies as well as those of the standard thermometers;
4◦ the comparison of the new prototypes with the fundamental standards of non-metric
weights and measures used in different countries and in the sciences;
5◦ the calibration and comparison of geodetic standards;
6◦ the comparison of precision standards and scales whose verification may be re-
quested, either by Governments, by learned societies or even by des artistes et des
savants(7 ).
In 1921 its responsibilities were extended by the addition of Article 7 (1921) of the
Convention as follows:
After the Committee will have carried out the work of coordination of measurements
related to electrical units, and when the General Conference shall have so decided by
unanimous vote, the Bureau will be charged with the establishment and the conservation
of standards of electrical units and their official copies as well as the comparison with
these standards of national standards or other precision standards.
The Bureau is also charged with determinations related to physical constants for
which more accurate knowledge might serve to increase the precision and ensure better
uniformity in the fields to which the units mentioned above belong (Article 6 and 1st
paragraph of Article 7).
Finally, it is charged with the work of coordinating similar determinations made in other
institutes.
Taking the stated missions together, the structure and the functioning of the Metre
Convention is clear:
(7 ) At the time of the signature of the Convention “artistes” and “savants” referred respectively
to craftsmen (artisans) who made precision standards and scales and to individual scientists.
– It is a Convention to assure the international unification and perfection of the

metric system,
– The General Conference is given overall responsibility: “to discuss and instigate
measures necessary for the propagation and continued improvement of the metric
system” Article 7 (1875) of the Regulations
– The International Committee is given a broad responsibility for running the af-
fairs of the Convention in Articles 3 of the Convention (the International Bureau)
and article 10 (1921) of the Regulations: The International Committee directs all
metrological work the High Contracting parties shall decide to have carried out in
common.
– The International Bureau is charged with carrying out the scientific work decided
by the international Committee and the General Conference in Articles 6 (1875)
and 7 (1921).
The specific scientific work was always understood to evolve (Dumas specifically stated
this during the Diplomatic Conference) and from time to time the General Conference
agrees changes to the work carried out at the International Bureau as the science of
metrology advances and needs change.
In brief, there are three organs of the Convention: The General Conference is respon-
sible for the main mission of the Convention, the International Committee carries out the
organization of work decided in common and the International Bureau does the scientific
work.
This is essentially what was proposed by the committee for Preparatory Research in
1872, formally proposed by the International Metre Commission in 1873 and adopted by
the Diplomatic Conference in 1875(8 ).
(8 ) In 2007 the Bureau of the CIPM (i.e., its President, Secretary and Vice-Presidents) adopted
a proposal from the BIPM that the intention of those who set up the Convention in 1875 was
to create an international organization named the BIPM; that this organization, the BIPM,
comprised only two organs: the General Conference and the International Committee and that
in the Convention there is no mention of a scientific institute (CIPM Report 2007). This position
was adopted apparently in ignorance of the preamble to the Convention of 1875 and the title of
the Convention of 1921 and it is inconsistent with the texts given here, concerning the intentions
of those who in 1875 set up the Convention, taken from the Report of the International Metre
Commission 1873, the Diplomatic Conference of 1875 and Conventions of 1875 and 1921. It
is also inconsistent with the description of the Convention and its organs given in the BIPM
Centenary volume of 1975, The Bureau international des poids et measures 1875-1975 published
by the BIPM and also published in English as NBS Special Publication 420, The International
Bureau of Weights and Measures 1875-1975, which are wholly consistent with the description
of the Convention and its organs given in this article.
218 T. J. Quinn
The legal status of the Bureau in France was set by Article 3 of the Regulations which
states:
The French Government at the request of the International Committee will take
the necessary steps to have the Bureau recognized as an establishment of public utility
(établissement d’utilité public).
An “establishment of public utility” is a particular French term originating in the
middle of the 19th century referring to such entities as hospitals, schools and museums
which have a legal personality but not funded by the French State as opposed to such
public establishments (établissement public) which are those with a legal personality
financed by the French State. For the International Bureau this took place by a decree
of the French Government dated 28 October 1876 in response to a letter to the French
Foreign Minister of 30 August from the President and Secretary of the International
Committee asking that Article 3 of the Regulations be implemented in order, among
other things, that the Bureau would have legal personality so that it could carry out
its financial and other duties. With the development of the practice of international
organizations, such a recognition was complemented in 1969 by what is now called an
Accord de Siege (Seat or Headquarters Agreement). This is a much more comprehensive
document signed by the President of the CIPM and the French Foreign Minister and
gave the Bureau the necessary legal and diplomatic protection by then established for
international bodies. It was updated in 2007.
In 1907, the 4th General Conference on Weights and Measures modified an Article in
the Regulations concerning financial reporting to the General Conference. The details
are of little interest now except for the introduction of the word “Service” apparently
to represent an International Service of Weights and Measures provided by the BIPM.
It had been used in 1887 to refer to the new hydrogen temperature scale established by
the BIPM. Much earlier, however, in a letter dated 23 December 1878 the President of
the CIPM informed the French Foreign Minister and other contracting Governments that
“starting in 1879 the BIPM will be ready to compare the provisional new metric standards
with those of the services of weights and measures of contracting Governments”. In
France the Service of weights and measures was provided by the Conservatoire National
des Arts et Métiers in Paris, i.e., the National services of weights and measures are
services provided by the National Bureaus of Weights and Measures (whatever their
official name happened to be) and the International Service is that provided by the
BIPM. This is confirmed on page 24 of the Report of the 4th General Conference in
1907 [16] where we find “. . . the International Bureau has continued to be busy with the
first periodic comparisons of prototype kilograms, both the official copies belonging to the
International service of weights and measures and those kilogram prototypes belonging
to the different national services of weights and measures.” Thus it is clear that the term
Service in Article 19 (1907) does indeed refer to the International Service of weights and
measures supplied by the BIPM. It appears in a number of CIPM and CGPM reports
during the first half of the 20th century but was never used by any Director of the BIPM
and died out in the 1940s.
The new Convention of 1921 made significant modifications to the text of 1875. After
much discussion at the beginning of the 20th century, instigated particularly by the then
Director of the National Bureau of Standards, to expand the role of the BIPM beyond
length and mass, agreement was reached for this to be done. Specifically, the Convention
was modified to include specifically work on electrical standards but more generally
any work in the future that the international committee and member states considered
should be done in common. The addition of electrical standards was a major step since
it brought under the auspices of the Metre Convention one of the most important areas
of scientific and industrial activity which up to then had been dealt with by a succession
of ad hoc Congresses.
Since 1921, the terms of the Convention and its Regulations have remained unchanged
and still remain the rules and procedures that govern today’s operation of the Interna-
tional Bureau, the International Committee and the General Conference. The range of
scientific work carried out by the BIPM has, of course considerable widened and broad-
ened and the range of activities covered by the CPM has also much increased. The
Convention remains the formal basis upon which Member States act together on matters
related to measurement science and measurement standards. That this is still the case
is an indication of the wisdom and foresight of those who drew it up. At the time of
writing this article there were 58 States Parties to the Convention —the modern formal
term for High Contracting Parties or Contracting Governments, terms that appear in the
Convention. Member States is also a modern term but because in 1875 the Convention
was given a name (to assure the international unification and perfection of the metric
system) but not the international organization created to carry out this mission, it does
not fit very well.
Formally they are referred to either as States Party to the Convention, High Con-
tracting Parties or contracting Governments — all terms found in the Convention. In
modern terms they would be Member States of (name of the Organization), but in the
case of the Metre Convention I see no objection to referring to Member States of the
“Convention to assure the international unification and perfection of the metric system”,
or in short Member States of the Metre Convention(9 ).
The CIPM since 1927 has created a number of Consultative Committees to advise it
on the various domains of metrology. There are now ten such Committees including one,
the Consultative Committee for Units, specially charged with advising the CIPM on the
International System of Units, SI.
Figure 1 is an organigram of the structure of the Convention that includes all of the
above but also recent additions such as Associate States of the General Conference and
the CIPM MRA, discussed later.
(9 ) Article 6 (1921) states that States that have not paid their subscription for six years are
excluded from the Convention. In my view this implies that they were previously “included”
in the Convention, which seems to imply they can be referred to as Members States of the
Convention.
220
The Metre Convention in 2017

Is a combination of two Conventions:
The Convention of 1875:
Its stated mission was “to assure the international unification and perfection of the metric system”
and the Convention of 1921:
having the title“International Convention to bring modifications to the Convention signed in Paris on 20 th of May 1875 to assure the
international unification and perfection of the metric system and to its annexed Regulations”
General Conference on Weights and Measures CGPM

delegates of Member States, now takes place every four years
Mission in Article 7 of the Regulations (1875):“to discuss and instigate measures necessary for the propagation and
perfection of the metric system “, election of International Committee plus scientific, financial and other questions
International Committee for Weights and Measures CIPM ,

annual meeting , 18 members elected by the CGPM
Tasks in Article 3 of Convention (1875): exclusive direction and supervision of the BIPM, and various Articles of the
Regulations, notably Article 10 (1921) directs all metrological work decided by Member States to be carried out in
common and initiates and coordinates cooperation as necessary
Fig. 1. – An organigram of the organs of the Metre Convention.
5. – The creation of the BIPM and early scientific work

International Bureau of Weights and Measures BIPM
The permanent scientific international metrology institute at Sèvres
Tasks: originally Article 6 (1875) of the Convention then Article 7 of the Convention of 1921; to include
conservation and international comparison of standards and other scientific and coordination work to meet
the changing needs of Member States as decided by successive General Conferences.
the most modern apparatus of the time and staffing it with the best possible scientists
T. J. Quinn
every help was given in its creation. The cost of the new institute, 400 000 francs and
of the Metre Convention. Everyone agreed as to the importance of the new institute and
The creation of the BIPM, the construction of its laboratories and equipping it with
was a relatively straightforward operation compared with the drawing up and signature
its annual cost of 75 000 francs(10 ) was met directly by the contributions of the Member
States, as was the eventual cost of the new international prototypes and the 40 copies of
the new Pt-Ir metres and kilograms ordered by the Member States.
The site offered by the French Government for the installation of the International
Bureau was the Pavillon de Breteuil in the Parc de Saint Cloud on the Western outskirts
of Paris [17]. The Pavillon de Breteuil, like the Chateau de Saint Cloud, long predates
1870. It had been inaugurated on 11 August 1672 by Louis XIV, King of France, and
given to his brother, Monsieur, Duc d’Orleans, then living at the Chateau de Saint Cloud
with his wife Henrietta, sister of Charles II of England. Monsieur had had it constructed
by Thomas Gobert, the King’s architect and engineer, at the southern extremity of the
Allée de Mail which led from the Chateau towards Sèvres. Begun in 1670 and finished
at the end of the following year, the Trianon de Saint Cloud, as it was then called, stood
halfway up a hillside overlooking the Seine. Enlarged in 1743, renamed the Pavillon
de Breteuil just before the Revolution in honour of the Baron de Breteuil the King’s
First Minister, it had been badly damaged in the Franco-Prussian war of 1870. In this
dilapidated state it was offered to the International Committee by the French Government
for the installation of the International Bureau.
The laboratory building, called the Observatoire, was completed very quickly and was
ready in 1878. One cannot but admire what was achieved. One of the best metrology
laboratories in the world had been constructed that would continue to serve the purpose
for which it had been designed up to the present day, some one hundred and thirty years
later. The problem of temperature control was a serious challenge notably because the
Metre was defined at zero degrees Centigrade and it was decided that the laboratories
should be capable of operating at or a little below 7 ◦ C. The conclusion one must come
to about the new BIPM laboratories in 1878 is that the designers did extremely well for
a very modest cost.
As regards equipment, the best comparators and best balances of the day were pur-
chased and installed in the new laboratories. In addition to the Director, a small scientific
staff, of only two people plus assistants as specified in the Convention, was recruited.
Article 19 of the Regulations annexed to the Convention requires the International
Committee to make an annual report to governments on the scientific, technical, and
administrative work that it and the International Bureau had accomplished. The first of
these annual reports was written in September 1877 and covered all that had been done
since 1875. Subsequently, and every year since then, such a report has been written.
The early ones provide a very full account of almost everything that had been achieved,
including copies of all the letters to and from member governments as well as details
of the scientific work, laboratories, and instruments. In addition, of course, a report on
each meeting of the International Committee was published.
An important activity of any scientific organization is the publication of the results
(10 ) The annual salary of the first Director was 15 000 francs, that of senior scientists 6000
francs, assistants and mechanic/caretaker 3000 francs and office boy 1500 francs.
222 T. J. Quinn
of its scientific work. The Committee decided quite early, in 1879, that there should be
a special publication entitled Travaux et Mémoires du Bureau International des Poids
et Mesures. The Travaux et Mémoires were to include details of all the work on the
calibration and evaluation of the new metric prototypes as well as articles by members
of the International Committee. Planned to appear as an annual volume, the articles re-
porting the work carried out under the responsibility of the Committee would be refereed
by members of the International Committee, while other articles of a general scientific
nature, either by members of the Committee or members of the BIPM staff, would fall
under the individual responsibility of each author. The committee decided that in all
its reports the results of measurements would be accompanied by an estimate of the
uncertainty evaluated as the probable error according to the law of Gauss.
The first volume appeared in 1881 followed by a further six up to 1890. The Travaux et
Mémoires continued in its original form up until Volume 21 in 1952, when the Committee
decided that it should cease publication. After that, the Bureau’s publications were
increasingly published in the open scientific literature.
During the first ten years of operation, the work was devoted to collaboration with
the French Section on the fabrication of the new international prototypes and the fabri-
cation and calibration of the 40 national prototypes. It soon became clear that for the
proper characterization of the new metre standards, a new temperature scale was needed
because, although the metre was defined at 0 ◦ C, it was necessary to know the thermal
expansion coefficient of the material so that they could be used at higher temperatures.
For this, a temperature scale was needed, soundly linked to thermodynamics, which could
be adopted internationally. Up to then, no such scale existed.
Work at the Bureau led to a temperature scale based on a hydrogen gas thermometer
carried by a set of Tonnelot mercury-in-glass thermometers — at the time the best
available. This was published in the Report of the 1887 meeting of the CIPM as the
Normal Hydrogen Scale and became the first internationally adopted scale. It covered
the range from −25 ◦ C to 100 ◦ C and was subsequently widely used, being replaced only
in 1927 when an International Temperature Scale of 1927 was adopted covering a much
wider range of temperature. The ITS-27 was based on a set of fixed points from the
boiling point of oxygen at about −183 ◦ C up to the freezing point of gold at 1063 ◦ C,
with interpolation between the fix points by means of platinum resistance thermometer
from −183 ◦ C to 660 ◦ C, the platinum/rhodium thermocouple from 660 ◦ C to 1063 ◦ C

and the optical pyrometer for extrapolation to higher temperatures, the scale usually
being carried by tungsten ribbon lamps. The ITS-27 was replaced by the IPTS-48, the
IPTS-68 and finally the ITS-90, the current international temperature scale [18].
By about 1887, ten years after the laboratories of the Bureau had been opened,
everything was ready for the formal adoption of the new International Prototypes of the
Metre and the Kilogram and the distribution of the national copies. The first General
Conference on Weights and Measures was called for October 1889. The opening session
was held at the French Foreign Ministry where the Diplomatic Conference of 1875 had
been held. Subsequent working sessions were at the International Bureau. This pattern
continued for many years, with the working sessions of the first nine General Conferences,
up to that in 1954, taking place at the International Bureau. From 1960 until 2007
all sessions took place at the Conference Centre of the French Foreign Ministry, with
essentially no financial charge on the BIPM other than a slowly increasing charge for
interpreters into English and Russian. Since then, in 2011, and 2014 the Conferences
have been organized wholly by, and paid wholly by, the International Bureau and held
at convenient Conference Centres in Paris and Versailles.
At the time of the 1st General Conference the scientific staff consisted of the Direc-
tor, René Benoit and two senior assistants, named Charles-Eduard Guillaume and Pierre
Chappuis. Both were remarkable men, Chappuis had established the hydrogen temper-
ature scale and Guillaume was beginning his work on the alloys of nickel (for length
standards) that eventually led him to discover an alloy that had almost zero thermal
expansion at room temperature —to be named invar. For this he received the Nobel
Prize for Physics in 1920.
In addition to the scientific work just mentioned, the International Bureau quickly
became in much demand for the provision of calibrations. These calibrations of length
standards, weights, and thermometers, other than the official national prototypes, began
with fairly small numbers, but as time went on and the reputation of the Bureau grew,
they began to occupy an increasing proportion of the time of the scientific staff. They
were, however, in no sense secondary to the main purpose of the Bureau. Among the ar-
guments advanced at the Diplomatic Conference on the Metre for a permanent scientific
institution was that there then existed no institute that could either give reliable calibra-
tions of standards to institutes and universities or provide the foundation for comparable
scientific results worldwide. That the signatories to the Convention took advantage of
the capabilities growing at the International Bureau is shown by the enormous diver-
sity and geographical spread of those asking for calibrations. These were not only for
work related to the international prototypes of the metre but also for thermometers and
weights. In 1887 the number was quite small but by 1895 it had risen to fifty per year
and by 1905 it had become about one hundred. This activity continued at a high level
until the creation of increasing numbers of national laboratories were founded which then
undertook calibrations for their own countries.
6. – The BIPM since 1921
The next big change to the work carried out at the Bureau was the introduction of
electrical standards in the late 1920s following the changes to the Convention in 1921.
Work on photometric standards were introduced in 1939 so that by the time of the
Second World War, the Bureau was active in length, mass, temperature, electrical and
photometric standards. To accommodate these extensions, the laboratory space had to
be increased and in 1929 an extension to the Observatoire was made, thanks in large
part to a grant from the Rockefeller Foundation in the USA.
Finally, in the years since the Second World War the Bureau has evolved to include
scientific work on standards of ionizing radiation in 1964, International Atomic Time in
1988 and chemistry in 1999. On the site a number of new buildings have been erected:
224 T. J. Quinn
1964 for ionizing radiation work, later also used for the chemistry programme, in 1984
for the length work when it became devoted to laser wavelength and frequency standards
after the re-definition of the metre in 1983; in 1988 for a new library and offices and
in 2002 for a large meeting room and new mechanical workshop. In order to fund the
extension into metrology in chemistry, the International Committee decided in 2002 to
cease work in photometry and in length standards. Reports of current scientific work are
to be found in the Annual Report of the Director and details are on the BIPM website.
Staff numbers remained small, some 15 people, until the 1930s but increased with the
introduction of new work after the Second World War and by the 1990s reached a level
of about 75 people where it has remained to the present day.
7. – The CIPM MRA
A major advance in the international coordination of national measurement standards

was the CIPM MRA, set up by the CIPM in 1999, under the authority given to it by
Article 10 (1921) of the Regulations appended to the Metre Convention(11 ). Its full name
is “Mutual Recognition Arrangement for national measurement standards and calibration
certificates issued by national metrology institutes”. It was to be signed by the Directors
of National Metrology Institutes with the approval of the appropriate authorities in their
country. These countries are primarily Member States of the Metre Convention but also
Associate States and Economies of the General Conference on Weights and Measures.
This second category was created by the 21st General Conference at the time it approved
the action of the CIPM in setting up the CIPM MRA.
The stated objectives of the CIPM MRA are:
– To establish the degree of equivalence of national measurement standards main-
tained by the national metrology institutes (NMIs).
– To provide for mutual recognition of calibration and measurement certificates issued
by NMIs.
– Thereby to provide governments and other parties with a secure technical founda-
tion for wider agreements related to international trade, commerce, and regulatory
affairs.
The process for meeting these objectives are:

– International comparisons of measurements, to be known as key comparisons.
– Supplementary comparisons of measurements.
– Quality systems and demonstrations of competence by NMIs.
The outcome of all this are:
– Statements of the measurement capabilities of each NMI in a database maintained
by the BIPM and publicly available on the web.
(11 ) The CIPM MRA is fully described on the BIPM website.
This can be summed up as a process to establish mutual confidence between national

metrology institutes as to the accuracy and reliability of their standards and calibration
and measurement certificates. It was also a process that was designed to give additional
confidence to the outside users of the services of the national laboratories.
The CIPM MRA has now been signed by the Directors of the national metrology
institutes of 57 member States, 40 Associate States and Economies of the General Con-
ference, 4 international organizations plus some 150 other institutes designated by the
signatory institutes. The results of the key comparisons and the lists of calibration and
measurement capabilities are on the BIPM public website.
8. – Conclusions
Since the signing of the Metre Convention on 20 May 1875, metrology has expanded
its domain and its importance for society. Today it is part of the essential but largely hid-
den infrastructure of the modern world. We need it for high-technology manufacturing,
human health and safety, the protection of the environment, global climate studies and
the basic science that underpins all these. Highly accurate measurements are no longer
the preserve of the physical sciences and engineering, many areas of chemistry, molecular
biology and medicine are now dependent on the ability to make accurate quantitative
measurements. International trade in all manufactured and agricultural products are
strictly controlled by regulations that need accurate metrology for their implementation.
Good metrology improves the quality and efficiency of manufactured products. There
are a multitude of examples of this from modern motor vehicles of much improved re-
liability and fuel efficiency to passenger aircraft that can fly nearly half-way round the
world non-stop. We have become accustomed to products and services that only work
because of extreme accuracy in measurement. Satellite navigation is the most striking of
these. GPS works only because atomic clocks in satellites are stable and can be adjusted
to run with common time scales accurate to nanoseconds with their positions in space
known to within a metre or so. Reliable long-term monitoring and prediction of criti-
cal parameters of the world’s climate needs accurate measurements linked to reference
standards that are themselves stable in the long term.
All of this is assured by the network of National Metrology Institutes cooperating

together under the auspices of the Metre Convention. On the eve of the transformation
of the International System of Units, the founders of the Convention could be rightly
proud of what has been achieved through the organization they created. The mission
they gave to the Convention on 20 May 1875 “to assure the international unification and
perfection of the metric system” (now the International System of Units, SI), remains
the guiding principle of the Convention and in November 2018 the transformation of the
SI will set the Convention ready to meet the challenges of the future.
Bibliography
The two principal sources for the content of this articles are:
226 T. J. Quinn
1. Quinn T. J. From Artefacts to Atoms, the BIPM and the Search for Ultimate
Measurement Standards (Oxford University Press, New York) 2011, 436 pp.
2. The BIPM website www.bipm.org on which can be found the principal official
documents mentioned in this article, namely, the proceedings of the Diplomatic
Conference on the Metre, 1875, the Reports of successive General Conferences on
Weights and Measures from the 1st in 1889 to the most recent, the 25th in 2014 and
Reports of the meetings of the International Committee for Weights and Measures
from 1875 to the present day as well as a wealth of information on the current
activities of the CIPM, its Consultative Committees and the BIPM.
REFERENCES
[1] Cox E. F., “The metric system: a quarter-century of acceptance (1851–1876)”, Osiris, 13
(1958) 358.
[2] Torge W., “The International Association of Geodesy 1862 to 1922: from a regional
project to an international organization”, J. Geodes., 78 (2005) 558.
[3] Procès verbaux de la Conférence Géodésique International pour la masure des degrés en
Europe réunie à Berlin du 30 septembre au 7 octobre 1867 (Imprimerie G. Guillaume fils,
Neuchatel) 1867 (French translation by Hirsch of the original German, part of this English
version is from a History of the International Association of Geodesy by J.J. Levallois).
[4] Bigourdan G., Le Système Mètrique des Poids et Mesures (Gauthier-Villars, Paris) 1901,
pp. 243–244.
[5] Comptes Rendus de l’Académie des Sciences, LXIX (1869) 514.
[6] Commission international du Mètre, Comité des recherches préparatoires, avril 1872
(Imprimerie Viéville et Capiomont, Paris) 1872, p. 90.
[7] Commission international du Mètre, Réunions générales de 1872 (Imprimerie Nationale,
Paris) 1872, p. 231.
[8] Commission international du Mètre Comité permanent, Procès-verbaux des Séances 1872-
1873, (Imprimerie Nationale, Paris) 1873, p. 40, and also Commission international du
Mètre Comité permanent, Procès-verbaux des Séances 1874, (Imprimerie Nationale, Paris)
1874, p. 74.
[9] Documents Diplomatique de La Conférence du Mètre, (Imprimerie Nationale, Paris) 1875,
p. 149, available on the BIPM website www.bipm.org.
[10] Quinn T. J., From Actefacto to Atoms, the BIPM and the Search for Ultimate
Measurements Standards (Oxford University Press, New York) 2011, pp. 61, 63.
[11] Airy papers (1873-1876), in Royal Greenwich Observatory Archives, RGO 6/367,
Cambridge University Library.
[12] Quinn T. J., in ref. [10], pp. 103–206.
[13] Documents Diplomatique de La Conférence du Mètre, in ref. [9], pp. 91-93.
[14] Documents Diplomatique de La Conférence du Mètre, in ref. [9], p. 144.
[15] Comptes Rendus des Séances de la Sixième Conference Général des Poids et Mesures
Réunie à Paris en 1921 (Gauthier-Villars, Paris) 1921, available on the BIPM website
www.bipm.org.
[16] Comptes Rendus des Séances de la Quatrième Conference Général des Poids et Mesures
Réunie à Paris en 1907 (Gauthier-Villars, Paris) 1907, available on the BIPM website
www.bipm.org.
[17] Quinn T. J., in ref. [10], pp. 95–97.
[18] Quinn T. J., Temperature, 2nd edition (Academic Press, London) 1990, p. 495.
DOI 10.3254/978-1-61499-818-1-227
The development of units of measurement from

the origin of the metric system in the 18th century
to proposals for redefinition of the SI in 2018
T. J. Quinn(∗ )
92 rue Brancas, 92310 Sèvres, France
The idea of referring all measurements to a unit of length taken from nature was seized upon
by mathematicians as soon as the existence of such a unit and the possibility of determining it
became known. They saw it as the only way to exclude all that was arbitrary from a system of
measurement and to conserve it unchanged, so that no event or revolution in the world could
cast uncertainty upon it. They felt that with such a system, belonging exclusively to no one
nation, one could hope that it would be adopted by all.(1 )
Summary. — This article is about the development of the metric system from its
origins at the time of the French Revolution to the present day. In November 2018,
the 26th General Conference on Weights and Measures (CGPM) will be invited
to adopt a new definition of the International System of Units, SI, based on fixed
numerical values of a set of seven defining constants, broadly the fundamental con-
stants of physics. From these, new definitions of the seven base units of the SI will
be deduced. It will then be a little more than two hundred years since a Commit-
tee of the l’Académie Royale des Sciences made a proposal to base a new unit of
length on a fraction of the meridian of Paris and thereby initiated the creation of the
metric system. The redefinition in 2018 will at last put into practice their original
proposal for a system independent of time and place, accessible to all and belonging
to no one nation. The key is the new possibility of replacing the present definition
of the kilogram, an artefact of platinum-iridium kept at the International Bureau
of Weights and Measures at Sèvres, by one based on a fixed numerical value of a
fundamental constant, the Planck constant.
(∗ ) Emeritus Director of BIPM.

(1 ) Author’s translation of the original French text of the Report made to the Académie Royale
des Sciences see ref. [1]

227
228 T. J. Quinn
1. – Introduction
This is the second of two articles related to the Metre Convention, the International
Bureau of Weights and Measures and the development of the metric system from its
origins at the time of the French Revolution to the present day. The main points in this
second article will be:
– Original proposals of 1791 for a system of measurement based on nature.
– The practical implementation leading to the Metre and Kilogram of the Archives
of France.
– Maxwell, Michelson and successive definitions of the Metre.
– Development of ideas related to absolute measurements in the early 19th century

and he need for units of measurement for the growing electrical industries in the
middle of the century.
– The new International prototypes of the Metre and Kilogram at the BIPM in
1889 and measurements of the metre in terms of the wavelength of light by A.A.
Michelson.
– Developments in other base units in the 20th century up until 1971.
– The changes since 1971 that finally opened the way to units based on constants of
physics.
– Discovery of the quantum-Hall effect and prospects for a new definition of the
kilogram using the Kibble balance and the x-ray crystal density of silicon.
– The final form of the proposed new definitions, notably the kilogram
– The practical advantages of units defined in terms of constants of nature
2. – The origin of the metric system in 1791

The Report of the Académie Committee dated 19 March 1791, from which the extract
at the head of this article is taken, represents the origin of the decimal metric system.
It proclaims the high ideals with which the metric system was created at the end of the
18th century. The paragraphs of the Report that followed set out what appeared to be
a practical realization of what they had planned, they read:
One can reduce to three the units that seem most appropriate as the base; the length of a
pendulum, the quarter of the length of the equator and finally the length of a quarter of
a meridian.
The length of a pendulum has the advantage of being the easiest to determine and, in
consequence, the easiest to verify if some accident happens that renders it necessary.
The development of units of measurement etc. 229
Furthermore, those who wish to adopt this measure already adopted by another country,
or having adopted it wish to verify it, would not be obliged to send observers to the place
where it was originally established.
In addition, the law of the length of a pendulum is well known, confirmed by experiment
and can be used without fearing small errors.
The idea of using the length of a pendulum that beat seconds as the unit of length
for the new system of measurement had been proposed to the National Assembly by
Talleyrand and the Committee of the Académie had been set up to examine it. By the
time Talleyrand put his proposal to the National Assembly in 1791 [2] it appeared that
a broad consensus had been reached that the new unit of length would be the length of a
pendulum that beat seconds at a latitude of 45 degrees. As is well known, the Académie
Committee, having set out so clearly all the reasons why the first option, a unit of length
based on a pendulum, met all the conditions set out in the first paragraphs of its Report,
decided to adopt the third. Why did they do this? The Report went on:
However, we should observe that the unit so determined includes within itself something
arbitrary. The second of time is one eighty six thousand four hundredths of a day and
it is in consequence an arbitrary division of this natural unit. Doing it In this way, to
fix a unit of length we are employing not only something heterogeneous (time) but also
something arbitrary[. . . ]. Besides, we have to admit that a unit of measure on the Earth
itself has another advantage that it is perfectly analogous to all real measurements in
everyday life.
Thus it was decided to choose the length of the quarter of the meridian as the base
for the new unit of length — for practical reasons the meridian passing through Paris.
This is what they did and the subsequent heroic measurement of the arc of the
meridian from Dunkerque to Barcelona passing through Paris by Méchain and Delambre
published in Bigourdan 1912 [3] and Alder [4] is one of the great stories of the history
of science.
3. – The Metre and Kilogram of the Archives 1796

The result of the meridian measurements was an artefact that became known as the
Metre of the Archives. This was a bar, nominally of platinum but in fact containing small
amounts of rhodium and iridium, 25 mm wide, 4 mm thick whose distance between the
flat and parallel ends was defined as one metre exactly. It is known as an end standard,
having the advantage of simplicity but the disadvantage that the flat ends are susceptible
to damage. This why, some 70 years later, there were calls for it to be replaced by a line
standard in which the length is defined by the distance between two fine lines engraved
on the flat surface of the standard, close to but not at the ends.
The length of the metre of the Metre of the Archives was set to be equal to 443.296
lines of the Toise de Pérou. This was their best estimate of one ten millionth of the
distance from the pole to the equator through the Paris meridian. It had been deduced
230 T. J. Quinn
from their triangulation measurements and an assumed value for the flattening of the
Earth at the poles, which was required to extrapolate their measurements to cover the
whole distance from the pole to the equator. The toise was the traditional unit of length
in France at the time. A toise was made up of 6 feet (Pieds du Roi ), each foot of
12 inches each inch of 12 lines and each line of 12 points. From the beginning it had
been understood that the unit of length, the metre, would be the basis upon which all
other units would be deduced. The kilogram, unit of mass or weight as it was called
then, was defined as the weight of one cubic decimetre of water at its maximum density
(which occurs at about 4 degrees Celsius). Remarkable experiments were carried out to
determine the weight of one cubic decimetre of water, first by Lavoisier before he was
sent to the Guillotine on 8 May 1794 by the Revolutionary judges, and afterwards by
Lefèvre-Gineau and Fabbroni (described by Bigourdan in ref. [3]). The final result was
thus the two artefacts Metre and Kilogram of the Archives. The Kilogram was a cylinder
of the same material as the Metre, about 39 mm high and 39 mm in diameter. The new
standards were deposited in the Archives of the Republic on 22 June 1799 where they
remain to this day.
The initial high ideals of those who created the metric system, in which it would
belong to no one nation and “Furthermore, those who wish to adopt this measure already
adopted by another country, or having adopted it wish to verify it, would not be obliged
to send observers to the place where it was originally established ” actually ended up as
two artefacts locked in a safe in Paris!
The lesson to be drawn is that units based on natural phenomena must be such that
they can be realized in practice within a reasonable time and with reasonable effort
otherwise one ends up with artefacts that are supposed to represent the inaccessible
ideals. This was much in the minds of those who, much more recently, set up the new
definition of the SI in 2018 based on seven defining constants chosen from nature.
4. – The Berlin Conference of 1867 and proposals for an international Euro-

pean bureau of weights and measures
Although the implementation metric system in France was slow, from 1812 pre-metric
units began to be allowed again not to be finally forbidden until 1837, it began to spread
outside France [5]. Stimulated by France exhibiting a set of metric standards from the
Conservatoire des Arts et Métiers at the Great Exhibition of 1851 in London, many
countries began to use metric measures based on standards provided by France. These
were not standards calibrated directly against the Metre and Kilogram of the Archives
but against copies at the Conservatoire made at the same time as the standards of the
Archives and deposited not only at the Conservatoire but also at the Paris Observatory
and the Académie des Sciences. In 1848 the Conservatoire had taken on the role of
intermediary for foreign links to the official French standards, although at the time of
the signing of the Metre Convention in 1875 there had arose some doubts as to which
were the real copies of the standards of the Archives [6]. While for most purposes the
situation was perfectly satisfactory, it was not so for geodesy.
The story of how the needs of geodesy in Europe led to two Conferences in Berlin,
1864 and 1867, has already been told in the accompanying article [7] on the origins of
the Metre Convention and will not be repeated here. For the purposes of the unit of
length however, the Recommendations of the 1867 Conference are important.
The main accomplishments of the 1867 Conference were to set up a formal orga-
nization for Central European collaboration in geodesy and set out a list of scientific
recommendations for future geodesy in Europe. This was the birth of what is today the
International Association for Geodesy [8].
Included in the scientific recommendations were ones to use the toise of Bessel for all
geodetic calculations, to insist that all measurements should be compared with Bessel’s
toise, that its length should be compared with the metre and that in the future results
should be given in both toise and metres. In addition, it was recommended that a
campaign of height measurements above sea level be carried out in Europe. The prime
importance for geodesy of agreement on length standards thus became very clear. The
adoption of these recommendations that marked the beginning of formal moves that led
to the signing of the Metre Convention in Paris in 1875 [9].
Among the Recommendations adopted by the 1867 Conference were the following:
– The Conference recommends that a new European prototype(2 ) be constructed in
order to define the common unit of measurement for all the countries of Europe
and for all time as precisely and stable as possible. The length of this metre should
differ as little as possible from that of the metre at the archives in Paris and should
in any case be compared with it as accurately as possible. In the construction of the
new prototype standard, ease and accuracy of the necessary comparisons should be
the primary objective.
– The construction of the new prototype metre and the manufacture and comparison
of copies for the various countries should be entrusted to an international commis-
sion on which these countries should be represented.
– The Conference favours the establishment of a European international bureau of
weights and measures.
5. – The construction of the new prototypes of the Metre and Kilogram
In 1875 while all the discussions leading up to and during the Diplomatic Conference
were taking place, the French Section of the International Metre Commission was working
on the practical matter of actually constructing new international prototypes of the Metre
and Kilogram.
(2 ) Here and elsewhere the word “prototype” is used with its original 16th century meaning as
an original of which something else is a copy. The subsequent new prototypes of the metre and
kilogram deposited at the International Bureau of Weights and Measures were exactly this until
displaced by references to constants of nature.
232 T. J. Quinn
The general principles of the design, material and production of the new standards
had been worked out in some detail during the meetings of the International Commission
in 1872 and 73. The new standards would be made from a casting of platinum-10%
iridium, chosen to represent closely the actual composition of the Metre and Kilogram
of the Archives. This was nominally pure platinum but in fact, as in all early platinum
objects, it contained significant amounts of iridium as found in the ore. The new Metre,
and its copies for States signing the Convention would be of a special X cross section
with lines engraved on the central neutral plane set to be one metre apart. The new
Kilogram would be a cylinder of the same alloy 39 mm high and 39 mm in diameter.
Following the meetings of the Commission it was decided that a total of 40 new metres
and kilograms should be made to meet the demand of the States that had shown interest
and to leave some over for future Member States. For this it was calculated that 250
kilograms of platinum-10% iridium would be needed. The actual new prototypes, of the
Metre and the Kilogram would be chosen from among these 40 when they had all been
compared with each other and with the Metre and Kilogram of the Archives. The new
prototypes would be those that had the length and weight closest to those of the Archive
originals.
The whole story of the casting of the alloy and construction of the new prototypes is
one of a saga of high science and national pride on the part of the French Section. Good
relations already existed between Henri Saint Clare Deville, the famous French chemist
who had developed the original process of making pure platinum, and George Matthey
the founder of the firm in London bearing his name which was the major producer
of platinum objects for industry. Instead of passing the order for the 250 kilograms
to Matthey, which would have been the sensible thing to do, it was decided to build
a special furnace in the courtyard of the Conservatoire National des Arts et Métiers
and do it there. Predictably, the casting failed and the 250 kilograms of alloys were
contaminated with a small but significant amount of iron. After much argument and
dispute within the French Section and with the International Committee which finally
refused to accept the contaminated casting, the order was finally placed with Matthey
in London. The new international prototypes were thus, in the end, made from an alloy
cast in London. The drawing of the metre bars and the tracing of the lines as well as
the machining and adjusting of the kilograms were all done in Paris. The work of the
French Section of the International Metre Commission was extensively documented and
provides a lasting record of the huge amount of work done [10]. This is also described in
some detail by Quinn (see Bibliography)
The new prototypes and the national standards were formally adopted at the 1st
General Conference on Weights and Measures(3 ) held at the BIPM in 1889 [11]. The
(3 ) CGPM: Conference Général des Poids et Mesures, in English: General Conference on

Weights and Measures often referred to in the text as General Conference; CIPM: Comité
international des poids et mesures, in English: International Committee for Weights and Mea-
sures, often referred to in the text as International Committee; BIPM: Bureau international des
poids et mesures, in English International Bureau of Weights and Measures, often referred to in
new International Prototype of the Metre served as the world reference until 1960 when it
was replaced by a definition based on the certain number of the wavelengths of the orange
emission line of krypton [12]. The International Prototype of the Kilogram remains the
world reference today but, as we shall see, we plan finally to replace it in 2018 by a
definition based on constants of physics.
6. – Next steps: Maxwell, Michelson and successive definitions of the Metre
Having established new international standards of the metre and the kilogram, mem-
bers of the International Committee were well aware of the ultimate goal of producing
measurement standards that stood above simple artefacts, albeit better ones than the
Metre and Kilogram of the Archives. In the report of their meeting in 1891 [13] we find
“From the very beginning of the International Committee it has been generally recognized
to be of fundamental importance to determine the relation between the metric units and
some basic fundamental constants that one can deduce from natural phenomena” a clear
but unstated reference to 1791 Académie Report.
The original ideas of the Committee of the Académie in 1791 had clearly not been
forgotten because only a few years earlier James Clerk Maxwell in his Presidential ad-
dress given to the Mathematics and Physics Section of the British Association for the
Advancement of Science, Liverpool 1870, had said [14]:
Yet, after all, the dimensions of our earth and its time of rotation, though, relative to our
present means of comparison, very permanent, are not so by any physical necessity. The
earth might contract by cooling, or it might be enlarged by a layer of meteorites falling on
it, or its rate of rotation might slowly slacken and yet it would continue to be as much a
planet as before. But a molecule, say of hydrogen, if either its mass or time of vibration
were to be altered in the least, would no longer be a molecule of hydrogen.
If then, we wish to obtain standards of length, time and mass which shall be absolutely
permanent, we must seek them not in the dimensions, or the motion or the mass of
our planet but in the wavelength, the period of vibration, and the absolute mass of these
imperishable and unalterable and perfectly similar molecules.
In his inaugural lecture as Cavendish Professor at Cambridge in 1871, Maxwell set

out his intention to pursue the subject of accurate experimental techniques and the
attainment of accurate results [15, 16]. While this was much criticized at the time as
being a subject unsuitable for university attention, it was and remained an important
theme of his work.
Coming back to 1891, the CIPM decided to invite A.A. Michelson, then at Clark
University in Worcester, Massachusetts USA, to come to the BIPM with his optical in-
terferometer and measure the length of the International Prototype of the Metre in terms
of the wavelength of light. By then his reputation as a fine experimental physicist had
the text as International Bureau.
234 T. J. Quinn
been established, principally by his measurement of the speed of light and invention of his
eponymous interferometer with which, in collaboration with E.W. Morley he had demon-
strated the constancy of the speed of light in all directions which in turn demonstrated
the absence of a stationary aether through which the Earth moved. He later received the
Nobel Prize for physics for his “optical precision instruments and the spectroscopic and
metrological investigations carried out with their aid” [17].
Michelson built a special interferometer and at the BIPM in 1892 in collaboration
with the then Director, René Benoit, accomplished the task as requested — remarkably
quickly. The result obtained was that the metre equals 1 553 163.8 times the wavelength
of the red emission line of cadmium [18]. This was much improved a few years later
in 1906 by Benoı̂t and Fabry who found the number 1 553 164.13 with an estimated
uncertainty equivalent to a few tenths of a micrometre [19].
Put another way, their result was equivalent to saying that the wavelength of this light
is 0.64384696 micrometres. This same value was taken by the 7th General Conference in
1927 to define the Angström [20] so that λcd = 6438.4696 Å.
Finally in 1960, when the metre was redefined in terms of the wavelength of the orange
line of krypton [21], the value taken was chosen to be consistent with the 1927 value for
the Angström and hence consistent with Benoı̂t and Fabry’s result in 1906 which was
itself very close to Michelson’s in 1892.
Why did they wait nearly seventy years before making this important change, which
with hindsight they could have done in 1892? During the measurements of Benoit
and Fabry and in the years soon after, they perceived that the emission line of cad-
mium was not as fine as it should be, as was the case with many other spectral emis-
sion lines. This problem was only resolved with the discovery of isotopes by Fred-
erick Soddy in 1913 and their effect made visible in mass spectrographs by Aston a
little later. It was only after the Second World War that nearly pure isotopes be-
came easily available so that in 1960 a new definition of the metre could finally be
adopted using emission light of the stable isotope of krypton 86. They were thus prob-
ably right not to be precipitate in 1892 in redefining the metre. It was also the case
that the stability and reproducibility of the platinum-iridium metres constructed in
the 1880s were remarkably good. The instruments used to compare them were also
remarkable, comparisons could be made with uncertainties below one tenth of a mi-
crometre, so there was no pressing practical need to adopt a new definition at that
time.
As regards the kilogram, very little happened from 1889 until the present day. This
is perhaps an astonishing statement to make in view of the advances in science that have
taken place since then. As we shall see, however, it has turned out to be a formidable
challenge to devise a method of linking mass to a constant of nature. Up until recently
there were no developments in science that gave any prospect of replacing the artefact
definition by one that would represent the aspirations of either the savants of the 18th
century or Maxwell in his statement of 1870. Although some knowledge of the atomic
structure of matter was gained during the nineteenth century, a detailed knowledge of
atomic structure and the ability to measure physical properties of atomic phenomena only
emerged in the twentieth. Quantum effects linked to a macroscopic mass remained non-
existent. It was only in 1980 that the prospect changed with the discovery of a quantum
standard of electrical resistance. This combined with the quantum standard of electrical
potential by Brian Josephson in 1962 opened the possibility of an electromagnetic power,
thus an electromagnetic force linked to quantum standards being developed. This we
come to in a later section.
7. – Electrical standards
During the nineteenth and early twentieth centuries developments had also taken
place in another area of science where the necessity for units was primordial, namely
electricity. For electrical units it all really started in 1832 with Carl Friedrick Gauss who
was not only interested in geodesy and the shape of the Earth as we have seen earlier
but also in its magnetic field. He wrote a memorandum entitled “Intensity of the Earth’s
magnetic field in absolute measure”(4 ). By absolute measure he meant in terms of the
three units of mass, length and time, as opposed to simply in terms of another magnet,
which itself might be variable. This use of the word absolute later came to mean a
measurement in terms of these three basic units.
The term “absolute measurement” is not used in modern metrology, the term does not
appear in the International Vocabulary of basic terms in metrology (VIM3, 3rd edition,
2008). Nevertheless, the concept remains and is well described for historians of science
by Darrigol [22] and Mitchell [23]. In metrology the term “primary measurement” is
used instead as we shall see later. The two areas of metrology in which such absolute
measurements took longest to establish were in the areas of temperature and electricity.
We shall come back to this later in sect. 10.2 in respect of the practical realization of
units of the new definition of the SI. The fundamental weakness of the present SI is, of
course, the definition of the kilogram which is based on the mass of an artefact which,
by its very nature, is still subject to change. In this sense, absolute measurements of any
quantity related to mass are still not possible in the present SI. This will be resolved by
the new definition of the kilogram proposed for 2018 when measurements made in terms
of the any of the base units of the SI, based on the constants of nature, can reasonably
be called absolute in the original sense of Gauss.
Soon after Gauss, Wilhelm Weber extended Gauss’s ideas to electrical quantities when
in 1840 he published his absolute measurements of electric current together with Gauss
and then followed by absolute measurements of electrical resistance in 1851 [24]. In the
1860s the British Association for the Advancement of Science (BAAS) created an Elec-
trical Standards Committee under the stewardship of James Clerk Maxwell and William
Thomson, and the telegraph engineer Fleeming Jenkin [25-27]. The main impetus at the
time was the urgent need for reliable standards and measurements related to the technol-
(4 ) There is an extensive literature on the development of international electrical standards in

the 19th and early 20th centuries, here I give bibliographic references only for those that concern
directly international activities concerning electrical units.
236 T. J. Quinn
ogy of undersea telegraphic cables. The Electrical Standards Committee in its reports of
the 1860s systematized, endorsed and disseminated absolute units first defined by Gauss
and Weber. Thomson’s commitment to energy physics, led it also to define and add
an absolute unit of work [28]. The BAAS in 1863 formulated the outline of a coherent
system of units in which there were base units and derived units. They used the word
coherent to mean without “useless coefficients”, about the same meaning as it has today.
In 1873 the BAAS Nomenclature Committee introduced the “cgs system” or centimetre,
gram, second system, [29]. The Ohm was named by the Standards Committee and the
Volt and Farad by the Nomenclature Committee and finally fixed internationally by the
first Electrical Congress in 1881, see below.
The BAAS Committees that developed all this, guided by Thomson, were strongly
of the opinion that the absolute units devised by Gauss and Weber, based on the three
fundamental units for mass, length and time, were the best proposed. Prefixes from
mega to micro were also introduced. There were two problems, however, with the cgs
system: one was that there were two categories of electrical units, electrostatic, if one
started with the force between electric charges, and electromagnetic, if one started with
the force between wires carrying electric currents; the other was that the sizes of the units
in both these categories were much too small for practical use. The BAAS therefore also
proposed two “practical units”: the ohm, equal to 109 times the cgs electromagnetic unit
of resistance and close to the resistance of a column of mercury 1 mm2 in section; and the
volt, equal to 108 times the electromagnetic unit of electromotive force and very close
to that of a Daniell cell, a practical source of electromotive force, or voltage in modern
terms.
The Congress of Electricians in 1881 [30], the first major international meeting to
consider international adoption of electrical units, had before it proposals to adopt the
BAAS practical units for international use and this is what happened. The Congress
went further and adopted practical units not only for the volt and the ohm, but also for
the ampere as unit of electrical current and the farad as the unit of electric capacitance.
The Chicago Electrical Congress in 1893 [31] then defined two International Units, one
for the ohm and one for the volt. The International Ohm was represented by a 106.300
centimetres column of mercury of cross section such that its mass was 14.4521 grams at a
temperature of melting ice. The International Volt was defined as the electrical potential
developed by a standard Daniell electrochemical cell, later replaced by a standard Weston

cell which was more robust. These perhaps rather strange definitions were designed to
give values of the ohm and volt that were as close as possible to the defined cgs practical
units. They were, however, much more practical in the real sense as they soon took on
a life of their own and were fabricated by many laboratories to become in fact the basis
of most electrical measurements in the world.
In 1927 when the BIPM took up the maintenance of electrical standards these were
the standards that were subsequently maintained up until the 1980s when the new quan-
tum standards replaced them. Thus starting in 1927 the basic electrical standards used
worldwide were those of the volt and the ohm maintained at the BIPM by means of sets
of Weston standard cells and 1 ohm resistance standards.
The discovery of the Josephson effect in 1962 and the quantum-Hall effect in 1980
brought fundamental changes to the maintenance of electrical standards. No longer were
they to be based on material artefacts but on constants of physics, the Planck constant
h and the elementary charge e. They remained, however practical representations of the
theoretical units because direct realization of the SI unit, the ampere, cannot be achieved
with the required precision. We come to this later in sect. 9.
8. – Developments in other base units until 1971
For the other basic standards, notably for time and light and temperature, not much
changed until the middle of the 20th century. The second remained 1/86 400 of the
mean solar day and the standard of light intensity remained one based on a luminous
source, first a standard candle and later the light from a radiator at the temperature of
melting platinum, about 1770 ◦ C. The unit of temperature for a long time was defined as
a fraction of the temperature interval between the freezing and boiling points of water.
However, its relation to the absolute zero of temperature was only fixed in 1954 [31, 32]
when the kelvin was defined as 1/273.16 of the thermodynamic temperature of the triple
point of water, 273.16 K exactly. For practical temperature measurement however there
exists an International Practical temperature Scale, ITS-90 for the same reason as in
electrical metrology, the theoretical unit cannot be realized with the same precision as
practical measurements with a platinum resistance thermometer for example.
At the 11th General Conference in 1960 three significant changes took place [33].
The metre was finally redefined in terms of the wavelength of the orange emission line
of krypton 86, the metric system was made really international by formalizing it as the
Système International d’Unites, SI, the International System of Units, SI, and the unit
of time, the second, was redefined as a fraction of the period of the orbit of the Earth
around the Sun instead of the rotation of the Earth.
The SI as defined in 1960 was made up of six base units, the metre, kilogram, second,
ampere, degree kelvin, and candela corresponding to the six base quantities length, mass,
time, electric current, temperature and luminous intensity. Included were with the names
of multiples and submultiples, two supplementary units, the radian and steradian and
a set of some thirty derived units with their definitions in terms of the six base units.
Since 1960 the structure of the SI has remained almost the same but one additional base
unit, the mole, the corresponding quantity amount of substance, was added by the 14th
General Conference in 1971 [34]. The candela was redefined in 1979 in terms of a fixed
value for the luminous efficiency of a particular radiation near the peak of the sensitivity
of the human eye [35] and the metre was again redefined in 1983 in terms of a fixed
numerical value for the speed of light [36]. The new base unit the mole was defined
as the amount of substance that contains as many particles as there are atoms in 12
grams of carbon 12. The name of the quantity measured in moles, amount of substance,
remains obscure to most people who are not professional chemists. The category of
supplementary units was abolished in 1980, the units of angle becoming simply derived
units of the SI.
238 T. J. Quinn
The change in definition of the second in 1960 was intended to hold off the impending
redefinition in terms of atomic clocks which the astronomers viewed with great misgiving.
The movements of the heavenly bodies were perennial whereas electronic boxes were
certainly not! However, the new definition, the so-called ephemeris second held back the
tide of advancing science for only seven years. The first atomic clock had been made
at the NPL in 1955 and it soon became obvious that its precision and accuracy far
surpassed anything that could be deduced from either the rotation of the Earth or its
orbit around the Sun. In 1967 the second was again redefined but this time in terms of
a microwave transition of the atom of caesium 133 [37] realized by what became known
as atomic clocks. This remains the definition today although the accuracy of realization
has improved enormously since 1967 when it was a few parts in 1010 . The world’s best
caesium clocks now have accuracies estimated to be parts in 1015 with a corresponding
accuracy in the world’s reference time scale International Atomic Time. This is the basis
of Coordinated Universal Time, the civil time throughout the world.
At the time of the introduction of the mole as the seventh base unit by the 14th
General Conference in 1971, the redefinition of the remaining base units in terms of con-
stants of nature still seemed to be a far distant prospect. In introducing the definition
to the Conference, the Secretary of the International Committee, Jan de Boer, made the
following remarks [38]:
“As far as the unit of mass is concerned, the choice of an atomic definition, for example
the mass of a proton or the unified atomic mass unit would seem natural; but such a
proposal seems to me still a far cry from practical because of the necessity of determining
to high precision the mass of the proton.”
As regards electrical units:
“Here again one could imagine the elementary charge of the proton as the natural and
fundamental electrical unit to serve as the base of a universal system of units; but in this
case as well it is the requirements of metrology that render such a proposition impracti-
cable for the high precision measurement of electrical quantities.”
He ended by saying:
‘‘Naturally, one might ask also in the case of the mole would it not be preferable to replace
the definition of the mole given here by a molecular one; but as in the cases of the unit
of mass and of electric current this would require determinations such as the absolute
counting of molecules or the measurement of the mass of molecules that are not possible
with the required precision.”
Thus, even in 1971 just two hundred years after the creation of the metric system, we
were still some way from achieving the high ideals set out in 1791
9. – The changes since 1971 that finally opened the way to units based on
constants of physics
Jan de Boer’s remarks remained valid for a further 20 years before everything changed
following advances in physics unforeseen in 1971. The new physics was the quantum-Hall
effect discovered in 1980 by Klaus von Klitzing [39], which allows an electrical resistance
to be developed exactly proportional to h/e2 . This, together with the Josephson effect,
discovered in 1962 [40] that allows a voltage to be developed exactly proportional to
2e/h, made it possible to measure electrical power in terms of two of these fundamental
constants of physics. In 1990 the 19th General Conference adopted conventional values
for these two constants RK = h/e2 , the von Klitzing constant and KJ = 2e/h the
Josephson constant [41]. These have been used ever since to realize the highest accuracy
electrical standards at the level of parts in 1010 . Conventional values were needed because
the absolute accuracy of the values of h and e at that time were of the order of parts in
106 . The consequence was that formally, the highest precision electrical measurements
did not give SI values. It was to bring them back into the formal SI that was the deciding
factor in the choice of definition of the kilogram, as we shall see. The new technology
that allowed the practical application of these two quantum effects to establish a unit of
mass was the invention in 1974 of the watt balance by Bryan Kibble at the NPL [42].
With this an electrical power can be compared with a mechanical power using a simple
beam balance, leading to a unit of mass based on the Planck constant.
Another path, but from quite a different area of science, was through ideas originating
in the 1960s [43, 44] that led to the high-accuracy measurement of the lattice spacing of
silicon which in turn allowed an accurate determination of the Avogadro constant. This
in turn opened the way to counting the number of silicon atoms in a known mass of
silicon and hence the possibility of a definition of the unit of mass in terms of the mass
of an atom.
There were thus two possibilities for the definition of the kilogram, one in which we
take the Planck constant h as the basis of our new definition and the other in which we
take the mass of an atom. While a definition based on a fixed mass of an atom would be
intuitively simpler than one based on a fixed value of the Planck constant, the decision
has been taken that a Plank constant definition is to be preferred. Before coming to this
choice we look in turn at the two experiments.
10. – The Kibble balance(5 )

A Kibble balance is quite simple and works in the following way: from one arm of
balance a mass m is suspended and from the other, assumed here to be exactly equal
in length to the first, a coil of wire is hung between the poles of a magnet. An electric
current is passed through the coils such that the downward gravitational force on the
mass, Fm = mg, where g is the local acceleration due to gravity, is balanced by the
electromagnetic force on the coil. Under these conditions Fe = BLI, where B is the
intensity of the magnetic field, L is the length of coil in the field and I the current. We
can write mg = BLI. Unfortunately, this simple equation does not allow us to link mass
(5 ) The original name was watt balance, invented by Bryan Kibble in 1974. It was renamed
the Kibble balance following his untimely death in April 2016.
240 T. J. Quinn
to an electric current with the accuracy needed to use it to produce a new definition
of the kilogram — namely a few parts in 108 . Even with the greatest effort we cannot
measure either B or L with anything like this accuracy.
The clever idea of Bryan Kibble was to add a second phase to the experiment in
which the electric current is switched off and the coil made to move vertically through
the magnetic field with a velocity v by tilting the arm of the balance. In this second
phase, the coil moves through the magnetic field so that a voltage U is induced across its
terminals. The relation between the velocity v and voltage is simply U = BLv. Provided
that the field and coil remain exactly the same in the two phases these two equations
can be combined to eliminate B and L to arrive at the simple equation
(1) mgv = IU.
One can now see how the quantum-Hall effect, which gives an electrical resistance
proportional to h/e2 in conjunction with the Josephson effect, which gives a voltage pro-
portional to 2e/h, allows us to measure the product IU in terms of these two fundamental
constants. In terms of voltage and resistance our equation thus becomes
(2) mgv = U 2 /R = Ch/4,
where C is a constant that includes the frequency of the microwave radiation needed to
make the Josephson effect work and some integers related to the details of the operation of
these two quantum effects. The frequency can of course me measured with an uncertainty
of parts in 1011 quite easily and the integers are exact. Thus by a measurement of
frequency, velocity and the acceleration due to gravity, which can be measured to parts
in 109 , we have a link between mass and the Planck constant.
At the level of parts in 108 the watt balance is not a straightforward device as it
relies on the coil always being aligned with respect to the gravitational vertical and there
being an exact equality between the magnitude and orientation of the field during the
two phases of the experiment with the coil moving exactly vertically. These conditions
are not easy to meet. The link between mass and h also rests upon the quantum-Hall
and Josephson relations being exact at this level.
11. – The x-ray crystal density method
For the x-ray crystal density method using silicon the basic equation linking the
measured quantities is also very simple:
(3) NA = nM (Si)/(ρa3 ),
where n is the number of atoms per units cell, ideally equal to eight if there are no
defects in the crystal, M (Si) is the molar mass of the silicon, ρ is the measured density
of the sample and a is the lattice spacing of the silicon. For a sphere of single crystal
silicon, the principal measured quantities in such an experiment are thus these plus the
volume V of the sphere and its mass m, to arrive at its density m/V . In addition, it
is necessary to characterize the surface and measure the thickness of the surface layer
of silicon dioxide (about 5 nm) and ensure that the crystal is nearly perfect without
too many crystal defects. The measurements of the lattice spacing and volume of the
sphere must be made at temperatures known relative to each other to within about
1 mK because, obviously, the lattice spacing a measured in a separate experiment must
correspond to that in the sphere so that its volume V refers to the same temperature. An
equality of temperature to 1 mK is not easy to obtain when the two measurements are
made in different institutes particularly as the sphere is a solid body of silicon 10 cm in
diameter and the lattice spacing is measured in thin wafers of silicon less than 1 mm thick.
The most difficult parameter turned out to be the molar mass of the sample. This
was done originally by measuring the relative proportions of the three natural isotopes of
silicon, namely silicon 28 (the major component), silicon 29 (5%) and silicon 30 (3%). It
was found impossible to do this with sufficient accuracy. In the end, large single crystals
were eventually obtained made for almost pure silicon 28 such that the uncertainties in
the much smaller proportions of the other two isotopes had a negligible effect on the
molar mass of the whole [45]. A new definition of the kilogram would thus be possible
based on an assigned fixed mass of an atom of silicon.
12. – Comparison of the two methods
Having done both experiments, the key questions were a) do the two experiments lead
to consistent results and b) which constants should we choose to redefine the kilogram?
It turns out that a value for h can obtained both via a watt balance and via the
x-ray crystal density experiment — exactly the experiments we want to compare for our
proposed new definition of the kilogram. At first sight it may not be obvious that a value
of h can be found from the x-ray crystal density experiment that gives a value for the
Avogadro constant, but the two are linked by the following equation:
(4) NA h = [cα2 /2R∞ ][Mu Ar (e)],

where c and Mu , the atomic mass unit, are exact, α the fine structure constant is known
to parts in 109 , R∞ the Rydberg is known to parts in 1012 and Ar (e), the relative atomic
mass of the electron is known to parts in 1010 . The right hand side of this equation is
thus known to parts in 109 . Using eq. (3), the value of h from the x-ray crystal density
method is thus
(5) h(silicon) = [cα2 /2R∞ ][Mu Ar (e)]ρa3 /nM (Si),
to be compared with that from the watt balance:
h(watt balance) = 4mgv/C.
242 T. J. Quinn
An essential prerequisite for any new definition of the kilogram based either on h
or the mass of an atom through NA , is that the values of h obtained via watt balance
experiments and x-ray crystal density experiments agree. This is now the case to within
a few parts in 108 [46-48].
Coming back to the question of how to formulate a new definition of the kilogram, we
can see that each of the two experiments could be the basis of the new definition. The
x-ray crystal density method of determining NA allows us to determine the number of
atoms in a given mass of silicon, it is obviously also a way of defining the unit of mass
as simply a specified number of atoms of silicon. The watt balance experiment allows us
to determine mass in terms of the Planck constant.
The question of which path to choose was much discussed and strong views were
expressed in favour of a definition based on the mass of an atom, since such a def-
inition would be much easier to understand. The disadvantage of such a definition,
however, is that by so doing the Planck constant reverts to being a constant calcu-
lated from combinations of other constants. In equation (2) for example the unknown
would no longer be m but h. The deciding argument in favour of a definition based
on h was related to the future accuracy of the electrical units, the volt and the ohm,
already realized in practice through the Josephson and quantum-Hall effects. In defining
the kilogram by fixing h and e rather than the mass of an atom, the Josephson and
von Klitzing constants would take exact values and hence there would no longer be a
limit to the future accuracy of realization of electrical units. Since such realizations
are already at the level of parts in 1011 , indeed recent comparisons of the quantum-
Hall effect in graphene and the traditional Ga-As showed agreement to about 8 parts
in 1011 [49]. If the mass of an atom definition were to be chosen, the accuracy of
electrical units would always be limited by the accuracy of the values of h and e calcu-
lated via other constants. One of the cornerstones of the new SI is intended to be that
by the way it is defined no limits will be set on the accuracy of future realization of
SI units.
Finally in this section, what does this agreement between the values of the Planck
constant obtained by the watt balance and x-ray crystal density methods tell us? The
most important outcome is the demonstration that the Josephson and quantum-Hall re-
lations correctly represent macroscopic voltages and resistances — something that had
not been demonstrated at this level of accuracy before. It is also demonstrates a remark-
able level of consistency among the measured values of fundamental constants using a
wide variety of methods based on an equally wide variety of equations of physics. One
can further conclude that classical and quantum physics in this area are consistent to a
few parts in 108 . It also tells us that we can use with confidence the value of the Planck
constant thus determined to define our unit of mass, the kilogram, hence the choice of h
to define the kilogram.
This whole project, which has been underway for some 20 years and included many
National Metrology Institutes, is a remarkable demonstration not only of high-level
metrology, but also of the soundness of our understanding of large areas of physics [50].
13. – Present plans for a fully constants-based SI for 2018

.
13 1. What does it mean to fix the numerical value of a constant of nature? . – The
24th General Conference on Weights and Measures in 2011 adopted a proposal from the
International Committee for Weights and Measures for a revision of the SI that would
be based on a set of fixed numerical values of seven defining constants.
One might reasonably ask how is it possible to fix the numerical value of a constant of
physics, such as the speed of light or Planck constant when their magnitude is evidently
a constant of nature and not for us to fix? The answer is that the numerical value we
attribute to any constant is a function of the size of the unit in which we choose to
measure it. For example, we know that the speed of light is expressed in the following
way:
c = numerical value × unit,
but we can express it in many different units, for example:
c = 299 792 458 metres per second,

or c = 983 571 056.4 feet per second.
But, since the value of c is a constant of nature:

a) If we define the units of length and time independently, then we must determine
the numerical value of c by experiment, and it will have an uncertainty. That was the
situation before 1983, when both the metre and the second were independently defined.
b) If, however, the second is independently defined in terms of the frequency of the
caesium transition, and we choose to fix a numerical value of c, then the effect is to
define the size of the unit, equal to 299 792 458 in the case of the metre. This is the
current definition of the metre, since the change in 1983. The numerical value now has
zero uncertainty.
In the case of the Planck constant which is a little more complex we can write:
h = numerical value × unit,

h = 6.626 0703 × 10−34 kg m2 s−1 .
But, since the value of h is a constant of nature:

a) If we define the units of mass, length and time independently, taking the present
definitions of the kilogram, metre and second, then we must determine the numerical
value of h by experiment, and it will have an uncertainty. This is the present situa-
tion, and is what we are doing at present with watt balances and x-ray crystal density
experiments.
b) If, however, we first define the units of length and time independently and then we
choose to fix the numerical value of h, then the effect is to fix the size of the kilogram.
This is the proposed new definition of the kilogram. The numerical value will have zero
244 T. J. Quinn
uncertainty. We just have to make sure we choose the right value of h i.e., one that
is really consistent with the present definition of the kilogram so that there will be no
step change in the value of the kilogram when the new definition is adopted. This same
remark applies to the numerical values of the other defining constants in the same way.
.
13 2. The form of the proposed new definition of the SI and its base units. – The key
words of the Resolution proposed to the 26th CGPM in 2018 are:
“[. . . ]that henceforth the International System of Units, the SI, is the system of units
in which:
– the unperturbed ground state hyperfine splitting frequency of the caesium 133
atom, ΔνCs , is 9 192 631 770 Hz,
– the speed of light in vacuum c is 299 792 458 m/s,
– the Planck constant h is 6.626 069 XX ×10−34 J s,
– the elementary charge e is 1.602 176 XX ×10−19 C,
– the Boltzmann constant k is 1.380 648 X ×10−23 J/K,
– the Avogadro constant NA is 6.022 141 XX ×1023 mol−1 ,
– the luminous efficacy Kcd of monochromatic radiation of frequency 540 ×1012 hertz
is 683 lm/W.”
In this text the experiments leading to the final values of some of the constants have
not been completed hence some of the numerical values terminate with XX, digits yet
to be finalized. At the time of writing this article agreement has been reached but final
approval by the CODATA Task Group will not be given until mid-2017 in order to take
into account the results of new watt balance and silicon density experiments that may
become available. However, this is the form of the proposal being prepared for the 26th
General Conference due to meet in November 2018.
The effect of setting the seven defining constants in this way is to fix the scale or size
of the units of the SI but it does not explicitly give the definitions of the seven base units.
Thus following this set of defining constants there will be the individual definitions of the
set of seven base units. These will be the same units as today, namely, the second, metre,
kilogram, ampere, kelvin, mole and candela but defined in terms of the fixed numerical
values of these constants.
The base units of the SI remain of course those that correspond to the base quantities
of the International System of Quantities adopted by ISO/IEC and which form the basis
of most of the equations of physics. Changing the definitions of the base units does not
of course change in any way the set of seven base quantities and dimensions which are
conventionally considered to be independent of each other.
To illustrate the way in which the new definitions will be formulated, the following
is the draft of the proposed new definition of the kilogram at the time of writing this
article:
– The kilogram, symbol kg, is the SI unit of mass. It is defined by taking the fixed
numerical value of the Planck constant h to be 6.626 070 040xx ×10−34 when
expressed in the unit J s, which is equal to kg m2 s−1 , where the metre and the
second are defined in terms of c and ΔνCs .
Where c and ΔνCs are defined in the list of defining constants given above.
Each definition will call upon the constants relevant to the units in question. The
final form of the definitions will be adopted by the International Committee for Weights
and Measures in October 2017. Documents, including draft definitions will be available
on the BIPM website. Once again, for the kilogram, the precepts of 1791 will be met in
that its definition “will be taken from nature” and no one “would not be obliged to send
observers to the place where it was originally established”. This is a significant change.
.
13 3. Advantages of units defined in terms of constants of nature: explicit constant
definitions rather than an explicit unit definitions. – Traditionally, a unit for a given
quantity was taken to be a particular example of that quantity chosen to be of a con-
venient size. Definitions of this sort are referred to as explicit unit definitions. The
definition always refers to the quantity of which the unit is an example. For instance:
“The metre is the length . . . ”. The kilogram is the mass. . . ” and so on. The prac-
tical realization of units defined in this way requires experiments directly or indirectly
linked to the particular conditions or states specified in the definitions. In consequence,
the accuracy of realization of such definitions can never be better than the accuracy of
realization of the particular conditions or states specified in the definitions.
The advantage of such definitions is that they are easy to understand but the disad-
vantage is that the accuracy or realization as one moves away from the specified state
or condition necessarily falls away. An evident example of this is in the measurement of
mass. The highest accuracy is at masses close to 1 kilogram, at masses near 1 milligram
or 500 kilograms the accuracy is much lower. With the new definition no such limitation
exists. This is the case also for other units.
In the new formulation of the SI the basis of the definition is quite different. We begin
with a set of seven defining constants, chosen so that each one has a unit or combination
of units such that taken together they cover the whole range of the seven base units of
the SI. There is no one-to-one correspondence between the defining constants and base
units, except for that of the second, as in most cases more than one constant is called
upon for a definition. The metre, for example, calls upon the speed of light and the
caesium frequency and the kilogram calls upon these two plus the Planck constant. Such
definitions are known as explicit constant definitions.
The practical realization and the making of measurements is also quite different.
Instead of each definition specifying a particular condition or state, which as we have
seen sets a fundamental limit to the accuracy of realization, it is now open to us to choose
any convenient equation of physics that links the particular constant or constants to the
quantity we want to measure. This is a much more general way of defining the basic
units of measurement. It is one that is not limited by today’s science or technology as
future developments may lead to at present unknown equations that could result in quite
246 T. J. Quinn
different ways of realizing units to much higher accuracy. Defined in this way, there is,
in principle, no practical limit to the accuracy with which a unit can be realized. This is
the overriding advantage of such explicit constant definitions. The exception remains the
definition of the second in which the original microwave transition of caesium remains,
for the time being, the basis of the definition.
For the kilogram, the unit whose definition has undergone the most fundamental
change, realization can be through any equation of physics that links mass, the Planck
constant, the velocity of light and the caesium frequency. As we have seen, two such
realizations have been used in the work leading up to the redefinition, the Kibble balance
and the x-ray crystal density of silicon. But there may be others. Another possibility
for measuring mass through the new definition, but this time at the microscopic level, is
through measurements of atomic recoil using the relation that includes h/m. All these
provide a striking illustration of the generality of the new way of defining units. It must
be admitted however that explicit constant definitions are more subtle and some, notably
the kilogram, are more difficult to understand
The experimental methods used for the realization of units using the equations of
physics as illustrated above are known as primary methods. The essential characteristic of
a primary method is that it allows a quantity to be measured directly from the definition
of its unit without prior knowledge of that unit by using only constants that themselves do
not contain that unit. The watt balance and silicon crystal density methods are examples
of primary method for the measurement of mass. An example for temperature is the
acoustic thermometer in which the temperature of a gas is obtained from measurements
of the speed of sound, using an equation that includes only the speed of sound, the molar
mass of the gas, the Boltzmann constant and the ratio of the specific heats of the gas.
The new definitions are thus more general, less likely to be overtaken by new science
and do not in themselves limit the accuracy with which they can now or in the future be
realized.
Bibliography
The two principal sources for the content of this articles are:
1. Quinn T. J. From Artefacts to Atoms, the BIPM and the Search for Ultimate
Measurement Standards (Oxford University Press, New York) 2011, 436 pp.
2. The BIPM website www.bipm.org on which can be found the principal official
documents mentioned in this article, namely, the proceedings of the Diplomatic
Conference on the Metre, 1875, the Reports of successive General Conferences on
Weights and Measures from the 1st in 1889 to the most recent, the 25th in 2014 and
Reports of the meetings of the International Committee for Weights and Measures
from 1875 to the present day as well as a wealth of information on the current
activities of the CIPM, its Consultative Committees and the BIPM.
REFERENCES
Note: The question of units of measurement was much discussed in the 19th and early
20th centuries and historians of science have taken a great interest in this subject. I have
therefore given rather more references to the history of science literature than is usual as
it might interest readers not familiar with it.
[1] “On the choice of a unit of measurement” by J. C. Borda, J. L. Lagrange, P. S. Laplace, G.
Monge and J. A. N. Condorcet, 19 March 1791 published in Histoire de l’Académie Royale
des Sciences, Volume de 1788, pp 7-16. Note that the date of the Report postdates that
of the volume year indicating that, as often happened at that time, the volume was put
together some years after.
[2] Tallyrand, (1790) Proposition faites à l’Assemblée National sur les poids et mesures par
M. l’evèque d’Autun, 20 pp, Paris, Imprimerie nationale
[3] Delambre H. B. J., Base du Système métrique décimale, edited by Bigourdan G.
(Gauthier-Villars, Paris) 1912, Vol. I, 1806, Vol. II, 1807 et Vol. III, 1810.
[4] Alder K., The Measure of All Things. The Seven-Year Odyssey and Hidden Error that
Transformed the World (Little, Brown and Company) 2002.
[5] Cox E. F., “The metric system: a quarter-century of acceptance (1851–1876)”, Osiris, 13
(1958) 358.
[6] Wolf M. C., “Recherches Historiques sur les Etalons de Poids et Mesures de l’Observatoire
et les appareils qui ont servi a les construire”, Annales de l’Observatoire de Paris, 17 (1883).
[7] Quinn T. J., The Metre Convention and the creation of the BIPM (article in this volume).
[8] Torge W., “The International Association of Geodesy 1862 to 1922: from a regional
project to an international organization”, J. Geod., 78 (2005) 558.
[9] Geodesy Conference, Procès verbaux de la Conférence Géodésique International pour
la masure des degrés en Europe réunie à Berlin du 30 septembre au 7 octobre 1867
(Imprimerie G. Guillaume fils, Nuechatel) 1867 (French translation by Hirsch of the original
German, part of this English version is from a History of the International Association of
Geodesy by J.J. Levallois).
[10] Metre Commission, Commission international du Mètre, Organisation, 1 Sept 1870 Session
de 1870 (Imprimerie Nationale, Paris) 1870, 8 pp.: Metre Commission, Commission
international du Mètre Comité permanent, Session de 1870, Procès-verbaux des Séances
(Librairie Polytechnique de J. Baudry, Editeur, Paris) 1870, 51 pp.: Metre Commission
Commission international du Mètre, Comité des recherches préparatoires, avril 1872 ;
(Imprimerie Viéville et Capiomont, Paris) 1872, 90 pp.: Metre Commission, Commission
international du Mètre, Réunions générales de1872 (Imprimerie Nationale, Paris) 1872,
231 pp.: Metre Commission Commission international du Mètre Comité permanent,

Procès-verbaux des Séances 1872-1873 (Imprimerie Nationale, Paris) 1873, 40 pp.: Metre
Commission Commission international du Mètre Comité permanent, Procès-verbaux des
Séances 1874 (Imprimerie Nationale, Paris) 1874, 36, pp. 74.
[11] Comptes Rendus des Séances de la Prèmiere Conference Général des Poids et Mesures
Réunie à Paris en 1889 (Gauthier-Villars, Paris) 1889.
[12] Comptes Rendus des Séances de la Onzième Conference Général des Poids et Mesures,
Paris 11-12 Octobre 1960 (Gauthier-Villars, Paris) 1960, Resolutions 6, 7 and 8.
[13] Comité international des poids et mesures, Procès-verbaux des Séances 1891 (Gauthier-
Villars, Paris) 1891, p. 106.
[14] Maxwell J. C., Transactions of the Sections, Notices and Abstracts of miscellaneous
Communications to the Sections, Mathematics and Physics (British Association for the
Advancement of Science, Liverpool) 1870, p. 7.
248 T. J. Quinn
[15] Mahon B., The Man who Changed Everything. The life of James Clerk Maxwell (John
Wiley) 2003, pp. 148, 149.
[16] Schaffer S., “Metrology, metrication, and Victorian values”, in Victorian Science in
Context, Chicago/London edited by B. Lightman (University of Chicago Press) 1997,
pp. 438–474.
[17] Livingston D. M., The Master of Light: A Biography of Albert A. Michelson (Levy,
Harriet Lane) 1996.
[18] Travaux et mémoires du bureau international des poids et mesures, Vol. 11, 1895
Détermination expérimentale de la valeur du mètre en longueur d’ondes lumineuses, by
Michelson 186 pp +51 pp Tables, BIPM.
[19] Travaux et mémoires du bureau international des poids et mesures, Vol. 15, 1913 Nouvelle
détermination du rapport des longueurs d’onde fondamentales avec l’unité métrique by
Benoit, Fabry and Pérot 134 pp.
[20] Comptes Rendus des Séances de la Septieme Conference Général des Poids et Mesures
Réunie à Paris en 1927 (Gauthier-Villars, Paris) 1927, pp. 85-86.
[21] See ref. [12].
[22] Darrigol O., Electrodynamics from Ampère to Einstein (OUP, Oxford) 2000, pp. 49–66.
[23] Mitchell, “Making sense of absolute measurement, James Clerk Maxwell, Fleming Jenkin
and the construction of the dimensional formula”, Studies History Philos. Mod. Phys., in
press.
[24] Roche J., The Mathematics of Measurement (Athlone, London) 2000, chapters 4, 8, 9,
appendix.
[25] Hunt B., “The ohm is where the art is: British telegraph engineers and the development
of electrical standards”, Osiris, 2nd series, 9, (1994) pp. 48.
[26] Kershaw M., “The International Electrical Units: a Failure in Standardisation?”, Studies
History Philos. Sci., 38 (2007) 108.
[27] Lagerstrom L., Constructing Uniformity. The Standardization of International
Electromagnetic Measures, 1860–1912, Unpublished doctoral thesis, University of
California, Berkeley (1992).
[28] Smith C. and Wise N., Energy and Empire. A Biographical Study of Lord Kelvin
(Cambridge University Press, Cambridge) 1989, pp. 649–97, chapter 19, 20.
[29] Everett J. D., “First Report of the Committee for the Selection and Nomenclature
of Dynamical and Electrical Units”, in Report of the Forty-Third Meeting of the British
Association for the Advancement of Science; held at Bradford in September 1873 (AAS,
London) 1874, pp. 222–225.
[30] Masson G. (Editor), Congrès International des Electriciens, Paris 1881, comptes rendus
des travaux (Ministère des Postes et Télégraphes, Paris) 1882.
[31] Chicago Congress (1893), Proceedings of the International Electrical Congress, Chicago
1893 (American Institute of Electrical Engineers, New York) 1894.
[32] CGPM Comptes Rendus des Séances de la Dixième Conférence Général des Poids et
Mesures Réunie à Paris en 1954 (Gauthier-Villars, Paris) 1954, Resolution 3.
[33] See ref. [12].
[34] CGPM Comptes Rendus des Séances de la Quatorzième Conference Général des Poids et
Mesures, Paris 1971 (BIPM, Sèvres) 1971, p. 54.
[35] CGPM Comptes Rendus des Séances de la 16e Conférence Général des Poids et Mesures,
Paris 1979 (BIPM, Sèvres) 1979, Resolutions 3 and 4.
Paris 1983 (BIPM, Sèvres) 1983, Resolutions 1 and 2.
[37] CGPM Comptes Rendus des Séances de la Douzième Conference Général des Poids et
Mesures, Paris 10-16 Octobre 1967, 15 Octobre 1968 (BIPM, Sèvres) 1967, 1968.
[38] See ref. [34].

[39] von Klitzing, “The quantized Hall effect”, Rev. Mod. Phys., 58 (1986) 519.
[40] Josephson B. D., “Possible new effects in superconductive tunnelling”, Phys. Rev. Lett.,
1 (1962) 251.
Paris 1991 (BIPM, Sèvres) 1991, Resolution 2.
[42] Kibble B. P., A measurement of the gyromagnetic ratio of the proton by the string field
method in Atomic masses and Fundamental Constants 5 edited by Sanders J. H. and
Wapstra H. (Plenum, New York) 1976, pp. 545–551.
[43] Bonse U. and Hart M., “An x-ray interferometer”, Appl. Phys. Lett., 6 (1965) 155.
[44] Deslattes R. D. et al., “Determination of the Avogadro constant”, Phys. Rev. Lett., 33
(1974) 463.
[45] Becker P. et al., “Large scale production of highly enriched 28 Si for the precise
determination of the Avogadro constant”, Meas. Sci. Technol., 17 (2006) 1854.
[46] Schlamminger S. et al., “Determination of the Planck constant using a watt balance with
a superconducting magnet system at the National Institute of Standards and Technology”,
Metrologia, 51 (2014) S 15.
[47] Azuma Y. et al., “Improved measurement results for the Avogadro constant using 28Si-
enriched crystals”, Metrologia, 52 (2015) 360.
[48] Sanchez C. A. et al., “A determination of the Planck constant using the NRC watt
balance”, Metrologia, 51 (2014) S5.
[49] Janssen T. J. B. M. et al., “Precision comparison of the quantum-Hall effect in graphene
and gallium arsenide”, Metrologia, 49 (2012) 294.
[50] Becker P., “History and progress in the accurate determination of the Avogadro
constant”, Rep. Prog. Phys., 64 (2001) 1945.
DOI 10.3254/978-1-61499-818-1-251
Frequency combs applications and optical frequency

standards
Th. Udem
Max-Planck Institute für Quantenoptik
Hans-Kopfermann Straße 1, 85748 Garching, Germany
F. Riehle
Physikalisch-Technische-Bundesanstalt
Bundesallee 100, 38116 Braunschweig, Germany
Summary. — My presentation on “Frequency Combs” was intended to be a tutorial

of the basic technology without an extended reference to the latest developments
in this field. I repeated parts of my contributions to the course CLXVI “Metrology
and Fundamental Constants” that was held at the same school in 2006. Since I did
not cover new material, the proceedings of 2006 is reproduced here.
A laser frequency comb allows the conversion of the very rapid oscillations of visible
light of some 100’s of THz down to frequencies that can be handled with conventional
electronics, say below 100 GHz. This capability has enabled the most precise laser spec-
troscopy experiments yet that allowed to test quantum electrodynamics, to determine
fundamental constants and to search for possible slow changes of these constants. Using
an optical frequency reference in combination with a laser frequency comb has made it
possible to construct all optical atomic clocks, that are about to outperform the current
cesium atomic clocks.
251
252 Th. Udem and F. Riehle
1. – Introduction
If one considers the attainable measurement accuracy of different physical quantities

it turns out that time intervals and frequencies are to be determined with the utmost
precision. Other physical dimensions, such as length mass or charge, can only be de-
termined with orders of magnitude less accuracy. The intrinsic high precision comes
about because counting, unlike any other physical measurement, has zero uncertainty
connected to it. The only uncertainty in determining a frequency in hertz or oscillations
per second, lies in the determination of the second. But this is about as good as it can
be, because atomic clocks that are used to determine the second are the most precise
instruments. In this sense frequency and time measurements are equivalent.
To exploit this potential it has been a top priority in metrology to convert other
physical measurables into a time or frequency equivalent. The simplest example of such
a conversion is to assign the speed of light c with a fixed defined value, which was done
within the International System of Units (SI) in 1983. Since then the conversion of an
optical wavelength λ to an optical frequency ω = 2πc/λ can be done without loss of
accuracy(1 ). The method requires a means to count optical frequencies, because only
then one can use a precise interferometer to extract the wavelength. Therefore it is no
surprise hat this redefinition had to wait until it became possible to count the frequency of
light. At that time the idea was to calibrate iodine stabilized HeNe lasers with harmonic
frequency chains that linked them to a cesium atomic clock. These frequency chains
where complex devices that got operational only in few places and worked continuously
only for short time intervals [1].
The optical frequency comb vastly simplified these efforts [2-8]. Even commercially
available cesium clocks are usually more accurate than an iodine stabilized HeNe laser.
Therefore these lasers are no longer required if a more accurate radio frequency (RF)
source and a frequency comb is available. On the other hand, another class of optical
standards based on trapped ions or atoms have been improving at a faster pace than
the cesium clocks so that the frequency comb can be used to calibrate a RF source. As
will be discussed in sect. 5 this led to the first all optical atomic clocks [9-11]. After
agreeing on a particular ion or atom with a suitable clock transition, it seems likely that
the definition of the SI second will be adjusted accordingly.

By definition an optical frequency comb consists of many continuous wave laser modes,
equidistant in frequency space, that can be used like ruler to determine large frequency
differences between lasers. By measuring the frequency separation between a laser at a
frequency f and its second harmonic 2f , the lasers absolute frequency f = 2f − f is
determined [12]. A frequency comb used for that purpose must span a complete optical
octave. Then, not only f and 2f are known relative to the comb mode spacing, but all
the modes in between, providing an octave full of calibrated laser lines at once.
(1 ) A second example is the utilization of the Josephson of effect to convert voltages into
frequencies and vice versa.
Frequency combs applications and optical frequency standards 253
Since light from lasers has been used to gather nearly all high-precision data about
atoms, the analysis of this light is key to a better understanding of the microscopic world.
To test quantum electrodynamics and to determine the Rydberg constant, transition
frequencies in atomic hydrogen have been measured [13, 14]. In fact the spectroscopy of
the narrowest line in hydrogen, the 1S-2S transition, was the motivation for setting up the
first frequency comb [2-4]. From these measurements the Rydberg constant became the
most accurately measured fundamental constant [14]. The frequency comb also helped
to determined the fine-structure constant from atomic recoil shifts [15, 16] using precise
values of the Rydberg constant and was the key for laboratory searches for possible slow
variations of these constants [17-19]. Even if there is not yet a physical theory that would
predict such a slow change there are some arguments, rather philosophically in nature,
that they should be there. The frequency combs are now providing a sharp tool for at
least setting up stringent upper limits.
2. – Frequency combs from mode locked lasers
Frequency combs can be produced with fast and efficient electro-optik modulators that
impose a large number of side bands on a single-frequency continuous laser [20]. The
factor that limited the achievable width of the generated frequency comb was dispersion
of the modulator crystal. After dispersion compensation was introduced [21], boosting
the intensity with external optical amplifier allowed spectral broadening up to 30 THz
bandwidth with the process of self-phase modulation [22].
Even more bandwidth can be generated with a device that already included dispersion
compensation, gain and self-phase modulation: the Kerr lens mode-locked laser. Such a
laser stabilizes the relative phase of many longitudinal cavity modes such that a solitary
short pulse is formed. In the time domain this pulse propagates with its group velocity vg
back and forth between the end mirrors of the resonator. After each round trip a copy of
the pulse is obtained at one of these mirrors that is partially transparent like in any other
conventional laser. Because of its periodicity, the pulse train generated this way produces
a spectrum that consists of discrete modes that can be identified with the longitudinal
cavity modes. The process of Kerr lense mode locking (KLM) introduced in the early
90s [23], allows to generate pulses of 10 femtoseconds (fs) duration and below in a rather
simple way. For Fourier-limited pulses the bandwidth is given by the inverse of the pulse
duration, with a correction factor of order unity that depends on the pulse shape and
the way spectral and temporal widths are measured [24]. A 10 fs for example will have
a bandwidth of about 100 THz, which is close to the optical carrier frequency, typically
at 375 THz (800 nm) for a laser operated near the gain maximum of titanium-sapphire,
the most commonly used gain medium in these lasers. By virtue of the repeating pulses,
this broad spectrum forms the envelope of a frequency comb. Depending on the length
of the laser cavity, that determines the pulse round trip time, the pulse repetition rate
ωr is typically on the order of 100 MHz but 4 MHz through 2 GHz repetition rates have
been used. In any case ωr is a radio frequency readily measured and stabilized. The
usefulness of this comb critically depends on how constant the mode spacing is across
the spectrum and to what precision it agrees with the readily measurable repetition rate.
These questions will be addressed in the next two section.
.
2 1. Derivation from cavity boundary conditions. – As in any laser the modes with
wave number k(ωn ) and frequencies ωn must obey the following boundary conditions of
the resonator of length L:
(1) 2Lk(ωn ) = n2π.
Here n is a large integer number that measures the number of half wavelengths within the
resonator. Besides this propagation phase shift there might be additional phase shifts,
caused for example by diffraction (Guoy phase) or by the mirror coatings that are thought
of being included in the above dispersion relation. This phase shift may depend on the
wavelength of the mode, i.e. on the mode number n, and is not simple to determine
accurately in practice. So we are seeking a description that lumps all the low-accuracy
quantities into readily measurable radio frequencies. For this purpose dispersion is best
be included by the following expansion of the wave vector about some mean frequency
ωm , not necessarily a cavity mode according to eq. (1):
k (ωm )
(2) 2L k(ωm ) + k (ωm )(ωn − ωm ) + (ωn − ωm )2 + . . . = 2πn.
2
The mode separation Δω ≡ ωn+1 − ωn is obtained by subtracting this formula from itself
with n being replaced by n + 1:
k (ωm )
(3) 2L k (ωm )Δω + (ωn+1 − ωm )2 − (ωn − ωm )2 + . . . = 2π.
2
To obtain a constant mode spacing, as the most important requirement for optical fre-
quency combs, Δω must be independent of n. This is the case if and only if all con-
tributions of the expansion of k(ω) beyond the group velocity term k (ωm ) = 1/v g (ωm )
exactly vanish. The unwanted perturbing terms, or “higher-order dispersion” terms, that
contradict a constant mode spacing are those that deform the pulse as it travels inside the
laser resonator. Therefore the mere observation of a stable undeformed pulse envelope
stored in the laser cavity leads to a frequency comb with constant mode spacing given
by
vg 1
(4) Δω = 2π with v g =
2L k (ωm )
with the round trip averaged group velocity(2 ) given by v g . Having all derivatives beyond
k (ωm ) vanishing, means that k (ωm ) and v g must independent of frequency. Therefore
this derivation is independent of the particular choice of ωm , provided it resides within
the laser spectrum. The group velocity determines the cavity round trip time T of the
pulse and therefore the pulse repetition rate ωr = Δω:
vg ωr
(5) T −1 = = .
2L 2π
The frequencies of the modes ωn of any frequency comb with a constant mode spacing
ωr can be expressed by [3, 8, 25, 26]
(6) ωn = nωr + ωCE
with a yet unknown frequency offset ωCE common to all modes. As a convention we
will now number the modes such that 0 ≤ ωCE ≤ ωr . This means that ωCE , like ωr
resides in the radio frequency domain. Using (6) to measure the optical frequencies ωn
requires the measurement of ωr , ωCE and n. The pulse repetition rate can be measured
anywhere in the beam of the mode-locked laser. To determine the comb offset requires
some more effort as will be detailed in sect. 3. In practice the beam of a continuous wave
laser whose frequency is should be determined is superimposed with the beam containing
the frequency comb on a photo detector to record a beat note with the nearest comb
mode. Knowing ωr and ωCE the only thing missing is the mode number n. This may be
determined by a coarse and simple wavelength measurement or by repeating the same
measurement with slightly different repetition rates [27].
Some insight on the nature of the frequency offset ωCE is obtained by resolving (6) for
ωCE and using the cavity averaged phase velocity of the n-th mode v p (ωn ) = ωn /k(ωn ).
With the expansion of the wave vector in (2) and the fact that that it ends with the
group velocity term, one derives

vg
(7) ωCE = ωn − nωr = ωm 1 −
v p (ωm )
For this the frequency offset is independent of n, i.e. common to all modes, and vanishes
if the cavity averaged group and phase velocities are identical. In such a case the pulse
train possesses a strictly periodic field that produces a frequency comb containing only
integer multiples of ωr . In general though this condition is not fulfilled and the comb
offset frequency is related to the difference of the group and phase round trip time. For
an unchirped pulse, that has a well-defined carrier frequency ωc , it makes sense to expand
(2 ) The usual textbook approach ignores dispersion either as a whole (Δω = 2πc/2L) or just
includes a constant refractive index: Δω = 2πvph /2L with vph and c being the phase velocities
with and without dispersive material, respectively.
the wave vector about ωm = ωc so that eq. (7) becomes

2L 2L
(8) ωCE T = Δϕ with Δϕ = ωc − .
vg vp
As discussed in more detail in the next section, it follows that the pulse envelope contin-
uously shifts relative to the carrier wave in one direction. The shift per pulse round trip
Δϕ is given by the advance of the carrier phase during a phase round trip time 2L/v p in
a frame that travels with the pulse. The shift Δϕ per pulse round trip time T = 2L/v g
fixes the frequency comb offset. Hence it has been dubbed carrier-envelope (CE) offset
frequency.
How precise the condition of vanishing higher-order dispersion is fulfilled in practice
could be estimated by knowing that an irregular phase variations between the modes on
the order of 2π are sufficient to completely destroy the stored pulse in the time domain.
The phase between adjacent modes of a proper frequency comb advances as ωr t, i.e.
typically by some 108 times 2π a second. An extra random cycle per second would
offset the modes by only 1 Hz from the perfectly regular grid, but destroy the pulse in
the same time. Compared to the optical carrier frequency this corresponds to a relative
uncertainty of 3 parts in 1015 at most, which is already close to the best cesium atomic
clocks. Experimentally one observes the same pulse for a much longer time, in some
lasers even for months. In fact no deviations have been detected yet at a sensitivity of a
few parts in 1016 [28, 7, 29, 30].
It should be noted that a more appropriate derivation of the frequency comb must
include non-linear shifts of the group and phase velocities because this is used to lock
the modes in the first place. In fact it has been shown that initiating the mode-locking
mechanism significantly shifts the cavity modes of a laser [31]. Of course the question
remains how this mode-locking process forces the cancellation of all pulse deforming
dispersive contributions with this precision. Even though this is beyond the scope of this
article and details may be found elsewhere [32, 24], the simplest explanation is given in
the time domain: Mode locking, i.e. synchronizing the longitudinal cavity modes to sum
up for a short pulse, is most commonly achieved with the help of the Kerr effect which
is expressed in the time domain through
(9) n(t) = n0 + n2 I(t).
Here n2 denotes a small intensity I(t) dependence of the refractive index n(t). Generally
a Kerr coefficient with a fs response time is very small, so that it becomes noticeable only
for the high peak intensity of short pulses. Kerr lens mode locking uses this effect in two
ways. The radially intensity variation of a Gaussian mode produces a lens that becomes
part of the laser resonator for a short pulse. This resonator is designed such that it has
larger losses without such a lens, i.e. for a superposition of randomly phased modes. In
addition self phase modulation can compensate the some de-phasing of the modes. A
wave packet that does not deform as it travels because higher-order dispersive terms are
compensated by self phase modulation, is called a soliton [33,34]. The nice feature about
this cancellation is that it is self-adjusting: Slightly larger peak intensity extends the
pulse duration and reduces the peak intensity leaving the pulse energy constant and vice
versa. So that neither the pulse intensity nor the dispersion has to be matched exactly
(which would not be possible) to generate a soliton. Of course it helps to pre-compensate
higher-order dispersion as good as possible before initiating mode locking. It appears
that for the shortest pulses the emission bandwidth is basically given by the bandwidth
this pre-compensation is achieved [35, 36]. Once mode locking is initiated, mostly by
a mechanical disturbance of the laser cavity, the modes are pulled on the regular grid
described by eq. (6). Another way of thinking about this process is that self phase
modulation that occurs with the repetition frequency, produces side bands on each mode
that will injection lock the neighboring modes by mode pulling. This pulling produces
the regular grid of modes and is limited by the injection locking range [37], i.e. how well
the pre-compensation of higher-order dispersion was done.
.
2 2. Derivation from the pulse train. – Rather than considering intracavity dispersion,
the cavity length and so on, one may simply analyze the emitted pulse train with little
consideration how it is generated. For this one assumes that the electric field E(t),
measured for example at the output coupling mirror, can be written as the product of a
periodic envelope function A(t) and a carrier wave C(t):
(10) E(t) = A(t)C(t) + c.c.
The envelope function defines the pulse repetition time T = 2π/ωr by demanding A(t) =
A(t − T ). The only thing about dispersion that should be added for this description, is
that there might be a difference between the group velocity and the phase velocity inside
the laser cavity. This will shift the carrier with respect to the envelope by a certain
amount after each round trip. The electric field is therefore in general not periodic with
T . To obtain the spectrum of E(t) the Fourier integral has to be calculated:
+∞
(11) Ẽ(ω) = E(t)eiωt dt.

−∞
Separate Fourier transforms of A(t) and C(t) are given by

+∞ +∞
(12) Ã(ω) = δ (ω − nωr ) Ãn and C̃(ω) = C(t)eiωt dt.
n=−∞ −∞
A periodic frequency chirp imposed on the pulses is accounted for by allowing a complex
envelope function A(t). Thus the “carrier” C(t) is defined to be whatever part of the
electric field that is non-periodic with T . The convolution theorem allows us to calculate
A(1)
convolution with A(1)

1
C(1) E (1 )
'1c
1c 1 1c - 1r 1c 1c + 1r 1
Fig. 1. – The spectral shape of the carrier function (left), assumed to be narrower than the
pulse repetition frequency (Δωc ωr ), and the resulting spectrum according to eq. (13) after
modulation by the envelope function (right).
the Fourier transform of E(t) from Ã(ω) and C̃(ω):

1
+∞ +∞
1
(13) Ẽ(ω) = Ã(ω )C̃(ω − ω )dω + c.c. = Ãn C̃ (ω − nωr ) + c.c.
2π −∞ 2π n=−∞
The sum represents a periodic spectrum in frequency space. If the spectral width of the
carrier wave Δωc is much smaller than the mode separation ωr , it represents a regularly
spaced comb of laser modes just like eq. (6), with identical spectral line shapes, namely
the line shape of C̃(ω) (see fig. 1). If C̃(ω) is centered at say ωc , than the comb is shifted
from containing only exact harmonics of ωr by ωc . The center frequencies of the mode
members are calculated from the mode number n [3, 8, 25, 26]:
(14) ωn = nωr + ωc .
The measurement of the frequency offset ωc [2-8]. as described below usually yields a
value modulo ωr , so that renumbering the modes will restrict the offset frequency to
smaller values than the repetition frequency and again yields eq. (6).
The individual modes can be separated with a suitable spectrometer if the spectral
width of the carrier function is narrower than the mode separation: Δωc ωr . This
condition is easy to satisfy, even with a free-running titanium-sapphire laser. If a single
mode is selected from the frequency comb, one obtains a continuous wave. However it
is easy to show that a grating with sufficient resolution would be at least as large as
the laser cavity, which appears unrealistic for a typical laser with a 2 m cavity length.
Fortunately for experiments performed so far it has never been necessary to resolve a
single mode in the optical domain as explained in sect. 3.
Now let us consider two instructive examples of possible carrier functions. If
the carrier wave is monochromatic C(t) = e−iωc t−iϕ , its spectrum will be δ-shaped
E (t) '7 2'7
t
Fourier
transformation
E (1)
1CE
1r 1c 1
Fig. 2. – Consecutive un-chirped pulses (A(t) real) with carrier frequency ωc and the correspond-
ing spectrum (not to scale). Because the carrier propagates with a different velocity within the
laser cavity than the envelope (phase- and group velocity), the electric field does not repeat
itself after one round trip. A pulse-to-pulse phase shift Δϕ results in an offset frequency of
ωCE = Δϕ/T . The mode spacing is given by the repetition rate ωr . The width of the spectral
envelope is given by the inverse pulse duration up to a factor order unity that depends on the
pulse shape (the time bandwidth product of a Gaussian pulse for example is 0.441 [24]).
and centered at the carrier frequency ωc . The individual modes are also δ-functions
C̃(ω) = δ(ω − ωc )e−iϕ . The frequency offset (14) is identified with the carrier frequency.
According to eq. (10) each round trip will shift the carrier wave with respect to the
envelope by Δϕ = arg(C(t − T )) − arg(C(t)) = ωc T so that the frequency offset is given
by ωCE = Δϕ/T [3,8,25,26]. In a typical laser cavity this pulse-to-pulse carrier-envelope
phase shift is much larger than 2π, but measurements usually yield a value modulo 2π.
The restriction 0 ≤ Δϕ ≤ 2π is synonymous with the restriction 0 ≤ ωCE ≤ ωr introduced
earlier. Figure 2 sketches this situation in the time domain for a chirp free-pulse train.
As a second example consider a train of half-cycle pulses like
t−kT 2
e−( τ ) .
(15) E(t) = E0
k
In this case the electric field would be repetitive with the round trip time. Therefore
C(t) is a constant and its Fourier transform is a delta-function centered as ωc = 0. If it
becomes possible to build a laser able to produce a stable pulse train of that kind, all the
comb frequencies would become exact harmonics of the pulse repetition rate. Obviously,
this would be an ideal situation for optical frequency metrology(3 ).
(3 ) It should be noted though that this is a rather academic example because such a pulse would
be deformed quickly upon propagation since it contains vastly distinct frequency components
with different diffraction. For example, the DC component that this carrier certainly has, would
not propagate at all.
As these examples are instructive it is important to note that one neither relies on
assuming a strictly periodic electric field nor that the pulses are unchirped. The strict
periodicity of the spectrum as stated in eq. (13), and the possibility to generate beat
notes between continuous lasers and single modes [38], are the only requirement that
enables precise optical to radio frequency conversions.
In a real laser the carrier wave will not be a clean sine wave as in the above example.
The mere periodicity of the field, allowing a pulse-to-pulse carrier envelope phase shift,
already guarantees the comb-like spectrum. Very few effects can disturb that property.
In particular, for an operational frequency comb, both ωr and ωCE will be servo controlled
so that slow drifts are compensated. The property that the comb method really relies
on, is the mode spacing being constant across the spectrum. As explained above, even
a small deviation from this condition will have very quick and devastating effects on the
pulse envelope. Not even an indefinitely increasing chirp could disturb the mode spacing
constancy, as this can be seen as a constantly drifting carrier frequency that does not
perturb the spectral periodicity but shifts the comb as a whole. However, the phase
of individual modes can fluctuate about an average value required for staying in lock
with the rest of the comb. This will cause noise that can broaden individual modes as
discussed in the next section.
.
2 3. Linewidth of a single mode. – The modes of a frequency comb have to be under-
stood as continuous laser modes. As such they possess a linewidth which is of interest
here. Of course as usual in such a case, several limiting factors are effective at the same
time. It is instructive to derive the Fourier limited linewidth that is due to observing the
pulse train for a limited number of pulses only. Following a derivation by Siegman [37](4 )
the linewidth of a train of N pulses can be derived. In accordance with the previous sec-
tion we assume that the pulse train consists of identical pulses E(t) separated in time by
T and subjected to a pulse-to-pulse phase shift of eiΔϕ :
N −1
E0 imΔϕ
(16) E(t) = √ e E(t − mT ).
N m=0
For decent pulse shapes, that fall off at least as ∝ 1/t from the maximum, this series
converges even as N goes to infinity. Using the shift theorem
(17) F T {E(t − τ )} = e−iωτ F T {E(t)}
one can relate the Fourier transform of the pulse train E(t) to the Fourier transform of
a single pulse Ẽ(ω). With the sum formula for the geometric series this becomes
N −1
E0 Ẽ(ω) −im(ωT −Δϕ) E0 Ẽ(ω) 1 − e−iN (ωT −Δϕ)
(18) Ẽ(ω) = √ e = √ .
N m=0 N 1 − e−i(ωT −Δϕ)
(4 ) The carrier envelope phase shift was ignored in [37] but can easily be accounted for here.
I(1)
IN (1)
(1-cos(Nx))/(1-cos(x))
x 1
1n
3
n 2S
1 ) 2S
2)2S
3)2S
1n+
1n+
1n+
(n+
(n+
(n+
Fig. 3. – Left: The function (1 − cos(N x))/(1 − cos(x)) normalized to the peak for N = 2, 3 and
10 pulses. The maxima are at x = ωn T − Δϕ = 2πn with integer n. From that and redefining
ωr = 2π/T and ωCE = Δϕ/T we derive again the frequency comb equation (6). Right: The
resulting frequency comb spectrum is given by the spectrum of a single pulse multiplied with
the comb function at the left-hand side.
Now the intensity spectrum for N pulses IN (ω) may be calculated from the spectrum of
a single pulse I(ω) ∝ |Ẽ(ω)|2 :
1 − cos(N (ωT − Δϕ))

(19) IN (ω) = I(ω).
N (1 − cos(ωT − Δϕ))
Figure 3 sketches this result for several N . As the spectrum of single pulse is truly
a continuum, the modes are emerging and becoming sharper as more pulses are added.
This is similar to the diffraction from a grating that becomes sharper as more grating lines
are illuminated. The spectral width
√ of a single mode can now be calculated from (19) and
may be approximated by Δω ≈ 24/T N for the pulse train observation time N T . From
this√the Fourier limited line width is reduced after one second of observation time N T
to 24/2π Hz ≈ 0.78 Hz and further decreases as the inverse observation time. In the
limit of an infinite number of pulses pulse the spectral shape of the modes approximate
delta-functions with x = ωT − Δϕ ≈ 2πn
1 1 − cos(N x) 1 1 − cos(N x)
(20) lim ≈ lim = δ(x).
2π N →∞ N (1 − cos(x)) π N →∞ N x2
The whole frequency comb becomes an equidistant array of delta-functions:

(21) IN (ω) → I(ω) δ(ωT − Δϕ − 2πn).
n
The Fourier limit to the linewidth derived here is important when only a number
of pulses can be used for example for direct comb spectroscopy as will be discussed in
.
subsect. 4 5. On the other hand, when a large number of pulses contribute to the signal,
for example when measuring the carrier envelope beat note, other limits enter. In most
of these cases acoustic vibrations seem to set the limit as can be seen by observing
that the noise of the repetition rate of an unstabilized laser dies off very steeply for
frequencies above typical ambient acoustic vibrations around 1 kHz [39]. Even if these
fluctuations are controlled as they can be with the best continuous wave lasers, a limit
set by quantum mechanics in terms of the power-dependent Schawlow-Townes formula
applies. Remarkably the total power of all modes enters this formula to determine the
line width of a single mode [40]. In fact subhertz linewidths have been measured across
the entire frequency comb when it is stabilized appropriately [41, 42].
.
2 4. Generating an octave spanning comb. – The spectral width of a pulse train emitted
by a fs laser can be significantly broadened in a single-mode fiber [34] by self-phase
modulation. According to eq. (9) assuming a single-mode carrier wave, a pulse that has
propagated the length l acquires a self-induced phase shift of
1
(22) ΦNL (t) = −n2 I(t)ωc l/c with I(t) = cε0 |A(t)|2 .
2
For fused silica the non-linear coefficient is comparatively small but almost instantaneous
even on the time scale of fs pulses. This means that different parts of the pulse travel
at different speed. The result is a frequency chirp across the pulse without affecting
its duration. The pulse is no longer at the Fourier limit so that the spectrum is much
broader than the inverse pulse duration where the extra frequencies are determined by
the time derivative of the self-induced phase shift Φ̇NL (t). Self-phase modulation modifies
the envelope function in eq. (10) according to
(23) A(t) −→ A(t)eiΦNL (t) .
Because ΦNL (t) has the same periodicity as A(t) the comb structure of the spectrum is
.
maintained and the derivations of subsect. 2 2 remain valid because periodicity of A(t)
was the only assumption made. An optical fiber is most appropriate for this process
because it can maintain the necessary small focus area over a virtually unlimited length.
In practice, however, other pulse reshaping mechanisms, both linear and non-linear, are
present so that the above explanation is too simple.
Higher-order dispersion is usually limiting the effectiveness of self-phase modulation as
it increases the pulse duration and therefore lowers the peak intensity after a propagation
length of a few mm or cm for fs pulses. On can get a better picture if pulse broadening
due to group velocity dispersion k (ωc ) is included. To measure the relative importance
of the √
two processes, the dispersion length LD (the length that broadens the pulse by a
factor 2) and the non-linear length LNL (the length that corresponds to the peak phase
shift ΦNL (t = 0) = 1) are used [34]:
4 ln(2)τ 2 cAf
(24) LD = LNL = ,
|k (ωc )| n2 ωc P0
where τ0 , Af and P0 = (1/2)Af cε0 |A(t = 0)|2 are the initial pulse duration, the effective
fiber core area and the pulse peak power. In the dispersion dominant regime LD LNL
the pulses will disperse before any significant non-linear interaction can take place. For
LD > LNL spectral broadening could be thought as effectively taking place for a length LD
even though the details are more involved. The total non-linear phase shift can therefore
be approximated by the number of non-linear lengths within one dispersion length. As
this phase shift occurs roughly within one pulse duration τ , the spectral broadening is
estimated to be ΔωNL = LNL /LD τ . As an example consider a silica single mode fiber
(Newport F-SF) with Af = 26 μm2 , k (ωc ) = 281 fs/cm2 and n2 = 3.2 × 10−16 cm2 /W
that is seeded with τ = 73 fs Gaussian pulses (FWHM intensity) at 905 nm with 225 mW
average power and a repetition rate of 76 MHz [3, 4]. In this case the dispersion length
becomes 6.1 cm and the non-linear length 35 mm. The expected spectral broadening of
LNL /LD τ = 2π × 44 THz is indeed very close to the observed value [3].
It turns out that within this model the spectral broadening is independent of the
pulse duration τ because P0 ∝ τ . Therefore using shorter pulses may not be effective
for extending the spectral bandwidth beyond an optical octave as required for simple
self-referencing (see sect. 3). However, very efficient spectral broadening can be obtained
in microstructure fiber(5 ) that can be manufactured with k (ωc ) ≈ 0 around a design
wavelength [43-45]. In this case the pulses are broadened by other processes (linear and
non-linear) than group velocity dispersion as they propagate along the fiber. Eventually
this will also terminate self-phase modulation and the dispersive length has to be replaced
appropriately in the above analysis. At this point a whole set of effects enter such as
Raman and Brillouin scattering, optical wave breaking and modulation instability [34].
Some of these processes even spoil the usefulness of the broadened frequency combs as
the amplify noise.
A microstructure fiber uses an array of submicron-sized air holes that surround the
fiber core and run the length of a silica fiber to obtain a desired effective dispersion. This
can be used to maintain the high peak power over an extended propagation length and
to significantly increase the spectral broadening. With these fibers it became possible
(5 ) Some authors refer to these fibers as photonic crystal fibers that need to be distinguished
form photonic bandgap fibers. The latter use Bragg diffraction to guide the light, while the
fibers discussed here use the traditional index step, with the refractive index determined by the
air filling factor.
-10
-20
power per mode [dBm]

-30
-40
-50
-60
-70
-80
-90
-100
-110
400 500 600 700 800 900 1000 1100 1200 1300
wavelength [nm]
Fig. 4. – Left: SEM image of a the core of a microstructure fiber made at the University of
Bath, UK [43]. The light is guided in the central part but the evanescent part of the wave
penetrates into the air holes that run parallel to the fiber core and lower the effective refractive
index without any doping. The guiding mechanism is the same as in a conventional single
mode fiber. Right: Power per mode on a logarithmic scale (0 dBm = 1 mW). The lighter
30 nm (14 THz–3 dB) wide spectrum displays the laser intensity and the darker octave spanning
spectrum (532 nm through 1064 nm) is observed after the microstructure fiber that was 30 cm
long. The laser was operated at ωr = 2π × 750 MHz (modes not resolved) with 25 fs pulse
duration. An average power of 180 mW was coupled through the microstructure fiber [47].
to broaden low peak power, high repetition rate lasers to beyond one optical octave as
fig. 4 shows.
A variant of the microstructure fibers are regular single-mode fibers that have been
pulled in a flame to form a tapered section of a few cm lengths [46]. When the diameter
of the taper becomes comparable to the core diameter of the microstructure fibers, pretty
much the same properties are observed. In the tapered section the action of the fiber
core is taken over by the whole fiber. The original fiber core then is much too small
to have any influence on the light propagation. The fraction of evanescent field around
the taper and along with it the dispersion characteristics can be adjusted by choosing a
suitable taper diameter.
The peak intensity that can be reached with a given mode-locked laser does not only
depend on the pulse duration but critically on the repetition rate. Because most laser
have pretty much the same average output power, a lower repetition rate concentrates
this available power into fewer pulses per second. Comparing different laser systems the
repetition rates in use cover more than 12 orders of magnitude from the most powerful
lasers (one laser shot per 1000 s) to highly repetitive lasers at ωr = 2π × 2 GHz [48]. It
has been long known that with enough peak intensity one can produce very wide spectra
that where initially called “white light continuum”. Unfortunately for a long time this
was only possible at a repetition rate of around 1 kHz that indeed justifies the name:
The generated spectrum could be called a “continuum” as there was not much hope to
resolve the modes in any way for self-referencing or by a beat note with another cw laser.
Because of its high efficiency the microstructure fiber allowed to generate an octave wide
spectra with repetition rates up to 1 GHz that conveniently allowed the beat notes to be
separated as described in sect. 3. In addition a large mode spacing puts more power in
each mode improving the signal to noise of the beat notes.
The laser that quickly became the working horse in the field for this reason was a
rather compact titanium-sapphire ring laser with typical repetition rates of 500 MHz to
1 GHz [48]. The ring design solved another problem that is frequently encountered when
coupling a laser into an optical fiber. Optical feedback from the fiber may disturb the
laser operation and even prevent mode locking in some cases. The standard solution to
this problem would be to place an optical isolator between the fiber and and the laser.
In this case however such a device would have almost certainly enough group velocity
dispersion to prevent any subsequent spectral broadening if this is not compensated for.
In a ring laser the pulses reflected back from the fiber travel in the opposite direction
and do not talk to the laser pulses unless they meet inside the laser crystal. The latter
can be prevented by observing the distance of the fiber from the laser. A disadvantage
of these lasers is that they are not easy to align and have so far not become turn key
systems that can be operated unattended for a long time say in an all optical atomic
clock (see below).
Even though microstructure fibers have allowed the simple f − 2f self-referencing
for the first time, they also have some drawbacks. To achieve the desired properties,
the microstructure fibers need to have a rather tiny core. The coupling to this core
causes problems due to mechanical instabilities and temperature drifts even with low
level and stable mounts. Another problem is the observed but not fully understood
strong polarization dependence of the fibers broadening action. The possibility of long-
term continuous operation is not so much of an issue for spectroscopy, because data
taking in these experiments usually do not last very long and they need some attention
on there own. However, this possibility seem to be a key requirement operating an
all optical atomic clock. So far the only way to operate such a system unattended for
hours was to use a set of additional servo systems that continuously measure and correct
deviations from the fiber coupling and polarization [49].
Another problem with spectral broadening by self phase modulation in general is
a excess noise level of the beat notes well above the shot noise limit [50, 51]. In fact
using the 73 fs laser mentioned above with a microstructure fiber, a two-octave–spanning

spectrum is generated within a few centimeters of fiber. However this spectrum does not
separate into modes that could be detected by a beat note measurement with a cw laser
but consist of noise [52]. One possible explanation is the Raman effect that produces
strong gain about 13 THz to the red from the pump wavelength. If this gain is not seeded,
it may trigger an avalanche of photons from the vacuum, that bear no phase relationship
to the carrier wave and coherence is lost(6 ). For sufficiently broad input spectra the
(6 ) This process is called “stimulated Raman scattering” because all but the first photon is
produced by stimulated emission. It should be noted though that the first spontaneous photon
destroys the coherence of the whole process.
Raman gain is seeded coherently with modes from the frequency comb amplifying the
low-frequency modes at the cost of the high-frequency modes. For longer pulses, say
below 0.441/(13 THz) = 34 fs for Gaussian pulse shape, less seeding occurs. In fact a
more detailed calculation predicts that enough coherence is maintained by self-phase
modulation if the seeding the pulses are shorter than 50 fs [53].
By going to shorter pulses for the the seed laser this problem can be handled but
the alignment issue remains. On the other hand, lasers that reach an octave-spanning
spectrum without using any external self-phase modulation can solve this problem [54-57].
So far however, these lasers seem to be rather delicate to handle so that one alignment
problem is replaced with another. An interesting alternative are lasers that avoid the use
of microstructure fibers in another way. For wide-band but not quite octave-spanning
lasers, a 2f − 3f self-referencing becomes possible by doubling the blue wing of the
spectrum and beat it with the tripled red wing [58, 59]. Such a system can remain phase
locked unattended for several hours without the burden of having extra servo systems.
Related but somewhat simpler seems to be a laser that produces pulses short enough so
that a little bit of self-phase modulation generated in single pass through a difference
frequency generating crystal provides sufficient spectral broadening [60].
Yet another class of frequency combs that can stay in lock for even longer times are
fs fiber lasers [61]. The most common type is the erbium-doped fiber laser that emits
within the telecom band around 1500 nm. For this reason advanced and cheap optical
components are available to build such a laser. The mode-locking mechanism is similar
to the Kerr lens method, except that non-linear polarization rotation is used to favor
the pulsed high peak intensity operation. Up to a short free-space section that can
be build very stable, these lasers have no adjustable parts. Bulk fused silica has its
zero group velocity dispersion at around 1.2 μm but this can be shifted to 1.5 μm in an
optical fiber. If, in addition, the radial dependence of the refractive index is designed to
obtain a small core area Af , the fiber becomes what is called a highly non-linear fiber
(HNLF) without any microstructure. These HNLF’s are commercially available and can
be spliced directly to a fs fiber laser. This virtually eliminates the remaining alignment
sensitive parts as the free space frequency doubling stage and bat note detection can be
build rather robust. Continuous stabilized operation for many hours [62, 63] have been
reported. The Max-Planck Institute für Quantenoptik in Garching/Germany operates

a fiber based self-referenced frequency comb that stays locked without interruption for
months. A significantly large jitter of the observed CE-beat note has been observed in
these lasers and can either be suppressed by using low noise pump lasers [64] or eliminated
with a fast servo system.
3. – Self-referencing
The measurement of ωCE fixes the position of the whole frequency comb and is called
self-referencing. The method relies on measuring the frequency gap between different
harmonics derived from the same laser or frequency comb. The first crude demonstra-
tion [2] employed the 4th and the 3.5th harmonic of a f = 88.4 THz (3.39 μm) laser to
determine ωCE according to 4ωn − 3.5ωn = (4n − 3.5n )ωr + 0.5ωCE = 0.5ωCE with
4n − 3.5n = 0. To achieve the condition of the latter equation, both n and n have to be
active modes of the frequency comb. The required bandwidth is 0.5f = 44.2 THz which
is what the 73 fs laser together with a single-mode fiber as discussed in the previous
section can generate.
A much simpler approach is to fix the absolute position of the frequency comb by
measuring the gap between ωn and ω2n of modes taken directly from the frequency
comb [4-8]. In this case the carrier-envelope offset frequency ωCE is directly produced
by beating the frequency doubled(7 ) red wing of the comb 2ωn with the blue side of the
comb at ω2n : 2ωn − ωn = (2n − n )ωr + ωCE = ωCE where again the mode numbers n
and n are chosen such that (2n − n ) = 0. This approach requires an octave spanning
comb, i.e. a bandwidth of 375 THz if centered at the titanium-sapphire gain maximum
at 800 nm.
Figure 5 sketches the f − 2f self-referencing method. The spectrum of a titanium-
sapphire mode-locked laser is first broadened to more than one optical octave with a
microstructure fiber. A broad-band λ/2 wave plate allows to choose the polarization
with the most efficient spectral broadening. After the fiber a dichroic mirror separates
the infrared (“red”) part from the green (“blue”). The former is frequency doubled in a
non-linear crystal and reunited with the green part to create a wealth of beat notes, all
at ωCE . These beat notes emerge as frequency difference between 2ωn − ω2n according to
eq. (6) for various values of n. The number of contributing modes is given by the phase
matching bandwidth Δνpm of the doubling crystal and can easily exceed 1 THz. To bring
all these beat notes at ωCE in phase, so that they all add constructively an adjustable
delay in form of a pair of glass wedges or corner cubes is used. It is straightforward to
show that the condition for a common phase of all these beat notes is that the green and
the doubled infrared pulse reach the photo detector at the same time. The adjustable
delay allows to compensate for different group delays, including the fiber. In practice
the delay needs to be correct within cΔνpm which is 300 μm for Δνpm = 1 THz. Outside
this range a beat note at ωCE is usually not detectable.
Whereas the half wave plates in the two interferometer arms are used to adjust for
whatever polarization exits the microstructure fiber, the half wave plate between the two
polarizing beam splitters helps to find the optimum relative intensity of the two beating
pulses. It can be shown that the maximum signal to noise ratio is obtained for equal
intensities reaching the detector within the optical bandwidth that contributes to the
beat note [3]. In practice this condition is most conveniently adjusted by observing the
signal-to-noise ratio of the ωCE beat note with a radio frequency spectrum analyzer. For
this purpose a low-cost analog device that operates up to ωr is usually sufficient.
A grating is used to prevent the extra optical power, that does not contribute to the
(7 ) It should be noted that this does not simply mean the doubling of each individual mode, but
the general sum frequencies generation of all modes. Otherwise the mode spacing, and therefore
the repetition rate, would be doubled as well.
beatnote at 1C E
2(n1r 1C E ) 2n1r 1C E
E (1 )
x2
n=2n
n
1
5-30 cm microstructure fiber
titanium-sapphire
mode locked laser
O/2
detector 1C E detector 1r
delay
dichroic
“blue” beam
splitter
O/2
O/2 O/2
“blue” “red”
grating polarizing beam frequency

splitters doubling
Fig. 5. – Top: f − 2f self-referencing by detecting a beat note at ωCE between the frequency
doubled “red” wing 2(nωr + ωCE ) of the frequency comb and the “blue” modes at 2nωr + ωCE .
Bottom: Layout of the self-referencing scheme. See text for details.
signal but adds to the noise level, from reaching the detector. Typically only a large
relative bandwidth of say 1 THz/375 THz needs to be selected so that a very moderate
resolution illuminating 375 lines is sufficient. For this reason it is usually not necessary to
use a slit between the grating and the photo detector. Sufficient resolution can be reached
with a small low-cost 1200 lines per mm grating illuminated with a beam collimated with
×10 microscope objective out of the microstructured fiber.
When detecting the beat note as described above, more than one frequency component
is obtained for two reasons. First of all any beat note, even between two cw lasers,
generates two components because the radio frequency domain cannot decide which of
the two optical frequencies is larger than the other. Secondly, observing the beat notes
0
low pass filter
21r 1CE
1r 1CE
1r 1CE
radio frequency power [dB]
21r
-10
1r
1CE
-20
-30
-40
-50
-60
-70
-80
-90
-100
Fourier frequency
Fig. 6. – Radio frequency spectrum produced by a self-referencing non-linear interferometer such

as the f − 2f interferometer shown in fig. 5. A low-pass filter with a cut-off at 0.5ωr selects the
component at ±ωCE .
between frequency combs, not only the desired component k = 2n − n = 0 is registered,

but all integer values of k, positive and negative contribute, up to the bandwidth of the
photo detector. This leads to a set of radio frequency beat notes at kωr ± ωCE for k =
. . . − 1, 0, +1 . . .. In addition the repetition rate, including its harmonics will most likely
give the strongest components. After carefully adjusting the nonlinear interferometer,
spatially and spectrally, and scanning the delay line for the proper pulse arrival times,
the radio frequency spectrum may look like the one shown in fig. 6. A low-pass filter with
a cut-off frequency of 0.5ωr selects exactly one beat note at ±ωCE . The design of such a
filter may be tricky, mostly depending on how much stronger the repetition rate signal
exceeds the beat note at ωCE . The sketch in fig. 6 gives a feeling on how steep this filter
needs to be at the cut-off in order to suppress the unwanted components below the noise
level. Such a suppression is required for taking the full advantage of the signal-to-noise
ratio. For this reason it is desirable to work at higher repetition rates. At ωr around
2π × 800 MHz, as used mostly for the ring titanium-sapphire lasers described above, the
filter requirements are much more relaxed than say at 80 MHz. In addition, a larger
repetition rate concentrates more power in each mode further improving the beat notes
with the frequency comb. It should be noted though, that currently higher repetition
rates cannot be used because the associated lower peak power will make it difficult to
.
achieve spectral broadening beyond one optical octave as detailed in subsect. 2 4.
As described, both degrees of freedom ωr and ωCE of the frequency comb can be
measured up to a sign in ωCE that will be discussed below. For stabilization of these
frequencies, say relative to a radio frequency reference, it is necessary to be able to control
them. Again the repetition rate turns out to be simpler. By mounting one of the lasers
cavity mirrors on a piezo electric transducer allows to control the pulse round trip time.
Another option is offered by mode-locked lasers that use prism pairs to compensate the
intracavity group velocity dispersion. In this case tipping the mirror at the dispersive
end where the cavity modes are spatially separated, changes the relative cavity lengths
of the individual modes and thereby the mode spacing in frequency space [7]. In practice
the detected repetition frequency is mixed with the radio frequency reference, i.e. the
frequency difference is generated, low-pass filtered and with appropriate gain send back
to the piezo electric transducer. When properly designed such a phase-locked loop forces
one oscillator, the repetition rate, to stay in phase with another, the radio frequency
reference. Because these servo systems are standard components in many electronic
devices such as FM radio receivers, a large amount of literature exists on their design
and stability analysis [65].
Setting up a phase-locked loop for the repetition rate therefore seems rather
straightforward. However, some caution concerning the servo bandwidth needs to be
observed. It turns out that the large frequency multiplication factor n in eq. (6) may
also multiplies the noise of the reference oscillator. The phase noise power for direct fre-
quency multiplication by n increases proportional to n2 [66], so that a factor of n = 106 ,
that would take us from a 100 MHz radio frequency signal to a 100 THz optical signal, in-
creases the noise by 120 dB. On this basis it has been predicted that, using even the best
available reference oscillator, it is impossible to multiply in a single step from the radio
frequency domain into the optical [67]. The frequency comb does just that but avoids the
predicted carrier collapse. In this case the laser acts as a flywheel in the optical that does
not follow the fast phase fluctuations of the reference oscillator but averages them out.
In this sense the n2 multiplication law does not apply, because it assumes a phase stiff
frequency multiplication that would correspond to an infinite servo bandwidth. Fortu-
nately a typical free-running titanium-sapphire mode-locked laser shows very good phase
stability of the pulse train on its own. For averaging times shorter than typical acoustic
vibrations of several ms period, such a laser shows better phase stability than a high-
quality synthesizer. It is therefore essential to use a moderate servo bandwidth for phase
locking the repetition rate of a few 100 Hz at most. A small servo bandwidth may be im-
plemented electronically by appropriate filtering or mechanically by using larger masses
than the usual tiny mirrors mounted on piezo transducers for high servo speed. In some
case a complete one inch mirror mount has been moved for controlling the repetition
rate [15].
Controlling the carrier envelope frequency requires some effort. Experimentally it
turned out that the energy of the pulse stored inside the mode locked laser has a strong
influence on ωCE . After initial explanations of this effects turned out to be too crude,
more appropriate mechanisms have been found [68, 69]. Conventional soliton theory [33]
predicts a dependence of the phase velocity but no dependence of the group velocity on
the pulse peak intensity. Any difference in the cavity round trip phase delay and the
cavity round trip group delay results in a pulse to pulse carrier envelope phase shift and
therefore a non-vanishing ωCE . However, the intensity dependence of that effect may
turn out to have the wrong sign [70]. The reason is that higher-order effects, usually
neglected in the conventional soliton theory, play an important role. The Raman effect
in the titanium-sapphire crystal produces an intensity-dependent redshift that in turn
affects the group round trip time. In general this leads to an extra term in eq. (8) for
the pulse to pulse carrier envelope phase shift per round trip [69]:

2L 2L
(25) Δϕ = ωc − + BIp .
vg vp
Because this phase shift is directly proportional to ωCE this equation also describes its
dependence on the pulse peak power Ip . The magnitude of the parameter B may best
be determined experimentally, as it turns out to depend on the operating parameters of
the mode-locked laser. In some cases it even changes in sign as the pump laser intensity
is changed [51].
To phase lock the carrier envelope offset frequency ωCE , one uses an actuator, in
most cases an acousto-optic modulator, that drains an adjustable part of the pump laser
power. Electro-optic modulators have also been used, but they have the disadvantage
that they need to a bias voltage that wastes some of the pump energy to work in the
linear regime. To servo control the phase of the ωCE component usually requires much
more servo bandwidth than locking the repetition rate. How much is needed in practice
depends on the type of laser, the intensity and beam pointing stability of the pump
laser and the phase detector in use. Mode-locked lasers that use pairs of prisms to
compensate for group velocity dispersion generally show a much larger carrier envelope
frequency noise. This is because intensity fluctuations slightly change the pulse round trip
path because of the intensity-dependent refraction of the titanium-sapphire crystal [71].
Even small variations of the beam pointing result in a varying prism intersection. It
should be noted that already 50 μm of extra BK7 glass in the path, shifts the carrier
envelope phase by 2π and the carrier envelope frequency by ωr . In most cases the carrier
envelope frequency fluctuations seem to be dominated by the pump laser noise, so that
stabilizing ωCE with a modulator as described above even reduces this noise. Today
titanium-sapphire lasers are mostly pumped by frequency doubled solid-state lasers that
seem to show some differences between the models currently on the market [72]. Fiber-
based mode-locked lasers used to have significantly larger noise in the carrier envelope
beat note than titanium-sapphire lasers before the semiconductor pump lasers have been
stabilized carefully [64].
In most cases a simple mixer is not sufficient to detect the phase of ωCE relative to a
reference oscillator as the expected in-loop phase fluctuations are usually much larger as
for the ωr servo. Prescalers or forward-backward counting digital phase detectors may be
used to allow for larger phase fluctuations, that in turn allow the use of moderate speed
(several 10 kHz) electronics. A complete circuit that has been used for that purpose very
successfully is published in [73]. Stabilizing the carrier envelope frequency, even though
it generally requires faster electronics, does not have the stability and accuracy issues
that enter via the repetition rate due to the large factor n in eq. (6). Any fluctuation or
inaccuracy in ωCE just adds to the optical frequencies rather than in the radio frequency
domain where it is subsequently multiplied by n.
None of the controls discussed here acts solely on either frequency ωCE and ωr . In
general a linear combination of the two is affected. In practice this turns out to be not
important because the different speeds of the two servo systems ensure that they do not
influence each other.
Measuring the frequency of an unknown cw laser at ωL with a stabilized frequency
comb, involves the creation of yet another beat note ωb with the comb. For this purpose
the beam of the cw laser is matched with the beam that contains the frequency comb,
say with similar optics components as used for creating the carrier envelope beat note.
A dichroic beam splitter, just before the grating in fig. 5, could be used to reflect out
the spectral region of the frequency comb around ωL without effecting the beat note
at ωCE . This beam would then be fed into another set-up consisting of two polarizing
beam splitters, one half wave plate, a grating and a photo detector for an optimum
signal-to-noise ratio. The frequency of the cw laser is then given by
(26) ωL = nωr ± ωCE ± ωb ,
where the same considerations as above apply for the sign of the beat note ωb . These
signs may be determined by introducing small changes to one of the frequencies with a
known sign while observing the sign of changes in another frequency. For example the
repetition rate cold be increased by picking a slightly different frequency of the reference
oscillator. If ωL stays constant we expect ωb to decrease (increase) if the “+” sign (“−”
sign) is correct.
The last quantity that needs to be determined is the mode number n. If the optical
frequency ωL is already known to a precision better than the mode spacing, the mode
number can simply be determined by solving the corresponding equation (26) for n and
allowing for an integer solution only. A coarse measurement could be provided by a wave
meter for example if its resolution and accuracy is trusted to be better than the mode
spacing of the frequency comb. If this is not possible, at least two measurements of ωL
with two different and properly chosen repetition rates may leave only one physically
meaningful value for ωL [27].
4. – Scientific applications
In this section a few experiments in fundamental research where optical frequency

measurements have been applied are discussed. Before the introduction of optical fre-
quency combs only a few measurements of visible light could be carried out. Since then
not only the available data has multiplied but also its accuracy has improved significantly.
Maybe even more important, the frequency combs have enabled the construction of all
optical atomic clocks that are treated in sect. 5.
.
4 1. Hydrogen and drifting constants. – The possibility to readily count optical fre-
quencies has opened up new experimental possibilities. High-precision measurements on
hydrogen have allowed for improved tests of the predictions of quantum electrodynamics
and the determination of the Rydberg constant [14]. As the simplest of all stable atomic
systems, the hydrogen atom, provides the unique possibility to confront theoretical pre-
dictions with experimental results. To explore the full capacity of such a fundamental
dye laser
486 nm BBO
servo
AOM
vacuum vessel
253 nm
cold
reference cavity nozzle
LD-detector
frequency comb
H
enhancement
resonator 243 nm
Fig. 7. – Exciting the hydrogen 1S-2S transition with two counterpropagating photons in a
standing-wave field at 243 nm. This radiation is obtained from a dye laser frequency doubled in
a BBO crystal and stabilized to a reference cavity. While scanning the hydrogen resonance the
frequency of this laser is measured with a frequency comb to be 2 466 061 413 187 074 (34) Hz
for the hyperfine centroid [74].
test, one should aim for the highest possible accuracy so that measuring a frequency is
imperative as explained in the introduction. At the same time one should use a narrow
transition line that can be well controlled in terms of systematic frequency shifts. The
narrowest line starting from the 1S ground state in hydrogen is the 1S-2S two-photon
transition with a natural line width of 1.3 Hz and a line Q of 2 × 1015 .
At the Max-Planck-Institute für Quantenoptik (MPQ) in Garching measurements of
this transitions frequency around 2466 THz has been improved over many years [13, 19,
74]. The hydrogen spectrometer used for that purpose is sketched in fig. 7. It consists of a
highly stable frequency-doubled dye laser whose 243 nm radiation is enhanced in a linear
cavity located in a vacuum vessel. The emission linewidth of the dye laser is narrowed
to about 60 Hz by stabilizing it to an external reference cavity. This stabilization also

reduces the drift rate below 1 Hz per second. The linear enhancement cavity ensures that
the exciting light field is made of two counterpropagating laser fields, so that the Doppler
effect is cancelled to first order. Hydrogen atoms are produced in a gas discharge and
ejected from a copper nozzle kept at a temperature of 6 K. After propagating the length
of 13 cm, the excited atoms are detected by quenching them to the ground state in an
electric field. The Lyman-α photon at 121 nm released in this process is then detected
with a photomultiplier.
The optical transition frequency was determined in 1999 [13] and in 2003 [74] with a
frequency comb that was referenced to a transportable cesium fountain clock from LNE-
SYRTE, Paris [75]. At this time the repeated measurement did not yield an improved
10-5
relative uncertainty 10-6 4 Hg+ vs Cs @ NIST
10-7
yr ]
10-8 2
-1
10-9
-14
10-10 0
PP [10
10-11
H vs Cs @ MPQ
10-12 -2
10-13 Yb+ vs Cs @ PTB
10-14 -4
10-15
1910 1930 1950 1970 1990 2010 -6 -4 -2 0 2 4 6
-15 -1
year DD [10 yr ]
Fig. 8. – Left: a century of hydrogen data. Since the early 1950’s when quantum electrodynamics
was developed, measurements have been improved by almost 7 orders of magnitude and still no
serious discrepancy has been discovered. Right: from the observation of transitions in single
trapped ytterbium and mercury ions and the 1S-2S transition over several years, upper limits
on small possible variations of the electromagnetic and the strong interaction can be found. The
latter appears here as a variation of the cesium nuclear magnetic moment μ measured in units
of Bohrs magneton.
value for the 1S-2S transition frequency. Lacking a suitable laser cooling method is a
particular problem for the light hydrogen atom. Even after thermalizing with the cold
nozzle to 6 Kelvin, the average atomic velocity is v = 360 m/s causing a second-order
Doppler effect of 0.5(v/c)2 = 7 × 10−13 that needs to be accounted for. In addition the
brief interaction of the atoms with the laser cause a variety of problems and can distort
and shift the lineshape in an unpredictable way on the 10−14 level. In addition, because
of the limited interaction time, larger laser power has to be used for a sufficient excitation
rate. This increases to ac Stark shift and cause problems when extrapolating to zero laser
power to find the unperturbed transition frequency. Nevertheless the inaccuracy is only
about an order of magnitude larger than the best atomic clocks. A historic summary
of hydrogen spectroscopy is given at the left side of fig. 8. As a remarkable aspect it
should be noted that before the introduction of quantum electrodynamics basically every
order of magnitude that measurement improved, required a new theory or at least some
refinement. Quantum electrodynamics by now has resisted a gain of almost 7 orders of
magnitude without such a refinement. This is probably unprecedented for any physical
theory. The development is summarized at the left panel of fig. 8.
The two comparisons of the hydrogen 1S-2S frequency with the LNE-SYRTE fountain
clock may also be used to derive upper limits of possible slow variations of the fundamen-
tal interactions. The question of a possible time variation of fundamental constants(8 )
was first raised in 1937 by P. A. M. Dirac, where he speculated that fundamental con-
(8 ) For a review on this subject see ref. [76].
stants could change their values during the lifetime of the universe [77]. The traditional
way to search for such a phenomenon is to determine the value of say the fine-structure
constant as it was effective billions of years ago. For this purpose atomic absorption lines
of interstellar clouds back-illuminated by distant quasars have been analyzed [78-81]. In
a related method, the only known natural nuclear fission reactor that became critical
some 2 billion years ago at Oklo, Gabon has been investigated [82-84]. Analyzing the
fission products the responsible cross-sections and from that the fine-structure constant
at the time this reactor was active can be deduced.
Now using frequency combs, high-precision optical standards can be compared with
the best cesium clocks on a regular basis. Besides the hydrogen transition, the best mon-
itoring data of this type so far derives from comparisons of narrow lines in single trapped
mercury [17] and ytterbium [18] ions with the best cesium atomic clocks. Remarkably,
the precision of these laboratory measurement makes it possible to reach the same sen-
sitivity within a few years of monitoring that astronomical and geological observations
require billions of years of look back time. The laboratory comparisons address some ad-
ditional issues: Geological and astronomical observations may be affected by systematic
effects that lead to partially contradicting results (see refs. [78,81,85,86] for contradicting
results on quasar absorption and [83, 84] for Oklo phenomenon data).
In the laboratory systematics can be investigated or challenged, and if in doubt,
experiments may be repeated given the relative short time intervals. So far atomic tran-
sition frequencies have been compared with the cesium ground-state hyperfine splitting,
which is proportional to its nuclear magnetic moment. The latter is determined by the
strong interaction, but unlike the electronic structure, this interaction cannot easily be
expressed in terms of the coupling constant. Lacking an accepted model describing the
drift of the fundamental constants, it is a good advice trying to analyze the drift data
with as few assumptions as possible. In previous analysis of the Oklo phenomenon, for
example, it was assumed that all coupling constants but the fine-structure constant are
real constant in time. However, if grand unification is a valid theory, at least at some
energy scale, all coupling constants should merge and drift in a coordinated way [87].
Therefore the possibility that the cesium ground-state hyperfine splitting, i.e. the pace
of the fountain clocks may have changed, should not be ruled out by the analysis. In
this sense any of the optical frequencies monitored relative to the cesium clock can only
put limits on the relative drift of the electromagnetic and the strong coupling constant.
Fortunately the three mentioned comparisons show different functional dependences on
the fine-structure constant leading to different slopes in a two-dimensional plot that dis-
plays the relative drift rates of the coupling constants (see right panel of fig. 8). The
region compatible with this data is consistent with no drift at all, at a sensitivity level
only a factor 2 away from the best astronomical observations that have been detecting a
statistically significant variation [19]. It should be noted though that without a model for
the drift, linearity in time cannot be assumed, so that the laboratory measurements do
not even compare with the astronomical observations as they probe on different epochs.
In the near future direct comparisons between different optical transitions will provide
much better data because the cesium clock drops out and some optical transitions are
getting more accurate than even the best cesium fountain clocks [9, 88]. The frequency
comb also allows this type of frequency comparison by locking one of the optical modes
ωn to the first optical reference and measuring the second with another mode ωn . Ide-
ally the carrier envelope offset frequency is stabilized to an integer fraction 1/m of the
repetition rate, so that the ratio of the two optical frequencies derives as [89].
ωn n + 1/m
(27) = .
ωn n + 1/m
No radio frequency enters in this comparison and possible beat notes can also be refer-
enced to the repetition rate. Probably the most advanced experiment of this type would
be the comparison of narrow transitions in aluminum and mercury ions operated in the
group of J. Bergquist at NIST, Boulder. These standards are now reaching a reproducible
within a few parts in 1017 [9]. To put this in perspective, it should be noted that the
gravitational red shift at the Earth’s surface is g/c2 = 1.1 × 10−16 m−1 .
.
4 2. Fine structure constant. – Besides helping to detect a possible variation of the
fine-structure constant α, the frequency combs are also useful to determine its actual
value. All experiments to determine α have in common that a quantity that depends
on it, is measured. The fine-structure constant is then determined by inverting the
theoretical expression which usually comes as a power series in α.
As mentioned above, precision spectroscopy of hydrogen has led to an accurate ex-
perimental value for the Rydberg constant. Using all the available hydrogen data, i.e.
the 1S-2S transition frequency and other less precise measurements [14], a value with an
uncertainty of only 7 parts in 1012 is obtained [14, 90]. The Rydberg constant R∞ can
be traced back to other constants according to
α 2 me c
(28) R∞ = ,
2h
so that the fine-structure constant α is derived as precise as me /h, the electron mass
divided by Planck’s constant, is known. This is because the speed of light c has a defined
value within the SI units and enters with no uncertainty.

Currently the most precise measurement of the fine-structure constant has an uncer-
tainty of 7 parts in 1010 [91]. This measurement is based on an experimental value of
the electrons gyromagnetic ratio or more precisely its deviation from 2. This quantity
can be calculated with quantum electrodynamics with comparable accuracy in terms of
a power series in α. By comparison with the measured value, the fine-structure constant
is determined.
Because α scales the strength of all electromagnetic interactions, it can in principle
be determined with a large number of different experiments. Currently the second best
method is based on the recoil an atom, such as cesium [92] or rubidium [93], experiences
when absorbing an optical photon. Momentum conservation requires that the transition
6010
- 137.03] x 106
6005
AC-Josephson
6000
quantum Hall
electron g-2
cesium h/m
rubidium h/m
neutron h/m
5995
myonium
[D-1
HFS
5990
5985
Fig. 9. – The agreement of various values, derived from different experiments is a crucial test for
quantum electrodynamics. Currently the values nicely agree within their assigned uncertainty,
except the one derived from the Josephson effect. For further details see [91, 90].
frequency is shifted by the kinetic energy associated with the photon momentum h̄k and
the atomic mass M :
ΔE h̄k 2 h̄ ω 2
(29) Δω = = = .
h̄ 2M M 2c2
Measuring this recoil shift Δω and the optical transition frequency ω yields a value for
M/h. The recoil shift is typically only a few kHz on top of the transition frequency of sev-
eral 100 THz. Therefore high-resolution optical frequency measurements are mandatory.
To obtain α from (28) one additionally needs to know the mass ration me /M . Atomic
mass rations can be measured very precise by comparing their cyclotron frequencies in
Penning traps [94]. In fact such an effort was the motivation for the first frequency
comb measurement performed with a femtosecond laser [15]. Note that all ingredients
for deriving α via eq. (29) are obtained from precision frequency measurements, two of
which optical frequencies.
Even though less precise than the gyromagnetic ratio, the recoil measurements are
important when it comes to testing of quantum electrodynamics. In general any theory
that uses N parameters can only be said to have a predictive character if at least N + 1
different experimental outcomes can be verified. The first N measurements are only
fixing the parameters. In this context here the available data may be interpreted in a
way where the gyromagnetic ratio fixes α without any verification of the theory. The
recoil measurements are then interpreted as a test quantum electrodynamics, or the other
way around. Currently the gyromagnetic ratio fits very well with recoil measurements
in cesium and rubidium and other measurements. As fig. 9 shows, the only exemption
may be a value derived from the Josephson effect, which does not seem to fit within its
assigned error bar.
Finally it should be mentioned that of course quantum electrodynamics could be cor-
rect while calculations and/or measurements have some undetected errors. Calculating
the electron gyromagnetic ratio as a function of α has so far used 891 Feynman diagrams.
One might think that the recoil measurements does not show this problem, but deriving
the Rydberg constant from hydrogen data involves a similar complex evaluation [95].
.
4 3. Optical frequencies in astronomy. – In connection with cosmological search for a
variation of what we believe are fundamental constants, optical frequency measurements
are required on samples in the sky and on Earth as reference. Yet another type of obser-
vation relies on precise optical frequency measurements. To detect extrasolar planets the
most powerful method has been to measure the changing recoil velocity of its star during
the orbital period. These recoils velocities are rather small unless a massive planet in close
orbit is considered. This is the reason why mostly “hot Jupiters” are among the roughly
200 extrasolar planets detected so far. The lightest of those planets possesses about 10
Earth masses. The wobble that our planet imposes on the motion of our Sun has a veloc-
ity amplitude of only vE = 9 cm/s with a period of one year of course. Because Earth and
Sun maintain their distance as they go around their common center of mass, this motion
is invisible from Earth. On the other hand, it can be detected at other stars, where it is
superimposed with the center-of-mass motion of that system of typically 100’s of km/s
and the motion of the Earth around the Sun. To detect Earth-like planets that orbit
sun-like suns with the recoil velocity method, a relative Doppler shift of vE /c = 3×10−10
needs to be measurable. Converted to visible radiation of say 500 THz this requires a
resolution of 150 kHz and the same reproducibility after half the orbital time.
Spectral lines from atoms and ions from interstellar clouds and the surface of stars
are subject to strong line broadening due to collisions and the Doppler broadening of
typically several GHz due to their thermal motion. They are measured with telescopes
like the Very Large Telescope operated by the European Southern Observatory which
can be connected to an Echelle-type spectrometer for high resolution. Given these rather
broad lines, the required spectral or velocity resolution can be obtained only by using
the statistics of many lines observed simultaneously. It requires a spectrometer with very
small irregularities in the calibration curve. So far the rather irregular line spectrum Tr-
Ar lamps have been used for that purpose, with large spectral gaps. Using a frequency
comb for this purpose appears to be the optimum tool, both in terms of providing an
equidistant dense calibration and for allowing long-term reproducibility that goes well
beyond the typical life time of an individual spectrometer [96]. The latter property
derives from the possibility to reference to a precise clock. In this case even a simple
GPS disciplined rubidium clock suffices for the required 3 × 10−10 reproducibility to
detect Earth-like extrasolar planets.
In the long run one may even think about direct heterodyne detection with a fre-
quency comb. In this scheme the star light is mixed with the frequency comb on a fast
photo detector producing a radio frequency spectrum identical to the optical spectrum
but shifted by the optical frequency. Signal processing can then be done with a radio
frequency spectrum analyzer rather than a with an optical spectrometer. This type of
detection is known for producing shot noise limited signals and is used to demonstrate
noise levels below the shot noise limit with squeezed light. The frequency comb can
cos-pulse
spectrometer signal [a.u.]

unstabilized
0 fs
t
-5 fs +5 fs
cos-pulse
sine-pulse
sine-pulse
0 fs
t
-5 fs +5 fs
80 90 100 110 120 130 140 150
photon energy [eV]
Fig. 10. – Left: infrared driving pulses of high intensity and the calculated intensity of high
harmonic radiation around the short wavelength cut-off spectral region for two different values
of the carrier envelope phase [97]. Whereas a “sine” pulse creates two high harmonic pulses,
a “cos” produces a single isolated attosecond pulse [98]. Right: measured spectra (different
intensity) for the two values of the carrier envelope phase. It is seen that the cut-off region
looses its periodicity with the carrier wave as expected for single isolated attosecond pulses.
Figures adapted from ref. [97, 99].
provide a large number of optical local oscillators to shift any optical component within
its span to the radio frequency domain.
.
4 4. Reconstructing pulse transients and generating attosecond pulses. – The appli-
cation of the frequency combs has also enabled advances in another field as it allows
the possibility for stabilizing the carrier envelope phase [3, 6]. According to fig. 2 and
eq. (8), a pulse train with a vanishing carrier envelop frequency ωCE has a fixed phase of
the carrier with respect to the envelope. This means that all the pulses have the same
electric field. With the technique of self-referencing this can be readily accomplished.
However, even though the electric-field transients of the pulses are identical in this case,
it is unknown what value the carrier envelope phase actually assumes upon stopping its
pulse-to-pulse slippage. In order to figure out at what position the carrier envelope phase
has actually stopped, all delays in the system including the optical path delays have to
be known. Taking into account that the latter progresses by 2π for each wavelength
propagation distance such an attempt seems unrealistic.
To detect the value of the carrier envelope phase very short pulses that drive processes
which depend on the electric field in high order are used. The left-hand side of fig. 10
shows two extreme values of the carrier envelope phase that correspond to a “cos” and a
“sine” pulse. Close observation of the field transients reveals that the peak electric field
slightly depends on the carrier envelope phase. In addition the duration of the optical
carrier cycle changes slightly due to the steep pulse envelope when measured say between
two field maxima. These effects are largely enhanced if the pulses are short and can be
detected using highly non-linear process such as “high harmonic generation” [100-102]
or “above threshold ionization” [103]. High harmonics are generated if pulses of high
intensity are focused into a noble-gas jet and are emitted in nicely collimated laser-
like beam. Very short wavelengths up to the soft–X-ray regime have been produced this
way [104]. Added to the left-hand side of fig. 10 is the calculated high harmonic radiation
at a narrow bandwidth around the high-frequency cut-off. Two important aspects are
realized from this: For “cos” driving pulse the high harmonic pulse is much shorter
than one cycle of the generating field. So by stabilizing the carrier envelope phase to
the proper value (“cos” drive pulses) single isolated attosecond could be produced for
the first time [97, 99]. Secondly the high-harmonic spectrum reveals whether the carrier
envelope phase has been fixed at the proper value or not. A “sine” drive pulse creates two
high-harmonic pulses, that have almost but not quite the time separation given by the
carrier frequency. For this reason the spectrum shows well-separated peaks even in the
cut-off spectral region. However the positions of these peaks are not exactly harmonics
of the infrared driving field. In contrast to that, a “cos” produces a shifted harmonic
spectrum that looses its periodicity in the cut-off region. This is because it belongs to a
single isolated attosecond pulse [98], as show in the upper left part of fig. 10. It should be
noted that all of this applies only when the high-harmonic radiation is properly filtered.
This should be obvious for two reasons: Only the highest harmonics posses enough non-
linearity to distinguish the carrier envelope phase settings. Secondly when attosecond
pulses are generated, their carrier frequency must be significantly larger than the inverse
pulse duration.
Not only that the detection and stabilization of the carrier envelope phase allowed
the production of attosecond pulses for the first time, it also allowed to completely
recover the electric-field transients of ultrashort pulses. The work in that direction relied
on measuring the pulses autocorrelation which, together with the determination of the
carrier envelope phase via spectral analysis of high harmonic allows the calculation of the
field transient [99]. A more direct measurement uses cross correlation between attosecond
pulses generated the way described above with the driving pulses. In this way the field
transient of the latter can be sampled with a temporal resolution significantly shorter
than one optical cycle [105]. Figure 11 shows the result of such a measurement.
.
4 5. Frequency comb spectroscopy. – While for high-resolution spectroscopy of tran-
sitions such as the hydrogen 1S-2S single-mode lasers are currently employed, many
transitions of fundamental interest occur at wavelengths too short for state-of-the-art
continuous wave lasers. The boundary is set by the transparency range of the existing
non-linear crystals. The crystal material that is useful for the shortest wavelength is
BBO (β-barium-borate), with a transparency cut-off at about 190 nm. Only with pulsed
lasers it is possible to efficiently convert into new regimes in the vacuum UV (200–10 nm),
extreme UV (30–1 nm) and possible even the soft X-ray (< 10 nm)(9 ). The process that
(9 ) These spectral ranges are not used in a unified way and partially overlap.
electron kinetic energy [ev]

90 electron counts [a.u.]
80
70
60
50
-8 -4 0 4 8 12
time delay [fs]
Fig. 11. – Cross-correlating attosecond pulses with the driving fundamental infrared laser pulses.
The attosecond pulses are used to ionize atoms and the liberated electrons are accelerated by
the instantaneous field of the infrared pulses. By changing the delay between the pulses, the
electric-field transient can be sampled with sufficient temporal resolution. Figure courtesy of E.
Goulielmakis et al. [105].
allows this is high-harmonic generation [100-102], that was already discussed in the last
section. For a long-time–pulsed lasers and high-resolution spectroscopy seemed to ex-
clude each other because of the large bandwidth associated with short pulses. However,
coherent train of pulses as compared to isolated or non-coherent pulses (say from a Q-
switched laser) have quite distinct spectra. The coherent pulse train generated from a
.
mode locked laser consists of narrow modes. It was shown in subsect. 2 3 that the Fourier
and the Schawlow-Townes limit of the linewidth of an isolated mode is identical to the
linewidth of a single-mode laser of the same type with the same average power. This
opens the possibility to combine the high peak powers of mode-locked lasers, suitable for
frequency conversion, with the properties of a sharp laser line produced by a single-mode
laser.
One obvious problem with this approach is the limited power per mode in a broad
frequency comb. Given an octave-spanning comb, as required for simple self-referencing,

this can easily drop below 100 nW in practice. In addition the unused modes may cause
excess ac Stark shift that poses a problem in high-accuracy measurements. Still these
problems can be handled as demonstrated in ref. [106]. Another solution, at least for the
UV spectral region, is to combine a fs mode-locked laser as a precise frequency reference
with a ps mode-locked laser for spectroscopy [107]. The latter type of lasers can be
converted in wavelength with efficiencies approaching unity even in single pass.
Yet an even better idea is to employ a two-photon transition as initially proposed
by Ye. F. Baklanov and V. P. Chebotayev [108]. At first glance it seems that a two-
photon transition does require even more power. However it is straightforward to see
that in this case the modes can sum up pairwise such that the full power of the frequency
|e> 1
1
4S r
...
flourescence [a.u.]
1eg
F=4
F=3
F=4
F=3
|g>
F=4
F=3
F=4
F=3
1 1
11 11 0 20 40 60
2 eg 2 eg comb detuning [MHz]
Fig. 12. – Left: the modes of a frequency comb add up to largely enhance the transition amplitude
between levels |g and |e if only one of the central modes is resonant with a two-photon transition
or that resonance occurs exactly half ways between two modes. Right: as an example, frequency
comb excitation of the 6S-8S two-photon resonance is shown as the frequency comb is tuned
across it. The hyperfine doublet with angular momentum F repeats with half the repetition
rate of ωr /2 = 2π × 41 MHz as expected from the resonance conditions [107].
comb contributes to the transition rate: suppose the frequency comb is tuned such that
one particular mode nωr + ωCE , say near the center, is resonant with the two-photon
transition. This means that two photons from this mode provide the necessary transition
energy of ωeg = 2(nωr +ωCE ). In this case the combination of modes with mode numbers
{(n−1, n+1), (n−2, n+2), (n−3, n+3), . . .} are also resonant as they sum up to the same
transition frequency. In fact all modes contribute to the transition rate in this way. The
same applies if the two-photon resonance occurs exactly half ways between two modes.
Figure 12 gives more details and a sample curve obtained from the 6S-8S two-photon
resonance in cesium at 822 nm. It can be shown for unchirped pulses that the total
two-photon transition rate is the same as if one would use a continuous laser with the
same average power [108]. If the pulses are chirped, transition amplitudes corresponding
to the various combinations of modes do no longer add up in phase so that the total
transition rate is lower [109].
Experimentally frequency comb spectroscopy has been pioneered by J. Eckstein [110]
and M. J. Snadden [111] and coworkers on sodium and rubidium, respectively, directly
with a mode-locked laser. While the former experiment was still a factor 2.5 short
of the 1.6 MHz natural linewidth, the latter reached the natural linewidth of 300 kHz.
The main difference between the two measurements is that Snadden and coworkers used
laser-cooled atoms in a magneto-optical trap.
This comparison reveals one disadvantage of the method. There may be a significant
time-of-flight broadening because the Doppler free signal only emerges from atoms within
the pulse collision volume for counterpropagating pulse trains. Atoms that fly through
this volume can only absorb a limited number of pulses that cause the line broadening
.
effects described in subsect. 2 3. To obtain a narrow resonance, it is therefore important
to apply many pulses, ideally for a time that exceeds the inverse natural linewidth. When
using fs pulses the collision volume may be smaller than 1 mm so that the atoms must
be laser cooled and/or trapped as in ref. [111] to reach the natural linewidth. Note that
the time-of-flight broadening may also be understood as a residual first-order Doppler
effect. This comes about because, unlike the single-mode case, the Doppler shift k1 v and
−k2 v for counterpropagating beams with wave numbers k1 and −k2 does not exactly
compensate for the motion of the atom with velocity v. However for any residual first-
order Doppler shift there exists a component obtained by exchanging the wave numbers
k1 k2 that possess the opposite shift. Hence a pure broadening and no shift results
just as in the time domain description. This is certainly an advantage.
In the meantime several other groups have used laser-cooled atoms for frequency comb
spectroscopy [112, 113]. S. Witte and coworkers used UV pulses at 212 nm generated in
conventional non-linear crystals to excite a two-photon transition in krypton [114]. In
this work only 3 pulses could be applied to the atoms so that the expected linewidth
was about one third of the repetition rate or 23 MHz. Indeed that is about the observed
linewidth which compares to the natural linewidth of 6.9 MHz. This experiment demon-
strates the principle, knowing that improvements by many orders of magnitude should
be possible. However, realizing these disadvantages of frequency comb spectroscopy, it
appears advisable to use a continuous wave laser whenever possible. On the other hand,
if there is no laser of this kind, frequency comb spectroscopy can become a very power-
ful method. One of the first demonstration in the short, so far inaccessible wavelength
range, uses the ninth harmonic of a titanium-sapphire laser generated in a xenon gas jet
to perform Ramsey-type spectroscopy on krypton at 88 nm [115]. The Ramsey method
is identical to frequency comb spectroscopy with two pulses and produces narrow lines
only when the two pulses come at a long delay (low repetition rate). In the meantime the
same Ramsey-type spectroscopy has also been demonstrated at a one-photon transition
in xenon at 125 nm [116]. The required radiation was obtained as the third harmonic
produced in a gas cell. In this experiment the “repetition rate”, i.e. the time separation
between the Ramsey pulses was varied to increase the resolution.
Using high harmonics from titanium-sapphire lasers several 100 octaves of laser radi-
ation may be addressed without gaps. Given the large tunability of these lasers allows
to shift between the harmonics seamlessly. Therefore it might become possible to use
a single laser system to cover everything from the near infrared to soft X-rays with
atomic clock resolution. A possibility that seems out of reach for single-mode lasers. For
such a general laser system there is yet another obstacle that needs to be solved. So
far long pulse trains of many pulses at very short wavelengths with a high repetition
rate necessary to resolve the modes have been contradicting requirements. This is be-
cause all methods up to very recently, that allowed to reach the necessary intensity for
high-harmonic generation of typically 5 × 1013 W/cm2 , effectively concentrate the avail-
able average power in fewer pulses per second. For this the repetition rate is typically
reduced to the kHz regime. The modes of the resulting dense frequency comb would
be very difficult to resolve. Even if the time-of-flight broadening could be reduced, say
by using trapped ions and transitions with narrow natural linewidths, the requirements
on the laser system would be difficult to achieve. For this reason a method allowing
the production of high-harmonic radiation with MHz repetition rates was sought. The
solution to this problem was to use an enhancement resonator for the driving pulses with
an intracavity gas jet for high-harmonic generation [117, 118]. This method is similar
to resonantly enhanced second-harmonic generation that has been used for many years.
However, there are several extra requirements that need to be fulfilled in order to res-
onantly enhance fs pulses. First of all the pulse round trip time in that cavity has to
be matched to repetition rate of the laser. In fact this is not difficult to accomplish and
standard continuous wave locking schemes can be applied to stabilize the frequency comb
to the modes of the enhancement resonator. Somewhat more difficult is it to prevent
the pulse stored in the enhancement resonator from reshaping, or more precisely to re-
shape it such that after one round trip it will match the next pulse from the laser. As
.
discussed in subsect. 2 1 this means that the carrier envelope phase shift of the cavity
has to be the same as for the laser and higher-order dispersion has to be suppressed. In
the frequency domain this simply means that the modes of the enhancement resonator
must be equidistant in frequency just as the frequency comb is. In this case all modes
can resonate at the same time. So far this problem has not been solved completely, so
that the high-harmonic intensity generated this way still seems too low for a reasonable
transition rate in stored ions. Fortunately this process scales very favorably with the
driving intensity I. The so-called plateau harmonics scale with number of fundamental
photons required for ionization, which is ∝ I 9 for Xe driven with a titanium-sapphire
laser around 800 nm.
5. – All optical clocks
Most likely the first time piece of mankind was the periodic movement of a shadow
cast by a fixed object. The daily period would measure the Earth rotation and the
superimposed annual period could be used to find the solstice and measure the length
of the year in units of days. Later other periodic phenomena have been used. For an
operational clock the periodic phenomena or oscillator has to be completed by a counter
that keeps track of the number of periods. The early sundials had a human operator for
counting the days. Later pendulum clocks used mechanical counters and todays precise
clocks such as quartz or atomic clocks have electronic counters. If we look at the history
of time keeping, it becomes obvious that clocks got more accurate as the oscillation
frequency increased. One simple reason for that is that higher oscillation frequencies can
slice time into finer intervals. Similar to a ruler that improves with the density of length
marks.
The two most important properties of clocks are accuracy and stability. The former
describes at what level two identically constructed clocks agree. The clocks stability,
on the other hand, measures how well a particular clock maintains its pace. Both,
accuracy and stability are determined by comparing at least two clocks and are limited
by many factors such as the oscillators sensitivity to external perturbations. An increased
oscillator frequency helps to improve stability. As an example consider the Earth rotation
represented by a sundial and a good quartz oscillator. Whereas the two oscillators may
be comparable in terms of accuracy, it would be much more difficult to measure a period

of one second with a sundial than with a quartz oscillator that typically vibrates 32 768
times a second. Even faster than the vibration of a quartz standard is the precession of
the Cs nuclear magnetic moment in the magnetic field of its electrons. This frequency
has been defined within the International Systems of Units (SI) to be exactly 9 192 631
770 Hz.
To quantify the mutual stability of two oscillators a statistical analysis of a set of
subsequent phase comparisons, obtained by averaging over time intervals τ , is used. For
this purpose one may not use the standard deviation because it does not converge if the
two clocks have slightly different frequencies, which is almost always the case. Instead the
statistical analysis is done in terms of the Allan variance σ(τ ) [119]. Having eliminated
all other sources of instability an atomic oscillator, that obeys the laws of quantum
mechanics, possesses the following expression for the Allan variance:

Δω0 Tc
(30) σ(τ ) = .
πω0 2N τ
Here the frequency and the linewidth of the transition are given by ω0 and Δω0 , respec-
tively and the number of oscillators (atoms or ions) and the interrogation time is given by
N and Tc . Modern Cs atomic clock can be operated close to that limit [75]. Of course the
Allan variance will not reduce indefinitely for longer averaging times as suggested by the
quantum limit. At some point varying systematic effects will set in and prevent further
reduction. Current state-of-the-art Cs clocks need to be averaged for hours to reach that
limit at around one part in 1015 . If one sets out to reach an accuracy of one part in 1018 ,
which is the predicted systematic uncertainty for some optical transitions, the necessary
averaging times would extend by 6 orders of magnitude according to eq. (30). This of
course is no longer useful for any practical purpose.
Fortunately the same equation suggests that one can make up for this by increas-
ing the transition frequency ω0 as illustrated above. At first glance it may seem that
reducing the linewidth Δω0 by choosing a narrow transition would serve the same pur-
pose. In principle one could pick an almost arbitrary narrow transition, such as between
ground-state hyperfine levels that basically live forever or highly forbidden transition such
as the Yb+ 2 S1/2 → 2 F7/2 transition, whose upper level lives for about ten years [120].
However this means that there will be one photon of information per lifetime available at
most, assuming 100% detection efficiency. For this reason a much better strategy is to use
a larger transition frequency such as an optical transition leaving the linewidth, i.e. the
inverse transition rate fixed. Ideally one would choose an optical transition that repre-
sents an oscillator at frequency of several hundreds of terahertz. Operating at these high
frequencies would have been possible after tremendous advances in laser spectroscopy
in the 1970s that ultimately resulted in trapped atom and ion standards [121] in the
1980s. What was missing though was an efficient counter that keeps track of these fast
oscillations. When it became possible to count these oscillations with the so-called har-
monic frequency chains in the early 1970s [122], physicists started to seriously thinking
100,000 10 -9
clock uncertainty [ns per day] 10 -11

1,000
relative uncertainty
optical standards
10 10 -13
0.1 NBS/NIST/PTB Cs clocks 10 -15
0.001 10 -17
1960 1970 1980 1990 2000 2010
year
Fig. 13. – The historic comparison of radio frequency standards reveals that just about now the
optical standards seem to take over the lead. The current lowest uncertainty is reached with
the Hg+ clock at NIST [9] with an estimated standard uncertainty of 7 parts in 1017 .
of running an optical clock. However, working with these counters was so tedious that
most of the harmonic frequency chains never reached the stage where they could operate
continuously even for minutes. So it was decided to use them only to calibrate some
chosen frequencies, like iodine or methane stabilized HeNe lasers, that could then be
reproduced in other labs that could not provide the tremendous resources required for
setting up a harmonic frequency chain. The calibrated lasers where then mostly used
as wavelength references in interferometers for the realization of the meter and in some
scientific experiments as frequency references.
With the introduction of femtosecond frequency combs a reliable running optical
clock became reality. In particular the fiber-based frequency combs can now run for
months with out touching them [63]. In addition excess noise observed in these lasers
has recently been suppressed to the level of titanium-sapphire laser [64]. One of the first
set-ups that deserved the term “optical clock” used a transition at 1064 THz in a trapped
single mercury ion operated at NIST [89]. For a device deserving this name one would
ask for some requirements on its accuracy and its ability to be operated long enough to
calibrate a hydrogen maser for example. The NIST Hg+ clock can be operated repeatedly
with an uncertainty low enough such that comparison with an ensemble of very stable
hydrogen masers resulted in an Allan variance of the latter [9]. By now many national
standard institutes are working on optical atomic clocks. The Physikalisch-Technische
Bundesanstalt, Germany (PTB) on Yb+ [123], the National Physics Laboratory, UK
(NPL) on Sr+ and Yb+ [11] and the Laboratoire national de métrologie et d’essais—
Système de Références Temps-Espace, France (LNE-SYRTE) on neutral Sr [124] to name
a few. The tremendous progress of the optical standards made since frequency combs
where introduced is summarized in fig. 13.
To operate a frequency comb as a clockwork mechanism one might stabilize its nearest
mode ωk to the clock transition. Similar to the frequency ratio scheme of eq. (27), the
carrier envelope offset frequency and possibly any local oscillator used in the stabilization,
can be locked to an integer fraction m of the repetition rate [89].
(31) ωk = kωr + ωCE = ωr (k + 1/m).
The countable clock output is then obtained in form of the pulse repetition rate that
is given by the optical reference ωk divided by k + 1/m. No other reference frequency
besides the optical reference is required at this point as it should be for a real clock.
In particular there is no radio frequency source required other than the repetition rate.
To use the repetition rate in this way as a radio frequency generator is advantageous as
it can be shown that, on a short time scale, this frequency is more stable than a good
synthesizer even for a free-running laser [39].
6. – Optical frequency standards
Since the redefinition of the unit of length, the meter, in the International Sys-
tems of Units (SI) in 1983 by the 17th General Conference on Weights and Measures
(CGPM) [125] the speed of light is fixed by definition to c = 299 792 458 m/s. With the
general relation c = λ × ν any radiation emitting device whose frequency ν and hence its
wavelength λ is stable and can be traced back to the frequency of the Cs clock can be used
as a frequency standard or a wavelength standard. Such frequency standards have been
used for decades for interferometric length metrology. The International Committee of
Weights and Measures (CIPM) has therefore recommended a list of evaluated frequency
standards, i.e. the so-called Mise en Pratique for the definition of the Metre [126] which
has been updated on several occasions [125, 127, 128].
The first frequency standards in this list where mainly gas lasers whose frequencies
were stabilized to the saturated absorption of suitable transitions in molecules. The
selected molecules by natural coincidence have transitions whose frequencies are located
in the tuning range of the laser. The most prominent examples are the CH4 stabilized
He-Ne laser at 3.39 μm, the OsO4 stabilized CO2 laser at 10.3 μm and the I2 stabilized
Ne-He or Nd:YAG lasers at 632 nm and 532 nm, respectively. While the first two of these
lasers probably have the best reproducibility of all lasers stabilized this way, the much
more compact and transportable iodine stabilized laser has been the working horse in
length metrology for quite some time. The best of these lasers are reproducible within a
few parts in 1013 and therefore cannot compete with the current Cs atomic clocks.
The possibility to cool atoms and ions by laser radiation has led to optical frequency
standards that could rival the best microwave clocks. In these laser-cooled quantum
absorbers suitable narrow transitions can be interrogated that are no longer excessively
broadened and shifted by effects associated with the velocity of the absorbers, e.g., the
short interaction time and the influence of the second-order Doppler effect.
Table I. – Optical clock transitions recommended by the CIPM.
Ion/Atom Transition Frequency/ Fractional

wavelength uncertainty
88
Sr+ 5s 2 S1/2 -4d 2 D5/2 444 779 044 095 484 Hz 7 × 10−15
674 nm
171
Yb+ 6s 2 S1/2 -5d 2 D3/2 688 358 979 309 308 Hz 9 × 10−15
435 nm
199
Hg+ 5d10 6s 2 S1/2 -5d9 6s2 2 D5/2 1 064 721 609 899 145 Hz 3 × 10−15
282 nm
87
Sr 5s 2 S1/2 -4d 2 D5/2 429 228 004 229 877 Hz 1.5 × 10−14
698 nm
Optical frequency standards employing ions or neutral atoms as the frequency refer-
ences show distinctive differences. Ions can be stored in a so-called radio frequency trap
where an alternating electric voltage is applied to a suitable arrangement of electrodes
that results in a net force onto the ions directed towards the field-free center of the ion
trap. Since alike charged ions repel each other only a single ion can be kept in the center
whereas the non-vanishing field outside the center in general shifts the transition fre-
quency. Most accurate ion frequency standards therefore store only a single ion which,
however, leads to a very weak signal.
To produce a much stronger signal and an associated much higher stability as com-
pared to single ions (N = 1 in eq. (30)) the second type of laser-cooled optical frequency
standards uses clouds of millions of neutral atoms. In this case though collisions between
the atoms may result in uncontrollable shifts in frequency. In other words, so far (but
not necessarily in the long run) single-ion standards have high statistical but low sys-
tematic uncertainties whereas neutral-atom standards have excellent short-term stability
but poorer systematic properties.
The recent progress with optical frequency standards in the mean time motivated
the CIPM to recommend in autumn 2006 four optical frequency standards (see table I)
that can be used as “secondary representations of the second” and be included into
the new list of “Recommended frequency standard values for applications including the
practical realisation of the metre and secondary representations of the second”. The
phrase “secondary representations” takes into account that the frequency of such an
optical frequency standard can never surpass the accuracy of the primary standard of time
and frequency, the Cs clock, but that it might be very important also for time keeping
to use optical clocks because of their better stability. We will discuss the frequency
standards of table I in more detail in the following subsections.
The uncertainties given in table I are in general larger than the uncertainties presented
in the relevant publications since additional contributions have been taken into account
caused for example by the uncertainty in linking to different Cs clocks.
Fig. 14. – Excerpt from the energy level diagram of Yb+ . The transition at 369 nm is used for
cooling the ion and for detecting quantum jumps (see text). Dashed lines (411 nm, 435 nm and
467 nm) represent transitions that were proposed for optical clocks.
.
6 1. Optical frequency standards based on single ions. – The trapping and preparation
of ions and their use for frequency standards has been described in detail in a number of
reviews and textbooks (see, e.g., [129-132]). An important feature however is the special
kind of interrogation of single ions that will be discussed briefly in the following.
.
6 1.1. Clock interrogation by use of quantum jumps. A properly chosen ion can get close
to the ideal situation where a single unperturbed particle at rest is probed. The drawback
though is that a single ion does not produce a strong signal by scattering many photons
per second on the clock transition. However, the method of quantum jumps or elec-
tron shelving proposed by Dehmelt [133,134] and demonstrated later in Wineland’s [135]
and Dehmelt’s group [136] is capable to detect a transition with unity probability. The
technique is often applied to so-called V-systems where, as in fig. 14, a strong (cooling)
transition and a weak (clock) transition are connected at the ground state. When the ion
is irradiated by radiation whose frequency is in resonance with the strong transition up to
about 108 absorption-emission cycles lead to the same number of emitted photons. Even
with a limited detection probability of the order of 10−3 about 105 photons per second
can be detected. If, however, the ion is excited on the weak clock transition by resonant
radiation, the excited electron is “shelved” in the excited state. Consequently the fluores-
cence on the strong transition cannot be excited before the ion makes a transition to the
ground state again. Thus the fluorescence light from the strong transition monitors the
“quantum jumps” of the ion into the excited state and back as sudden intensity changes.
.
6 1.2. Yb+ and Hg+ single-ion standards. The clock transitions of the 171 Yb+ and
199
Hg+ ions given in table I are quadrupole transitions with natural linewidths of 3.1 Hz
and 1.1 Hz, respectively. The level scheme of both ions is basically very similar and
hence both systems will be treated here at the same time. In Yb+ the transition at
369 nm is used for cooling the ion and for detecting quantum jumps. The 2 S1/2 -2 D3/2
transition at 435 nm is preferred, rather than the 2 S1/2 -2 D5/2 transition at 411 nm since
the 2 D5/2 level decay may occur into the long-lived 2 F7/2 that has a natural lifetime of
about ten years. This level is connected to the ground state by an 467 nm octupole clock
transition with a natural linewidth of the transition in the nanohertz range [120] and is
also investigated as an optical frequency standard.
Two independent 171 Yb+ single-ion optical frequency standards operating at the
435 nm line were compared at PTB to look for systematic frequency shifts. A signif-
icant contribution to the systematic uncertainty could result from the interaction of the
atomic electric quadrupole moment with the gradient of an electric stray field in the trap.
Two methods can be used to eliminate the quadrupole shift. One uses the fact that the
quadrupole shift averages to zero in three mutually orthogonal directions of the magnetic
quantization field [137]. A second method uses the fact that the average over all Zeeman
and quadrupole shifts of all magnetic sublevels is zero [138].
Another possible systematic shift can result from the quadratic Stark shift. In the case
of the 171 Yb+ clock transition the scalar polarizability of the ground state, αS (2 S1/2 ), and
from the scalar and tensorial polarizabilities of the 2 D3/2 state, αS (2 D3/2 ) and αT (2 D3/2 )
contribute to this effect. The contribution has been measured and is corrected for.
For the unperturbed clock transition a mean relative frequency difference of the fre-
quencies between both standards of (3.8±6.1)×10−16 was observed [88]. This agreement
is similar to that of the most accurate comparisons between cesium fountain clocks. The
frequency of the 171 Yb+ standard was measured with a relative systematic uncertainty
of 3 × 10−15 and a statistical uncertainty of 0.6 × 10−15 using a femtosecond frequency
comb generator based on an Er3+ -doped fiber laser [139].
A similar level scheme of the cooling and clock transitions is present in the optical
frequency standard based on the 1.06×1015 Hz (282 nm) electric quadrupole transition in
a single trapped 199 Hg+ ion that was developed at NIST. This frequency standard uses a
cryogenic spherical electromagnetic (Paul trap) to trap a single Hg ion for up to 100 days.
The frequency measurements over a period of more than five years between the 199 Hg+
standard and the Cs primary clock transition agreed to better than 1 × 10−15 over three
years [9]. A fractional frequency instability of about 7 × 10−15 at 1 s was demonstrated
that follows the τ −1/2 dependence down to the 10−17 regime. The uncertainty budget
presented from the NIST groups leads to a total fractional uncertainty of 7.2 × 10−17
and represents the smallest uncertainty to find the unperturbed line center reported for
an optical clock. This uncertainty is an order of magnitude smaller as the uncertainty
of the best cesium atomic clocks. As a result the frequency (in SI Hz) of such an optical
clock can never surpass the uncertainty of the latter one.
.
6 1.3. 88 Sr+ single-ion standard. In contrast to the Yb+ and Hg+ single-ion standards
discussed so far the 88 Sr+ 5s 2 S1/2 -4d 2 D5/2 at 674 nm 444 779 044 095 484 Hz (fig. 15) of
the even isotope is lacking hyperfine structure. Thus there is no magnetic-field insensitive
mF =0 → mF =0 transition and a small external field splits the clock transition with a
natural linewidth of 0.4 Hz into five pairs of Zeeman components.
40
Fig. 15. – Sr+ energy level scheme.
This standard is investigated at the British National Physical Laboratory (NPL) [140]
and at the National Research Council (NRC) [138] of Canada. In order to eliminate the
first-order Zeeman shift, these groups probe alternately a symmetric pair of Zeeman
components.
.
6 2. Neutral atom optical frequency standards. – Neutral-atom–based frequency stan-
dards can be divided in three groups: the first one is based on free atoms in an atomic
beam effusing from a nozzle and collimated by diaphragms.
.
6 2.1. Atomic beam standards. The most advanced beam standard in the optical regime
are the hydrogen standard based on the 1S-2S transition in atomic hydrogen [13, 19, 74]
described above and the Ca standard based on the 1 S0 -3 P1 transition (see fig. 16) in
atomic Ca [141-143]. The large velocity of the atoms leads to a large second-order
Doppler effect which leads to a considerable broadening and shift of the interrogated
line. Even though techniques that suppress the first-order Doppler effect were applied,
residuals resulting from non-ideal alignment or curvature of the interrogating optical

beams (residual first-order Doppler effect) occur that seem to prevent the use of atomic
beam for use in optical frequency standards with the highest accuracy.
.
6 2.2. Neutral atom standards based on ballistic atoms. The large influence of the first-
and second-order Doppler effects can be reduced very efficiently by laser cooling the
atoms, e.g., in a magneto-optical trap. After shutting off the magneto-optical trap the
released atoms have an initial velocity between a few cm/s and a m/s and follow ballistic
paths in the gravitational field. Due to the free fall the interrogation of these atoms is
limited to a few milliseconds in typical laser beams. The broadening of the interrogated
line associated with such an interaction time corresponds to a few hundred hertz. In
general, fountain geometries could be utilized to increase the interaction time to about
Fig. 16. – Excerpt of the 40 Ca energy level scheme. The transition at 423 nm is used for cooling
the atoms. The clock transition (657 nm) is also used for second-stage cooling in combination
with the quench laser at 423 nm or 553 nm.
a second [144]. Such a technique has been proposed a long time ago but has never been
realized up to now.
For neutral ballistic atom standards therefore mostly alkaline earth atoms like
Mg [145], Ca [146-149], Sr [150] have been investigated since there the intercombina-
tion transitions 1 S0 -3 P1 with their natural linewidths between a few tens of hertz and a
few kilohertz are well adapted to the achievable interaction time-limited linewidth.
The most investigations and probably the best results have been obtained with cal-
cium. The Ca intercombination transition 1 S0 -3 P1 (see fig. 16) has a natural width of
0.37 kHz which has been resolved to below the natural linewidth. In [148] a thermal
beam of Ca atoms from an oven at 900 K is decelerated using a Zeeman slower from a
velocity of 600 m/s to a velocity of about 40 m/s. Then the slow atoms are collimated
and deflected by 10 degrees towards a magneto-optical trap (MOT) with a 2D molasses.
This set-up avoids the black-body radiation from the hot oven which would lead to a
so-called black-body shift [151]. At the first cooling step Ca atoms are cooled on the al-
lowed 423 nm transition 1 S0 -3 P1 (see fig. 16) and captured in a MOT. In 10 ms about 107
atoms are collected in a cloud at a temperature close to the Doppler limit of 0.8 mK for
this cooling transition. A second cooling stage is performed using the forbidden 657 nm
transition (see fig. 16). For efficient cooling the scattering rate is increased by quenching
the upper metastable 3 P1 state by excitation of the 453 nm transition to the 4s4d 1 D2
state. With this quench cooling method the atoms are further cooled down to 12 μK,
corresponding to an r.m.s. velocity of 10 cm/s. Then the clock transition is interrogated
using a sequence of four pulses cut from a single cw laser beam. The first pair of pulses
and the second one are antiparallel to allow for a first-order Doppler free excitation. The
resolution is determined by the time between the first and the second and the third and
the fourth pulse. This sequence represents the time domain analogue to the well-known
optical separated field excitation [152] or a Ramsey-Bordé atom interferometer [153].
The necessary radiation to interrogate the clock transition with a spectral linewidth of
approximately 1 Hz was provided by a master-slave diode laser system [154].
The frequency of the clock transition at 657 nm was referenced to the caesium fountain
clock utilizing a femtosecond comb generator to be 455 986 240 494 144 (5.3) Hz in PTB
with a fractional uncertainty of 1.2 × 10−14 and later at NIST with 6.6 × 10−15 [149], one
of the lowest uncertainties reported to date for a neutral atom optical standard. These
authors also concluded that this type of standard could be capable to achieve a fractional
uncertainty in the 10−16 regime.
A fractional frequency instability of 4 × 10−15 at 1 s averaging time was achieved.
As has been pointed out [39, 146] the quantum projection-noise limit [151] for the Ca
standard corresponds to a fractional instability or Allan deviation of σy (τ ) < 10−16 for
τ = 1 s. To approach this limit, a frequency instability of the spectroscopy laser of better
than 3 × 10−16 for the duration of the atom interferometry 2T ≈ 1.3 ms is necessary
since the so-called Dick effect [155] degrades the stability through the aliasing of the
laser frequency noise in a discontinuous interrogation. The Dick effect for an optical
frequency standard with Ramsey-Bordé interrogation was first addressed by Quessada et
al. [156]. Hence, the relatively short interaction time achievable with neutral atoms in
free flight limits the accuracy and the stability.
.
6 2.3. Optical lattice clocks. A possible way to combine the advantages of trapped
single ions and the large number of neutral atoms, i.e. the long storage times and the
Fig. 17. – Atoms are trapped in the potential minima located at the maxima of a standing wave
of the lattice laser.
Fig. 18. – The calculated ac Stark shift potential energies U for a given power P and a given
waist w of the lattice laser beam for the ground-state 1 S0 and the excited state 3 P0 are equal
at the so-called “magic wavelength” (circle).
good signal-to-noise ratio, respectively, was pointed out by Katori [157, 158]. A large
array of micro traps for atoms (see fig. 17) can be generated in a standing wave laser
field capable of holding maybe a million of atoms. Even though these optical lattices
have been investigated for a long time they did not seem to be appropriate to be used
in an optical frequency standard because of the large ac Stark shifts associated with the
laser field. The remedy for this effect is to tune the lattice laser to a so-called “magic
wavelength” where the upper and lower clock levels are shifted by the same amount,
leaving the clock transition frequency unaltered (see fig. 18). The big advantage here
Fig. 19. – Excerpt from the energy level scheme of 40 Sr. The transitions at 461 nm and 689 nm
are used for cooling the atoms. The dashed line (698 nm) is the clock transition.
is that the ac Stark shift can be controlled by controlling the lattice laser frequency
rather than by measuring the lattice laser intensity. The former can be determined very
accurately, in particular if one has an optical clock at hand.
Several atomic species are investigated for use as lattice clocks, e.g., 87 Sr [10,159-161],
Yb [162], and in the future also Hg. The best results have been obtained with 87 Sr which
will be discussed in the following. The strongly forbidden J = 0 → J = 0 transition 1 S0 -
3
P0 in Sr (see fig. 19) at 689 nm becomes weakly allowed through hyperfine mixing leading
to a natural linewidth of about 1 mHz. The atoms are cooled on the 461 nm transition to
several hundred microkelvin. To further reduce the temperature cooling on the 1 S0 -3 P1
transition is performed before the atoms are transferred into a dipole trap operated at
the magic wavelength near 813.4 nm. In a one-dimensional lattice Boyd et al. [160] stored
typically 2 × 104 atoms distributed across about 80 lattice sites. Interrogating the atoms
several effects contribute to the uncertainty to find the unperturbed line center. These
authors gave an uncertainty budget where the perturbations due to the residual ac Stark
shift of the lattice laser, the effect of the magnetic field necessary to separate the Zeeman
levels, and possible collisions contribute the largest effects ending up with a fractional
uncertainty below 1 × 10−15 . The frequency of this transition has been determined now
by three independent groups [159-161]. The measurement of Boyd et al. [160] gave 429
228 004 229 874.0 (1.1) Hz. The associated uncertainty is already much smaller as the
one recommended very recently by the CIPM (see table I) indicating the current speed
of the evolution of optical clocks and frequency standards.
7. – Conclusions
After only a few years the measurement of optical frequencies by optical frequency
combs is a mature technology that has led to a variety of novel applications. We are now
in a situation where optical frequency standards and clocks are beginning to outperform
traditional microwave clocks and most likely will lead in the future to a new definition of
the SI unit second probably based on an optical frequency standard. Optical clocks in the
visible part of the spectrum by no means need be the end of the evolution. Transitions
in the UV or even in the XUV might give superior stability, as this important property
scales as the transition frequency in use. It can be expected that the evolution in this
field also in the future will keep its current pace since novel ideas like reading out of
clock transitions by quantum information techniques [163, 164] or the use of nuclear
transitions [165] will continuously shift the frontiers in optical frequency metrology.
REFERENCES
[1] Hall J. L., IEEE Sel. Top. Quantum Electron., 6 (2000) 1136.
[2] Udem Th. et al., in Proceedings of the 1999 Joint Meeting of the European Frequency
and Time Forum and the IEEE International Frequency Control Symposium, Besançon
France, IEEE catalog no. 99CH36313 (1999), pp. 602-625.
[3] Reichert J. et al., Opt. Commun., 172 (1999) 59.
[4] Reichert J. et al., Phys. Rev. Lett., 84 (2000) 3232.

[5] Diddams S. A. et al., Phys. Rev. Lett., 84 (2000) 5102.
[6] Jones D. J. et al., Science, 288 (2000) 635.
[7] Holzwarth R. et al., Phys. Rev. Lett., 85 (2000) 2264.
[8] Udem Th., Holzwarth R. and Hänsch T. W., Nature, 416 (2002) 233.
[9] Oskay W. H. et al., Phys. Rev. Lett., 97 (2006) 020801.
[10] Takamoto M. et al., Nature, 435 (2005) 321.
[11] Gill P. and Margolis H., Physics World, May (2005) 35.
[12] High Resolution Spectroscopy of Hydrogen, in The Hydrogen Atom, edited by Bassani
G. F., Inguscio M. and Hänsch T. W. (Springer Verlag, Berlin, Heidelberg, New
York) 1989, pp. 93-102.
[13] Niering M. et al., Phys. Rev. Lett., 84 (2000) 5496.
[14] de Beauvoir B. et al., Eur. Phys. Lett. D, 12 (2000) 61.
[15] Udem Th. et al., Phys. Rev. Lett., 82 (1999) 3568.
[16] Gerginov V. et al., Phys. Rev. A, 73 (2006) 032504.
[17] Bize S. et al., Phys. Rev. Lett., 90 (2003) 150802.
[18] Peik E. et al., Phys. Rev. Lett., 93 (2004) 170801.
[19] Zimmermann M. et al., Laser Phys., 15 (2005) 997.
[20] Kourogi M. et al., IEEE J. Quantum Electron., 31 (1995) 2120.
[21] Brothers L. R. and Wong N. C., Opt. Lett., 22 (1997) 1015.
[22] Imai K. et al., IEEE J. Quantum Electron., 34 (1998) 1998.
[23] Spence D. E., Kean P. N. and Sibbett W., Opt. Lett., 16 (1991) 42.
[24] Diels J. C. and Rudolph W., Ultrashort Laser Pulse Phenomena (Elsevier, Amsterdam,
Heidelberg, Tokyo) 2006.
[25] Eckstein J. N., High Resolution Spectroscopy using Multiple Coherent Pulses (Thesis,
Stanford University, USA) 1978.
[26] Frequency Standards in the Optical Spectrum, in The Hydrogen Atom, edited by Bassani
G. F., Inguscio M. and Hänsch T. W. (Springer Verlag, Berlin, Heidelberg, New York)
1989, pp. 123-133.
[27] Holzwarth R. et al., Appl. Phys. B, 73 (2001) 269.
[28] Udem Th. et al., Opt. Lett., 24 (1999) 881.
[29] Diddams S. A. et al., Opt. Lett., 27 (2002) 58.
[30] Ma L. S. et al., Science, 303 (2004) 1843.
[31] Stenger J. and Telle H. R., Opt. Lett., 25 (2000) 1553.
[32] Krausz F. et al., IEEE J. Quantum Electron, 28 (1992) 2097.
[33] Hasegawa A. and Tappert F., Appl. Phys. Lett., 23 (1973) 142.
[34] Agrawal G. P., Nonlinear Fiber Optics (Academic Press, New York) 2001.
[35] Sutter D. H. et al., Opt. Lett., 24 (1999) 631.

[36] Morgner U. et al., Opt. Lett., 24 (1999) 411.
[37] Siegman A. E., Lasers (University Science Books, Mill Valley Ca USA) 1986.
[38] Bramwell S. R., Kane D. M. and Ferguson A. I., Opt. Commun., 56 (1985) 112.
[39] Hollberg L. et al., see fig. 9 in IEEE J. Quantum Electron., 37 (2001) 1502.
[40] Rush D. W., Ho P. T. and Burdge G. L., Opt. Commun., 52 (1984) 41.
[41] Bartels A. et al., Opt. Lett., 29 (2004) 1081.
[42] Swann W. C. et al., Opt. Lett., 31 (2006) 3046.
[43] Knight J. C. et al., Opt. Lett., 21 (1996) 1547.
[44] Ranka J. K. et al., Opt. Lett., 25 (2000) 25.
[45] Russell P. St. J., Science, 299 (2003) 358.
[46] Birks T. A., Wadsworth W. J. and Russell P. St. J., Opt. Lett., 25 (2000) 1415.
[47] Holzwarth R. et al., Laser Phys., 11 (2001) 1100.
[48] Bartels A., Dekorsy T. and Kurz H., Opt. Lett., 24 (1999) 996.
[49] Hoi M. et al., Phys. Rev. Lett., 96 (2006) 243401.
[50] Corwn K. L. et al., Phys. Rev. Lett., 90 (2003) 113904.
[51] Holman K. W. et al., Opt. Lett., 28 (2003) 851.
[52] Holzwarth R., PhD thesis Ludwig-Maximilians-Universität Munich (2001).
[53] Dudley J. M. et al., J. Opt. Soc. Am. B, 19 (2002) 765.
[54] Ell R. et al., Opt. Lett., 26 (2001) 373.
[55] Fortier T. M. et al., Opt. Lett., 28 (2003) 2198.
[56] Matos L. et al., Opt. Lett., 29 (2004) 1683.
[57] Fortier T. M. et al., Opt. Lett., 31 (2006) 1011.
[58] Morgner U. et al., Phys. Rev. Lett., 86 (2001) 5462.
[59] Diddams S. A. et al., IEEE J. Quantum Electron, 9 (2003) 1072.
[60] Fuji T. et al., Opt. Lett., 30 (2005) 332.
[61] Nelson L. E. et al., Appl. Phys. B, 65 (1997) 277.
[62] Kubina P. et al., Opt. Express, 13 (2005) 909.
[63] Adler F. et al., Opt. Express, 12 (2004) 5872.
[64] McFerran J. J. et al., Opt. Lett., 31 (2006) 1997.
[65] See, for example, Gardener F. M., Phaselock Techniques (John Wiley & Sons, New
York) 1979.
[66] Walls F. L. and DeMarchi A., IEEE Trans. Instrum. Meas., 24 (1975) 210.
[67] Telle H. R., in Frequency Control of Semiconductor Lasers, edited by Ohtsu M. (Wiley,
New York) 1996, pp. 137-167.
[68] Private communication T. Brabec.
[69] Haus H. A. and Ippen E. P., Opt. Lett., 26 (2001) 1654.
[70] Xu L. et al., Opt. Lett., 21 (1996) 2008.
[71] Helbing F. W. et al., Appl. Phys. B, 74 (2002) S35.
[72] Witte S. et al., Appl. Phys. B, 78 (2004) 5.
[73] Prevedelli M., Freegarde T. and Hänsch T. W., Appl. Phys. B, 60 (1995) S241.
[74] Fischer M. et al., Phys. Rev. Lett., 92 (2004) 230802.
[75] Santarelli G. et al., Phys. Rev. Lett., 82 (1999) 4619.
[76] Uzan J. P., Rev. Mod. Phys., 75 (2003) 403.
[77] Dirac P. A. M., Nature (London), 139 (1937) 323.
[78] Webb J. K. et al., Phys. Rev. Lett., 87 (2001) 091301.
[79] Murphy M. T., Webb J. K. and Flambaum V. V., Mon. Not. R. Astron. Soc., 345
(2003) 609.
[80] Quast R. et al., Astron. Astrophys., 417 (2004) 853.
[81] Srianand R. et al., Phys. Rev. Lett., 92 (2004) 121302.
[82] Shlyakhter A. I., Nature, 264 (1976) 340.

[83] Fujii Y. et al., Nucl. Phys. B, 573 (2000) 377.
[84] Lamoreaux S. K. and Torgerson J. R., Phys. Rev. D, 69 (2004) 121701(R).
[85] Murphy M. T. et al., arXiv:astro-ph/0612407v1.
[86] Murphy M. T., Webb J. K. and Flambaum V. V., arXiv:astro-ph/0611080 v3.
[87] Calmet X. and Fritzsch H., Phys. Lett. B, 540 (2002) 173.
[88] Schneider T., Peik E. and Tamm Chr., Phys. Rev. Lett., 94 (2005) 230801.
[89] Diddams S. A. et al., Science, 293 (2001) 825.
[90] Mohr P. J. and Taylor B. N., Rev. Mod. Phys., 77 (2005) 1.
[91] Gabrielse G. et al., Phys. Rev. Lett., 97 (2006) 030802.
[92] Wicht A. et al., Phys. Scr., T02 (2002) 82.
[93] Cladé P. et al., Phys. Rev. Lett., 96 (2006) 03301.
[94] Bradley et al., Phys. Rev. Lett., 83 (1999) 4510.
[95] Pachucki K. and Jentschura U. D., Phys. Rev. Lett., 91 (2003) 113005.
[96] Murphy M. T. et al., sumitted to Mon. Not. R. Astron. Soc.
[97] Apolonski A. et al., Phys. Rev. Lett., 85 (2000) 740.
[98] Kienberger R. et al., Nature, 427 (2004) 817.
[99] Baltuška A. et al., Nature, 421 (2003) 611.
[100] Wahlström C. G. et al., Phys. Rev. A, 48 (1993) 4709.
[101] Brabec T. and Krausz F., Rev. Mod. Phys., 72 (2000) 545.
[102] Eden J. G., Prog. Quantum Electron., 28 (2004) 197.
[103] Paulus G. G. et al., Phys. Rev. Lett., 85 (2000) 253004.
[104] Seres J. et al., Nature, 433 (2005) 596.
[105] Goulielmakis E. et al., Science, 305 (2004) 1267.
[106] Gerginov V. et al., Opt. Lett., 30 (2005) 1734.
[107] Fendel P. et al., Opt. Lett., 32 (2007) 701.
[108] Baklanov Ye. F. and Chebotayev V. P., Appl. Phys. Lett., 12 (1977) 97.
[109] Meshulach D. and Silberberg Y., Nature, 396 (1998) 239.
[110] Eckstein J., Ferguson A. I. and Hänsch T. W., Phys. Rev. Lett., 40 (1978) 847.
[111] Snadden M. J. et al., Opt. Commun., 125 (1996) 70.
[112] Marian A. et al., Science, 306 (2004) 2063.
[113] Fortier T. M. et al., Phys. Rev. Lett., 97 (2006) 163905.
[114] Witte S. et al., Science, 307 (2005) 400.
[115] Cavalieri S. et al., Phys. Rev. Lett., 89 (2002) 133002.
[116] Zinkstok R. Th. et al., Phys. Rev. A, 73 (2006) 061801(R).
[117] Gohle Ch. et al., Nature, 436 (2005) 234.
[118] Jones R. J., Moll K. Thorpe M. and Ye J., Phys. Rev. Lett., 94 (2005) 193201.
[119] Barnes J. A. et al., IEEE Trans. Instrum. Meas., 20 (1971) 105.
[120] Webster S. A. et al., Phys. Rev. A., 65 (2002) 052501.
[121] See, for example, contributions of Madej A. A., Bernard J. E., Riehle F. and
Helmcke J., in Frequency Measurement and Control, edited by Luiten A. N. (Springer
Verlag, Berlin, Heidelberg, New York) 2001.
[122] Evenson K. M. et al., Appl. Phys. Lett., 22 (1973) 192.
[123] Peik E., Schneider T. and Tamm Chr., J. Phys. B, 39 (2006) 145.
[124] Brusch A. et al., Phys. Rev. Lett., 96 (2006) 103003.
[125] Quinn T. J., Metrologia, 40 (2003) 103.
[126] Editor’s note: Documents concerning the new definition of the metre, Metrologia, 19
(1984) 163.
[127] Quinn T. J., Metrologia, 30 (1993/1994) 523.
[128] Quinn T. J., Metrologia, 36 (1999) 211.
[129] Wineland D. J. et al., Phys. Rev. A, 36 (1987) 2220.

[130] Blatt R., Gill P. and Thompson R. C., J. Mod. Opt., 39 (1992) 193.
[131] Madej A. A. and Bernard J. E., Frequency Measurement and Control: Topics in
Applied Physics, edited by A. N. Luiten, 79 (2001) 153.
[132] Riehle F., Frequency Standards: Basics and Applications (Wiley-VCH, Weinheim) 2004.
[133] Dehmelt H., Bull. Am. Phys. Soc., 20 (1975) 60.
[134] Wineland D. J. and Dehmelt H., Bull. Am. Phys. Soc., 20 (1975) 637.
[135] Wineland D. J. et al., Opt. Lett., 5 (1980) 245.
[136] Nagourney W., Sandberg J. and Dehmelt H., Phys. Rev. Lett., 56 (1986) 2797.
[137] Itano W. M., J. Res. Natl. Inst. Stand. Technol., 105 (2000) 829.
[138] Dubé P. et al., Phys. Rev. Lett., 95 (2005) 033001.
[139] Tamm Chr. et al., to be published in IEEE Trans. Instr. Meas.
[140] Margolis H. S. et al., Science, 306 (2004) 1355.
[141] Ito N. et al., Opt. Commun., 109 (1994) 414.

[142] Zibrov A. S. et al., Appl. Phys. B, 59 (1994) 327.
[143] Kersten P. et al., Appl. Phys. B, 68 (1999) 27.
[144] Beausoleil R. G. and Hänsch T. W., Phys. Rev. A, 33 (1986) 1661.
[145] Sterr U. et al., Appl. Phys. B, 54 (1992) 341.
[146] Wilpers G. et al., Phys. Rev. Lett., 89 (2002) 230801.
[147] Sterr U. et al., C. R. Physique, 5 (2005) 845.
[148] Sterr U. et al., Phys. Rev. A, 72 (2005) 062111.
[149] Wilpers G., Oaetes C. W. and Hollberg L. W., Appl. Phys. B, 85 (2006) 31.
[150] Sorrentino F. et al., arXiv:physics/0609133 v1 (2006).
[151] Itano W. M. et al., Phys. Rev. A, 47 (1993) 3554.
[152] Bordé Ch. J. et al., Phys. Rev. A, 30 (1984) 1836.
[153] Bordé Ch. J., Phys. Lett. A, 30 (1989) 10.
[154] Stoehr H. et al., Opt. Lett., 31 (2006) 736.
[155] Dick G. J., Proceedings of 19th Annual Precise Time and Time Interval (PTTI)
Applications and Planning Meeting, Redondo Beach, CA, 1987 (U. S. Naval Observatory)
1987, pp. 133-147.
[156] Quessada A. et al., J. Opt. B: Quantum Semiclassical Opt., 5 (2003) S150.
[157] Katori H., Ido T. and Kuwata-Gonokami M., J. Phys. Soc. Jpn., 68 (1999) 2479.
[158] Katori H., in Proc. 6th Symposium on Frequency Standards and Metrology, edited by
P. Gill (World Scientific, Singapore) 2002, p. 323.
[159] Takamoto M. et al., J. Phys. Soc. Jpn., 75 (2006) 104302.
[160] Boyd M. M. et al., Phys. Rev. Lett., 98 (2007) 083002.
[161] Le Targat R. et al., Phys. Rev. Lett., 97 (2006) 130801.
[162] Barber Z. W. et al., Phys. Rev. Lett., 96 (2006) 083002.
[163] Wineland D. J. et al., in Proc. 6th Symposium on Frequency Standards and Metrology,
edited by P. Gill (World Scientific, Singapore) 2002, p. 361.
[164] Schmidt P. O. et al., Science, 309 (2005) 749.
[165] Peik E. and Tamm Ch., Europhys. Lett., 61 (2003) 161.
DOI 10.3254/978-1-61499-818-1-301
The SI and quantum metrology(∗ )

Carl J. Williams(∗∗ )
Physical Measurement Laboratory (PML), U.S. National Institute of Standards and Technology
Gaithersburg, MD, USA
Summary. — The first lecture begins with a brief overview of the origin of the
metric system and the international system of units (SI). Even from the earliest
days the concept was to create a system of units based on nature. However, as
one can easily see, we actually created in practice an artifact-based system of units
that has since 1960 been slowly transforming to a set of units that are increasingly
based on nature. The first standard was the unit of time followed by length and
for all practical purpose electrical units have been quantum based since the early
1990 s. With the planned redefinition in 2018 the new SI or quantum SI will be
based on a set of principles that are based on the Standard Model of Physics. This
lecture provides a description of how the SI began and how we have evolved to the
present day. Toward the end of the lecture a summary of quantum standards and
metrology today is provided before presenting a brief picture of the future that will
include integrated embedded standards that provide far better metrology for both
the factory floor and consumer technology.
The United States’ National Institute of Standards and Technology (NIST) is playing
a major role in the worldwide effort to base the International System of Units (SI)
on fundamental constants. In many cases, units will be redefined in terms of invariant
quantum phenomena. So to a great degree, the progress of modern metrology will depend
on our ability to make increasingly sensitive, accurate, and repeatable measurements of
quantum processes.
(∗ ) Official contribution of the National Institute of Standards and Technology, not subject to
copyright in the United States.
(∗∗ ) E-mail: carl.williams@nist.gov
301
302 Carl J. Williams
Within NIST, that effort is the responsibility of an organization called the Physical
Measurement Laboratory (PML). PML’s core mission is to realize, disseminate, and
advance the SI in the United States. PML seeks to ensure that the SI is maintained and
improved, realized in practice, disseminated for routine uses, and disseminated for new
and novel uses. And it is at the forefront of the measurement science required to enable
the impending redefinition of the SI units, planned for 2018.
1. – An old new idea
The redefinition is very much in keeping with the philosophy that underlay the cre-
ation of the decimal Metric System on December 10, 1799 — namely, that humanity
should have a set of measurement units based on nature and thus realizable in the same
way from region to region or country to country. The French Revolution celebrated
“natural law” over the rule of hereditary kings; as a result, it was very important for the
metric units to be based on natural standards, rather than the foot size or arm length of
an arbitrary local monarch.
Thus the length of the “meter” was defined as 1/10 000 000 the distance between the
equator and the North Pole, along a meridian that ran through Paris. An international
committee (including representatives from France, the Netherlands, Denmark, Spain,
Italy, and Switzerland) certified the surveyors’ results and adopted an artifact standard
to memorialize what the “meter” was. Similarly, the kilogram was specified as the weight
of 1000 cubic centimeters of water at the melting point of ice, and a physical standard
was adopted to memorialize that, as well.
Both standards were, in principle, based on nature, but were in reality artifacts.
They were, however, very good artifacts. The actual meter is 0.02% too short (0.2 mm)
due to a miscalculation of the flattening of the Earth. (The distance ended up being
10 001.9657 km.)
During the 1800s, the Metric System came into common use, to a greater or a lesser
extent, for various purposes in various countries. In the United States, Congress explicitly
allowed the use of the Metric System in all commercial dealings and contracts in the
Metric Act of 1866. While the author of the legislation was enthusiastic about the Metric
System and looked forward to it eventually becoming the usual system of measurement
in the U.S., the law itself merely made the Metric System lawful throughout the United
States, provided official definitions and conversion tables, and preempted certain state
laws that had forbidden its use.
In 1878, the United States became a signatory to the Convention of the Metre (es-
tablished by representatives of 17 nations in 1875, and often called the Treaty of the
Meter in the United States), a diplomatic treaty that created the CGPM (Conférence
générale des poids et mesures), the CIPM (Comité international des poids et mesures)
— an advisory body to the CGPM, and the BIPM (Bureau international des poids et
mesures) — an organization based at Sèvres, France with custody of the International
Prototype Kilogram and responsibility for providing metrology services for the GCPM
and CIPM, among other tasks.
The SI and quantum metrology 303
At the dawn of SI in 1960, the CGPM moved away from the artifact standard of the
meter, the prototype meter bar kept at the BIPM, and instead redefined the “meter” in
terms of a standard wavelength of light.
This changed the “meter” from being an arbitrary length of a classical object to a
specific multiple of a discrete, quantum mechanical object.
Then, in 1983, the definition of the meter changed yet again. Given that the standard
for time (the second—itself based on a quantum-mechanical standard) afforded much
greater precision, the CGPM opted to make the speed of light a defined constant (no
uncertainty), allowing a high-precision connection between time and distance.
The 1983 definition allows realization of the meter using interferometry of any laser
line, not just the wavelength of light from a krypton discharge lamp.
The Convention of the Metre was initially concerned only with weights and measures.
It was amended in 1921 to coordinate measures of electrical units, establish and keep
standards of electrical units and their “test copies”, determine the physical constants,
and coordinate “similar determinations affecting other institutions”. In 1954 the CGPM
adopted six base units (meter, kilogram, second, ampere, Kelvin, and candela). In 1960,
CGPM adopted the name “Système International d’Unités” (SI) for this system adding
the seventh and final base unit, the mole, in 1971. The process to revise and improve the
units in a way that benefits the system as a whole and bases them on nature had truly
begun.
2. – Quantum standards and redefinition
The establishment of the SI was by no means the end of the process set in motion
in 1799. Rather it was, as Winston Churchill might have put it, only the end of the
beginning. The rapid technological advances that occurred at the dawn of the Space
Age (e.g., lasers, fast electronics) accelerated a process of continuous improvement in the
definitions and the realizations of the units of measurement, notably in terms of invariant
quantum phenomena for several of the base units.
Interestingly, Max Planck had suggested that possibility as early as 1900:
“The two constants [the Planck constant, h, and the Boltzmann constant, k]. . . which
occur in the equation for radiative entropy offer the possibility of establishing a system
of units for length, mass, time, and temperature which are independent of specific bod-
ies or materials and which necessarily maintain their meaning for all time and for all
civilizations, even those which are extraterrestrial and non-human”.
Planck’s language is remarkably similar to that used by the Marquis de Condorcet
when he transferred the original French length and mass standards to the Archives de la
Republique in 1799. More on the new SI can be found in Dave Newell’s Physics Today
article [1].
In that spirit, at the dawn of SI in 1960, the CGPM moved away from the artifact
standard of the meter, the prototype meter bar kept at the BIPM, and instead redefined
the meter in terms of a standard wavelength of light. This changed the meter from being
an arbitrary length of a classical object to a specific multiple of a discrete, quantum

mechanical object:
The meter is the length equal to 1 650 763.73 wavelengths in vacuum of the radiation
corresponding to the transition between the levels 2p10 and 5d5 of the krypton-86 atom.
Then, in 1983, the definition of the meter changed yet again. Given that the standard
for time (the second) afforded much greater precision, the CGPM opted to make the speed
of light a defined constant (no uncertainty), allowing a high-precision connection between
time and distance: the meter was redefined as the length of the path travelled by light
in vacuum during a time interval of 1/299 792 458 of a second.
By then, of course, the second had already been given a quantum mechanical basis. In
1967, it was defined as the duration of 9 192 631 770 periods of the radiation corresponding
to the transition between the two hyperfine levels of the ground state of the 133 Cs atom.
This remains the definition today. The world’s best fountain clock, NIST’s F2 (placed
in service in 2014) can measure this frequency to an uncertainty of one part in 1016 , the
equivalent of 1 second in 300 million years.
By then, quantum standards for the volt (via the Josephson effect, Nobel Prize 1973)
and the ohm (via the quantum Hall effect, Nobel Prize 1985) had proven so robust that
in 1987 the CGPM (Resolution 6) established conventional electrical units!
3. – Quantum standards and quantum metrology today
Continuous progress in measurement science and technology promises near-term im-

provements on the already remarkable accuracy of quantum-based standards.
.
3 1. Time. – For example, the cesium fountain clock operates at microwave frequency,
permitting minimal measurable time intervals of about 1/9 000 000 000 second (9 GHz.)
But the advent of optical clocks — which rely on visible-light frequencies in the range
of hundreds of terahertz (1012 Hz) generated by trapped ions or atoms in atomic lat-
tices — can make possible time resolution at least two orders of magnitude finer, with
uncertainties in the range of parts in 1018 .
Determining absolute optical frequencies, however, is extremely difficult, and the
full potential of optical-clock technology could not be realized without the invention of
frequency combs, pioneered and continuously improved at NIST. Frequency combs use
the output of mode-locked lasers to provide an absolute scale for determining optical
frequencies.
Since 2005, optical frequency standards using ions or neutral atoms of Ca, Sr, Yb,
Al, Hg, Sr, and Yb have shown better fractional uncertainty and estimated systematic
uncertainty than the primary cesium standard, and are now under consideration for a
possible redefinition of the second being discussed for 2026.
.
3 2. Electrical units. – The measurement and realization of key electrical units have
already been transformed by the application of quantum science, and the proposed re-
definition of the ampere will be quantum-based.
The worldwide standard for both AC and DC voltage relies on the Josephson effect
(Josephson 1962, Nobel Prize 1973). NIST can now build 10 V programmable Josephson
DC voltage chips employing 32 microwave channels and 300 000 Josephson junctions.
These voltage standards are typically cooled by liquid helium, but a new NIST design
incorporates a fully automated electronic cryocooler and enables the user to employ
commercial instrumentation.
In addition, precisely controlled AC voltages up to 2 V are now possible using a NIST
design for junction arrays driven by an arbitrary waveform synthesizer. The latest ex-
perimental models reach a quantum accuracy up to 1 MHz with precision control over
amplitude and phase of arbitrary waveforms via pulse pattern and timing. Using 12 800
Josephson junctions per array, the junctions act as perfect quantizers, converting ar-
bitrary input pulses to output voltage pulses, thus making the device an intrinsically
accurate, quantum voltage standard.
The internationally recognized electrical resistance standard is based on the quan-
tum Hall effect (QHE: von Klitzing 1980, Nobel Prize 1985), which displays exquisitely
precise quantization. Conventional instruments at NIST and elsewhere employ gallium
arsenide structures for the two-dimensional electron gas and run at 12.9 kΩ at cryogenic
temperatures.
They are the present internationally recognized standard for the ohm. However, they
are difficult to scale and operation requires specialized equipment and training. As a
result, there is considerable interest in using graphene instead of GaAs: QHE has been
observed in graphene at room temperature. Graphene-based devices, under development
at NIST, NPL, PTB and other laboratories, are also much more easily scalable, and hold
promise as the future basis for realization of the ohm.
NIST has developed novel techniques to grow graphene on SiC and to process the ma-
terial into high-current QHR devices. The devices are compatible with NIST’s existing
highly customized measurement infrastructure, and can be used directly with commer-
cially available room temperature bridge systems. The latest versions are not yet robust,
but rapid progress is expected.
The ampere — the only electrical SI base unit — has not yet been placed on a
quantum footing. Its current definition, dating from 1946, is
The ampere is that constant current which, if maintained in two straight parallel
conductors of infinite length, of negligible circular cross-section, and placed 1 meter
apart in vacuum, would produce between these conductors a force equal to 2 × 10−7
newton per meter of length. (9th CGPM, Resolution 2)
The ampere has mostly not been realized because the definition is infinitely com-
plex. Approaches to realizing it have been demonstrated [2] with the most common
being counting of electrons or more accurately measuring the flux of single charges in
single-electron transistors (SETs). In practice one uses SETS made from gallium ar-
senide, aluminum, and more recently in silicon to create a conductance channel and then
measures the current.
Twenty years ago, NIST demonstrated an aluminum SET at 1 pA at an uncertainty of

1.5 × 10−8 [3]. A few years later using Ohm’s law (V = IR) and taking the voltage from
the Josephson effect and resistance from the QHE, NIST demonstrated that we could
close the Quantum Metrology Triangle at an uncertainty of 1 × 10−6 [4]. The resulting
experiment was extraordinarily demanding and difficult to pull off due to the existence
of charge noise in the aluminum SET. NIST temporarily abandoned this effort due to its
difficulty.
Today, numerous NMI’s around the world are again exploring SETs as a potential
current standard. NIST is at present developing single-electron transport structures in
silicon. It is expected that each channel will produce 100 pA; 100 of the channels placed
in parallel reaches 10 nA. PTB as part of the Quantum Ampere project in Europe has
recently demonstrated a gallium arsenide SET at 100 pA at an uncertainty of 2 × 10−8 .
In the next few years it is expected that multiple NMI’s will close the quantum metrology
triangle and surpass the NIST result from the last Century.
According to the proposed redefinition, the ampere’s magnitude will be set by fixing
the numerical value of the elementary charge making the use of SETs a direct realiza-
tion. This will clearly hasten investigation of the quantum metrology triangle and lead
to instrumentation that combines two legs of the triangle into a single device that in
principle would generate all electrical quantities from quantum standards. Integration of
two of these quantum standards on a single device would revolutionize electrical instru-
mentation, and could result in an on-chip application of Ohm’s law making it possible
to create self-calibrating instruments.
The NIST Quantum Watt already uses a voltage reference from a Programmable
Josephson Voltage Standard applied to a power meter (5 A, 120 V). Voltage and current
are adjusted to minimize difference between sampled and reference voltage to measure
power.
Capacitance does not currently have a quantum basis. The most accurate standard is
the calculable capacitor. NIST is in the process of building a new unit as a replacement
or back-up for the existing apparatus. The new model will be continuously tunable, and
is expected to become the most linear capacitance standard. It can possibly be used to
measure the non-linearity of capacitance bridges and could lead to a calibration method
to correct the bridges’ nonlinearity.

Similarly, impedance does not now have a quantum basis. But there is interest in using
phenomena observed in alternating-current applications of the quantum Hall resistance
— often attributed to capacitance effects that alter longitudinal resistance — as the
foundation for what could eventually become an impedance standard.
These projects are for the future. But it seems clear now that eventually electrical
units will be brought back into the SI and conventional units will be abrogated.
.
3 3. From instrumenting the SI to embedded measurements. – The “New SI” —com-
prising the impending redefinitions of 2018— will be the quantum SI. Of the seven base
units, two (meter and second) are already based on quantum phenomena, and the candela
is defined in terms of an invariant equivalent to a fundamental constant. The kilogram,
ampere, kelvin, and mole will be defined in terms of the Planck constant, elementary
charge, Boltzmann constant, and Avogadro constant, respectively.
This action will inevitably change the way in which NIST (and possibly other NMIs)
fulfills its mission to disseminate the SI. NIST traditionally has provided that service in
one of four ways:
1) The customer sends NIST an artifact. NIST measures it and returns it. Examples:
gauge blocks and other artifacts used in dimensional metrology.
2) The customer sends NIST an instrument. NIST calibrates it and returns it. Ex-
amples: thermometers and pressure gauges.
3) The customer contacts NIST and orders one or more of NIST’s 1200 Standard
Reference Materials with fully characterized engineering, chemical, and physical
properties. Examples: nanomaterials, thermoelectric materials, organic and inor-
ganic materials.
4) NIST and one or more collaborators jointly observe and measure phenomena. Ex-
ample: performance of the GPS satellite constellation, precision transfer of time
and frequency standards.
The first two activities can sometimes be very time-consuming. Some limitations
of the traditional model include: significant down-time for customer’s standard or de-
vice; need for back-up standards/devices; expensive NIST staff time and administrative
burden; need for periodic recalibrations; customer uncertainty about whether a device re-
mains in calibration specifications; limited range of available calibrations/measurements
that may be relatively inflexible compared to changing customer needs; and inability to
easily deploy standards/devices on the factory floor and integrate them with existing
systems.
This situation will change with the advent of more quantum-mechanical standards
for calibrations and measurements. At present, NIST offers 591 measurement services in
eight metrology areas. Some of those are quantum-based on quantum-mechanical stan-
dards: length, voltage, resistance, power and energy, force, time dissemination, frequency
measurement, oscillator characterization, and noise measurement.
In the very near future, however, NIST will also be able to make quantum-based
measurements for electromagnetic field strength, radioactivity, mass, acoustics and vi-
bration, acceleration, photometry, laser power and energy, thermometry, and pressure
and vacuum.
At the same time, NIST has made a long-term commitment to developing quantum-
based standards — many of them self-calibrating — that can be maintained at the user’s
location 24/7/365 without periodic return to NIST, providing NMI-quality measurements
available directly where the user needs them. This capability, which will feature many
dual-platform standards and sensors that permit measurement of multiple units on the
same device, has the potential for much broader and more flexible range of measurements
and standards.
Achieving this goal will require design and testing of complementary technologies
simultaneously. For example, the success of NIST’s Chip-Scale Atomic Clock (CSAC)
required significant advances in solid-state lasers (in particular, vertical cavity surface
emitting lasers), microelectromechanical system (MEMS) fabrication techniques, and
precision control of atomic vapors in microcells. Once the requisite technologies are in
place, development can move swiftly. In the case of the CSAC, NIST developed a working
prototype in 2004 [5] and the device was commercialized by 2011.
Because these devices have uncertainties of a few parts in 1011 , models with improved
long-term stability could provide a distributed back-up for the Global Positioning System,
among other uses.
NIST is now at work on numerous technologies that may enable disruptive change,
among them: solid-state lasers, MEMS fabrication, nanoelectronics, microfluidics, inte-
grated photonics, superconducting systems, and various quantum-based standards and
phenomena.
The desired outcome is a 21st century toolkit can enable the development of a new gen-
eration of artifacts and instruments with capabilities that far exceed those traditionally
used for traceability. Indeed, in some cases, they might rival the capabilities of an NMI.
The implications of the new quantum metrology are as exciting as they are broad. For
example, the role and character of NMIs may change, moving away from conventional
calibration and into more monitoring and research.
At the same time, widely deployed quantum standards will require changes in legal
metrology that can accommodate new definitions of traceability. Social impacts will
range from tiny embedded sensors that monitor conditions in major infrastructure such
as bridges to dependable devices that can limit overexposure to x-rays and other ionizing
radiation on-site in clinical settings.
In conclusion the pending redefinition of the SI will create a far more robust system of
units, based on the so-called Standard Model of Physics — hence the colloquial reference
to the Quantum SI. This “New SI” will realize the goal of the original decimal Metric
System of 1799. This summary is both brief and very NIST centric. However it is essential
to state clearly that this is a worldwide effort resulting from many great scientists from
academia, scientific organizations, and National Measurement Institutes (NMIs) around
the world. While NIST is often a leader in pushing measurement science, numerous other
NMI’s both make original contributions and frequently lead the way.
REFERENCES
[1] Newell D., Phys. Today, 67, No. 7 (2014) 35.

[2] Driscoll R. L., J. Res. Natl. Bur. Stand., 60 (1958) 287.
[3] Keller M. W., Martinis J. M., Zimmerman N. M. and Steinbach A. H., Appl. Phys.
Lett., 69 (1996) 1804.
[4] Keller M. W., Eichenberger A. L., Martinis J. M. and Zimmerman N. M., Science,
285 (1999) 1706.
[5] Knappe S., Schwindt P. D. D., Shah V., Hollberg L. and Kitching J., Opt. Express,
13 (2005) 1249.
DOI 10.3254/978-1-61499-818-1-309
The future of quantum-based measurements

and the SI(∗ )
Summary. — The second lecture in this series starts by providing a definition of

quantum technology, quantum metrology, and quantum-based measurements along
with a short summary of the properties required of the devices and underlying
technology. These properties include that the devices must be deployable, usable,
flexible, and manufacturable. The lecture then provides a brief introduction to
single-photon technologies including their potential application as a source of cer-
tifiable random numbers. The lecture continues with an explanation of how this
technology may be the basis for a future redefinition of the candela. This is followed
by an overview of photon pressure and its use for calibrating small mass and force.
The lecture then presents new concepts for measuring ultra-high vacuum and how
atom-based sensors provide a means for measuring electric and magnetic fields in
addition to chip-scale atomic clocks. The lecture concludes with a brief description
of the dissemination of the quantum SI including the development of embedded
“chip-scale” metrology in the broader infrastructure.
1. – Introduction
The U.S. National Institute of Standards and Technology (NIST), along with other
NMIs, is preparing for a new era in the history of metrology made possible by the growing
use of quantum-based standards and measurements. This development will dramatically
alter many of our conventional practices, and will influence how NMIs approach their
missions.
309
That is as it should be. Metrology is not about simply maintaining the status quo and
our traditional responsibility to realize, maintain, and disseminate units. Metrology must
constantly evolve to meet the ever-changing needs of commerce, science, and engineering.
It is the infra-technology that drives and supports the world economy, figuring critically
in 80% of all global trade.
Quantum metrology is desirable in itself because it promises a new level of precision
and accuracy. But is also in keeping with the philosophy behind the upcoming redefinition
of the SI units, expected in 2018, in which units will be based on fundamental constants
that are based on quantum phenomena.
What will quantum metrology mean in practical terms? NIST has developed some
definitions that help guide and focus our efforts. They may be useful to others.
Quantum technology: Comprises materials, devices, and systems that provide indi-
vidual sensitivity to quanta in the form of photons, spin, charge, and other quantized
quantities. It also enables technology that creates high-efficiency detection and trans-
port of quanta — e.g. ultraviolet (UV) fiber for improved UV lasers, materials with less
charge-trapping or fewer stray electron spins, ion traps with lower decoherence, and other
methods to minimize noise and prolong coherence times.
Quantum metrology: Exploits quantum technology to improve measurement science to
make better detectors, sources, magnetic or electric field sensors, improved quantum Hall
resistance (QHR) devices, improved single-electron transistors (SETs), and associated
devices. It includes clocks, magnetometers, self-correcting interferometers, and systems
for the transduction of signals — both classical and quantum. And it embraces squeezing
technologies and techniques that allow measurement beyond the Standard Quantum
Limit (SQL) which — together with transduction — enables amplification and parametric
amplification of signals beyond the SQL.
Quantum-based measurement: Utilization of quantum technology, quantum metrol-
ogy, superposition, entanglement, and/or squeezing to improve a physical measurement.
The emerging technologies that will make these advances possible are likely to enable
disruptive change. At a minimum, the following are expected to have a significant impact:
– Solid state lasers (e.g., VCSELs) that offer quantum-based precision tuning, are
deployable in sub-millimeter spaces, and are extremely well-suited to coupling with
optical fiber.
– Microelectromechanical systems (MEMS) — especially micro- and nano-
fabrication, nanoelectronics, microfluidics, and integrated photonics — which will
enable ensembles of highly sensitive standards and sensors to operate in very small
dimensions.
– Superconducting systems, including increasingly sophisticated arrangements of
Josephson junctions for electrical metrology.
– Quantum-based standards and phenomena, based on advances in knowledge of
fundamental atomic and molecular properties, properties of new materials, and
ultracold systems.
The future of quantum-based measurements and the SI 311
NIST will utilize these and other innovations to develop devices for a new paradigm in
the dissemination of SI-traceable standards in which users will be provided with standards
and measurement tools that require little, if any, calibration, and are available on-site
at any time. The success of this “embedded standards” effort will begin to transform
the role of NIST — and perhaps other NMIs — away from an emphasis on measure-
ment and calibration of user artifacts and instruments to a concentration on monitoring
measurement activity and providing basic metrology research.
NIST has determined that the new generation of devices must meet four essential
criteria:
– They must be deployable directly in a user’s factory, laboratory, equipment system,
or even in homes.
– They must be maximally usable, which in general will mean devices that are rel-
atively small, have low power consumption, are sufficiently rugged, are easily op-
erated, and can be conveniently integrated into the user’s existing systems and
instrumentation.
– They must be flexible in the sense of having the capability to provide a range
of SI-traceable measurements and standards (often quantum-based) while being
adaptable to a variety of customer needs and applications.
– They must be manufacturable, with the potential for production costs commensu-
rate with the intended applications, whether low cost for large markets and broad
deployment or higher costs that are still acceptable for high-value applications.
NIST has identified 12 initial areas for embedded standards research and development
that will have maximum impact on modern, quantum-based metrology:
– Photonic sensing of thermodynamic quantities. These projects include methods
of harnessing interferometry and other photonic procedures to sense variation in
temperature (as heat-related dimensional changes alter the propagation of light)
and pressure (which affects the index of refraction in gases, and thus the behavior
of light passing through the gas).
– Quantum-based electrical standards. Work in this area comprises Josephson voltage

and Hall-effect resistance standards, as well as controlling the movement of single
electrons to realize the ampere and — potentially — quantum-based measurement
of impedance and capacitance.
– Atom-based measurements in vapor cells. This technology is the basis for NIST’s
patented chip-scale atomic clock, in which a VCSEL laser is tuned to resonance with
a gas of cesium-133 atoms. The behavior of light as it passes through tiny volumes
of trapped gas is also being used to make exquisitely sensitive magnetometers, and
can be adapted to standards for length, temperature, and current and voltage.
– Optical/photonic input and output. Underlying many of the key technologies re-
quired for a new generation of devices is the ability to transfer light-based signals
rapidly and with excellent fidelity. The quantum effects that are measured by chip-
scale devices are typically small, faint, or both. Progress will depend critically on
optimizing input and output of light.
– Optomechanics and magnetic sensing. The interaction of light with various sorts
of mechanical systems in micro- and nano-scaled devices has many potential uses.
It can be employed to quantify ultra-small forces using radiation pressure, or to
modulate light signals in response to quantized vibrations. Signal transduction
between and among mechanical and optical components may also be an important
component of future communications systems.
– Mass/force/acceleration metrology. NIST’s “electronic milligram” project seeks to

realize very small masses using electrostatic forces. Quantum accelerometers use
interferometry to measure displacement of a mass during acceleration. Ultra-small
forces can be determined using radiation pressure.
– Quantum optics and radiometry. Many phenomena of interest are manifest in very
few, or even individual, photons, and the ability to precisely control a stream of
single photons is of great interest to quantum communications and encryption.
Developing the technology to detect, count, and differentiate single photons is an
ongoing activity at NIST, and will be crucial to the new generation of devices.
– Microfluidics. Manipulating and characterizing fluid behavior on the scale of nano-

liters, and investigating streams of objects as small as individual cells and clusters
of molecules, is of increasingly urgent interest to biophysics, chemical synthesis,
and medical research.
– Electromagnetic field metrology. Conventional methods of measuring EM fields

entail inherent calibration difficulties. NIST is exploring self-calibrating quantum
methods that rely on using light to measure the effect of a field on a volume of
Rydberg atoms.
– Integration of physical and chemical/material/biological measurements.

– Connectivity of distributed embedded standards. Many kinds of measurements, such

as local strains in materials, require interconnection and integrated comparison of
readings from multiple sensors and standards. How to do so is an active subject of
research at NIST.
– Fundamental quantum-based research.
These areas of exploration were chosen for a variety of reasons. Some represent clearly
defined measurement needs. Some group closely related measurement technologies. Some
represent “building block” technologies with broad relevance to many technology areas,
and some represent “stretch goals” that will require long-term attention. Many share
more than one of these attributes.
As these technologies evolve and improve, NIST anticipates that in the near future
several calibration and measurement services that are not currently quantum-based will
become so, including mass, acoustics and vibration, acceleration, electromagnetic field
strength, photometry, laser power and energy, thermometry, pressure and vacuum, and
radioactivity.
What follows is a brief survey of some of the areas in which NIST is engaged in
developing the science and technology to meet the new demands of quantum metrology.
2. – Single-photon sources & detectors and quantum communication
Quantum communications will require the ability to transmit streams of single pho-
tons and detect them with the lowest possible error rate. NIST has a vigorous research
program in both photon sources and detectors. At present, clock synchronization tech-
niques enable transmission rates as high as GHz rate — both free space [1] and in fiber [2].
A key application is the transfer of quantum keys for encryption. To date, NIST has
demonstrated a continuous one-time-pad encryption with a quantum key at a data rate
> 4 MB/s [3]. That is approximately 100 greater than previous demonstrations. Key
processing uses a multi-threaded Forward Error Correction algorithm. Taken together,
these advances make possible broadband applications of quantum encryption.
The challenges involved in this work are extraordinary, raising questions such as:
How do you pull a single photon in the near infrared or the green out of space in broad
daylight? And what are the physical limitations of transfer and detection?
Further progress will depend on many factors. One is the continued development
of superconducting photon detectors. NIST has demonstrated a transition edge sensor
(TES) with a quantum efficiency (QE) greater than 99% [4], and a high-speed single-
photon superconducting detector with a QE above 93% [5]. One advanced TES was
recently used to set a distance record for quantum key distribution [6]. Meanwhile, the
search for improved detector materials is ongoing.
TES technology can also be used to determine the number of photons in a beam.
Even when the actual total number would swamp the sensor, a variable attenuator can
be used to reduce the beam intensity by an exactly known amount. The TES output
signal can then be analyzed with a statistical clustering algorithm to deliver a photon
number [7]. This technique could provide a quantum definition of the candela.
Alternatively, NIST’s TES detectors are capable of counting the number of as many as
1 000 photons in a single pulse of light, with an accuracy limited mainly by the quantum
noise in the laser source. This is accomplished by tracking the relaxation time of the
sensor after excitation — that is, the amount of time in microseconds that it takes for
the TES to return from the elevated resistance state to the transition-edge state [8]. This
interval varies with photon number.
NIST has also developed ultra-sensitive InGaAs single-photon avalanche diodes, bi-
ased by a multi-harmonic RF interferometer, gated at 1.25 GHz, and receptive to light
from 950 nm to 1650 nm, which includes familiar telecomm wavelengths [9]. The devices
are capable of detecting avalanches of fewer than 7 femtocoulombs, or about 42 000 elec-
trons. The unit achieves the lowest noise and the highest efficiency in room-temperature
high-speed detectors. Applications include quantum communications and LIDAR signal
processing, as well as various potential medical uses.
For example, using a technique called cap analysis gene expression, researchers can
get a snapshot of a particular region of genetic material. When it reacts with a particular
biomarker, the resulting signal is captured by a low-noise, high-sensitivity NIST detector.
This technique holds promise for non-invasive testing in early detection of Alzheimer’s
disease, investigation of apoptosis, and toxicity studies.
In addition, ultra-sensitive detectors play a role in quantum physics research such
as the recent demonstration of a loophole-free Bell test of local realism by scientists at
NIST [10] and collaborators elsewhere [11]. That work will eventually lead to the NIST
photonic quantum random number generator, which will become a source for NIST’s
“random number beacon” that will provide a series of numbers that are provably unknown
prior to their generation.
Single-photon sources and detectors also enable the manipulation and analysis of
squeezed light and other techniques that reduce the inherent uncertainty of quantum
measurements. For example, by reducing the noise in the amplitude of a squeezed light
beam at the expense of phase noise, researchers at NIST obtained a measurement that
was about a factor of 10 below the standard quantum (shot noise) limit [12]. The
technology can enable better measurement of optical absorption from trace contaminants,
new methods for absolute calibration of optical detectors, and “noiseless” amplifiers.
Related experiments with entangled photons in twin images [13] should allow beyond-
classical image resolution, and may have implications for quantum memory.
Finally, improvements in phase-sensitive amplifiers offer the possibility of noiseless
amplification for a host of applications from defense to medical imaging [14]. At present,
results are typically monochromatic and low-resolution, but the technique preserves the
signal-to-noise ratio even at maximum gain.
3. – Photon pressure and small mass and force
Traditionally, it has been extremely challenging to measure very small forces in the
range of piconewtons and smaller with adequate resolution. But the impending SI
redefinition provides an opportunity to employ new techniques for scales of small masses
and forces. Because the value of the Planck constant (h) will be fixed, as will the speed
of light, photon momentum can be a force standard (to within knowledge of the photon
frequency). NIST has been developing new methods and technologies to that end.
In general, the process typically involves measuring the force that an incident laser
beam imparts to a mirror. Assuming perfect reflectivity and normal incidence, a force
of a 10W laser beam used for marking has an equivalent mass of about 6.7 micrograms;
a 1 kW laser beam used for welding and cutting has a mass equivalence around 670
micrograms; and so forth [15]. In a recent experiment, NIST demonstrated this
equivalence by measuring a 3-Watt laser using both a standard thermopile detector and
NIST’s Electrostatic Force Balance [16]. The initial agreement was to better than 1%.
The upcoming SI redefinitions in terms of fundamental constants will allow new tech-
niques for scaling of small masses and forces. For example, fixing a value for the Planck
Constant, together with a fixed speed of light, means that photon momentum can be a
force standard (to within knowledge of photon frequency).
NIST has used this idea to develop an experimental optomechanical system that bal-
ances mechanical force with photon pressure force. It incorporates a superluminescent
diode to provide carefully calibrated photon momentum force with a Fabry-Perot inter-
ferometer to indicate displacement. This arrangement permits very low uncertainty for
small force measurements, and can scale down to the single-photon level. At present,
the device can measure RMS force for two different amplitudes of modulated laser power
with a resolution of 14 fN, and holds great promise for the calibration of atomic force
microscopy [17].
Photon pressure is also critical to another key NIST objective: calibrated laser weld-
ing. At present, it is difficult to characterize the welding process exactly in real time.
The complex, multi-physics process (light-matter interaction at the extreme) has meant
that measurements are often more art than science.
Because of that situation, laser welding has remained a promising but underuti-
lized solution, limiting users’ ability to improve manufacturing efficiency (through Im-
proved yield, real-time quality control, and process portability) and slowing the advent
of new manufacturing capabilities such as enabling laser welding of hard-to-weld or “un-
weldable” materials. The problem is that too low power can create a weak weld and too
high can induce undesirable melting and weakening of surrounding material.
NIST’s laser-welding program aims to eliminate these problems by exploiting our
ability to measure the very small forces exerted by light. Monitoring the incident beam
as it reflects from a mirror/scale before the weld occurs makes it possible to measure in
real time — using sensitive commercial scales — the small forces that are proportional
to laser power [18].
Our goal is for the relevant measurements to be available with sufficient accuracy that
a welding operation could be completely specified by the required settings. A process
could then be set up by simply dialing in the desired parameters, making it portable to
any other welding workstation with no additional setup time.
The impact could be substantial. Welding is involved in 50% of the US GNP, and
laser welding could eventually replace one-fourth of conventional welding processes.
4. – Quantum-based pressure and vacuum standards
Until very recently, even the most modern and sophisticated manometers relied on
basically the same design as Evangelista Torricelli’s original mercury barometer of 1643,
with only incremental improvements. Commercial manometers costing upward of USD
500 000 use that four-century-old technology. Today’s U.S. national standard, the NIST
Ultrasonic Interferometer Hg Manometer, employs a column 3 meters high containing
230 kg of mercury.
NIST has pioneered and developed an optical-based standard for pressure that re-
quires no mercury, replaces legacy technology, has no theoretical size limit, is ∼ 100
times faster than conventional equipment, has ∼ 35 times better resolution, and is far
more accurate than off-the-shelf commercial units [19].
The device uses a system of two Fabry-Perot cavities of identical length into which
laser light is directed. One cavity contains a volume of helium or nitrogen gas; the other
is a hard vacuum. An interferometer compares the standing waves formed by the light in
each of the cavities. The difference will depend on the refractive index of the gas, which
is in turn dependent on its density. Thus the all-optical system provides a sensitive,
quantum-based measure of pressure in the range of 0.1 mPa to 1000 kPa.
Recent comparisons between the experimental optical-cavity device and the U.S. na-
tional standard found that the photonic sensor performs better than expected and, in
fact, outperforms the traditional mercury system at low pressure ranges. It can, in effect,
bring the SI to the factory floor in a unit that can be reduced to the size of a cell phone,
and bring NMI-quality measurement into physical and chemical laboratories at a fraction
of the cost of conventional instruments.
Because the device relies on the relationship among displacement, pressure, and tem-
perature, it can be used to measure any quantity when the other two are known. Thus it
becomes a thermometer and a length gauge to measure displacements to 2 ppb in air [20].
Finally, in a project very recently initiated, NIST is developing a Cold Atom Vacuum
Standard (CAVS) which involves using a population of ultracold atoms to sense pressure.
In this design, a collision between a background molecule and a trapped atom will result
in the atom being knocked out of the trap with near unit probability. We will exploit this
to make an absolute sensor of number density—CAVS will count background molecules
in the vacuum by measuring the lifetime of the cold atom trap.
5. – Quantum-based approaches to temperature
It is now possible to replace the extremely expensive and fragile legacy technology for
measuring temperature — e.g., the standard platinum resistance thermometer (SPRT) —
with quantum-based instruments that are free of the SPRT’s known shortcomings, such
as hysteresis and extreme susceptibility to even minor thermal or mechanical shock,
which resets the calibration.
The fast-expanding field of photonic thermometry, which measures temperature
through heat-induced changes to optical properties of material, promises robust, ultra-

small devices with very fast response times that can be manufactured at low cost with
conventional CMOS technology and can be embedded in structures of interest.
For example, Bragg gratings in optical fiber (FBG) or in waveguides will reflect or
transmit different wavelengths depending on how thermally induced expansion or shrink-
ing changes the spacing in the grating. That effect can be used to measure tempera-
ture [21]. Recent modifications offer even better accuracy: The latest silicon gratings
—which take advantage of the high thermo-optic coefficient of silicon— examined at
NIST have an 8-fold increase in temperature sensitivity over FBGs, while occupying a
space about 50 times smaller [22].
Alternatively, optical monitoring of temperature-induced vibration of nanostructures
(such as reflective cavities in silicon nitride beams or photonic crystals) offers the prospect
of stable, reproducible, linear temperature measurements. The resonance frequency
changes as a function of temperature, and because the signal is a frequency measure-
ment, it can be conducted at high precision.
Yet another design uses tiny ring resonators positioned adjacent to a fiber beam path.
Temperature changes in the system cause changes in the dimensions and thermo-optic
properties of the rings, and thus changes the wavelengths that are “trapped” in the
resonator ring and do not propagate down the beam path [23]. Monitoring those changes
is a sensitive, quantum-based measure of temperature.
Clearly, there are many potential routes to the “quantum kelvin”.
6. – Other atom-based sensors
In recent years, the ability to control and interrogate very small volumes of trapped
atomic gases has led to numerous possibilities for quantum-based measurements.
One of the most promising involves measurement of electric field strength. The very
core of all electromagnetic measurements is having accurate calibrated probes and an-
tennas. Traditionally, however, this poses a sort of chicken-or-egg dilemma. To calibrate
a probe, one must place it in a “known” field; but knowing the field requires a calibrated
probe.
NIST is developing a fundamentally new, quantum-based approach for the measure-
ment of E-fields, based on the interaction of radio frequency (RF) fields with alkali atoms
in a glass vapor cell. The atoms are excited optically to Rydberg states in which they
are extremely sensitive to E-fields via the Stark effect [24]. The applied RF-field alters
the resonant state of the atoms, and that change can be detected photonically.
The Rydberg atoms act like an RF-to-optical transducer, converting an RF E-field
to an optical-frequency response. Current experimental units have been shown to be
capable of detecting very weak fields (in the range of 1 mV/m) as well as strong fields
in a broadband range from 1 GHz to 500 GHz, and potentially as high as 1 THz. This
technology can enable direct SI-traceable measurement of RF E-field strength, and can
potentially be fashioned into small and compact probes.
Another use of atomic vapor cells is in sensors for magnetoencephalography, which
maps brain activity by recording magnetic fields produced by electrical currents occurring
naturally in the brain, using very sensitive magnetometers [25]. The same technique can
be employed to detect abnormal activity in heart muscle. Again, the devices employ
volumes of alkali atoms smaller than 1 mm3 .
The technology takes advantage of the fact that an applied magnetic field alters the
spin states of atoms in the field. NIST’s chip-scale atomic magnetometers use a beam of
polarized light to align the spins of the atoms in the vapor cell. As a magnetic field alters
those atomic spins, the amount of light that passes through the gas and into a sensor de-
creases. This design, which has femtotesla sensitivity, operates at room temperature and
could eventually replace some uses of the large and expensive superconducting quantum
interference detectors now in clinical use.
The magnetometer design is derived from earlier NIST work on the first chip-scale
atomic clock. Since that time, the miniaturization of optical frequency combs has made
it possible to fabricate a chip-scale optical-frequency clock [26]. That technology pro-
vides the technical path forward for sensor performance 1000 times better than currently
available for such devices as magnetometers, gyroscopes, and gravimeters. That kind
of improvement has obvious and profound implications for communication, navigation,
spectroscopy, and more.
7. – Quantum transduction
Another auspicious line of research involves the conversion of one form of quantized
energy into another. There is no doubt that future quantum communications systems
and networks will have to exchange information in multiple modalities, switching among
optical, electronic, and even mechanical domains. Moreover, many microelectromechan-
ical and optomechanical devices will depend critically on fully characterized conversion
phenomena. That is why NIST is at work on devising and testing methods of quantum
transduction.
One involves exploring the relationship between laser light and an array of gratings
(each only a fraction of a square millimeter, and each with slightly different grating spac-
ings) mounted on a thin, highly reflective, movable silicon-nitride membrane at one end
of an optical cavity. At the other end is a fixed mirror. Laser light bounces from one
end of the cavity to another, applying radiation pressure to the gratings on the mem-
brane, which functions as an acoustic resonator. The experiment reveals how the phonon
patterns of different grating configurations respond mechanically to that pressure [27].
Among other uses, the “phonon laser” can help researchers determine exactly what kinds
of radiation can damp motion in objects to place them in the ground state.
Another approach seeks to transfer quantum information from optical to electronic
domains using a micromechanical resonator as the intermediary. The system is based on
a two-stage process of transduction. First, the microwave “pump” or carrier wave, which
contains quantum information in the form of small variations in the carrier waveform, is
routed to a resonant circuit that contains a two-plate capacitor, the top plate of which
is a thin membrane that is free to vibrate like a drum head. In the second stage, the
membrane’s vibrations are coupled to light by placing the membrane between two facing
mirrors in an optical cavity through which a light beam passes. The membrane’s motion
changes the frequency of light that resonates in the cavity, thus transferring mechanical
changes into frequency changes in the beam of photons [28].
The same kind of technology can also be used to amplify extremely faint radio-
frequency and microwave signals. By coupling those signals to the motion of a nanoscale
membrane, and then shining a laser light onto the other side of the membrane and de-
tecting the output, the original signal can be detected as phase shifts in the reflected laser
light. This results in a versatile quantum interface between radio and optical frequencies.
These two examples of quantum transduction are only two examples of many that
have been done at NIST and the efforts world-wide are phenomenal. The ability to
transduce quantum signals will both push the range of new types of quantum-based
sensors and standards and also allow these to be miniaturized. It is likely that the future
will lead to highly accurate quantum-based standards to be part of every day society
where we transduce something hard to measure into a frequency or length that can be
accurately and readily measured.
8. – Implications
When the explosion of research and progress in quantum-based measurements reaches

its full potential, it will have transformative effects on the ways in which NMIs fulfill
their mission. It will likely shift an institution’s focus away from providing conventional
artifact measurement and instrument calibration toward deploying dual standards and
sensors whose performance is accurate, stable, and inherently repeatable because it is
based on invariant quantum phenomena. This form of SI dissemination is bound to be
disruptive, especially as commercialization makes quantum measurement devices widely
available.
NMIs’ traditional role of providing traceability will change if devices themselves are
directly based on quantum mechanics. Indeed, it is uncertain whether NMIs will even
have a role in traceability. The entire process of certification will be affected — and in
some cases, perhaps, eliminated. Today’s elaborate systems of mutual recognition may
have to be rethought.
NIST is anticipating and welcomes these potential changes. We are committed to:
– Quantum SI dissemination — Using the “21st century toolkit” to reinvent metrol-
ogy calibration standards and sensors
– Embedded (“chip-scale”) metrology — Employing miniaturized device minimizing
the need for classical calibration services by using quantum effects.
– Open metrology: teaching what we know — Providing training opportunities to our
customers to facilitate adoption of the Quantum SI.
As with my first lecture, this summary is both brief and very NIST centric. How-
ever it is essential to state clearly that the effort to transform measurement science using
quantum-based measurements is a worldwide effort resulting from many individuals from

academia, scientific organizations, and National Measurement Institutes (NMIs) around
the world. While NIST is a leader in pushing measurement science and quantum-based
measurements numerous other NMI’s have made and continue to make original contri-
butions and sometimes even lead the way.
REFERENCES
[1] Bienfang C., Gross A., Mink A., Hershman B., Nakassis A., Tang X., Lu R., Su
D., Clark C., Williams C., Hagley E. and Wen J., Opt. Express, 12 (2004) 2011.
[2] Ma L., Nam S., Xu H., Baek B., Chang T., Slattery O., Mink A. and Tang X.,
New J. Phys., 11 (2009) 045020.
[3] Mink A., Tang X., Ma L., Nakassis T., Hershman B., Bienfang J. C., Su D.,
Boisvert R., Clark C. W. and Williams C. J., Proc. SPIE, 6244 (2006) 62440M.
[4] Lita A. E., Miller A. J. and Nam S. W., Opt. Express, 16 (2008) 3032; Lita A. E.,
Calkins B., Pellouchoud L. A., Miller A. J. and Nam S., Proc. SPIE, 7681 (2010)
76810D.
[5] Marsili F., Verma V. B., Stern J. A., Harrington S., Lita A. E., Gerrits T.,
Vayshenker I., Baek B., Shaw M. D., Mirin R. P. and Nam S. W., Nat. Photon., 7
(2013) 210.
[6] Takesue H., Nam S. W., Zhang Q., Hadfield R. H., Honjo T., Tamaki K. and
Yamamoto Y., Nat. Photon., 1 (2007) 343.
[7] Levine Z. H., Glebov B. L., Pintar A. L. and Migdall A. L., Opt. Express, 23 (2015)
16372.
[8] Levine Z. H., Gerrits T., Migdall A. L., Samarov D. V., Calkins B., Lita A. E.
and Nam S. W., J. Opt. Soc. Am. B, 29 (2012) 2066.
[9] Restelli A., Bienfang J. C. and Migdall A. L., Appl. Phys. Lett., 102 (2013) 141104.
[10] Shalm L. K., Meyer-Scott E., Christensen B. G. et al., Phys. Rev. Lett., 115 (2015)
250402.
[11] Giustina M., Versteegh M. A. M., Wengerowsky S. et al., Phys. Rev. Lett., 115
(2015) 250401.
[12] McCormick C. F., Marino A. M., Boyer V. and Lett P. D., Phys. Rev. A, 78 (2008)
043816.
[13] Boyer V., Marino A. M., Pooser R. C. and Lett P. D., Science, 321 (2008) 544.
[14] Vogel U., Glasser R. T., Clark J. B. et al., New J. Phys., 16 (2014) 013011; Corzo
N. V., Marino A. M., Jones K. M. and Lett P. D., Phys. Rev. Lett., 109 (2012)
043602.
[15] Williams P. A., Hadler J. A., Lee R., Maring F. C. and Lehman J. H., Opt. Lett.,
38 (2013) 4248.
[16] Shaw G. A., Williams P., Stirling J., Guzmán-Cervantes F. and Lehman J., 2016
Conf. Prec. Elect. Meas., DOI: 10.1109/CPEM.2016.7540726.
[17] Melcher J., Stirling J., Cervantes F. G., Pratt J. R. and Shaw G. A., Appl. Phys.
Lett., 105 (2014) 233109.
[18] Williams P., Sowards J. and Simonds B., AWS Welding J., 30 (March 2016).
[19] Egan P., Stone J., Hendricks J., Ricker J., Scace G. and Strouse G., Opt. Lett.,
40 (2015) 3945.
[20] Hendricks J. H., Ricker J. E., Stone J. A. et al., 21st IMEKO World Congress 2015:
Meas. in Res. and Ind., p. 1574 (2015).
[21] Klimov N. N., Mittal S., Berger M. and Ahmed Z., Opt. Lett., 40 (2015) 3934.
[22] Klimov N. N., Purdy T. and Ahmed Z., Sensors & Transducers, 191 (2015) 67.
[23] Klimov N. N., Berger M. and Ahmed Z., Sensors & Transducers, 191 (2015) 63.
[24] Holloway C. L., Gordon J. A., Jefferts S. et al., IEEE Trans. Antennas Prop., 62
(2014) 6169.
[25] Sander T. H., Preusser J., Mhaskar R., Kitching J., Trahms L. and Knappe S.,
Biomed. Opt. Express, 3 (2012) 981.
[26] Papp S. B., Beha K., Del’Haye P. et al., Optica, 1 (2014) 10.
[27] Kemiktarak U., Durand M., Metcalfe M. and Lawall J., New J. Phys., 14 (2012)
125010.
[28] Andrews R. W., Peterson R. W., Purdy T. P. et al., Nature Phys., 10 (2014) 321.
MODULE III. PHYSICAL METROLOGY AND FUNDAMENTAL CONSTANTS
DOI 10.3254/978-1-61499-818-1-321
Space metrology and space-time warping

R. Battiston
Agenzia Spaziale Italiana - Roma, Italy
INFN-TIFPA and Dipartimento di Fisica, Università di Trento - Trento, Italy
Summary. — Gravity is the most mysterious of all interactions, even though it

was the first to be formalized. It is the only interaction for which the equivalence
principle holds between the gravitational charge and the inertial mass, with a num-
ber of far-reaching consequences. In addition, gravity is so weak with respect to all
other interactions that it is not surprising that most precise tests of general relativ-
ity are performed in outer space. General relativity tells us that gravity concerns
the curvature of space and time, so testing gravity means applying metrology at its
best. In this review I discuss some of the most important tests of general relativ-
ity performed in space, starting from the observables, discussing the experimental
methods, the current results and some of the the future developments.
1. – Introduction
Gravity is the most mysterious of all interactions, even though it was the first to be
formalized. Asides from the important problems of the “true” singularities and of the
unification with the quantum mechanics, gravity is the only interaction for which the
equivalence principle holds between the gravitational charge and the inertial mass, with
a number of far-reaching consequences for the dynamical properties of bodies subject
321
322 R. Battiston
to a gravitational force. This is what makes gravity so asymmetric with respect to all
other interactions. In addition gravity is so weak with respect to all other forces that
it is not surprising that most precise tests of general relativity are conducted in outer
space.
General relativity tells us that gravity concerns the curvature of space and time, so
testing gravity naturally implies discussing some of the best and most accurate examples
of metrology experiments and methods.
In what follows, I will review some of the most important tests of general relativity
performed in space, starting from the observables, discussing the experimental methods,
the status of current results and some of the expected developments. Since the field is
vast, I will not attempt to cover systematically all the topics but I will rather concentrate
on some significant examples which can provide a solid basis to understand this field of
research.
I will start from reviewing which are the observable effects of general relativity in
the Solar System, how they can be measured, and which are the anomalies which have
been observed and studied. In this context, I will discuss contributions to the general
understanding of gravity which have been provided by deep-space probes.
I will then address the perspectives opened by their observations for ground-based
interferometers.
Finally, I will move closer to Earth and discuss some topics related to high-precision
space geodesy and to its role in testing general relativity.
2. – Theoretical landscape of the 20th century
After the breaktrough of Newton [1] and the revolution of Einstein [2, 7], gravity
has been studied very extensively during the 20th century. Theoretical works related
to gravity and general relativity (GR henceforth) have flourished through the years.
Starting with the introduction of Poincaré’s gauge formalism [3], a stream of contri-
butions have provided improvements, interpretations or alternatives for the theory of
gravity (the following is not a complete list): Nordström [4, 5], Einstein and Fokker [6],
Whitehead [9], Cartan [10], Kaluza and Klein [11], Fierz and Pauli [12], Birkhoff [13],
Milne [14], Thiry [15], Papapetrou [16], Jordan [17], Littlewood and Bergmann [18],
Brans and Dicke [19], Yilmaz [33], Whithrow and Morduch [20], Kustaanheimo and Nuo-
tio [21], Page and Tupper [22], Bergmann [23], Deser and Laurent [24], Nordtvedt [25],
Wagoner [27], Bollini et al. [26], Rosen [28], Will and Nordtvedt [29], Ni [32], Hellings and
Nordtvedt [31], Ni [30], Yilmaz [34], Lightman and Lee [35], Lee, Lightman and Ni [40],
Rosen [39], Belinfante and Swihart [36], Lee et al. [37], Bekenstein [38], Barker [41],
Rastall [42], Coleman [43] and Hehl [44].
However only a small number of these theories survived till the end of last century,
due to theoretical issues or failing experimental verifications. Let us see how.
First of all, a theory of gravity should be:
Space metrology and space-time warping 323
– complete: not a law, but a theory; experimental expected results should be derived
from first principles
– self-consistent: get the same results no matter which mathematics tools or models
are used
– relativistic: non-gravitational laws are those of special relativity
– Newtonian: reduce to Newton’s equation in the limit of low gravity and low veloc-
ities.
Starting from this point of view some of the theories developed during the 20th century
can be immediately discarded:
– Newton [1] — non-relativistic: implicit action at a distance — incompatible with

special relativity
– Poincaré [3] and conformally flat theory of Whitrow and Morduch [20] since it is
incomplete and not in harmony with non-gravitational physics (Maxwell electro-
magnetism)
– Fierz and Pauli [12] “spin-2 field theory” since it is inconsistent, given that its field
equations imply that all gravitating bodies move along straight lines, while in the
equation of motion, gravity deflects bodies
– Birkhoff [13] — not Newtonian: it demands speed of sound equal the speed of light
– Milne [14] — incomplete, does not predict gravitational red-shift
– Kustaanheimo and Nuotio [21] — inconsistent, predicting gravitational redshift for

photons, but not for light waves
Second, in order to fulfill the Einstein Equivalence Principle (EEP), which predicts
a) uniqueness of Free Fall, b) local Lorentz invariance and c) local position invariance,
only metric theories are viable. This kills theories such as those proposed by Belinfante
and Swihart [36] and Kaluza and Klein [11], quasi linear theories like Deser and Lau-
rent [24] and Whitehead [9] as well as stratified theories with time-orthogonal time slices
by Einstein [2], Papapetrou [16], Whithrow and Morduch [20], Page and Tupper [22],
Rosen [28], Ni [32], Coleman [43].
Third, conformally-flat theories fail tests of time delay and deflection of light: in
particular, Nordstrøm [4, 5], Einstein and Fokker [6], Littlewood and Bergmann [18] and
Ni [32].
Fourth, scalar-Tensor theories are extremely constrained by Viking [45] and Cassini
[46], results. This eliminates proposals by Thiry [15], Jordan [17], Brans and Dicke [19],
Bergmann [23], Nordtvedt [25], Wagoner [27], Bekenstein [38], and Barker [41].
Other theories predict preferred frame effects on perihelion shift which do not match
observations of Mercury’s perihelion — Ni [30] and Lee, Lightman and Ni [40] — while
324 R. Battiston
bi-metric theories predict a dipole radiation which is not observed — this is the case of
Lightman and Lee [35], Rosen [39], Lee et al. [37] and Rastall [42].
Only a few theories resisted so far all attempts to challenge them, but even these
theories have some testable consequence. Will and Nordtvedt’s [29] and Hellings and
Nordtvedt’s [31] proposals are vector-tensor theories: deviations can only be significant
in the high-energy regime (e.g., Planck-scale energy). Cartan [10] and Hehl [44] introduce
instead a non-vanishing matter spin which is very difficult to observe.
Since Eddington [47], a standardized formalism has been developed, the Parametric
Post-Newtonian (PPN), based on a set of parameters used to evaluate alternative gravity
models. Each parameter measures a different aspect of how much a theory departs from
Newtonian gravity. Here we are referring only to deviation from Newtonian theory, that
is, measurements of weak-field effects.
The ten PPN parameters are called γ, β, η, α1 , α2 , α3 , ζ1 , ζ2 , ζ3 , ζ4 .
– γ is a measure of space curvature, taking value 0 for Newtonian gravity and 1 for
GR,
– β is a measure of non-linearity in the addition of gravitational fields, taking value
1 for GR,
– η is a check for preferred location effects,
– α1 , α2 , α3 measure the extent and nature of “preferred-frame” effects. Any theory
of gravity with at least one nonzero α is called a “preferred-frame effects theory”.
– ζ1 , ζ2 , ζ3 , ζ4 , α3 measure the extent and nature of breakdowns in global conservation
laws. A theory of gravity possesses 4 conservation laws for energy-momentum and
6 for angular momentum only if all five are zero.
Using this notation, we can list the different theories about gravity in the following
table I (note: ξ is computed from 3η = 12β − 3γ − 9 + 10ξ − 3α1 + 2α2 − 2ζ1 − ζ2 ).
This situation explains why at the conclusion of the 20th century there was a
widespread consensus that “ Among all bodies of physical law none has ever been found
that is simpler and more beautiful than Einstein’s geometric theory of gravity”(Misner,
Thorne and Wheeler [48]) and that “ Aesthetic or philosophical motives will therefore
continue to play a part in the widespread faith in Einstein’s theory, even if all tests verify
its predictions” (Malcolm MacCallum [49]).
Is that the final triumph for Einstein [7]? Maybe not: at the beginning of the new
century needs for a new theory of gravity have emerged since a) classical GR description
breaks down in regimes with large curvature, b) the need for quantized gravity requires
GR to be modified or extended, c) other challenges arose linked to the existence of Dark
Matter and to Dark Energy. These developments have motivated new tests of GR aimed
at finding to which extent it is really a fundamental theory. A summary of the proposed
techniques and the sensitivity ranges which could be explored is shown in fig. 1.
It is difficult, however, to predict at what level GR could be violated. Unfortunately
no reliable, predictive, alternative theory has been proposed yet and the theoretical
Table I. – Testing General Relativity theories [8].
γ β ξ α1 α2 α3 ζ1 ζ2 ζ3 ζ4
Einstein (1916) GR 1 1 0 0 0 0 0 0 0 0
Tensor theories
1+ω
Bergmann (1968) 2+ω
β 0 0 0 0 0 0 0 0
1+ω
Wagoner (1970) 2+ω
β 0 0 0 0 0 0 0 0
1+ω
Nordtvedt (1970) 2+ω
β 0 0 0 0 0 0 0 0
1+ω
Bekenstein (1977) 2+ω
β 0 0 0 0 0 0 0 0
1+ω
Brans-Dicke (1961) 2+ω
1 0 0 0 0 0 0 0 0
Vector-Tensor theories
Hellings-Nordtvedt (1973) γ β 0 α1 α2 0 0 0 0 0
Will-Nordtvedt (1972) 1 1 0 0 α2 0 0 0 0 0
Bimetric theories
c0
Rosen (1975) 1 1 0 0 c1
−1 0 0 0 0 0
Rastall (1979) 1 1 0 0 α2 0 0 0 0 0
Lightman-Lee (1973) γ β 0 α1 α2 0 0 0 0 0
Stratified theories
c
Lee-Lightman-Ni (1974) a c0 β ξ α1 α2 0 0 0 0 0
1
c
Ni (1973) a c0 bc0 0 α1 α2 0 0 0 0 0
1
Scalar Field theories
Einstein (1912) Not GR 0 0 −4 0 −2 0 −1 0 0(a)
Whitrow-Morduch (1965) 0 −1 −4 0 0 0 −3 0 0(a)
Rosen (1971) λ 3 + λ −4 − 4λ 0 −4 0 −1 0 0
4 4
Papetrou (1954a, 1954b) 1 1 −8 −4 0 0 2 0 0
Ni (1972) (stratified) 1 1 −8 0 0 0 2 0 0
Yilmaz (1958, 1962) 1 1 −8 0 −4 0 2 0 −1(a)
Page-Tupper (1968) γ β −4 − 4γ 0 −2 − 2γ 0 ζ2 0 ζ4
Nordström (1913) −1 1 0 0 0 0 0 0 0
2
Einstein-Fokker (1914) −1 1 0 0 0 0 0 0 0
2
Ni (1972) (flat) −1 1−q 0 0 0 0 ζ2 0 0(a)

Whitrow-Morduch (1960) −1 1−q 0 0 0 0 q 0 0(a)
Littlewood (1953) −1 1
2
0 0 0 0 −1 0 0(a)
Bergman (1956) −1 1
2
0 0 0 0 −1 0 0(a)
Nordström (1912) −1 1
2
0 0 0 0 0 0 0(a)
(a)
The theory is incomplete, and ζ4 can take one of two values. The value closest to zero is listed.
uncertainties are so large that every experiment able to improve over previous tests is
significant.
We should however keep in mind that, due to the difficulty of these tests, violations
of GR from a single experiment will be accepted with great caution (if not skepticism)
and confirmation with different techniques will be essential.
326 R. Battiston
Fig. 1. – Esa roadmap for General Relativity tests.
3. – Testing General Relativity
The observables studied in experiments testing General Gravity are:
– geodesic motion of test masses (deep-space probes, solar system bodies, free falling
masses),
– propagation of photons in a gravity field,

– measurements of angles, distances and velocities.
The techniques used to perform these tests in space can be grouped as follows.
Gravitational Red-Shift: slowing-down of clocks and time, near a mass.
– Gamma-ray frequency red-shift (equivalent to the slowing down of clocks), using the
Mossbauer effect at the Harvard physics laboratory tower, with estimated accuracy
of about 1%, 1960-1965 [50];
– GRAVITY PROBE A (GPBA), a NASA spacecraft, with estimated accuracy of

about 2 · 10−4 (comparison of two hydrogen-maser clocks, one on ground and the
other on a rocket at 10000 km of altitude), 1976-1980 [55];
– the GNSS (Global Navigation Satellite System) satellites (GPS, GLO- NASS,
GALILEO) routinely use the special and general relativistic corrections that
amount to about 39 μ s/day, i.e. about 11 km of correction (time multiplied times
the speed of light);
– the forthcoming space experiment ACES (Atomic Clock Ensemble in Space) [56],
using a PHARAO (Projet d’Horloge Atomique par Refroidissement d’Atomes en
Orbite) [57] cold trapped atom clock and an advanced hydrogen maser clock on the
International Space Station, is planned to reach an estimated accuracy of 2 · 10−6 ;
– the future space experiment SOC (Space Optical Clock) [58] of ESA on the Inter-
national Space Station, is planned to reach an estimated accuracy of 2 · 10−7 .
Weak equivalence principle: gravitational mass equal to inertial mass, not including
the gravitational binding energy.
– Galileo (about 1610), with pendulum (and inclined tables), Huygens and Newton
(1680) with pendulum, with estimated accuracy of about 10−3 ;
– Eotvos [59], using a torsion balance, with an estimated accuracy of about 5 · 10−9 ,
while Krotov and Dicke, Braginsky and Panov [60], have reached below the 10−12
level at the end of the last century;
– Lunar Laser Ranging (LLR), 1969-2004, using retro-reflectors on the Moon (Apollo
11, 14 and 15; Luna 17 and 21), reached in 2004 an estimated accuracy of about
1.4 · 10−13 ;
– the space experiment MICROSCOPE of CNES [61] is planned to reach an estimated

accuracy of about 10−15 in testing the weak equivalence principle.
Strong equivalence principle: the gravitational binding energy contributes in equal

amounts to the gravitational mass and to the inertial mass.
– Lunar Laser Ranging, 1969-2004 [62], reached an estimated accuracy of about 2 ·

10−4 in testing the strong equivalence principle.
Deflection of light by a mass: deflection of the path followed by photons near a mass.
– Eddington’s test [63] during a solar eclipse, with estimated accuracy of about 30%;
– VLBA and VLBI tests, from 1969 to 2009, with estimated accuracy of about 10−4 ;
– Hipparcos astrometry satellite test, with an estimated accuracy of about of 1.5 ·

10−3 ;
– GAIA, an astrometric orbiting telescope of ESA [64], could measure the light-
deflection to accuracy of about 106 (and the γ parameter at a level of about 2·10−6 );
– Gravitational Lensing was observed also with the Hubble space telescope.
328 R. Battiston
Time delay of light: delay in the propagation time of electromagnetic waves due to a
mass.
– Experiments with radio waves using Mercury and Venus as passive reflectors;
– Mariner 6 and 7, and Voyager spacecraft;
– Viking Mars landers and orbiters with an estimated accuracy of about 10−3 [45];
– Cassini spacecraft test with estimated accuracy of about 10−5 , and of about 2·10−5
for the γ parameter [46].
Precession of the periastron and of the perihelion of Mercury.
– Perihelion advance of Mercury tested to agree with its General Relativity prediction
with an estimated accuracy of less than 10−3 ;
– the Hulse and Taylor binary pulsar PSR 1913+16 [65] and other binary pulsars
provided extremely accurate tests of the periastron precession, in agreement with
General Relativity.
Test of the post-Newtonian parameters β and γ.
– The γ parameter measures the amount of space curvature produced by a mass:

within space experiments, light deflection and time delay are measured to agree
with its General Relativity value with an estimated accuracy of about 2 · 10−5 ;
– the β parameter measures (in a suitable gauge) the non-linearity of a gravitational

interaction; with Lunar Laser Ranging (and the space experiments of light deflec-
tion and time delay), it is measured to agree with its GR value with an estimated
accuracy of less than 10−4 .
Frame-dragging or dragging of inertial frames: spacetime dragging due to mass-energy

currents, such as that of a rotating mass.
– Using the LAGEOS (NASA) and LAGEOS 2 (ASI and NASA) satellites and the
GRACE (NASA and DLR) spacecraft for determining the Earth gravity field, esti-
mated accuracy of about 20% in 2004 [68]. GRACE Follow-On (NASA and DLR)
is planned to be launched within few years;
– using the Gravity Probe B (NASA) dedicated space mission, estimated accuracy
of about 20% in 2011-2015 [69];
– using the dedicated LARES satellite (ASI), together with LAGEOS and GRACE
data, the accuracy has improved to about 5% in 2016 [70].
4. – Deep-space navigation and the investigation of gravity

.
4 1. Observables used in deep-space navigation.
– Range (light travel time)
- Phase comparison of modulation tones or codes in coherent radio links
- Current accuracies (2-way):
- 1–3 m (incl. station bias)
- 0.2 m (Bepi Colombo Ka-band/multilink radio systems with wideband
code modulation and delay calibration)
– VLBI (angles)
- Time delay at two widely separated ground antennas

- Current accuracies:
- 2–4 nrad (up to 100 better with phase referencing — but absolute accuracy
limited by quasar position error)
– Range rate
- Phase comparison (carrier) in coherent radio links

- Current accuracies (2-way):
- 3·10−6 m/s at 1000 s integration times (Ka-band/multilink radio systems).
.
4 2. Radioscience GR tests. – In dealing with tests based on propagation of photons
in the solar well, we can parametrize the effects on the light deflection angle θgr (1a), on
the time delay Δt (1b) and on the frequency shift Δν
ν (1c) as a function of the parameter
γ which is 1 for GR.
Msun Rsun
(1a) θgr = 2(1 + γ) = 4x10−6 (1 + γ) rad,
b b
l0 + l1 + l01
(1b) Δt = (1 + γ)Msun ln ,
l0 + l1 − l01
Δν d Msun db
(1c) = Δt 4(1 + γ) .
ν dt b dt
Experimental noise affecting these measurements is related to
– uncertainties in the dynamical model (solar system ephemerides, asteroid masses),
– non-gravitational accelerations (onboard accelerometer),
– propagation noise (solar corona, interplanetary plasma, troposphere),
– spacecraft and ground instrumentation.
330 R. Battiston
Fig. 2. – General Relativity tests on PPN parameter γ.
Dynamical noise and non-gravitational effects must be reduced to a level compatible

with the accuracy of radio-metric measurements with eqs. (2a) and (2b)
1
(2a) σa = σν = 3 · 10−8 cm s−2 at t = 104 s,
τ
1
σa = 2 σρ = 1 · 10−13 cm s−2 at t = 107 s.
(2b)
τ
One of the experiments which has provided most of radioscience results on GR is

Cassini, launched in 1997 and having performed a series of important measurements
during 30 days in June 2002 while coasting towards Saturn on the opposite side of the
Sun with respect to the Earth (Solar Conjunction Experiment), reaching an accuracy on
γ = 1–2 parts in 105 . Figure 2 summarizes the results of measurements of γ in space.
.
4 3. Satellites orbit anomalies. – In addition to GR radioscience tests, it is possible
to test the orbits of deep-space probes all along their trajectories. This topic has gar-
nered the attention of researchers at the time of the observation of the so-called Pioneer
anomaly, consisting in deviations from predicted accelerations of the Pioneer 10 and
Fig. 3. – The Pioneer anomaly. Left: the anomaly as a function of the distance from the Sun.
Right: the Pioneer spacecraft.
Pioneer 11 spacecrafts after they passed about 20 astronomical units (AU) (3 · 109 km)
on their trajectories out of the Solar System. After significant excitement in the scientific
community, which lasted several years, this effect has been subsequently understood in
terms of anisotropic radiation pressure due to heat losses caused by the geometry of the
spacecraft. The anomaly resulted in an acceleration ap = (8.74 ± 1.33) · 10−8 cm/s2 ,
pointing towards the Sun and almost constant (fig. 3).
Various non-conventional hypotheses have been considered, including modifications
to the expected solar gravitation pull due to, e.g., a) dark matter, b) interplanetary dust,
c) non-Newtonian gravitation. These effects could be modeled in terms of a Yukawa-like
−r/λ
force, FY (r) = −k e r2 . Constraints from Cassini phase referencing determination and
from the accurate determination of planetary ephemerides killed these hypotheses. The
attention eventually shifted to the spacecraft thermal properties. The Pioneers satellites
are equipped with Radioisotope Thermoelectric Generators (RTG) based on a 238 Pu
with 88 y half life and a thermal power of 2500 W: 63 W of this power would, if radiated
anisotropically, produce an acceleration consistent with the Pioneer anomaly. This is

just 2.5% of the total RTG power at launch, which was reduced by 20% at the time of
the 1991 observation.
In order to test this hypothesis, a similar analysis was performed on the orbit of
Cassini. The precise determination of the orbit of Cassini requires a correction, due to
strongly anisotropic thermal emission from its 13 kW RTG, which is about 5 times larger
than the Pioneer anomalous acceleration. Through a series of experiments performed
during different phases of the Cassini mission, including changing the orientation of the
spacecraft, it was possible to derive an upper limit to the Pioneer-like anomaly affecting
Cassini, several times lower than the observed effect on Pioneer, thus supporting the
conclusion that the effect is due to a small, asymmetric heat dissipation.
332 R. Battiston
Table II. – Expected Bepi Colombo contributions (MORE experiment) to GR testing [71].
Parameter Present accuracy MORE

γ 2 × 10−5 2 × 10−6
−4
β 1 × 10 2 × 10−6
−4
η 5 × 10 8 × 10−6
J2 4 × 10 −8
2 × 10−9
Ġ/G 9 × 10−13 yr−1 3 × 10−13 yr−1
.
4 4. Future deep-space satellites. – GR predicts the Lense-Thirring [66] precession of
a giroscope around a spinning mass. It is the gravitomagnetic analogy of the Larmor
precession of a magnetic dipole in a magnetic field. This effect is small around the
Earth, where it has been recently observed for the first time (see later), but can be large
in the vicinity of large rotating bodies, as is the case for the Juno probe orbiting around
Jupiter [67]:
(1 + γ)G 3(J · r)r

(3a) Ω= 2 3
−J +
2c r r2
where J is Jupiter’s angular momentum and r is the Jupiter-Juno position vector. In

this case the SNR ratio for the L-T effect is expected to be of the order of 100.
Another deep-space mission which will provide accurate tests of GR is the ESA Bepi
Colombo mission to Mercury, as shown in table II.
5. – Gravitational waves
Gravitational Waves (GW) are the analogous of the electromagnetic waves for gravi-
tation. They propagate at the speed of light temporarily deforming the texture of space
time. Predicted by Albert Einstein [7] on the basis of his theory of General Relativity,
gravitational waves transport energy as gravitational radiation, and have been discovered
exactly 100 years later by ground based laser interferometers [72]. They are emitted by
massive bodies undergoing acceleration. A two-body orbiting system, with masses m1
and m2 , emits a power P :
dE 32 G4 (m1 m2 )2 (m1 + m2 )
(4a) P = =− .
dt 5 c5 r5
The emitted power is really small in most gravitating systems. For example, in the
case of the Sun-Earth system, it amounts to about 200 W, about 5 · 10−25 times less
that the electromagnetic power emitted by our star. The GW spectrum extends from
frequencies corresponding to the inverse of the age of the universe to few hundreds of Hz
(fig. 4).
Fig. 4. – Gravitational wave spectrum and detection techniques.
Their detection has only recently been demonstrated on ground but there are solid
reasons to believe that the S/N ratio will be much larger for space-borne inteferometers.
LISA is a three-arm space interferometer studied by ESA and NASA up to formulation
level for more than 10 years. These studies have evolved in the ESA eLISA/NGO two-
arm space interferometer, which is the current reference mission for the L3 ESA mission.
With the success of the LISA Pathfinder experiment, ESA is on track to develop eLISA
which could be operational towards the end of the next decade and detect signals coming
from super-massive black-hole mergers, compact objects captured by supermassive black
holes and compact binaries.
eLISA would measure a) the orbital period of the binary system, b) the chirp mass
3/5
(m1 m2 )
M = (m 1 +m2 )
1/5 , discriminating between white dwarf, neutron star and black hole bi-
naries and determining the distance for most binary sources with an accuracies better
than 1%.
.
5 1. The basic concept of a space-borne GW detector . – Measurement of space-time
curvature using light beams requires an emitter and a receiver which are perfectly free
falling (satellites). In flat space-time, the length of proper time between two light-wave
334 R. Battiston
crests is the same for the emitter and for the receiver. GW curvature gives oscillating
relative acceleration to local inertial frames if the wave-front is used as a reference:
it follows that the receiver sees frequency oscillating. Acceleration of receiver and/or
emitter relative to their respective inertial frame produces the same effect of a curvature
and should be carefully avoided.
In order to detect gravitational waves via the slowly-oscillating (1 hour), relative
motion they impose onto far apart free bodies, one needs a) an instrument to detect tiny
motion, of the size on atom peak-to-peak, ensuring b) that only gravitational waves can
put your test-bodies into oscillation and c) eliminating all other forces above the weight
of a bacterium.
The motion detector a) is provided by a laser interferometer, as for ground based
GW detectors, detecting relative velocities by measuring the Doppler effect through the
interferometric pattern variation. Using very stable laser light one can reach the accuracy
of 1 atom size in 1 hour.
The free bodies b) cannot be touched or supported, at least in an ordinary way. They
must be shielded against all other forces c): in particular, one needs to suppress gravity
of the Earth (and of the Sun). The gravity force can be be turned off by falling with
it. For all other forces, the satellite body would neutralize solar radiation and plasma
pressure, actively and precisely following the test mass inertial motion. In order to ensure
non-contacting (drag-free) behavior, the spacecraft position relative to the test mass is
measured by a local interferometer, and it is kept centered on the test mass by acting on
micro-Newton thrusters.
The specifications of the current GW interferometers designs which have been con-
sidered are
– LISA
- 3 arms, each 5 Mkm
√
- 10 pm/ Hz single-link interferometry @1 mHz
√
- Forces (per unit mass) on test masses < 3 fm/(s2 Hz) @ 0.1 mHz
- 3 non-contacting (“ drag-free”) satellites
– eLISA/NGO
- 2 arms each 1 Mkm
√
- 10 pm/ Hz single-link interferometry @1 mHz
√
- Forces (per unit mass) on test masses < 3 fm/(s2 Hz) @ 0.1 mHz
- 3 non-contacting (“drag-free”) satellites
– ESA L3 mission design

- 1 Mkm ≤ arm-length ≤ 5 Mkm
- 2 ≤ Number of arms ≤ 3
- 2(6) ≤ Mission duration ≤ 5(10) years
Fig. 5. – The GRS; left: reference mass housing, right: reference mass.
A basic concept of LISA/eLISA is that the satellites follow independent heliocentric

orbits and no formation keeping is needed. The constellation rotates within waves pro-
viding source location. In the case of the eLISA instrument, the implementation of the
requirements a)-c) is provided by the following main elements:
– the Gravitational Reference Sensor (GRS) with the test mass (also called Inertial
Sensor): the GRS is drag-free along sensitive direction, while the other degrees of
freedom are controlled via electrostatic forces through a 3–4 mm clearance between
test mass and electrodes (fig. 5);
– the Optical Bench (fig. 6) with the complete interferometry: it carries all needed
interferometry on a monolithic ultra-stable structure obtained by silica hydroxyl
bonding;
– a telescope allowing to exchange light with other satellites.
Fig. 6. – LISA-Pathfinder optical bench.
336 R. Battiston
Fig. 7. – eLISA sensitivity to Gravitational Waves.
.
5 2. LISA Pathfinder . – In order to test in space most of the techniques needed for a
LISA class space interferometer, the LISA Pathfinder mission has been built, launched in
December 3rd 2015 and successfully operated in space for about eight months, starting
from March 1st 2016.
The LISA Pathfinder is based on squeezing of one arm of the final interferometer
to within a O(1 m) optical bench. This was implemented by removing the long-arm
interferometer and replacing the long-arm laser beam reference with a second (quasi-)
free test mass. In this miniature implementation of one LISA arm two Au-Pt test masses
and two interferometers were placed on the same optical bench. The two masses were
not contacting the satellite but the second test mass was forced to follow the first at very
low frequency by electrostatic forces (this is different from LISA).
LISA Pathfinder can be seen as a remotely controlled gravitational laboratory oper-
ating in space conditions. The GRS consists in two light test masses (2 kg, 46 mm) with
a very high-density homogeneity ( 1 μ m pores), so that the position of the CoG at
geometrical center is known within ±2 μ m. It has a very low magnetic susceptibility

χ = −(2.3 ± 0.2) · 10−5 as well as a negligible magnetic moment < 4 nAm2
Many subtle physical effects apply unwanted forces to test-bodies [52], such as:
– impact with the few molecules that still surround the bodies in high vacuum [51,53];
– spontaneous electric fields generated by surrounding bodies;
– fluctuating electrical charge from cosmic rays [54];
– changing gravitation generated by thermal deformation of satellite;
– impact with wandering photons;
– fluctuations of the interplanetary magnetic field;
These effects have been studied over the years in the laboratory, pushing forward
knowledge in different fields of physics. The results published by the LISA-PF [73]
shows that the mission has been very successful, exceeding the predicted accuracy and
demonstrating that sub-femto-g differential accelerometry can be achieved, which is an
improvement of orders of magnitude in the field of experimental gravitation. LISA-PF
results confirm the projected eLISA sensitivity to the bulk of GW sources present in our
galaxy (blue line in fig. 7): a green light for an eLISA class mission at the end of the
next decade.
6. – Testing GR within the Earth gravity field
The Earth gravity field, although weak with respect to the other astrophysical sources,
can be used for precision tests of General Relativity, to the extent to which it is precisely
measured and modeled. Space Geodesy addresses this issue, by reconstructing the shape
and the intensity of the Earth magnetic field using satellite borne measurements combined
with ground-based Very Large Baseline Interferometry (VLBI) and Lunar Laser Ranging
(LLR).
.
6 1. Satellite geodesy. – In recent year a few satellites devoted to the precise deter-
mination of the Earth gravity field have provided substantial progress in this field. The
measurement is based on the measurements of acceleration differences in 3-D over the
short baseline of the gradiometer instrument. In short the principles are (fig. 8):
– SST-hl: 3-D accelerometry corresponds to gravity acceleration,
– SST-ll: intersatellite link corresponds to acceleration differences between the Low

Earth Orbit satellites,
– SSGl: gradient of gravity components corresponds to acceleration gradient.
In a mathematical sense, this is a transition from the first derivative of the gravita-
tional potential (SST-hl) via the difference of first derivative over a long baseline (SST-ll)
to the second derivative (gradiometry).
The three most important satellites which have provided the data on which current
Earth gravity field models are based are:
– CHAMP (CHAllenging Mini-satellite Payload for geophysical research and appli-

cation) [74] is a DLR mission based on the SST-hl (Satellite-Satellite Tracking in
the high-low mode) principle (fig. 8). Launched on July 15, 2000 its two main
objectives are the mapping of the magnetic field and of the gravity field of the
Earth. The CHAMP spacecraft is tracked by the GPS constellation, which would
ultimately yield an orbit accuracy of one to three centimeters. Launched in summer
2000.
– GRACE (Gravity Recovery And Climate Experiment) [75], is a NASA mission

complementary to GOCE, based on the SST-ll (Satellite-Satellite Tracking in the
338 R. Battiston
Fig. 8. – Concepts of Satellite-to-Satellite Tracking (SST) modes. Left: high-low mode (SST-hl),
the orbiter is tracked by GPS relative to a network of ground stations, while non-gravitational
forces on the low orbiter are measured by accelerometers; center: low-low (SST-ll) mode com-
bined with SST-hl, where the relative motion between two low orbiters travelling on the same
orbit at a distance of a few hundreds km is measured by an inter-satellite link; right: satel-
lite gradiometry combined with SST-hl, where the second-order derivative of the gravitational
potential is measured in the low orbiting satellite by differential accelerometry.
low-low mode). It consists in flying two satellites separated by 150 to 300 km from
each other, on the same mean orbit; the satellites are each tracked by GPS (like
CHAMP) and also measure their relative distance and velocity with micrometric
accuracy: this observable may be viewed as an extremely fine measurement of the
gravitational potential difference at the satellites. The mission concept is espe-
cially suitable for the monitoring of time variations (from monthly to biennial) of
the gravity field over the 300 to 5000 km spatial scale, with a corresponding contri-
bution to the mean (static) field where an order of magnitude gain over the present
knowledge is expected at such scales. Launched in 2002.
– GOCE (Gravity field and steady-state Ocean Circulation Explorer) [76]. For
the first time, a gradiometer was flown on board a satellite on a very low alti-
tude orbit (between 230 and 250 km) with drag-free control. The gradiometer
instrument delivers
√ gravity gradients in spacecraft axes with an accuracy of a few
milli-Eötvös/ Hz (1 E = 10−9 s−2 ) in the 0.005–0.1 Hz measurement bandwidth.
GOCE provides a global gravity model with a resolution of 100 km at least (65 km
at most) with a total uncertainty (i.e. up to the 100 km resolution) of 2.5 mm on
the geoid and 0.08 mgal on the gravity. Launched in 2009.
.
6 2. Very Long Baseline Interferometry. – Very-long-baseline interferometry (VLBI)
is a type of astronomical interferometry used in radio astronomy. In VLBI a signal from
an astronomical radio source, such as a quasar, is collected at multiple radio telescopes
on Earth. The distance between the radio telescopes is then calculated using the time
difference between the arrivals of the radio signal at different telescopes. This allows
observations of an object that are made simultaneously by many radio telescopes to be
combined, emulating a telescope with a size equal to the maximum separation between
the telescopes.
Data received at each antenna in the array include arrival times provided by a local
atomic clock, such as a hydrogen maser. At a later time, an image is produced by
correlating such data with data from other antennas that recorded the same radio signal.
The resolution achievable using interferometry is proportional to the observing frequency.
The VLBI technique enables the distance between telescopes to be much greater than
that possible with conventional interferometry, which requires antennas to be physically
connected by coaxial cable, waveguide, optical fiber, or other type of transmission line.
Greater telescope separations are possible in VLBI, allowing it to produce images with
superior resolution.
VLBI is most well known for imaging distant cosmic radio sources, spacecraft tracking,
and for applications in astrometry. However, since the VLBI technique measures the time
differences between the arrival of radio waves at separate antennas, it can also be used
“in reverse” to perform Earth rotation studies, study very precisely (within millimeters)
the movements of tectonic plates, and perform other types of geodesic measurements.
Using VLBI in this manner requires large numbers of time difference measurements from
distant sources (such as quasars) observed with a global network of antennas over a given
period of time.
.
6 3. Laser Lunar Ranging. – The ongoing Lunar Laser Ranging Experiment measures
the distance between Earth and the Moon using laser ranging. Lasers on Earth are aimed
at retroreflectors planted on the Moon during the Apollo program (11, 14, and 15) and
the two Lunokhod missions [62]. The time for the reflected light to return is measured.
The Apollo 15 array is three times the size of the arrays left by the two earlier Apollo
missions: its size made it the target of three quarters of all the sample measurements
taken in the first 25 years of the experiment. Improvements in technology since then have
resulted in greater use of the smaller arrays, from sites such as the Côte d’Azur Obser-
vatory in Grasse, France, the Apache Point Observatory Lunar Laser-ranging Operation
(APOLLO) at the Apache Point Observatory in New Mexico, and the G. Colombo Space
Geodesy Center of the Italian Space Agency (ASI) in Matera, Italy.
LLR provides accurate references for the Earth center of gravity, for the Earth-Moon
distance, and the Earth rotation axis and polar motion, which in turn are at the basis of
the accurate understanding of the Earth gravity field.
.
6 4. Frame dragging and Gravitomagnetism in the Earth gravity field. – LAGEOS
(Laser Geodynamics Satellite or Laser Geometric Environmental Observation Survey)
and LARES (Laser Relativity Satellite) are a series of scientific satellites designed to
provide an orbiting laser ranging benchmark for geodynamical studies of the Earth.
Each satellite is a high-density passive laser reflector in a very stable medium Earth
orbit (MEO).
There are two LAGEOS spacecrafts: LAGEOS-1 launched in 1976, and LAGEOS-2
launched in 1992, and one LARES spacecraft, launched in 2012 (fig. 9). Both LAGEOS
and LARES spacecrafts are routinely tracked by the ILRS network.
340 R. Battiston
Fig. 9. – The LAGEOS (left) and LARES (rigth) spacecrafts (to scale).
The LAGEOS spacecraft are aluminum-covered brass spheres with a diameter of 60 cm

and masses of 400 and 411 kg, covered with 426 cube-corner retroreflectors, giving them
the appearance of giant golf balls [77]. 422 of the retroreflectors are made of fused silica
glass while the remaining 4 are made of germanium, in order to obtain measurements in
the infrared for experimental studies of reflectivity and satellite orientation. They orbit
at an altitude of 5900 kilometers, well above low Earth orbit and well below geostationary
orbit, at orbital inclinations of 109.8 and 52.6 degrees, respectively.
The LARES spacecraft consists in a 36 cm diameter, 387 kg tungsten sphere, hosting
92 cube-corner retroreflectors and orbiting at 1450 km and 69.5 degrees of inclination [78].
Both LAGEOS and LARES have no on-board sensors or electronics and are not
attitude-controlled.
Measurements are made by transmitting pulsed laser beams from Earth ground sta-
tions to the satellites. The laser beams then return to Earth after hitting the reflecting
surfaces; the travel times are precisely measured, allowing ground stations in different
parts of the Earth to measure their locations to better than 1 cm in thousands of km.
A clever use of the precise orbit determination over time for both LAGEOS satellites
allowed for the first experimental confirmation, within a 20% accuracy, of the Lens-
Thirring effect on the weak Earth gravity field [68]. An improved result to within 5%
has been recently published by the same team [70] using both LAGEOS and LARES
spacecrafts.
7. – Conclusions
Tests of General Relativity in space rely on the availability of free-flying spacecraft

protected by stray accelerations and uncertainties in the masses of Solar System bodies.
Onboard accelerometers are of limited use (unless of LISA class or better), since they
must be bias-free and work to very low frequencies. Planetary orbiters, tied to central
body, could be made nearly immune to stray acceleration but are subject to uncertainties
in the masses of Solar System bodies. Advances in Solar System gravity tests have
been painfully slow. Progresses so far have relied upon piggy-back experiments (Viking,
Cassini, Bepi Colombo, GAIA) and have been able to rule out claims of violations of GR
at Solar System scales. We lack a predictive theoretical framework for violations of GR
and, consequently, space agencies are not willing to invest on dedicated missions, with the
exception of Gravity Waves detection, due to their key contribution to multi-messenger
astrophysics. In these circumstances, any experiment claiming a violation would not
be immediately accepted and concurrence of different measurements would be crucial.
However, modern astrophysics and cosmology hint for new physics and this could boost
the experimental efforts to test GR also at Solar System level. Indeed, violations at
cosmological scales will almost surely affect laws of gravity at short scales, maybe with
detectable effects in classical, accurate tests.
∗ ∗ ∗
The author thanks Pippo Bianco, Ignazio Ciufolini, Luciano Iess, Kip Thorne and
Stefano Vitale for fruitful discussions while preparing this talk.
REFERENCES
[1] Newton I., Philosophiae Naturalis Principia Mathematica (London) 1687.

[2] Einstein A., “Lichtgeschwindigkeit und Statik des Gravitationsfeldes”, Ann. Phys.
(Leipzig), 38 (1912) 355-369; 38 (1912) 443-458.
[3] Poincaré H., “Sur le probléme des trois corps et les équations de la dynamique”, Acta
Math., 13 (1890) 1-270 (Oeuvres VII 262-490).
[4] Nordström G., “Relativitätsprinzip und Gravitation”, Phys. Zeitschr., 13 (1912) 1126.
[5] Nordström G., “Zur Theorie der Gravitation vom Standpunkt des Relativitätsprinzips”,
Ann. Phys., 42 (1913) 53.

[6] Einstein A. and Fokker A. D., “Die Nordströmsche Gravitationstheorie vom
Standpunkt des absoluten Differentkalküls”, Ann. Phys., 44 (1914) 321-328.
[7] Einstein A., “Die Feldgleichungen der Gravitation”, Sitzungsberi. Preuss. Akad. Wiss.
Berlin, (1915) 844-847.
[8] Einstein A., “Die Grundlage der allgemeinen Relativitätstheorie”, Ann. Phys., 49 (1916)
769.
[9] Whitehead A. N., The Principles of Relativity (Cambridge University Press) 1922.
[10] Cartan E.,´ “Sur les variétés á connexion affine et la théorie de la relativité généralisée”,
Ann. Sci. École Norm. S., Sér. 3, 40 (1923) 325-412.
[11] Kaluza T., “Zum Unitätsproblem in der Physik Sitzungsber”, Preuss. Akad. Wiss.
Berlin. (Math. Phys.), (1921) 966-972; Klein O., “Quantentheorie und fünfdimensionale
Relativitätstheorie”, Z. Phys. A, 37 (1926) 895-906.
342 R. Battiston
[12] Fierz M. and Pauli W., “On relativistic wave equations for particles of arbitrary spin in
an electromagnetic field”, Proc. R. Soc. London, 173 (1939) 211-232.
[13] Birkhoff G. D., “Matter, electricity and gravitation in flat space-time”, Proc. Natl. Acad.
Sci. U.S.A., 29 (1943) 231-239.
[14] Milne E. A., Kinematic Relativity (Clarendon Press, Oxford) 1948.
[15] Thiry Y., “Les équations de la théorie unitaire de Kaluza”, C. R. Acad. Sci. (Paris), 226
(1948) 216.
[16] Papapetrou A., Zs. Phys., 139 (1954) 518; Math. Nach., 12 (1954) 129; 12 (1954) 143.
[17] Jordan P., Schwerkraft und Weltall (Vieweg, Braunschweig) 1955.
[18] Littlewood D. E., Proc. Cambridge Philos. Soc., 49 (1953) 90-96; Bergman O., “Scalar
field theory as a theory of gravitation”, Amer. J. Phys., 24 (1956) 39.
[19] Brans C. and Dicke R. H., “Mach’s principle and a relativistic theory of gravitation”,
Phys. Rev., 124 (1961) 925-935.
[20] Whitrow G. J. and Morduch G. E., “Relativistic theories of gravitation”, Vistas
Astron., 6 (1965) 1-67.
[21] Kustaanheimo P. E. and Nuotio V. S., Publ. Astron. Obs. Helsinki, 128 (1967) .
[22] Page C. and Tupper B. O. J., “Scalar gravitational theories with variable velocity of
light”, Mon. Not. R. Astron. Soc., 138 (1968) 67-72.
[23] Bergmann P. G., “Comments on the scalar-tensor theory”, Int. J. Theor. Phys., 1 (1968)
25-36.
[24] Deser S. and Laurent B. E., “Gravitation without self-interaction”, Ann. Phys., 50
(1968) 76-101.
[25] Nordtvedt Jr. K., “Post-Newtonian metric for a general class of scalar-tensor
gravitational theories with observational consequences”, Astrophys. J., 161 (1970) 1059.
[26] Bollini C. G., Giambiagi J. J. and Tiomno J., “A linear theory of gravitation”, Nuovo
Cim. Lett., 3 (1970) 65-70.
[27] Wagoner R. V., “Scalar-Tensor Theory and Gravitational Waves”, Phys. Rev. D, 1 (1970)
3209-3216.
[28] Rosen N., “Theory of gravitation”, Phys. Rev. D, 3 12 (1971) 2317.
[29] Will C. M. and Nordtvedt Jr. K., “Conservation laws and preferred frames in
relativistic gravity I”, Astrophys. J., 177 (1972) 757.
[30] Ni W. T., “A New Theory of Gravity”, Phys. Rev. D, 7 (1973) 2880-2883.
[31] Hellings R. and Nordtvedt K., “Vector-Metric Theory of Gravity”, Phys. Rev. D, 7
(1973) 3593-3602.
[32] Ni W. T., “Theoretical Frameworks for Testing Relativistic Gravity. IV a Compendium
of Metric Theories of Gravity and Their POST Newtonian Limits”, Astrophys. J., 176
(1972) 769.
[33] Yilmaz H., “New approach to general relativity”, Phys. Rev., 111 (1958) 1417.
[34] Yilmaz H., “New approach to relativity and gravitation”, Ann. Phys., 81 (1973) 179-200.
[35] Lightman A. and Lee D., “New Two-Metric Theory of Gravity with Prior Geometry”,
Phys. Rev. D, 8 (1973) 3293-3302.
[36] Belinfante F. J. and Swihart J. C., “Phenomenological linear theory of gravitation
Part I”, Ann. Phys., 1 (1957) 168; 2 (1957) 196.
[37] Lee D. L. et al., “Theoretical frameworks for testing relativistic gravity. V. Post-
Newtonian limit of Rosen’s theory”, Astrophys J., 206 (1976) 555-558.
[38] Bekenstein J. D., Phys. Rev. D, 15 (1977) 1458; Phys. Rev. D, 70 (2004) 083509.
[39] Rosen N., “A bimetric theory of gravitation II”, Gen. Relativ. Gravit., 6 (1975) 259-268.
[40] Lee D., Lightman A. and Ni W. T., “Conservation laws and variational principles in
metric theories of gravity”, Phys. Rev. D, 10 (1974) 1685-1700.
[41] Barker B. M., “General scalar-tensor theory of gravity with constant G”, Astrophys. J.,
219 (1978) 5.
[42] Rastall P., “The Newtonian theory of gravitation and its generalization”, Can. J. Phys.,
57 (1979) 944-973.
[43] Coleman S., Nucl. Phys. B, 307 (1988) 864; 310 (1988) 643.
[44] Hehl F. W. and Socorro J., “Gauge theory of gravity: electrically charged solutions
within the metric affine framework”, Acta Phys. Pol. B, 29 (1998) 1113-1120.
[45] Reasenberg R. D. et al., “Viking relativity experiment - Verification of signal retardation
by solar gravity”, Astr. J. Lett., 234 (1979) L219-L221.
[46] Bertotti B., Iess L. and Tortora P., “A test of general relativity using radio links
with the Cassini spacecraft”, Nature, 425 (2003) 374-376.
[47] Eddington A. S., The Mathematical Theory of Relativity (Cambridge University Press)
1922.
[48] Misner C. M., Thorne K. S. and Wheeler J. A., Gravitation (W. H. Freeman and
Co., New York) 1973.
[49] General Relativity and Gravitation: A Centennial Perspective, edited by Ashtekar A.,
Berger B. K., Isenberg J. and MacCallum M., 1976.
[50] Pound R. V. and Rebka Jr. G. A., “Gravitational Red-Shift in Nuclear Resonance”,
Phys. Rev. Lett., 3 (1959) 439-441.
[51] Carbone L. et al., “Achieving Geodetic Motion for LISA Test Masses: Ground Testing
Results”, Phys. Rev. Lett., 91 (2003) 151101; Erratum 91 (2003) 179903.
[52] Carbone L. et al., “Thermal gradient-induced forces on geodesic reference masses for
LISA”, Phys. Rev. Lett., 76 (2007) 102003.
[53] Cavalleri A., “Increased Brownian Force Noise from Molecular Impacts in a Constrained
Volume”, Phys. Rev. Lett., 103 (2009) 140601.
[54] Antonucci E. et al., “Interaction between Stray Electrostatic Fields and a Charged Free-
Falling Test Mass”, Phys. Rev. Lett., 108 (2012) 181101.
[55] Milliner J. B., Space Probe to Test Einstein’s “Space-Time Warp” Theory, 1976.
[56] Atomic Clock Ensemble in Space (ACES), ESA, Oct. 25, 2011,
http://www.esa.int/SPECIALS/HSF Research/SEMJSK0YDUF 0.html.
[57] PHARAO atomic clock agreement signed by ESA and CNES, ESA, Dec. 15, 2009,
http://www.esa.int/esaCP/SEMRDI9K73G index 0.html.
[58] Schiller S., The Space Optical Clocks Project: Development of high-performance
transportable and breadboard optical clocks and advanced subsystems, arXiv:1206.3765.
[59] Eötvös R. v., Math. Naturwiss. Berich. Ungarn, 8 (1890) 65.
[60] Roll P. G., Krotkov R. and Dicke R. H., Ann. Phys., 26 (1964) 442; Braginsky
V. B. and Panov V. I., Zh. Eksp Teor. Fiz., 61 (1971) 873; Braginsky V. B. and
Panov V. I., Sov. Phys. JETP, 34 (1974) 463.

[61] Bauer M., Space Microscope to test universality of freefall (European Space Agency) 26
April 2016.
[62] Battat J. B. R., Chandler J. F. and Stubbs C. W., “Testing for Lorentz
Violation: Constraints on Standard-Model Extension Parameters via Lunar Laser
Ranging”, Phys. Rev. Lett., 99 (2007) 241103; Battat J. B. R. et al., “The Apache
Point Observatory Lunar Laser-ranging Operation (APOLLO): Two Years of Millimeter-
Precision Measurements of the Earth-Moon Range”, Publ. Astron. Soc. Pac., 121 (2009)
29; Bender P. L. et al., “The Lunar Laser Ranging Experiment: Accurate ranges have
given a large improvement in the lunar orbit and new selenophysical information”, Science,
182 (1973) 229; Bertotti B., Ciufolini I. and Bender P. L., “New test of general
relativity: Measurement of de Sitter geodetic precession rate for lunar perigee”, Phys.
Rev. Lett., 58 (1987) 1062.
344 R. Battiston
[63] Dyson F. W., Eddington A. S. and Davidson C., “A determination of the deflection
of light by the Sun’s gravitational field, from observations made at the total eclipse of 29
May 1919”, Philos. Trans. R. Soc., 220A (1920) 291-333.
[64] Gaia overview, http://www.esa.int/Our Activities/Space Science/Gaia overview.
[65] Hulse R. A., The discovery of the binary pulsar, Rev. Mod. Phys, 66 (1994) 699; Taylor
J. H., Binary pulsars and relativistic gravity, Rev. Mod. Phys., 66 (1994) 711.
[66] Thirring H., Phys. Zeit., 19 (1918) 33.
[67] Iorio L., “Juno, the angular momentum of Jupiter and the Lense-Thirring effect”, New
Astron., 15 (2010) 554-560; Erratum 16 (2011) 17-18.
[68] Ciufolini I., Pavlis E., Chieppa F., Fernandes-Vieira E. and Perez-Mercader J.,
“Test of General Relativity and measurement of the Lense-Thirring effect with two Earth
satellites”, Science, 279 (1998) 2100.
[69] Everitt C. et al., “Gravity Probe B: Final Results of a Space Experiment to Test General
Relativity”, Phys. Rev. Lett., 106 (2011) 22.
[70] Ciufolini I. et al., “Test of general relativity using the LARES and LAGEOS satellites
and a GRACE Earth gravity model”, Eur. Phys. J. C., 76 (2016) 20.
[71] Milani et al., Phys. Rev. D, 66 (2002) 082001.
[72] Abbott B. P. et al., LIGO Scientific Collaboration and Virgo Collaboration “Observation
of Gravitational Waves from a Binary Black Hole Merger”, Phys. Rev. Lett., 116 (2016)
061102.
[73] Armano M. et al., “Sub-Femto-g Free Fall for Space-Based Gravitational Wave
Observatories: LISA Pathfinder Results” Phys. Rev. Lett., 116 (2016) 231101.
[74] Reigber Ch. et al., “A high quality global gravity field model from CHAMP GPS tracking
data and Accelerometry (EIGEN-1S)”, Geophys. Res. Lett., 29 (2002) 10.1029.
[75] Gravity map from GRACE Astron. Geophys., 44 (2003) 4-4.
[76] Drinkwater M. R. et al., “GOCE: ESA’s first Earth Explorer Core mission”, edited by
Beutler G. B., Drinkwater M., Rummel R. and von Steiger R., “Earth Gravity
Field from Space - from Sensors to Earth Sciences”, Space Sci. Ser. ISSI, 18 (2003) 419-432.
[77] LAGEOS Characteristics NASA Historical Data Book Volume III, Table 4-166 SP-4012.
[78] Ciufolini I., Paolozzi A. and Paris C., Overview of the LARES mission: orbit, error
analysis and technological aspects, J. Phys., Conf. Ser., 354 (2012) 1-9.
DOI 10.3254/978-1-61499-818-1-345
Measuring light
Giorgio Brida
INRIM, Istituto Nazionale di Ricerca Metrologica - Strada delle Cacce 91, Torino, Italy
Summary. — This paper is a review of the existing primary standards for the mea-
surement of the optical radiation; blackbody, synchrotron radiation and electrical
substitution radiometer. A summary of the recent and significant advances in this
filed with the carbon nanotube absorber in the electrical substitution radiometer
and the development of the predictable quantum efficiency detector is described.
Photon-counting principles for the realization of the radiometric and photometric
quantities are introduced and finally the route for the realization of the candela, the
measurement of the luminous intensity, is depicted.
1. – Introduction
Radiometry is the metrology domain devoted to electromagnetic radiation measure-

ments, including visible light. In particular, photometry measures the visible light taking
into account optical sensibility of human eye. Radiometry performs light measurements
over the complete electromagnetic spectrum while photometry is limited to visible range,
between 380 nm and 830 nm, where human eye is sensitive.
All radiometric quantities have a corresponding photometric quantity where power or
energy are conveniently weighted by visible perception of the average human eye. The
345
346 Giorgio Brida
photometric quantity Luminous Intensity is one of the seven base units of the Interna-
tional System (SI); the corresponding measurement unit is the candela (cd). Photometry
is essential to evaluate light sources and generally lighting devices, luminous signals, dis-
play and every application where the light has to be seen by human observers. Radiom-
etry, on the other hand, is important in all those applications where human being is not
involved, but a measurement in terms of energy and/or power is necessary, as for example
the characterisation of optical fibre devices, photovoltaic installations, photolithography
for nano-micro-fabrication, optical sensors for environment monitoring, high-power lasers
used in metal cutting and welding, single-photon methods for quantum technologies.
2. – Primary standards
A primary standard is an instrument that can generate or detect the energy (or
power) of an optical radiation by a direct reference to another measurable physical phe-
nomenon and does not need calibration against any other radiometric quantities. There
are two types of primary standards in general use for realizing radiometric units: primary
standards based on sources (blackbody source and synchrotron radiation) and primary
standards based on detectors (Electrical Substitution Radiometer).
.
2 1. Blackbody source. – The emitted power by a perfect blac-body cavity is given by
the Planck’s law:
c1 1
L(λ, T ) = ,
c2
λ5 exp( λT )−1
where L(λ, T ) is the spectral radiant power emitted by the blackbody, i.e. the power per
unit area per unit of wavelength interval, λ is the wavelength and T is the thermodynamic
temperature. The values of the constansts c1 and c2 are
c1 = 3.7415 · 10−16 W/m2 and c2 = 1.4388 · 10−2 m K.
The spectral power distribution of some blackbody sources, at different temperatures, is

shown in fig. 1. A blackbody source operated around room temperature has a peak in
the emitted spectrum in the infrared region around 10 μm; a blackbody source with a
peak in the visible region must be heated at higher temperature, in the range between
3000 K and 6000 K.
For metrological purposes, Planck’s law can be exploited in two different ways:
– A blackbody with known temperature can be used as a radiation-source standard

to realize radiometric units.
– A blackbody can also serve as a temperature standard, if the emitted (spectral or

total) radiance is measured with sufficient accuracy.
Hence, there is a close connection between blackbody-based radiometry and radiation

thermometry. Today, in high-accuracy applications, the thermodynamic temperature of
Measuring light 347
Fig. 1. – Spectral emission of blackbody sources at different temperatures.
the blackbody cavity is determined by use of a calibrated filtered detector, commonly

speaking referred as a “filter radiometer”. A blackbody at the freezing temperature
of platinum (2045 K) was at the core of a previous definition for the realization of the
candela (9th CGPM, 1948). In 1979, because of the experimental difficulties in realizing
a Planck radiator at high temperatures and the new possibilities offered by radiometry,
i.e., the measurement of optical radiation power, the 16th CGPM (1979) adopted a new
definition of the candela.
.
2 2. Synchrotron radiation source. – Electrons travelling at high velocity on a storage
ring emit synchrotron radiation. A synchrotron-radiation source provides a broadband,
continuous spectrum that ranges from infrared to x-ray wavelengths with high brightness
and low beam divergence. In addition, the properties of the synchrotron radiation (such
as spectral distribution, polarization, and angular divergence) can be accurately calcu-
lated from the machine parameters (geometry, electron current, magnetic field) through
the use of Schwinger equation (fig. 2) [1]. Synchrotron radiation covers a huge dynamic
range in photon flux of about 12 decades by properly adjusting the number of stored
electrons, in the range from maximum current to a single stored electron, without chang-
ing the shape of the emitted spectrum. These characteristics make synchrotron radia-
tion much superior to conventional laboratory light sources for UV, vacuum ultraviolet
(VUV), and x-ray radiometry.
.
2 3. Electrical Substitution Radiometer . – Electrical-substitution radiometers (ESRs)
serve as primary standards for optical power at many national metrology institutes
around the world. In the ESR the heating effect produced by optical radiation im-
pinging on an optical absorber is compared with the effect produced by electrical power
348 Giorgio Brida
Fig. 2. – Typical spectral power distribution for synchrotron radiation at different electron
energies.
heating. Figure 3 shows the key components of an ESR: an optical absorber, a ther-
mometer, an electrical heater and a thermal link. The optical absorber heats up when
irradiated by laser; to achieve absorption close to unity light traps or cavity design are
used for multiple reflections. Temperature rise of the optical absorber is measured with
a thermometer; with electrical heating we reproduce an equivalent temperature change.
Fig. 3. – The Electrical Substitution Radiometer (ESR) optical absorber (black-painted cavity),
heater, temperature sensor.
Measuring light 349
Fig. 4. – Electron microscope view of the multiwall Vertically-Aligned Carbon Nanotubes

(VACNT) (kind courtesy of NIST Boulder).
For the substitution principle, when the electrical power heating effect equates the
optical heating effect, then the electrical power is equal to the optical power impinging on
the absorber. In this way the optical power measurement is traced to SI electrical units.
This technique is today implemented in instruments operated at cryogenic temperatures
with the benefits for the reduced specific heat capacity and higher thermal diffusivity
of the absorber materials (shorter time constant), low background radiation, supercon-
ducting electrical connections [2, 3]. The ESR in use today are operated at liquid helium
(4.2 K), with relatively large optical absorbing cavity (about 10 s of time constant), in
laser based facility (collimated beam), at around 100 μW power level. The measurement
uncertainty is typically around 100 ppm.
3. – The state of the art in primary standard

The improvement of primary standards for radiometry follows two main streams:
carbon nanotube cryogenic radiometer under development by NIST, and Predictable
Quantum Efficiency Detector developed within a European consortium of National
Metrology Institute.
.
3 1. Carbon nanotube for radiometers. – A new cryogenic ESR has been developed by
NIST, gaining advantage from lithographic and micromachining fabrication techniques,
integrating together the optical absorber, thermometer, electrical heater and thermal
link in an engineered single chip [4]. The optical absorber is based on vertically aligned
carbon nanotubes (VACNT), the blackest known substance (fig. 4). However, typical
VACNT samples are not perfect optical absorbers and their reflectance increases at short
wavelengths. The reflectance of VACNT depends on many variables —including density,
350 Giorgio Brida
Fig. 5. – Planar bolometric radiometer designed for optical fiber power measurements. The chip
includes the optical absorber (VACNT), the thermal link (micro-machined Si leg), thermometer
and electrical heater (kind courtesy of NIST Boulder).
height, uniformity, and alignment. The measured optical reflectance at 1550 nm, was
6.5 × 10−4 which should be even reduced by exposure to oxygen plasma (ashing) below
100 ppm in visible [5-7]. Anyway, the absorption of VACNT is high enough to allow a
planar design for the optical absorber (avoiding light trap or cavity) and the integration
with the remaining key components of the ESR (thermometer, heater and thermal link)
on a single chip (fig. 5). This allow to design an ESR with shorter time constants (8 ms),
high reproducibility and low cost compared to traditional cryogenic radiometers.
.
3 2. Predictable Quantum Efficiency Detector . – The Predictable Quantum Efficiency
Detector (PQED) is a new radiometric standard based on the photoelectric effect in
silicon semiconductors working at liquid nitrogen temperature, developed under Euro-
pean Joint Research Projects. The spectral responsivity, R, of an ideal photoelectric
detector can be expressed using fundamental constants e, h, c and vacuum wavelength
λ, according to R(λ) = eλ/hc. This equation is based on the assumption that every
incident photon generates exactly one elementary charge carrier. The performance of
specially designed silicon detectors can approach the performance of the ideal quantum
detector in the visible wavelength range, although the photon-to-electron conversion is
always slightly affected by charge carrier losses and reflection. In 1978, Hansen developed
induced junction photodiodes to be used as an alternative to diffused p-n junction pho-
todiodes [8]. The advantage of these photodiodes is improved charge carrier collection
efficiency when used with reverse bias voltage. Geist et al. [9] proposed, using induced
junction photodiodes at cryogenic temperatures, to further reduce internal losses. Recent
modelling of internal carrier losses in induced junction photodiodes, by Gran et al. [10],
shows that when operating at low temperatures and in reverse bias mode, induced junc-
Measuring light 351
Fig. 6. – PQED photodiode on a PCB.
tion photodiodes should be capable of converting absorbed photons to electron-hole pairs

with carrier losses smaller than 1 ppm in the 400 nm to 600 nm wavelength range. More-
over, the simulations also indicate that photodiodes work well at room temperature over
the full visible wavelength range having a known external quantum efficiency within an
uncertainty of approximately 100 ppm [10]. A critical parameter in designing photodiodes
is the thickness of the silicon dioxide layer on top of the silicon substrate. The thicker
the oxide, the easier it is to obtain the silicon inversion layer. With a properly chosen
thickness, specular reflectance can be suppressed significantly when using p-polarized
incident light in a multiple reflection trap detector. The photodiodes produced (fig. 6)
have active areas of 11 × 22 mm2 with an oxide layer thickness of 220 nm or 300 nm.
Photodiode oxide thickness and its spatial uniformity were measured with the INRIM
ellipsometer facility [11]. To reduce reflectance losses, the induced junction photodiodes
were assembled into a light trapping configuration (fig. 7) designed to develop a quan-
tum detector with efficiency close to 100% [12]. PQED responsivity was predicted on the
basis of fundamental constants and material parameters [10, 13]. Two prototype PQEDs
were assembled and validated; the spectral responsivity of the PQEDs was measured
with cryogenic electrical substitution radiometers at the Physikalisch-Technische Bunde-

sanstalt (PTB) and the Czech Metrology Institute (CMI); the experimental results show
good agreement with the modelled response of the PQED to optical radiation, as well as
near-unity external quantum efficiency [14].
The outstanding performance of the PQED at 77 K offers an opportunity to push the
boundaries of fundamental physics. Comparing the results obtained from tests conducted
on high accuracy standards based on thermodynamics (i.e., cryogenic radiometers) and
electromagnetism (PQEDs) will play a significant role in searching for consistency be-
tween the two fields. Fundamental limitations can be tackled by basic research on the
interaction between light and matter, on new materials and new structures with revolu-
tionary photonic properties.
352 Giorgio Brida
Fig. 7. – 3D structure of the PQED.
4. – Counting photons
The “mise en pratique” for the definition of the candela and associated derived
units [15] includes information on the practical realization of units for photometric and
radiometric quantities using photon-number-based techniques because of recent advances
in the generation and manipulation of individual photons show great promise of produc-
ing radiant fluxes with a well-established number of photons.
.
4 1. Spontaneous Parametric Down-Conversion (SPDC). – SPDC is a quantum effect
consisting of the spontaneous decay, inside a non-linear crystal, of one photon from a
pump (p) beam, into a pair of photons arbitrarily referred to as signal (s) and idler (i)
(fig. 8) [16]. This decay process must obey the laws of energy conservation, ωp = ωs + ωi ,
and momentum conservation (or phase matching), kp = ks + ki , where ωp , ωi and ωi

are the optical frequencies, and kp , ks and ki the wave vectors (within the crystal) iden-
tified by their subscripts. The probability of spontaneous decay into a pair of correlated
photons is usually very low, of the order of 10−9 or lower, therefore with typical pump
power of the order of some milliwatts, the fluorescence emission lies at the level of the
photon-counting regime. As the photons are produced in pairs and because of restric-
tions imposed by energy and momentum conservation, the detection of the direction and
energy of one photon indicates the existence of a pair-correlated photon, with definite
energy and direction. Using coincidence-measurement techniques, these characteristics
allow absolute calibration of a photodetector without the need for an absolute radio-
metric reference [17, 18]. If two photon-counting detectors are placed downstream of the
Measuring light 353
Fig. 8. – Experimental set-up for the measurement of detection efficiencies by means of twin
photon pairs generated by SPDC.
non-linear crystal, along the directions of propagation of the correlated-photon pair, the
detection of an event by one detector guarantees the presence of a photon in the con-
jugate direction. If N is the total number of photon pairs emitted from the crystal in
a given time interval, Ns , Ni and Nc are the mean number of events recorded, in the
same time interval, by the signal detector, idler detector and in coincidence, respectively,
we have the following relationships: Ns = ηs N and Ni = ηi N , where ηs and ηi are the
detection efficiencies on the signal and idler arms, respectively. The number of events in
coincidence is Nc = ηs ηi N , owing to the statistical independence of the two detectors.
The detection efficiencies are easily derived from the three measured values: ηs = Nc /Ni ,
ηi = Nc /Ns .
.
4 2. Single-Photon Source (SPS). – In radiometry a predictable single-photon source
i.e., a source that emits a calculable number of photons at a specific rate and wavelength,
could act as standard source. The development of a new standard sources based on
the emission of one photon at a time is a relevant step forward for the photon-counting
metrology. The possibility to increase the emission rate of these sources will help to bridge
the gap between classical radiometry and photon-counting regime [19, 20]. In the last
years, a lot of resources have been devoted to identify new single-photon sources (SPS)
and improve the specifications of those known [21-23]. Three main types of SPS have
been investigated: a) defect centres in nano-diamond, b) quantum dot and c) molecule.
The defect centres in nano-diamond, typically nitrogen vacancy (NV) or silicon vacancy
(SiV), are stable emitters at room temperature, optically pumped, with emission rate
up to 106 photons/s; their main disadvantage is the low collection efficiency for emitters
in bulk material. Quantum dots are typically based on GaAs semiconductor able to
operate at telecomm wavelength and are electrically pumped; their disadvantage is the
354 Giorgio Brida
Fig. 9. – Traceability chain for the realization of the candela.
low-temperature operation and the low collection efficiency. Finally the molecule-based
SPS have shown the highest emission rate but again they must be operated at low
temperature in order to avoid degradation working at room temperature. The collection
efficiency of each of these emitters could be greatly enhanced using proper coupling
structures [24].
5. – Photometry – The realization of the candela
According to the 16th Conference of Weights and Measures of 1979, the SI unit, the
candela, is the luminous intensity, in a given direction, of a source that emits monochro-
matic radiation of frequency 540 × 1012 hertz and that has a radiant intensity in that
direction of 1/683 watt per steradian. An important consequence of this definition is the
possibility to realize the candela based on ESR and laser sources. Figure 9 shows a typi-
cal traceability chain from the ESR to the luminous intensity standard lamps: a transfer
detector is calibrated against a cryogenic radiometer at a number of discrete wavelength;
by means of proper interpolation techniques the spectral responsivity of the transfer de-
tector is extended over all the visible wavelength range. A silicon photodiode is spectrally
calibrated against the transfer detector and then is spectrally corrected adding properly
designed filters in such a way that the product of the relative spectral responsivity of the
silicon photodiode and the relative transmittance of the filter match the relative spectral
response of the average human eye V (λ), as close as possible. This filtered photodiode,
with a calibrated aperture area in front, realizes a reference photometers to be used to
calibrate the luminous intensity of standard lamps.
Measuring light 355
REFERENCES
[1] Schwinger J., “On the classical radiation of accelerated electrons”, Phys. Rev., 75 (1949)
1912.
[2] Martin J. E., Fox N. P. and Key P. J., “A cryogenic radiometer for absolute radiometric
measurements”, Metrologia, 21 (1985) 147.
[3] Hoyt C. C. and Foukal P. V., “Cryogenic radiometers and their application to
metrology”, Metrologia, 28 (1991) 163.
[4] Tomlin N. A., White M., Vayshenker I., Woods S. I. and Lehman J. H., Metrologia,
52 (2015) 376.
[5] Lehman J., Sanders A., Hanssen L., Wilthan B., Zeng J. and Jensen C., Nano
Lett., 10 (2010) 3261.
[6] Chunnilall C. J., Lehman J. H., Theocharous E. and Sanders A., Carbon, 50 (2012)
5348.
[7] Tomlin N., Curtin A., White M. and Lehman J., Carbon, 74 (2014) 329.
[8] Hansen T., “Silicon UV-photodiodes using natural inversion layers”, Phys. Scr., 18 (1978)
471.
[9] Geist J., Brida G. and Rastello M. L., “Prospects for improving the accuracy of
silicon photodiode self-calibration with custom cryogenic photodiodes”, Metrologia, 40
(2003) S132.
[10] Gran J., Kübarsepp T., Sildoja M., Manoocheri F., Ikonen E., and Müller I.,
“Simulations of a predictable quantum efficient detector with PC1D”, Metrologia, 49 (2012)
S130.
[11] White M., Lolli L., Brida G., Gran J. and Rajteri M., “Optical constants and spatial
uniformity of thermally grown oxide layer of custom, induced-junction, silicon photodiodes
for a predictable quantum efficient detector”, J. Appl. Phys., 113 (2013) 243509.
[12] Sildoja M., Manoocheri F. and Ikonen E., “Reflectance Calculations for a Predictable
Quantum Efficient Detector”, Metrologia, 46 (2009) S151.
[13] Sildoja M., Manoocheri F., Merimaa M., Ikonen E., Müller I., Werner L.,
Gran J., Kübarsepp T., Smı̂d M. and Rastello M. L., “Predictable quantum efficient
detector: I. Photodiodes and predicted responsivity”, Metrologia, 50 (2013) 385.
[14] Müller I., Johannsen U., Linke U., Socaciu-Siebert L., Smı̂d M., Porrovecchio
G., Sildoja M., Manoocheri F., Ikonen E., Gran J., Kübarsepp T., Brida G. and
Werner L., “Predictable quantum efficient detector: II. Characterization and confirmed
responsivity”, Metrologia, 50 (2013) 395.
[15] Zwinkels J., Sperling A., Goodman T., Campos Acosta J., Ohno Y., Rastello L.
M., Stock M. and Woolliams E., “Mise en pratique for the definition of the candela and
associated derived units for photometric and radiometric quantities in the International
System of Units (SI)”, Metrologia 53 (2016) G1.
[16] Klyshko D. N., Photons and Nonlinear Optics (Gordon and Breach, New York) 1988.
[17] Brida G., Castelletto S., Novero C. and Rastello M. L., Metrologia, 35 (1998)
397.
[18] Migdall A. L., Datla R. U., Sergienko A., Orszak J. S. and Shih Y. H., “Absolute
detector quantum-efficiency measurements using correlated photons”, Metrologia 32
(1995/96) 479.
[19] Cheung J. Y., Chunnilall C. J., Woolliams E. R., Fox N. P., Mountford J. R.,
Wang J. and Thomas P. J., “The quantum candela: a re-definition of the standard units
for optical radiation,” J. Mod. Opt., 54 (2007) 373.
356 Giorgio Brida
[20] Zwinkels J. C., Ikonen E., Fox N. P., Ulm G. and Rastello M. L., “Photometry,
radiometry and “the candela”: evolution in the classical and quantum world”, Metrologia,
47 (2010) R15.
[21] Oxborrow M. and Sinclair A., “Single-photon sources”, Contemp. Phys., 46 (2005)
173.
[22] Scheel S., “Single-photon sources – an introduction”, J. Mod. Opt., 56 (2009) 141.
[23] Buller G. S. and Collins R. J., “Single-photon generation and detection”, Meas. Sci.
Technol., 21 (2010) 012002.
[24] Chen X. W., Götzinger S. and Sandoghdar V., “99% efficiency in collecting photons
from a single emitter”, Opt. Lett., 36 (2011) 3545.
DOI 10.3254/978-1-61499-818-1-357
Why redefine the kelvin?

Michael de Podesta
National Physical Laboratory - Teddington, TW11 0LW, UK
Summary. — Every day people make millions of successful temperature measure-

ments. The people who make these measurements frequently identify ways in which
they would like temperature measurement to be improved. Typically people would
like results with lower uncertainty and at lower cost, particularly at extremes of
temperature. But nobody has ever expressed — to me at least — any dissatisfac-
tion with the definition of what we mean by “one degree”. Nonetheless, in 2018,
the CIPM will introduce a new definition of the kelvin and degree Celsius. In this
paper I discuss the background to this redefinition and explain why — despite not
addressing users’ pressing needs — the redefinition is desirable and indeed long
overdue.
1. – Measurement and unit quantities

.
1 1. Measurement and the role of a unit definition. – Measurement — in its most basic
conception — is simply the comparison of one object or state with another. Although
this may be a useful process when considered on a case-by-case basis, it gains power
when developed in two characteristic ways. Firstly when one of the objects or states is a
standard object or state, against which many different objects or states may be compared.
And secondly when the comparison is quantitative. Together these two characteristics
have created measurement in its modern sense, and enabled humanity to enumerate our
357
358 Michael de Podesta
experience of the universe, share our observations, and created the disciplines of science
and engineering.
The first development requires the ability to reproduce a standard object or state
with which all unknown objects or states can be compared: this is the main topic of
this article. This standard object or state effectively defines what we mean by “one” of
whatever it is we wish to measure: it is a unit definition. So, for example, a kilogram, is
defined as the mass of the International Prototype of the Kilogram (IPK) and so defines
what we mean by “one kilogram”. A slightly more complex example involves the kelvin
where we define the temperature of the triple point of water (TPW) to be 273.16 kelvin.
This effectively defines a unit of temperature by saying that the unit is such that there
are 273.16 of them between absolute zero and the temperature of the triple point of
water. This second example requires more physical understanding to make sense of it,
but it does unambiguously define a standard state.
In the jargon of measurement science the reproduction of a standard object or state
is known as “realising” a unit. And it is a key insight of measurement science that no
measurement can ever have a lower uncertainty than the uncertainty associated with
the realisation of the unit quantity. “Realising” a kilogram means making copies of the
IPK that are closely similar in mass. “Realising” the kelvin means reproducing of the
temperature of the triple point of water. In both cases, the ability to reproduce the
standard object or state represents a fundamental limit to measurement uncertainty.
The second development requires the use of instruments which respond in a pre-
dictable way to changes in the quantity being measured. This allows us to specify how
much bigger or smaller one property of an object or state is than the standard object
or state. This process of quantitative comparison becomes more difficult as we consider
comparisons between objects or states that differ greatly in magnitude from our standard
object or state. For example, electrical currents measured in practice vary from pico-
amperes (10−12 A) to mega-amperes (106 A); similarly length measurements cover the
range from nanometres (10−9 m) to millions of kilometres (109 m). Our ability to make
quantitative measurements across this astonishing range has evolved — and improved —
over the centuries in ways that would once have been unimaginable.
So measurement can be considered to consist of two processes: firstly, realisation of
standard objects or states based on the definition of a unit; and secondly quantitative
comparison against these standards. Alongside these two processes, we must also mention
the concept of measurement traceability. This is the idea that a measurement made at a
particular time can be traced back to a particular realisation of the unit quantity. This
enables the meaningful comparison of traceable measurements made in different places,
at different times, and using different techniques. If these diverse measurements can all be
shown to be traceable to the same unit definition then they can be fairly intercompared.
Our modern world implicitly assumes a high level of metrological traceability. This
capability is built upon a metrological infrastructure of which many people are unaware.
And the foundation of our current ability to make measurements and compare them
meaningfully with measurements at other times and places is the International System
of Units.
Why redefine the kelvin? 359
Fig. 1. – Representation of the links between the seven base units of the International System of
Units. Notice that the kilogram is a “source definition”, i.e. its definition affects the definition
of, for example, the mole. Notice that the kelvin definition is currently not connected to the
definition of any other SI base units.
.
1 2. The International System of Units. – The International System of Units, the
SI [1] defines seven “base” units that form the foundation of measurements in the physical
sciences. These consist of four “core” units built upon the metre-kilogram-second-ampere
(MKSA) system of units, and three additional units: the mole, candela and kelvin (fig. 1).
The SI is overseen by the International Committee on Weights and Measures (CIPM)
and a series of consultative committees, one for each unit. The consultative committees
contain measurement experts from states which are signatories of the Metre Convention
(1875). Overall, the role of the CIPM is to make sure that the definitions of the units
and the recommended procedures for their dissemination stay relevant to the needs of
users worldwide.
In recent years, CIPM has been concerned with two fundamental problems regarding
the definitions of the kilogram and the ampere. And their reflections have led CIPM to
additionally consider the definitions of the mole and the kelvin. The present definitions
. .
of the kilogram and ampere will be discussed in sects. 1 4 and 1 5, but first we consider
a more general question: what makes a “good” unit definition?

.
1 3. What makes a “good” unit definition? – The world of metrology and the commit-
tees of the CIPM are generally conservative in their outlook, and with good reason. The
SI forms the basis of every single measurement in science and engineering throughout
the entire world. Thus it would be unwise to embark upon sudden changes in the SI that
might have unintended consequences.
Nonetheless, in 2018 it is planned to embark upon profound changes in the definitions
of four of the base units of the SI: the kilogram, the mole, the ampere and the kelvin.
In this paper we will be mainly considering the kelvin definition, but before we focus on
the kelvin, let us consider what properties a “good” definition of a measurement unit
should have. The following list is not intended to be exhaustive, but rather indicative.
A unit definition should be: comprehensible; unambiguous; practical; and enduring. Let
us look at each of these properties in turn.
Comprehensible: In a way this requirement is unnecessary because all scientific or
engineering specifications should embody this property. But the requirement of unit
definitions is especially important since they will, hopefully, be of universal use. For ex-
ample, we need to consider whether the definition can be understood by school students.
Unambiguous: This feature is technically critical. Any ambiguity in the definition is
likely to lead to variability between different realisations of the same unit quantity.
Practical : If a specification of a unit does not enable copies of the unit to be made
without excessive cost or effort, it is likely to fall into disuse. As part of this category we
can also include the ease with which multiples or sub-multiples of the unit quantity can
be realised.
Enduring: The history of measurement science has seen a succession of systems of
units, each with a bold rationale reflecting political reality and scientific understanding.
For example the switch from “Imperial” units based on English legal metrology using
pounds, feet and inches, has gradually given way to the SI, a system based on interna-
tional agreement. This is partly for political reasons, and partly because of the technical
superiority of the SI.
The transitions between unit systems have occurred for political, scientific and tech-
nological reasons. However, whatever the reason for these changes, they are expensive
and disruptive, and ideally we would choose definitions that will not require replacement
or modification. This requirement is especially challenging in the face of the breath-
taking technological changes that have occurred in the last century. Somehow, we need
definitions which will remain appropriate in the face of technological change which — by
its nature — cannot be clearly foreseen.
When we look at scientific instruments that are 100 years old, they generally look
like they belong in a museum rather than a laboratory. This is true even when they are
still fully functional. And with this perspective, we have to imagine that in a further
100 years, our current “state of the art” measuring instruments will also look like museum
pieces. If we lengthen the period of comparison to 1000 years, the contrast with the past
and the challenge for the future becomes even more striking. And yet we would like to
build a system of units which in principle will not need to be changed over this period.
With these four requirements in mind (comprehensibility, ambiguity, practicality, and
endurance) but most particularly the last requirement of endurance, in the following
two sections we will consider the reasons that the CIPM are considering revising the
definitions of the kilogram and the ampere.
.
1 4. Historical Decay: The kilogram. – The SI defines what is meant by one kilo-
gram [1] as
The kilogram is the unit of mass; it is equal to the mass of the international proto-
type of the kilogram.
This definition is certainly comprehensible, and there appears at first to be no am-

biguity. However in practice, the International Prototype of the Kilogram (IPK) only
weighs one kilogram immediately after it has been rubbed with a goatskin cloth and
then steam-cleaned in a process known as “nettoyage et lavage” [2].
So perhaps there is some ambiguity. For example we might ask: What type of goatskin
is suitable? Or perhaps: For how long does the IPK have a mass of one kilogram after
cleaning? From our knowledge of surface physics, we anticipate that all solids in air
will have a surface layer of adsorbed molecules, typically including water. Does the IPK
weigh one kilogram with or without this native layer ? If we estimate that the surface
layer consists of water and is a single molecule thick (∼ 0.1 nm) then on a right circular
cylinder 39 mm in height and diameter, this layer will have a mass of approximately
1 microgram. One can argue that at the level of 1 microgram in a kilogram — 1 part in
109 of the mass — these questions represent potential ambiguities.
In fact most of these questions have been investigated (e.g., [3]) and the answers are
that this “native layer” must be removed, and that within about 1 microgram, after the
kilogram has been cleaned twice, it will stay “clean” for up to 3 weeks. So perhaps these
questions represent ambiguities at the level of 1 part in 1010 ?
And if the problem of surface contamination does not represent an ambiguity, then
it certainly presents practical problems. Of course, scientists and technologists have
developed solutions to these problems, but nonetheless, there are limits to the perfection
with which the mass of a secondary standard kilogram can be evaluated in terms of the
mass of the IPK: currently these amount to a few parts in 109 at the best laboratories
in the world.
It is also clear how multiples and sub-multiples of the kilogram can be created to
allow measurement comparisons at different scales. First copies of the kilogram can be
created, compared with each other and the IPK, and then weighed in combination to
create larger mass standards. Similarly submultiples of the kilogram can be created by
making standards that can be shown to be equal, but which when weighed together add
up to one kilogram. The fractional uncertainty of multiples and submultiples is larger
than for the kilogram, and for small masses, this can be quite significant. However, for
most “normal weighings” the definition of the kilogram is still highly practical.
And finally we come to the question of endurance. How has the definition of the
kilogram stood up to the test of time? The answer is that it has stood up remarkably
well, and there are no immediate practical problems which would lead us to change the
definition. However we have made observations that cause us to suspect that the IPK
has changed its mass in the 128 years since it was adopted [2-4]. And these small changes
have caused metrologists to reflect on the nature of the definition of units.
The evidence of drift in the mass of the IPK was acquired because the creators
of the IPK also manufactured an ensemble of nominally identical kilograms, some of
which were retained at BIPM and some of which were distributed to nations around
the world, and which have formed the basis of our globally-coherent system of mass
measurement. Periodically (1889, 1948, 1989 and 2014) these copies have been returned
to the International Bureau of Weights and Measures (BIPM) and (after nettoyage et
Fig. 2. – Changes in masses of some of the official copies of the IPK held at BIPM [4].
lavage) compared with the IPK. The evidence is that the copy masses have shifted in
mass with respect to the IPK. Figure 2 shows a comparison between some of the official
copies held at BIPM against the IPK. Since all the copies are kept in similar conditions
to the IPK, the implication is that the IPK itself has probably changed mass, but we
have no way even in principle of assessing the extent of that shift.
Based on general considerations we can estimate that the IPK is probably changing
mass by something on the order of 1 part in 1010 (0.1 micrograms) per year. This is
currently of little practical consequence, and if it were merely a problem of realising a
kilogram, then we might await some technological progress to improve matters. However
what is profoundly unsatisfactory is that the IPK represents a limit to the uncertainty
associated with the definition of what we mean by one kilogram.
The IPK was made to replace the “kilogram of the archive” which had been made of
pure platinum. Advances in smelting technology and metallurgy enabled the casting of
the IPK from a platinum-iridium alloy which has proved to be astonishingly resistant to
chemical reactions with the components of atmospheric air. However, even this amazingly
stable alloy, cast into an artefact which is cared for with extreme delicacy, does appear
to be chemically reacting with the atmosphere. We might think to replace the IPK with
a new definition, perhaps based on the mass of a particular silicon sphere [5] which might
be more stable. However to do this would be to fail to learn the lesson of history.
The lesson we should learn is that technological obsolescence and historical decay of
even the most remarkable artefacts are both inevitable. So if we wish to build a system
of measurement which will remain stable over a centennial or millennial timescale, then
we need to choose definitions of units that are not related to the current “state of
the art” technology because the “state of the art” will inevitably advance (or at least
change) with time. Similarly we should avoid definitions that are based around human
artefacts because these too will inevitably change over time. Instead we should search for
definitions that will endure — to the best of our current understanding — untarnished
by the passage of time.
.
1 5. Technological irrelevance: the ampere. – The 1961 definition of the ampere [1] is:
The ampere is that constant current which, if maintained in two straight parallel
conductors of infinite length, of negligible circular cross-section, and placed 1 metre apart
in vacuum, would produce between these conductors a force equal to 2 × 10−7 newton per
metre of length.
In the opinion of this author, the comprehensibility of this definition is poor. This is,
I think, because it appears to be describing a realisation of how to recreate one ampere
— but it is not. It is describing an entirely hypothetical arrangement and attempting to
convey what we mean by one ampere in terms of the force between two wires. Nonetheless,
we can consider that the definition scores poorly on comprehensibility.
The definition is carefully worded using concepts such as “infinitely long” conductors
of “negligible cross-section” to avoid ambiguity. This wording is chosen because one
limiting uncertainty in actual realisations of the ampere is that electrical current does
not flow uniformly along wires, and so the geometrical centre of a wire, may not be an
indicator of the location of the current flow.
This careful wording means that the definition is unambiguous in its specification
of what we mean by one ampere. However this definition implies “realisation” of the
ampere by methods involving measurement of forces, which in turn implies the use of force
balances. Thus on fig. 1 we can see that the definition of the ampere is affected by the
definition of kilogram. And initially considerable research effort was expended in order to
build calculable resistors and hence derive standard resistances and voltages. When this
definition was adopted in 1961, it was not foreseen that the world of electrical metrology
was about to change dramatically. Not just from the development of semiconductor
technology and the invention of analogue-to-digital converters with astonishing linearity,
but also from the discovery of astonishing “quantum” phenomena.
In 1962, the Josephson Effect was discovered. In one version of this effect, when
electromagnetic waves (typically microwaves) illuminate a junction between two super-
conductors, a DC voltage is generated across the junction. This DC voltage has a mag-
nitude proportional to the frequency of the radiation, with a constant of proportionality
known as the Josephson Constant, KJ . Significantly, as far as can be determined, KJ
has a value equal to exactly h/2e where h is the Planck constant and e is the elementary
electronic charge. Numerous tests have shown that KJ has the same value independent
of the material from which the junction is constructed. This discovery has allowed the
development of facilities which can generate arbitrary known voltages with extremely
low uncertainty.
Then in 1980, the Quantum Hall Effect (QHE) was discovered, in which the electrical
resistance of a semiconductor could be stabilised in integer multiples of h/e2 . The equiv-
alence of the resistance values has again been tested extensively in different materials.
This discovery has allowed the realisation of electrical resistances in terms of a “natural”
unit of electrical resistance.
The Josephson Effect and the QHE have transformed electrical metrology, allowing
realisations of the ampere, volt, and ohm with uncertainties at least 10 times better than
conventional “force-based” metrology [6]. So since 1990, electrical metrology worldwide

has been based on these two quantum-mechanical effects using our best estimates of h
and e.
So the unit definition scores poorly on practicality and endurance because the ampere
as presently realised in practice does not derive directly from the definition of the unit.
In essence the definition of the ampere became technologically irrelevant because it was
linked too closely to one particular mode of realisation. The lesson we should learn from
this experience, is that even a definition that precisely expresses “what we mean” by a
quantity can become irrelevant.
.
1 6. The definition of the kelvin. – In previous sections we looked at the definition of
the kilogram, and ampere, and inferred some general principles of what to avoid in unit
definitions. Here we consider the definition of the kelvin in the light of that discussion
and our requirements of comprehensibility, ambiguity, practicality, and endurance. Let
us remind ourselves of the current definition of the kelvin in the SI [1].
The kelvin, unit of thermodynamic temperature, is the fraction 1/273.16 of the ther-
modynamic temperature of the triple point of water.
At first glance, this definition holds up well. Although many people are unfamiliar
with the concept of a “triple point” on the phase diagram of a pure substance, the idea
can be readily explained. The concept can either be elucidated as a consequences of
Gibb’s phase rules, or more operationally as being the melting/freezing temperature at
a pressure equal to the vapour pressure of water at the melting/freezing temperature.
For many years the definition was also considered to be unambiguous, but around
the turn of the 21st Century, realisations of the definition were becoming good enough
to reveal variability due to the isotopic content of the water [7]. The definition was thus
amended to specify that by “water” the definition referred to water of a specified isotopic
composition known as Vienna Standard Mean Ocean Water (VSMOW) [8].
The definition is also eminently practical. Using simple glassware, a triple-point-of-
water cell can be constructed and filled with distilled water in even primitive scientific
laboratories. This will reproduce TTPW within 1 millikelvin. A cell with isotopically-
adjusted water would cost a few thousand pounds but would enable the realisation of
TTPW with an uncertainty approaching a few tens of microkelvin [9].

And finally, since cells can be made and re-made, and old cells checked against new
ones, there is no special artefact which could age and affect temperature measurements
world-wide without our knowledge.
In summary when considering the current definition in the light of our requirements
for comprehensibility, ambiguity, practicality, and endurance, the definition performs
well. It does however have a weakness which we can appreciate when we consider trying
to measure extremes of temperature.
In the technological world we inhabit, we encounter situations in which temperatures
vary over typically three orders of magnitude. The “normally encountered” extremes can
be exemplified by (say) the temperature of a superconducting magnet in a body scanner
(< 4.2 K) and the temperature of the filament of a light bulb (2500 K). Of course, higher
and lower temperatures are encountered but let us just consider these limits for now.
The definition of the kelvin establishes TTPW as being special amongst all temper-
atures. In this sense the definition is like the kilogram definition. However, physically
there is nothing special about either the mass of “one kilogram” or the temperature of
“the triple point of water”: they just happen to be convenient for measurements. At
least they are convenient when the thing we wish to measure differs by only a small
factor from the standard. But equally, they can be inconvenient when the thing we wish
to measure differs by a large factor from the standard.
As a consequence of the definition of the kelvin, if we want to work out how hot or
cold an object is in kelvins, we need to determine how much hotter or colder an object
is than TTPW . The temperatures mentioned correspond to roughly 0.01 TTPW and 10
TTPW and establishing precisely how much hotter or colder something is than TTPW at
such large ratios is very difficult. This difficulty arises because the physical principles
used to make the comparison differ dramatically in different temperature ranges. To see
this, consider how we establish the temperature at which (say) pure palladium metal
freezes.
We first create a crucible containing pure palladium metal. The crucible is shaped to
form a deep hollow cavity known as a blackbody. We then use a radiometer to measure
the spectral intensity of radiation emitted by the blackbody as the palladium slowly
cools around it. As the palladium freezes, the latent heat causes a pause — ideally a
plateau — in the curve of brightness versus time. In this way we can establish the spectral
intensity of light emitted by the blackbody at the temperature at which palladium freezes:
TPd . Ideally we would now measure the spectral intensity of light emitted in the same
wavelength band from a blackbody at TTPW . However, the spectral intensity at TTPW is
immeasurably small, and so we need to make the comparison against TTPW in (typically)
two stages.
To do this we perform the same experiment using a crucible with molten copper
or molten silver. The temperature at which (say) molten silver TAg freezes has been
previously estimated by constant volume gas thermometry which involves measurements
of the pressure of a gas in a bulb at TAg and at TTPW . In this way we determine
the ratio of temperatures TPd and TTPW in two steps with each step being based on
a different physical principle. The key point is that we are forced into this procedure
because the definition of the unit of temperatures gives a special status to TTPW . But
in fact physically, it is just a convenient temperature for many metrological operations.
Or in this case, an inconvenient temperature.
Some consider this difficulty to be connected to the status of temperature as an
intrinsic property of a material. By this we mean that if we increase the size of an
object, then although its linear dimensions and mass and heat capacity all scale with
the size of the object — these are known as extrinsic properties — the temperature
of the object does not. Thus for example mixing two equal volumes of water at 300 K
does not create water at 600 K. The intrinsic nature of temperature for example makes
it practically impossible to create an object with (say) twice or half the temperature
of a given object. And this certainly does not help in establishing the temperatures of
phenomena that occur at temperatures much higher or lower than TTPW .
However the real problem is that the definition of the kelvin is really an operational
definition that pre-supposes a particular mode of realisation involving a triple-point-
of-water cell. The definition makes no reference to what we fundamentally mean by
temperature. So although the definition is satisfactory when tested against our require-
ments for comprehensibility, ambiguity, practicality, and endurance, its weakness is that
it does not clearly distinguish between a definition of what we mean by one kelvin and
a mode of realisation in one particular temperature range.
2. – A new type of unit definition

.
2 1. Unit definitions. – Figure 1 represents each of the SI base units as a circle, and
links them together in a circle implying that the units are all somehow similar. I hope
the discussion in the preceding sections have made clear that defining each unit quantity
presents very distinct problems. I hope too that it has become clear that arriving at
unit definitions which are comprehensible and unambiguous yet practical and enduring
is challenging.
Stimulated by the need to find a solution to the kilogram problem, and mindful of the
need to refine the ampere, the CIPM have looked more widely at all the SI base units to
look for consequences of the redefinitions, and possible further improvements.
Many options were available, but after lengthy consideration, CIPM have decided to
step back from “old style” definitions which create a unit definition in terms of the best
technology available at a particular time, and instead create definitions based on the
most stable things we know: the physical constants of the natural world [10].
CIPM’s radical proposal is to avoid using objects created by humans as standards. In-
stead they propose to use features of the natural world — the constants of nature — as the
basis for all measurements. Their motivation is that the constants of nature are the most
stable things we know. For example, as I mentioned earlier, the kilogram artefact (the
IPK) is likely to be changing mass at a rate of approximately 1 part in 1010 every year.
In contrast, we have no reason to think that the natural constants of the universe have
changed, or are changing, their value. And in fact we can set experimental limits on their
maximum possible rate of change. For example, recent measurements of a combination
1 e2
of fundamental constants known as the fine structure constant (α = 4πε 0 hc
) have been
shown them to be changing at rate of less than 2 parts in 1017 per year: at least one million
times more stable than the kilogram [11-13]. Since the kilogram has proved satisfactory
for more than a century, an improvement by a factor 10 would probably have been suffi-
cient to justify a change in definition and to settle the question for around 1000 years. An
improvement by a factor of a million represents a breath-taking step forward in metrology.
So if we create a system of measurement defined in terms of constants of nature, then
the concept of what we mean by a unit should remain unchanged, subject only to the
constancy of the laws of physics.
.
2 2. Uncertainty. – In the new conception of the SI, the definitions of units are achieved
by associating a natural constant with the definition of a unit. Currently these natural
constants — such as the Planck constant, h, have to be measured in terms of the SI base
units. For example the units of the Planck constant are joule seconds and the joule —
the SI unit of energy — is related to the definitions of the metre, second and kilogram.
So the value we associate with h has an uncertainty when expressed in units of joule
seconds.
However, the new conception of the SI considers that in fact it is the Planck constant
which is really constant — hence its name! And thus the arrow of uncertainty should be
reversed. Instead of expressing the Planck constant in terms of the current SI units, we
should assume a value for the Planck constant and then define SI units in terms of the
Planck constant.
The uncertainty that was previously embodied in our measurements of the natural
constants will now be associated, not with the constants themselves, but instead with
our ability to realise copies of the SI units in terms of the natural constants. The benefit
of this reversal is that these unit definitions and values for the natural constants can
be exact, with no uncertainty, and they should remain unchanged even in the face of
unanticipated technological developments. We should then be able to take advantage of
developments in science and technology to improve the perfection with which we realise
the unit definitions, and hence gradually improve practical metrology.
This change crystallises the separation between definitions of units and realisations
of units.
.
2 3. Making the change. – This kind of change has been done before. In 1983 the
definition of the metre was changed to define a metre in terms of the speed of light in a
vacuum, c, a quantity which we believe is universally constant and unchanging [14]. The
procedure for making this change involved many years research to measure the speed
of light (in metres per second) in terms of the previous definition of one metre and one
second. When research groups independently measured c and their results agreed within
the estimated uncertainty Δc at a low enough level, then the switch was made to the
new definition.
Before the switch, the metre was defined as an exact length, specified as an exact
number of wavelengths of light of a specified origin, and the speed of light c expressed in
terms of this metre definition had an uncertainty Δc. After the redefinition, the speed
of light c was considered to have an exact value with no uncertainty, and it now became
our measurement standard. The metre was then defined as exactly the distance travelled
in vacuum by light during the time 1/c seconds.
The uncertainty of measurement now arises from our ability to realise this exact
definition in practice. As had been anticipated, over the 34 years since the redefinition,
our concept of the speed of light in vacuum has not changed, and so neither has our
definition of one metre. However, even over that relatively short period, our ability
to realise that definition in practice has improved significantly due to developments in
technology.
Fig. 3. – Following on fig. 1, the figure shows a representation of the links between seven base
units of the International System of Units after the 2018 redefinition. The grey circles contain
the natural constant most closely associated with unit. Notice that the kelvin definition is now
connected to the definition of three SI base units: the second, the metre and the kilogram, that
together define the unit of energy, the joule.
The plan for the 2018 redefinitions is similar. Since 2007 measurement scientists from
around the world have been re-measuring fundamental constants, particularly the Planck
constant h and the Boltzmann constant kB . Measurements will continue until July 2017,
but already (in 2016) measurements have reached a level of agreement and with low
enough uncertainty to allow the redefinition to go ahead with negligible disturbance to
the worldwide measurement system.
The SI will then define the measurements units by stating the exact value of constants
of nature in this system of measurement [10]. Specifically, the SI, is the system of units
in which
– the unperturbed ground-state hyperfine-splitting frequency of the caesium-133

atom ΔνCs is 9 192 631 770 Hz,
– the speed of light in vacuum c is 299 792 458 m/s,
– the Planck constant h is 6.626 070 040 ×10−34 J s,
– the elementary charge e is 1.602 176 620 8 ×10−19 C,
– the Boltzmann constant kB is 1.380 648 52 ×10−23 J/K,
– the Avogadro constant NA is 6.022 140 857 ×1023 mol−1 ,
– the luminous efficacy Kcd of monochromatic radiation of frequency 540 ×1012 hertz
is 683 lm/W
Fig. 4. – An alternative representation showing the same links between the base units after the
2018 redefinition as fig. 3. Although less elegant, this construction emphasises the clustering of
the three SI base units: the second, the metre and the kilogram, that together define the unit
of energy, the joule.
and the numerical values of these seven defining constants have no uncertainty. Note:
These numerical values are from the 2014 CODATA evaluation [15] and will be revised
using the latest results in 2017.
The values of these constants are sufficient to allow us to realise the base units of
the SI: the second, metre, kilogram, ampere, kelvin, mole and lumen. The latest tech-
niques for realising the measurement units will be described in a series of accompanying
documents known as ‘Mise en Pratique’s
The redefinitions change the links between the unit definitions shown in fig. 1 and the
revised dependencies and links to relevant natural constants are shown in fig. 3. Also
shown in fig. 4 are the same links, but with the second, metre and kilogram definitions
clustered together to show how they define the SI joule. This construction is less elegant
but shows the dependence of both the candela and the kelvin on the joule, a similarity
which can be difficult to see in fig. 3.
3. – The kelvin redefinition

.
3 1. The new definition. – Using the current definition, TTPW is considered to be ex-
actly 273.16 K (0.01 ◦ C). As the only exactly-known temperature, calibration at TTPW is
the starting point for the calibration of every high-quality thermometer. Indeed, because
of this definition, ultimately every temperature measurement is a comparison with TTPW .
After the redefinition the kelvin will be defined such that the Boltzmann constant kB
has the exact value 1.380 648 52 ×10−23 J K−1 (Note: this is the CODATA 2014 estimate
Fig. 5. – Currently (a) the triple point of water both defines what we mean by one kelvin,
and is also the primary means of realisation of the unit in practice. After redefinition (b) our
definition of one kelvin will be linked by a fixed value of the Boltzmann constant kB to the
statistical mechanical concept of average molecular energy, and so to the SI unit of energy, the
joule. The triple point of water will still be a useful temperature for practical thermometry,
alongside other fixed points.
and the number is likely to change slightly before the actual redefinition). This number
has been chosen to ensure that TTPW still has the same value (273.16 K) but it now has
an associated measurement uncertainty. Instead of being 273.160 000 000 . . . K exactly,
TTPW will in future be considered to be 273.160 000 K ± 0.000 156 K. This actual
uncertainty value may change slightly depending on measurement results published in
the year to July 2017.
The uncertainty in TTPW arises because after redefinition, TTPW no longer defines
what we mean by one degree. Instead it is now merely a convenient reproducible tem-
perature. Temperatures measured according to the new definition will ultimately be
comparisons of the average level of molecular energy with the SI unit of energy, the joule
(i.e. one kilogram metre second−2 ).
Currently the distinction between the definition of one kelvin and the way in which the
kelvin is realised (fig. 5) are not clearly separated. After the redefinition what we mean
by one kelvin will be clearly defined in terms of basic physics, while the temperature of
the triple point of water will become just one of many useful fixed point temperatures.
Figure 5 hints at some of the possible future benefits of the redefinition. Because the
definition of what we mean by one kelvin will be in terms of the Boltzmann constant,
realisations of the kelvin will no longer need to be based around TTPW . So establishing
traceability to kelvin no longer implies traceability to TTPW . One likely first step is that
the melting temperature of gallium TGa (at approximately 302.91 K) may be developed as
an alternative to TTPW as a reference temperature for electrical resistance thermometry.
But as we shall see, further changes are likely to follow, especially at the extremes of
temperature measurement.
Fig. 6. – Data from an intercomparison of triple-point-of-water cells from National Measurement

Institutes in several European countries [9]. Each point represents the estimated difference
between the realisation of TTPW at a particular institution and the reference value formed
from the weighted average of the estimates. Because of the current definition of the kelvin, no
temperature measurement can ever be made with lower uncertainty than the uncertainty with
which TTPW can be realised.
.
In sect. 1 1 we mentioned that the ability to reproduce the “standard object or state”
represents a fundamental limit to measurement uncertainty. One particular advantage
of the unit redefinition in terms of natural constants is that it removes any intrinsic
.
limits to accuracy caused by the definition of the unit. For example, in sect. 1 4 we saw
that there are limits to the uncertainty with which a kilogram may be realised which
are intrinsic to the nature of the IPK. Similarly, the definition of the ampere in terms
.
of forces (sect. 1 5) must again be limited by our ability to realise a kilogram. When it
comes to temperature, the current definition is limited by our ability to realise TTPW .
As evidenced by intercomparison data in fig. 6, this limit is currently a few tens of
microkelvin.
After the redefinition it will be possible in principle to make temperature measure-
ments with lower uncertainty. The words “in principle” indicate that we do not yet have
the technology to do this. But the redefinition makes sure that the definition of the kelvin
does not intrinsically provide any limit to the achievable uncertainty of measurement.
.
3 2. Impact of the kelvin redefinition.
.
3 2.1. Impact on ITS-90. The International Temperature Scale of 1990 (ITS-90) [16] is
not the topic of this article, but since ITS-90 is the procedure used worldwide to calibrate
thermometers, any impact of the redefinition on ITS-90 would be serious. However
because the redefinition does not change the value assigned to either the triple point of
water, or the temperature of fixed points of the ITS-90, the redefinition will have no
effect on calibrations carried out according to ITS-90.
Fig. 7. – Representation of how primary thermometers will operate before and after the kelvin
redefinition. Currently there is an arbitrarily chosen special temperature, and thermometers
that operate close to this temperature —such as Acoustic Gas Thermometers (AGTs) —can
produce uncertainties which are artificially reduced. This reduction in uncertainty does not
apply to thermometers which are limited to lower temperatures (such as Coulomb Blockade
Thermometers (CBTs)) or higher temperatures (such as Filter Radiometers (FR)).
Thus despite its profound philosophical significance, the immediate practical impact
of the redefinition for essentially all users will be exactly zero. This lack of any immediate
impact is deliberate: the value of the Boltzmann constant in the redefinition has been
extensively investigated using a wide range of physical principles [17] and the value
chosen so as keep the temperature of the triple point of water — and hence calibrations
according to ITS-90 — unchanged.
It may puzzle some readers that metrologists should be bothering to redefine the
kelvin when the immediate impact of the redefinition will be close to zero. The answer
is that the definitions of a kilogram, ampere, mole and kelvin are part of the foundations
of the SI, and the SI, like all complex structures, occasionally needs maintenance of its
foundations. As with physical structures, work on foundations is often expensive and

time consuming, and when completed there is little left to show for the expense and hard
work. However, one is then able to build on those foundations with greater confidence
knowing that in future, they will not “shift” or “crack” or “distort”. In short, the
redefinition is not about solving problems we have today, but is instead building a stable
foundation for centuries of future technological change.
.
3 2.2. Impact on thermometry at extreme temperatures. The effect of the redefinition
at extremes of temperature is illustrated in fig. 7, which shows three types of primary
thermometer operating at “low”, “medium” and “high” temperatures. The lower part
of the figure shows the situation before redefinition where we consider each temperature
measurement to be scaled to TTPW . The upper part of the figure shows the situation
after redefinition where there is no special temperature. Instead there will be exact values
of the Boltzmann constant and — because the Avogadro constant will also be defined
exactly — the molar gas constant defined as the product R = NA kB .
Consider for example the operation of an acoustic gas thermometer (AGT) [18]. These
devices are currently the most accurate form of primary thermometry [19], and they are
especially accurate when they operate close to the TTPW because (a) the ratios between
unknown temperatures and TTPW are close to unity and (b) there is (by definition)
no uncertainty in the value of TTPW . After the kelvin definition, the special status of
TTPW will disappear and although its value will not change it will acquire an additional
uncertainty. Thus measurements in this range will acquire a small additional uncer-
tainty. The additional uncertainty will be small. For the NPL-Cranfield AGT the un-
certainty of measurement at 303 K will increase from approximately u(k = 1) = 0.15 mK
to u(k = 1) = 0.21 mK.
However for filter radiometers (FR) [20], Coulomb blockade thermometers (CBT) [21],
and other thermometers operating far away from TTPW the redefinition will have a ben-
eficial effect. The operating range of these devices does not overlap with TTPW and so
scaling results to link back to TTPW introduces significant uncertainty to the measure-
ments. In the near future this uncertainty reduction is much less significant than other
practical limits to uncertainty of measurement, but in future it could conceivably become
a limiting uncertainty.
For example, consider the way in which the freezing temperature of the palladium
.
metal was established (sect. 1 6) by two-stage comparison back to TTPW . With the
new definition in place, a filter radiometer could use any piece of physics involving the
Boltzmann constant to measure the temperature. For example, it could still use the old
route of comparison in multiple steps to TTPW or perhaps a more convenient fixed-point
temperature, such as TGa . However, other routes to SI traceability are also possible.
For example, the temperature could be determined by measuring the absolute radiant
power in a known bandwidth emitted by a blackbody cavity. This is traceable to temper-
ature via a measurement of radiant power (joules per second) and the Planck radiation
law using a fixed value of Boltzmann constant. In a cryogenic radiometer this power can
be traceably compared by a substitution heating method with the SI watt and measure-
ments of volts and amperes. In some circumstances this route to traceability can have a
lower uncertainty than the multistage comparison with other fixed-point temperatures.
.
3 3. Why has this taken so long? – Having reflected on the kelvin redefinition for
several years, I have become convinced that there are no significant negative impacts,
and that as part of the wider redefinition of the SI, the redefinition makes excellent sense.
But one question has lingered in my mind: since the rationale seems clear and the
physics is well understood, why has this taken so long? And the answer I have found
has surprised me. I believe that it is in part because we learned to measure temperature
and apply the results practically before we understood what it was we were measuring!
This may at first seem unbelievable: Maxwell-Boltzmann statistics emerged in the
nineteenth Century! Yet below I will show that at key points in the nineteenth and
twentieth centuries, thermal phenomena were considered mysterious. Because of the

practical importance of accurate temperature measurement, practical thermometry de-
veloped along quite distinct lines from primary thermometry, and it is only now, at the
start of the twenty first century that we are finally and fundamentally linking the two
together again.
At the start of the nineteenth century, Jean Baptiste Fourier in the preface to his
Analytical Theory of Heat [22] explained that thermal phenomena were “different” from
mechanical phenomena:
“Newton comprised the system of the universe. It is recognised that the same principles
regulate all the movements of the stars, their form, the inequalities of their courses, the
equilibrium and oscillations of the seas, the harmonic vibrations of the air and sonorous
bodies, the transmission of light, capillary actions, the undulations of fluids[. . . ]
But whatever may be the range of mechanical theories, they do not apply to the effects
of heat. These make up a special order of phenomena which cannot be explained by the
principles of motion and equilibrium”.
So we can be sure that at the start of the nineteenth century, there was no consensus
on the nature of temperature and thermal phenomena. We can be sure that this state
existed in the middle part of the nineteenth century by considering the astonishing story
of John James Waterston
In 1845 Waterston submitted a paper entitled “On the Physics of Media that are
Composed of Free and Perfectly Elastic Molecules in a State of Motion” to “Philosophical
Transaction of the Royal Society of London”. In this paper he described a thoroughly
modern conception of the microscopic structure of a gas:
“[. . .]we must imagine a vast multitude of small particles of matter, perfectly alike in
every respect
[. . .]The quality of perfect elasticity being common to all the particles, the original amount
of vis viva[. . .] of the whole multitude must forever remain the same[. . .] as unchanged as
the matter that is associated with it.
The medium must in this way become endowed with a permanent state of elastic energy,
or disposition to expand, uniformly sustained in every part and communicating to it the

physical character of an elastic fluid.”
In this passage the archaic term vis viva is used to describe the kinetic energy of the
molecules 12 mv 2 . In the paper he showed that a medium of this kind would: a) obey
the laws of Mariott, Dalton and Gay-Lussac i.e. the medium would obey the ideal gas
equation; b) that the vis viva would be “like temperature” i.e. that temperature was
proportional to the average kinetic energy of the microscopic constituents of the gas; and
c) that vis viva would be equally distributed, even amongst molecules of differing mass,
which was the first statement of the equipartition principle of statistical mechanics.
He then calculated the “mechanical equivalent of heat” which had recently been mea-
sured by Joule, and stated that the “Absolute Zero” of temperature (∼ −480 ◦ F) was
when the vis viva of molecules is zero. In short, Waterston described a recognisably
modern conception of a gas.
That the paper was rejected with the comment “The paper is nothing but nonsense”
compels us to conclude that at this key period for the development of practical ther-
mometry, there was no widespread acceptance of the microscopic nature of heat at even
the foremost scientific society in the world. The paper was eventually published (posthu-
mously) in 1892 [23] with a foreword and commentary by Rayleigh stating that “The
omission to publish [this paper] at the time was a misfortune which probably retarded
the subject by 10 to 15 years”.
We can revisit the situation at the start of the twentieth century by considering
words from Josiah Willard Gibbs. In 1901 in the foreword to Elementary Principles in
Statistical Mechanics [24] he writes:
“In the present state of science, it hardly seems possible to frame a dynamic theory of
molecular action[. . . ]
Even if we confine our attention to the phenomena distinctively thermodynamic, we do
not escape difficulties in as simple a matter as the number of degrees of freedom of a
diatomic gas. It is well known that while theory would assign to the gas six degrees of
freedom per molecule, in our experiments on specific heat we cannot account for more
than five. Certainly, one is building on an insecure foundation who rests his
work on hypotheses concerning the constitution of matter”.
The emphasis here is mine. Even Gibbs one of the founding architects of statisti-
cal mechanics considered that our understanding of the microscopic nature of heat was
inadequate “for building a foundation”.
Finally we come to Jean Baptiste Perrin who was awarded the Nobel Prize in 1926
“for his work on the discontinuous structure of matter, and especially for his discovery
of sedimentation equilibrium”. Perrin investigated a wide range of physical phenomena
— from the radiation emitted by hot objects, to the way that small particles distributed
themselves at different heights in a liquid — and found that a wide range of phenomena
were described by the same numbers for what we now call the Avogadro constant and
the Boltzmann constant. Perrin concluded that [25]
“The discovery of such relationships marks the point where the underlying reality of
molecules becomes a part of our scientific consciousness”.
Recall that this was after the award of Nobel Prizes to Thomson (1906), Van de Waals
(1910), von Laue and the Braggs (1914 and 1915), Planck (1918), Einstein (1921) and
Bohr (1922). Their work related to the development of our modern conception of atoms
and even the internal structure of atoms. That it took until 1926 to award a Nobel Prize
to Perrin indicates to me that even well into the twentieth century there was scepticism
about the atomic hypothesis and it had only recently begun to be regarded as fact rather
than theory.
However while it was not until the start of the twentieth Century that the dynamical
theory of heat was fully accepted, there was great progress in practical thermometry
throughout the nineteenth century [26]. This progress was driven by the practical needs
of scientists and engineers. We will consider progress in just two particular fields: gas
thermometry and electrical resistance thermometry.
One of the problems of early thermometers, was the problem of assessing the lin-
earity of a thermometer. Unfortunately, there is no definitive way to do this without
already having a thermometer whose linearity one knew — the so-called “nomic” prob-
lem discussed by Chang [26]. From the 1840s onward, stimulated by developments by
Regnault, it became clear that the relationship between the temperature and pressure of
real gases formed a close approximation to what became known as an ideal gas. And so
gas thermometry, first with air and then later with hydrogen, led to an ability to measure
temperature linked to the basic properties of a gas rather than the particular selection
of liquid (say) chosen for a liquid-in-glass thermometer.
The development of a believably linear thermometer operating over a wide tempera-
ture range enabled the study of the temperature dependence of many physical properties
of materials. From these studies it became clear that the electrical resistance of metals
varied almost linearly with temperature and could be used as a convenient “secondary”
thermometer: a practical substitute for a “primary” gas thermometer. And as electrical
technology progressed, the resolution of electrical measurements improved rapidly and
quickly formed the basis for dissemination of practical temperature scales.
In the twentieth century, the BIPM have recommended a number of revisions in the
way these “practical scales” are implemented. They have tried to balance the need to
improve the practical scales — i.e. to make them read closer to the true (thermodynamic)
temperature, with the inconvenience of changing the scales. Thus there have been very
few significant revisions, with the main ones occurring in 1927, 1948, 1968 and 1990 [27].
The final revision established the International Temperature Scale of 1990 (ITS-90) [16,
28] that is currently used worldwide for all practical temperature measurements.
Significantly it is only recently that the uncertainty of measurement of primary ther-
mometry using gases has improved to the point such that — over some ranges of tem-
perature — it has become possible for a primary thermometer to measure true ther-
modynamic temperature with a lower uncertainty than a secondary thermometer can
reproduce a particular temperature using ITS-90 [29] (fig. 8).

The relevance of this in the context of the redefinition of the kelvin is that not only
is our understanding of temperature now free from historical scepticism, but also our
measurement techniques have evolved to the point that we can make primary temperature
measurements with a lower uncertainty than we can make secondary measurements. If in
1900, one had speculated that the uncertainty of measurement of the Boltzmann constant
(and primary thermometry in general) would improve by four orders of magnitude, this
would no doubt have seemed optimistic, but perhaps just achievable. However the specific
techniques which have been developed over the last century would have been completely
unforeseeable. In the same way, we now foresee future centuries of technical improvements
without knowing precisely what they will involve. However we can be sure that whatever
Fig. 8. – Uncertainty (in parts per million) of the reproducibility of conventional temperature
using the International Temperature Scales since 1900. Over the same period our knowledge
of the Boltzmann constant has improved by almost five orders of magnitude. Figure derived
from [29].
they are, the definition of the kelvin in terms of the Boltzmann constant will not provide
any limitation to the achievable improvements.
4. – Conclusions
As part of a wider redefinition of SI units and a reframing of concepts within the SI, the
kelvin and degree Celsius will be redefined in 2018. Although philosophically significant,
there will be almost no immediate impact on most users because thermometers calibrated
using the ITS-90 will not be affected by this redefinition. Nonetheless, the new definition
in terms of the Boltzmann constant satisfies our requirements of being comprehensible,
unambiguous, practical, and — most importantly — enduring.
It is important to bear in mind that the redefinition is not about solving problems we
have today. It is instead being undertaken to provide a stable foundation for centuries
of measurements which will be undertaken in the face of technological changes of which
we can only dream.
∗ ∗ ∗
This work was partly funded by the European Metrology Research Program (EMRP)
and the UK Department for Business, Energy and Industrial Strategy. The EMRP is
jointly funded by the EMRP participating countries within EURAMET and the Euro-
pean Union. The author would like to personally thank his NPL colleagues Robin Under-
wood, Stephanie Bell, Tom Gardiner, and Gavin Sutton for their help with this paper.
REFERENCES
[1] SI Brochure 8th edition, 2006; updated in 2014 http://www.bipm.org/en/publications/

si-brochure/.
[2] Davis Richard, The SI unit of mass, Metrologia, 40 (2003) 6.
[3] http://www.bipm.org/cc/CCM/Allowed/15/09B WGR-kg ExtraordinaryCalibrations.
pdf.
[4] Girard G., The third periodic verification of national prototypes of the kilogram (1988–
1992), Metrologia, 31 (1994) 317.
[5] Andreas B. et al., Counting the atoms in a 28 Si crystal for a new kilogram definition,
Metrologia, 48 (2011) S1 doi:10.1088/0026-1394/48/2/S01.
[6] BIPM Description of Electrical Units http://www.bipm.org/en/publications/mises-en-
pratique/electrical-units.html.
[7] Nicholas J. V., Dransfield T. D. and White D. R., Isotopic composition of water
used for triple point of water cells, Metrologia, 33 (1996) 265.
[8] Clarification of the definition of the kelvin, unit of thermodynamic temperature CIPM
RECOMMENDATION 2 (CI-2005).
[9] Report to the CCT on Key Comparison EUROMET.T-K7 (EUROMET Project 899)
Key comparison of water triple point cells Prepared by Peruzzi A. (coordinator),
Dobre M., Kerkhof O., Bosma R. Final Report (November 2008) http://kcdb.bipm.
org/AppendixB/appbresults/cct-k7/euromet.t-k7 final report.pdf.
[10] Draft SI Brochure 9th edition, http://www.bipm.org/en/measurement-units/new-si/
#communication.
[11] Safronova Marianna S., Viewpoint: Time Trials for Fundamental Constants, Physics,
7 (2014) 117.
[12] Huntemann N., Lipphardt B., Tamm Chr., Gerginov V., Weyers S. and Peik E.,
Improved Limit on a Temporal Variation of mp /me from Comparisons of Yb+ and Cs
Atomic Clocks Phys. Rev. Lett., 113 (2014) 210802.
[13] Godun R. M., Nisbet-Jones P. B. R., Jones J. M., King S. A., Johnson L. A. M.,
Margolis H. S., Szymaniec K., Lea S. N., Bongs K. and Gill P., Frequency Ratio
of Two Optical Clock Transitions in Yb+171 and Constraints on the Time Variation of
Fundamental Constants, Phys. Rev. Lett., 113 (2014) 210801.
[14] http://www.bipm.org/en/measurement-units/history-si/evolution-metre.html.
[15] Peter J. Mohr, David B. Newell, Barry N. Taylor, CODATA Recommended Values
of the Fundamental Physical Constants: 2014, http://arxiv.org/abs/1507.07956.
[16] Preston-Thomas H., The International Temperature Scale of 1990 (ITS-90), Metrologia,
27 (1990) 107 Erratum.
[17] Fischer J., Low uncertainty Boltzmann constant determinations and the kelvin
redefinition, Philos. Trans. R. Soc. A, 374 (2016) 2064.
[18] Moldover M. R., Gavioso R. M., Mehl J. B., Pitre L., de Podesta M. and Zhang
J. T., Metrologia, 51 (2014) R1.
[19] Underwood R., de Podesta M., Sutton G., Stanger L., Rusby R., Harris P.,
Morantz P. and Machin G., Philos. Trans. R. Soc. A, 374 (2015) 20150048.
[20] Sadli M., Machin G., Anhalt K., Bourson F., Briaudeau S., del Campo D., Diril
A., Kozlova O., Lowe D. H., Mantilla Amor J. M., Martin M. J., McEvoy H. C.,
Ojanen-Saloranta M., Pehlivan O., ¨ Rougié B. and Salim S. G. R., Dissemination
of thermodynamic temperature above the freezing point of silver, Philos. Trans. R. Soc.
A, 374 (2016) 20150043.
[21] Meschke M., Kemppinen A. and Pekola J. P., Accurate Coulomb blockade thermometry
up to 60 kelvin Philos. Trans. R. Soc. A, 374 (2016); DOI: 10.1098/rsta.2015.0052.
[22] Jean Baptiste Fourier, Analytical Theory of Heat, translated by Alexander Freeman,
reproduced by Encyclopaedia Britannica (1952), Library of Congress Catalogue Number:
55-10351: ISBN 0-85229-163-9.
[23] Waterston J. J., On the Physics of Media that are Composed of Free and Perfectly
Elastic Molecules in a State of Motion, Philos. Trans. R. Soc. London Ser. A, 183 (1892)
1.
[24] Willard Gibbs J., Elementary Principles in Statistical Mechanics (Dover) 1960, reprint
(Yale University Press) 1902.
[25] Perrin M. Jean, Brownian Motion and Molecular Reality, Translated by Soddy F., Ann.
Chem. Phys. 8th Series (1909).
[26] Chang Hasok, Inventing Temperature (Oxford University Press, New York) 2004.
[27] http://www.bipm.org/en/measurement-units/history-si/temperature-scales/.
[28] http://www.bipm.org/en/committees/cc/cct/publications-cc.html#kelvin-and-
temperature-scales Addition ITS90 link.
[29] White D. R. and Fischer J., The Boltzmann constant and the new kelvin, Metrologia,
52 (2015) S213.
DOI 10.3254/978-1-61499-818-1-381
Contributions of fundamental constants from atomic

physics to the redefinition of kilogram
H. Fleurbaey, C. Courvoisier, P. Cladé, S. Guellati-Khélifa, L. Julien,
F. Biraben and F. Nez
Laboratoire Kastler Brossel, UPMC-Sorbonne Universités, CNRS,
ENS-PSL Research University, Collége de France - 4 place Jussieu
75252 Paris Cedex 05, France
Summary. — The international system of units will be redefined in terms of fun-

damental constants in 2018, specially the kilogram which is still defined, nowadays,
by a material artefact. For many years, two experiments have been carried out to
define the kilogram based on an atomic or a fundamental physical constant: the
X-ray crystal density (XRCD) methods to determine the Avogadro constant (NA )
and the watt balance to determine the Planck constant (h). Both constants can
be compared using other fundamental constants issued from atomic physics. An
overview of the determinations of these fundamental constants from atomic physics,
since 1998, is given. Their contributions to the accuracy of the comparison between
the two methods will be underlined. The consequences of an eventual evolution of

their estimates will be also considered.
1. – Introduction
Nowadays, the kilogram, defined by a cylinder made of iridium and platinum alloy
is the last unit which is based on a materiel artefact. This International Prototype of
the Kilogram (IPK) is kept at Bureau International des Poids et Mesures in Sévres,
France. Its stability over time has been questioned ([1] and references therein). In
381
382 H. Fleurbaey, C. Courvoisier, P. Cladé, etc.
2005, a proposal was made to redefine the kilogram by fixing the value of the Planck
constant (h) or the Avogadro constant NA which are invariants of nature [2]. It is
now recognized that the kilogram will be defined in term of the Planck constant [3].
In 2013, the Consultative Committee for Mass and related quantities has proposed a
recommandation for the conditions under which a fixed numerical value of the Planck
constant can be adopted [4]. Those conditions refer to three independent determinations
either from watt balance or XRCD experiments, to their relative uncertainties (at least
one determination with a relative uncertainty lower than 2 × 10−8 ) and consistencies at
5 × 10−8 level. But no conditions are given for the factor which enables the comparison
between the determination of the Planck constant from XRCD experiments and the one
from the watt balance.
Indeed the two fundamental constants are linked with the relation
Ar (e)Mu 2
(1) hNA = cα ,
2R∞
in which Ar (e) is the atomic relative mass of the electron, Mu is the Molar mass constant
(Mu = 10−3 kg mol−1 [5]), R∞ the Rydberg constant, c the velocity of light and α the
fine-structure constant.
The Molar Planck constant “hNA ” [5] is regularly adjusted in the least squares ad-
justment of fundamental constants, done every 4 years by CODATA, taking into account
the correlation between the fundamental constants ([6] and references therein). An evo-
lution of this constant is given in fig. 1. An indication of the 10−8 level is represented. In
2006, a significant shift occurred. The Molar Planck constant is highly correlated to α


Fig. 1. – Evolution over time of the estimate of the Molar Planck constant from CODATA since
1998.
Contributions of fundamental constants from atomic physics etc. 383
""! #"" "

""!

!

"
!
! !
""
!
"

#
$
# $#
""!
""!

"

! !
! "
! "" $" !
Fig. 2. – Evolution of the estimate of the values of various constants contributing to the de-
termination of the Molar Planck constant over time since 1998. The main contribution of the
fine-structure constant is clearly visible. The quantities r are the 2014 correlation coefficients
of the quantities in the parenthesis [7].
(r(hNA , α) = 0.9979) and less to Ar (e) and R∞ (respectively, r(hNA , Ar (e)) = 0.0550 and
r(hNA , 2R∞ ) = −0.0123) in which the quantities r are the 2014 correlation coefficients
of the quantities in the parenthesis [7]. Figure 2 shows clearly the strong dependence of
the value of Molar Planck constant to the value of the fine-structure constant. In the
following parts of the paper, we will look at the determinations of Ar (e), R∞ and α since
1998 and at their current status and reliabilities.
2. – The Rydberg constant
The value of the Rydberg constant is provided by hydrogen (H)/deuterium (D) spec-
troscopy combined with quantum electrodynamics calculations (QED). The full expres-
sion of the energy levels is complex and can be sketched as a function f of n, l, j the
quantum numbers, α the fine structure constant, me /mp the electron-proton mass ratio
and rp the rms charge radius of the proton [8]:
(2) E(n, l, j) = hcR∞ f (n, l, j, α, me /mp , rp ).
The function f can be split in two sub-functions: a first one (f1 ) includes the Dirac
contribution and recoil term which have an exact expression and the second one (f2 )
contains non-exactly-known contributions such as higher-order recoil terms, the QED
contributions, and the corrections due to the finite size of the proton charge radius. The
main contribution to f1 is coming from the Dirac term which can be “simplified” using
the Bohr expression which scales the hydrogen spectra. For this contribution, the full
accuracy of the Rydberg constant is required. The function f2 is also proportional to
the Rydberg constant but its contribution to hydrogen spectra is much smaller than the
first one. Therefore a less accurate value of the Rydberg constant is required. In such a
way, eq. (3) below can be linearized as [9]:
hcR∞
(3) E(n, l, j) ≈ + L.S.(n, l, j, α, me /mp , R∞ , rp ),
n2
in which L.S. (Lamb shift) represents the non-well-known contribution.

Even if many spectroscopy studies are considered by the CODATA task group for
the evaluation of the Rydberg constant (from radio-frequency measurement of the 2S-2P
Lamb shift, to frequencies comparison of the 1S-2S/2S-4S/D or 1S-3S/2S-6D transitions
to absolute frequency measurement of the transition frequency of 1S-2S, 2S-8S/D, 2S-12D
lines; [6] and references therein), it is possible, with equation (3), to identify the hydrogen
lines which mainly contribute to the determination of the Rydberg constant: 1S-2S and
2S-8S/8D [9]. The 1S-2S transition frequency is determined by Hänsch’s group with an
uncertainty better than 100 Hz since 1998, now reduced to 10 Hz [10]. The second best
determination arises from 2S-nS/D spectroscopy (n = 8, 12) done in the Paris group
of. Biraben at ∼ 8 kHz level [9]. In fact since 1998, despite an improvement of the
1S-2S transition frequency determination, there has been no significant new second best
measurements of hydrogen lines, and therefore the knowledge of R∞ has remained almost
constant up to now (see fig. 2).
However, the situation may change because of the recent determination of the r.m.s.
charge radius of the proton (rp ) from spectroscopy of muonic hydrogen (made of a proton
(p) and a negative muon (μ− )) [11-13]. Until recently, rp was determined from scattering
experiments and H spectroscopy assuming the exactness of QED corrections included in
the L.S. function. These two determinations are in good agreement.
Because the muon is heavier than the electron (mμ = 207me ), the contribution of
the proton’s finite-size effect to the 2S-2P transition is larger for (μp) (2%) than for H
(0.014%). Therefore, this experiment is a good way to determine rp . The experiment has
been carried out at the muon source of the Paul Scherrer Institute in Switzerland. The
2S-2P transition of a muonic hydrogen atom is excited by a dedicated 6 μ m laser [14].

The excitation of the transition is monitored by looking at the 2P-1S decay at 2 keV.
The deduced value of the rms charge radius of the proton from muonic hydrogen rp (μp)
is smaller than the one obtained from H spectroscopy. This value is also in contradiction
with rp deduced from scattering experiments. However there is controversy on the anal-
ysis done on such experiments [15]. More astonishing, is the discrepancy between H and
(μp) determinations. The situation is known as the “proton size puzzle”. A simplified
picture of this puzzle is given in fig. 3.
The discrepancy between H and (μp) determinations of rp may arise from a systematic
effect in the 2S-nS/D frequency measurements. Many experiments are on development
to measure different lines in H [17-20]. In Paris, the 1S-3S transition has been studied
#)(*$$

+ #)('#()
, + #)&#)(%(

Fig. 3. – Two determinations of the rms charge radius of the proton considered in the 2014
adjustments of the fundamental constants and the one from muonic hydrogen spectroscopy [16].
for many years. A first absolute measurement was done in 2010 [21]. A new laser source
to excite this transition has been built recently [22]. The sensitivity of the measurements
is good enough to discriminate between the two values of rp [23]. However, we need
to study the dependence of the velocity distribution with the pressure in the vacuum
chamber before concluding on the frequency of this line.
What could be the value of the Rydberg constant in 2017? Assuming the exactness
of QED corrections included in function f2 and using the latest 1S-2S frequency determi-
nation [10] and the value of rp (μp), the Rydberg constant would be shifted by −110 kHz
from its current value. Scaling the Rydberg constant value in equation (1) would shift
insignificantly the Molar Planck constant from 3.9903127110(18) × 10−10 J s mol−1 to
3.9903127111(18) × 10−10 J s mol−1 .
The Rydberg constant and the associated “proton puzzle” will not be a limiting
factor for the comparison between the watt balance experiment and the XRCD methods,
in 2017.
3. – The relative atomic mass of the electron
The relative atomic mass of the electron is defined as
me
(4) Ar (e) = ,
mu
in which me is the mass of the electron and mu = m(12 C)/12 is the unified atomic mass
constant following the notation of [5] with m(12 C) the mass of a 12 C atom.
In 1995, Ar (e) has been determined using a Penning trap mass spectrometer [25],
and later from anti-protonic helium spectroscopy [26] and from the measurement of the
gyromagnetic factor of ions [27, 28].
.
3 1. Ar (e) from Penning trap mass spectrometer . – The principle of the experiment
is to measure the ratio of the cyclotron frequency of a ionized carbon atom fc (12 C6+ ) to
the one of an electron fc (e) in the same magnetic field B.
6eB
(5) fc (12 C6+ ) = ,
2πm(12 C6+ )
eB
(6) fc (e) = ,
2πme
fc (12 C6+ ) 6me 6Ar (e)
(7) = = ,
fc (e) m(12 C6+ ) Ar (12 C6+ )
in which e is the charge of the electron and m(12 C6+ ) is the mass of 12 C6+ ion.
The relative atomic mass of the 12 C6+ ion is deduced from the following equation
with the relative atomic mass of the 12 C atom (Ar (12 C) = 12 by definition) expressed as
Eb (12 C)
(8) Ar (12 C) = Ar (12 C6+ ) + 6Ar (e) − ,
mu c2
in which Eb (12 C) is the total binding energy of the 6 electrons of the 12 C atom. It is
calculated accurately enough [24].
The cyclotron frequencies ratio was measured by trapping alternatively a single ion
and between 5 to 13 electrons to obtain a good resolution. The experiment was delicate
as the trapping potentials have to be adjusted for two species which have charges of
opposite sign and large different masses. Moreover, the magnetic field B was slightly
drifting [25]. The relative atomic mass of the proton Ar (p) has also been measured in a
similar way but more precisely [29].
.
3 2. Ar (e) from antiprotonic helium spectroscopy. – This determination is done by
measuring the optical frequencies (265 nm to 726 nm) of transitions between circular
states of antiprotonic helium p̄He+ [26]. This atom is made of 4 He2+ (or 3 He2+ ) nucleus,
an electron in the ground state (1S) and an antiproton (p̄) in long lifetime circular states.
These optical frequencies depend on the antiproton-electron mass ratio (mp̄ /me ) and
on the antiproton-nucleus mass ratio. The masses of the nuclei are well known. Assuming
the validity of CPT theory, the relative atomic mass of the antiproton can be deduced
from the one of the proton as the ratio of the charge to the mass of an antiproton and
the one of the proton have been accurately compared [30]. As Ar (p) has been measured
more accurately than Ar (e) [29], Ar (e) can deduced from the determination of mp̄ /me
from p̄He+ spectroscopy.
.
3 3. Ar (e) from the Larmor frequency to cyclotron frequency ratio of ions. – This
method provides the most accurate determinations of Ar (e). The principle of the exper-
iment is to measure the ratio of the Larmor frequency fs (A X(Z−1)+ ) of the hydrogenic
ion A X(Z−1)+ to the cyclotron one fc (A X(Z−1)+ ) in the same magnetic field B.
The expressions of these two frequencies are
(Z − 1)eB
(9) fc (A X(Z−1)+ ) = ,
2πm(A X(Z−1)+ )
|ge (A X(Z−1)+ )|eB
(10) fs (A X(Z−1)+ ) = ,
4πme
in which ge (A X(Z−1)+ ) is the electron g-factor of this ion. Their ratio is
fs (A X(Z−1)+ ) |ge (A X(Z−1)+ )| Ar (A X(Z−1)+ )

(11) =
A
fc ( X (Z−1)+ ) 2(Z − 1) Ar (e)
Ar (A X(Z−1)+ ) is estimated from an equation similar to (8). Ar (e) is then deduced using
the theoretical estimation of ge (A X(Z−1)+ ).
The experiment is made with two Penning traps: a precision trap in which the mag-
netic field is homogeneous and an analysis trap above. The latter contains a Ni-ring
electrode which introduces an inhomogeneity in the magnetic field giving the possibility
to detect spin direction thanks to “continuous Stern et Gerlach effect” [31].
Spin-flip (Larmor) and cyclotron frequencies are measured, in the precision trap, at
the same time which cancels the fluctuation of B field at the first order. The spin flip is
induced by a tunable RF frequency in the precision trap. The ion is then transported in
the analysis trap in which the spin state is measured on the axial frequency and shifted
back in the precision trap. The spin flip resonance curve is the number of spin flips versus
the tunable RF frequency.
The expression ge (A X(Z−1)+ ) contains many contributions which are functions of the
fine structure constant α. The main one is the Dirac term which can be calculated
accurately enough using the CODATA estimate of α. The other ones are much smaller
and therefore do not require a sharp value of α to be calculated.
This method has been used for 12 C5+ and 16 O7+ ions [27] and then for 12 C5+ and
28 13+
Si ions [28]. The last determination of Ar (e) [28] has a relative uncertainty of
3 × 10−11 which is more than 10 times better than the 2010 CODATA estimate of Ar (e).
Therefore for the 2014 adjustment, this measurement determines entirely the CODATA
estimate of Ar (e).
The various determinations of Ar (e) are well under control and accurate enough, so
that Ar (e) will not be a limiting factor for the comparison between the watt balance
experiment and the XRCD methods.
4. – The fine-structure constant
Among the fundamental constants, the fine structure constant plays a particular
role. It is a dimensionless constant therefore independent of the system of units. The
fine-structure constant scales the electromagnetic interaction and has been measured
in various fields of physics [32]. Its present value is only determined by two types of
experiments: the measurement of atomic recoil and the anomaly of the gyromagnetic
factor of the electron.
.
4 1. Atomic recoil experiment, determination of h/m. – Presently, this method is the
only one which can provide a competitive determination of the fine-structure constant
but with a larger uncertainty than the one from the ae experiment. The determination
of the fine structure constant from the ratio h/mX (in which h is the Planck constant
and mX the mass of a particle or an atom) is derived from the expression of the Rydberg
constant (R∞ ):
2R∞ Ar (X) h
(12) α2 = .
c Ar (e) mX
As the Rydberg constant and the relative atomic mass ratio of the particle X (Ar (X))
(respectively, of the electron (Ar (e))) are sufficiently well known, and c the velocity
of light perfectly known, an accurate measurement of h/mX gives a determination of
α less dependent on quantum electrodynamics calculations than the one from g − 2
.
experiment (see sect. 4 2). The theoretical dependence on QED calculations is due to
the Rydberg constant (see sect. 2). However, at the 10−10 level, the confidence in the
Rydberg constant value is good.
In these experiments, the ratio h/matom is deduced from the measurement of the
atomic recoil velocity vr [33]. The recoil velocity is the resulting velocity of an atom
initially at rest after the absorption of a photon. This velocity is
(13) vr = h̄k/matom ,
in which k is the wave vector of the photon and h̄ the reduced Planck constant (h̄ = h/2π).
For a rubidium atom, vr = 6 mm/s. It has to be compared to the thermal velocity at
300 K: vth ≈ 300 m/s. Consequently all the experiments are done with laser cooled
atoms.
The pioneering experiment has been done at Stanford with caesium atoms [34]. A first
determination has been obtained with an atom interferometer realized with two pairs of
Raman π/2 pulses [35]. To improve the resolution, 30 additional Raman π pulses were
inserted between the pairs of π/2 pulses [34]. A preliminary result has been given in [36]:
−1
(14) αCs = 137.0360001(11)[7.7 × 10−9 ].
Currently, developments are done, in the group of Müller, to improve the sensitivity of
the atom interferometer and so to provide a more accurate determination of αCs [37, 38].
−1
A value of αCs can be derived from the proposal presented in [39]
−1
(15) αCs = 137.03599867(55)[4.0 × 10−9 ].
However this result may have to be slightly corrected and is not used in the last adjust-
ment of CODATA [16].
Since 2000, a second experiment has been developed in Paris, in the group of Biraben
with rubidium atoms [40-43]. The peculiarity of this experiment is the use of Bloch
oscillations instead of the 30 π pulses to improve the resolution on the determination
of vr . At each Bloch oscillation realized in a vertical accelerated optical lattice, the
momentum 2h̄k is transferred to the atoms. As this process is very efficient [44], up to
1600 photon momenta can be transferred to the atoms [42].
The latest measurement of h/m(87 Rb) has been obtained with a new dedicated vacuum
chamber [43]. The subsequent determination of α depends on the estimates of the other
quantities:
(16) α(−1
87 Rb) = 137.035998996(85)[6.2 × 10
−10
],
with R∞ = 10973731.568508(65) m−1 , Ar (e) = 5.48579909070(16) × 10−4 u and

Ar (87 Rb) = 86.9091805319(65) u [16].
More powerful laser sources have been developed to reduce the contributions of sys-
tematic effects [45]. A new determination of h/m(87 Rb) will be done in the near future
while new types of atom interferometers with large momentum beam splitter are also
studied [46-48].
.
4 2. Anomaly of the gyromagnetic factor of the electron. – For a long time, the mea-
surement of the anomaly of the gyromagnetic factor of the free electron associated with
QED calculations has provided the best determination of α. The gyromagnetic factor of
the electron ge is given by
(17) ge /2 = ωL /ωc ,
where ωL is the Larmor frequency and ωc the cyclotron frequency. The value of ge given
by the Dirac equation is 2. In fact, ge is slightly larger than 2 because of the interaction
of the electron with the vacuum quantum field. This anomaly of the gyromagnetic factor
of the electron ae is defined as
(ωL − ωc )
(18) ae = (ge − 2)/2 = .
ωc
The theoretical expression of ae takes into account the electromagnetic, hadronic, and
electroweak interactions:
(19) ae = ae (QED) + ae (hadronic) + ae (electroweak).
ae (QED) is calculated with the perturbative expansion of α/π:
α 2n
(20) ae (QED) = Σn ae(2n) .
π
(2n)
The complexity of the calculations of ae increases with the index n labeling the ex-
(2n)
pansion’s order. Up to n = 3, the ae terms can be calculated analytically while the
higher-order terms must be evaluated numerically. Only very recently, after many years
of efforts, the 10th-order term has been calculated [49]. The hadronic and electroweak
contributions can be also evaluated (see [49, 50]).
The need of an accurate theoretical evaluation came with the progresses of the ex-
perimental determinations of ae . Recent determinations have been done in a Penning
trap. In these experiments, an oscillating electromagnetic field induces transitions be-
tween energy levels of the electron interacting with a static magnetic field. The first
determination of ae with a single trapped electron has been realized in Washington
University [51]. The Penning trap is placed in a liquid helium bath at 4.2 K in a
strong magnetic field 5T. The anomaly is determined by measuring ωL − ωc and ωc
frequencies in the same magnetic field. As a Ni ring is added to form a magnetic
bottle gradient, the spin state can be determined by measuring the axial frequency.
This experiment was mainly limited by stray effects in the microwave cavity (power,
resonance). The deduced fine structure constant was having a relative uncertainty of
3.8 × 10−9 [51].
In Harvard University, a new trap has been designed. This cylindrical Penning trap
forms a microwave cavity that could inhibit spontaneous emission (by a factor up to 250)
consequently the width of the spin flip resonance is reduced. Moreover because the trap
is calculable, the inside radiation and magnetic fields can be accurately controlled. As the
trap is cooled at 100 mK, the electron motion is in its ground state. As in the Washington
experiment, the detection of quantum cyclotron state or quantum spin state is done with
the axial frequency. Now thanks to all the improvements, the one quantum change in
cyclotron motion is well controlled and resolved ( [52]). The signals are obtained by
measuring the quantum jumps per attempt to drive them as a function of the drive
frequency.
With this quantum jump spectroscopy, the ge factor of the electron is known with a
relative uncertainty of 2.8 × 10−13 [52, 53]:
(21) ge /2 = 1.00115965218073(28),
and the fine-structure constant from ae with a relative uncertainty of 2.5 × 10−10 [49]:
(22) αa−1
e
= 137.035999166(33).
This determination of α is currently the most accurate one. However as it relies on

complex QED calculations carried in only one group in the world, this determination is
strongly affected by any re-evaluation of the theoretical expression as it was between 2006
and 2010 CODATA adjustments. It must be noticed that the two measurements of the
electron anomaly done at Harvard in 2006 and 2008 [54, 53] are in very good agreement
with each other which demonstrates that this experiment is well under control.
/ +*%3
& 32'4 32

& 32'4 *, $ & *,'
& 32'4 *0

&*0'4 *0
& 32'4 +* $ &*0'
&+*'4 +* $ & *,'
& 32'4 +. $ &+*'
&+*'4 +.
$ $ &+*'4##&+.'
&+.'
&%+%+-1#*-' +*-
/"331/ /"332* /"332/ /"333* /"333/ 0"**** 0"***/ 0"**+* 0"**+/
Fig. 4. – Evolution of the most precise estimates of α deduced from various experiments which
have been considered to estimate the CODATA recommended values of the fundamental con-
stants. The indicated year refers to the year of the adjustment. An experimental input can be
used for many years but the associated value of α can change with time because of an evolu-
tion of the theory or an evolution of other fundamental constants used to deduce α from the
experimental quantity. For example, the fine-structure constant deduced from h/MRb in 2014
is shifted from the previous value indexed LKB10 because of the new estimation in 2014 of R∞
and Ar (e).
The evolution of the most precise determinations of α deduced from various experi-
ments which have been considered to estimate the CODATA recommended values of the
fundamental constants is given in fig. 4.
A fluctuation of 5 × 10−9 of α induces a shift of 1 × 10−8 of the Molar Planck constant
(see eq. (1)). It is clear that another independent determination of α would increase our
confidence in the knowledge of α. The situation may be enlightened with a competitive
determination of the g-factor of H- and Li-like ions of the same element [55, 56].
5. – Conclusion
Despite the “proton puzzle” which has stimulated a lot of activities in Physics, the
Rydberg constant is sufficiently well known for the redefinition of the kilogram in 2018.
The current knowledge of the relative atomic mass of the electron is determined by
only one experiment, but it is in very good agreement with previous determinations.
This quantity will not be a limiting factor in the estimate of the Molar Planck constant
and in the determination of the fine-structure constant from atomic recoil.
The case of the fine-structure constant is more delicate. Our current knowledge of α
is mostly determined by the measurement of ae combined with QED calculations. Over
the past few years, evolutions of these calculations have shifted significantly this value,
but only in the 5 × 10−9 range.
On the other hand, the atomic recoil type experiment should help to clarify the
situation, but unexpected systematics may also affect these measurements. It is clear
that new determinations of h/mX are necessary. Another independent determination of α
would increase our confidence in the knowledge of α. The situation may be enlightened
with a competitive determination of the g-factor of H- and Li-like ions of the same
element.
The experiments of Atomic Physics have brought the relative uncertainty of the Molar
Planck constant to 4.5 × 10−10 level. Even if the determination of α is not completely
satisfactory, a shift at 10−8 level would be extremely surprising.
After the revision of the International System of Units, NA and h will have a fixed
value and the molar mass of (12 C) won’t be equal to 12 g mol−1 . The quantity h/mu will
allow a realization of the kilogram based on the Planck constant by the XRCD method.
The special feature of h/mu has been underlined in a recent paper [57].
∗ ∗ ∗
H.F thanks the LABEX Cluster of Excellence FIRST-TF(ANR-10-LABX-48-01),
within the Program “Investissements d’Avenir” operated by the French National Re-
search Agency (ANR) for financial support. The authors thank the Labex First-TF and
l’Agence Nationale de la Recherche (ANR) for their financial support to our experiments
(PROCADIS Projet no. ANR-2010-BLANC: 041501 and INAQED Projet no. ANR-
2012-JS04-0009). The authors are indebted to the Service du Temps of LNE-SYRTE
laboratory for the time reference.
REFERENCES
[1] Stock M., Barat P., Davis R. S., Picard A. and Milton M. J. T., Metrologia, 52
(2015) 310.
[2] Mills I. M., Mohr P. J., Quinn T. J., Taylor B. N. and Williams E. R., Metrologia,
42 (2005) 71.
[3] http://www.bipm.org/en/CGPM/db/25/1/.
[4] www.bipm.org/cc/CCM/Allowed/14/31a-Recommandation-ccm-G1(2013).pdf.
[5] Mohr P. and Taylor B. N., Rev. Mod. Phys., 72 (2000) 351.
[6] Mohr P., Taylor B. N. and Newell D., Rev. Mod. Phys., 84 (2012) 1527.
[7] http://physics.nist.gov/cuu/Constants/index.html.
[8] Udem T., private communication.
[9] De Beauvoir B., Schwob C., Acef O., Jozefowski L., Hilico L., Nez F., Julien
L., Clairon A. and Biraben F., Eur. Phys. J. D, 12 (2000) 61.
[10] Matveev A., Parthey C. G., Predehl K., Alnis J., Beyer A., Holzwarth R., Udem
T., Wilken T., Kolachevsky N., Abgrall M., Rovera D., Salomon C., Laurent
P., Grosche G., Terra O., Legero T., Schnatz H., Weyers S., Altschul B. and
Hänsch T. W., Phys. Rev. Lett., 110 (2013) 230801.
[11] Antognini A., Nez F., Schuhmann K., Amaro F. D., Biraben F., Cardoso J. M. R.,
Covita D., Dax A., Dhawan S., Diepold M., Fernandes L., Giesen A., Gouvea
A., Graf T., Hänsch T. W., Indelicato P., Julien L., Kao C.-Y., Knowles P. E.,
Kottmann F., Le Bigot E., Liu Y.-W., Lopes J. A. M., Ludhova L., Monteiro
C. M. B., Mulhauser F., Nebel T., Rabinowitz P., dos Santos J. M. F., Schaller
L., Schwob C., Taqqu D., Veloso J. F. C. A., Vogelsang Jan and Pohl R., Science,
339 (2013) 417.
[12] Pohl R., Antognini A., Nez F., Amaro F. D., Biraben F., Cardoso J. M. R.,
Covita D., Dax A., Dhawan S., Fernandes L., Giesen A., Graf T., H änsch T. W.,
Indelicato P., Julien L., Kao C.-Y., Knowles P. E., Le Bigot E., Liu Y.-W., Lopes
J. A. M., Ludhova L., Monteiro C. M. B., Mulhauser F., Nebel T., Rabinowitz
P., dos Santos J. M. F., Schaller L., Schuhmann K., Schwob C., Taqqu D.,
Veloso J.F.C.A. and Kottmann F., Nature, 466 (2010) 213.
[13] Pohl R., Nez F., Fernandes L. M. P., Amaro F. D., Biraben F., Cardoso J. M. R.,
Covita D. S., Dax A., Dhawan S., Diepold M., Giesen A., Gouvea A. L., Graf
T., Hänsch T. W., Indelicato P., Julien L., Knowles P., Kottmann F., Le Bigot
E.-O., Liu Y.-W., Lopes J. A. M., Ludhova L., Monteiro C. M. B., Mulhauser F.,
Nebel T., Rabinowitz P., dos Santos J. M. F., Schaller L. A., Schuhmann K.,
Schwob C., Taqqu D., Veloso J. F. C. A. and Antognini A., Science, 353 (2016) 669.
[14] Antognini A., Amaro F. D., Biraben F., Cardoso J. M. R., Conde C. A. N.,
Covita D. S., Dax A., Dhawan S., Fernandes L. M. P., Hänsch T. W., Hughes
V. W., Huot O., Indelicato P., Julien L., Knowles P. E., Kottmann F., Liu Y.-
W., Lopes J. A. M., Ludhova L., Monteiro C. M. B., Mulhauser F., Nez F., Perry
B. N., Pohl R., Rabinowitz P., dos Santos J. M. F., Schaller L. A., Schwob C.,
Taqqu D., Veloso J. F. C. A., Opt. Commun., 253 (2005) 362.
[15] Carlson C. E., Prog. Part. Nucl. Phys., 82 (2015) 59.
[16] Mohr P., Taylor B. N. and Newell D., J. Phys. Chem. Ref. Data, 45 (2016) 043102.
[17] Beyer A., Alnis J., Khabarova K., Matveev A., Parthey C. G., Yost D. C., Pohl
R., Udem T., Hänsch T. W. and Kolachevsky N., Ann. Phys. (Berlin), 525 (2013)
671.
[18] Hessels E., Frontiers in Optics 2015, OSA Technical Digest (on line) (Optical Society of
America) 2015, paper LTu2G.2.
[19] Jentschura U. D., Mohr P. J., Tan J. N. and Wundt B. J., Phys. Rev. Lett., 100
(2008) 160404.
[20] Yost D. C., Matveev A., Grinin A., Peters E., Maisenbacher L., Beyer A., Pohl
R., Kolachevsky N., Khabarova K., Hänsch T. W. and Udem Th., Phys. Rev. A,
93 (2016) 042509.
[21] Arnoult O., Nez F., Julien L. and Biraben F., Eur. Phys. J. D, 60 (2010) 243.
[22] Galtier S., Nez F., Julien L. and Biraben F., Opt. Commun., 324 (2014) 34.
[23] Galtier S., Fleurbaey H., Thomas S., Julien L., Biraben F. and Nez F., J. Phys.
Chem. Ref. Data, 44 (2015) 031201.

[24] Linstrom P. and Mallard W. (Editors), NIST Chemistry Web-Book, NIST Standard
Reference Database Number 69, March 2003 Release http://webbook.nist.gov/
chemistry/ion-ser.html.
[25] Farnham D. L., Van Dyck R. S., Jr., and Schwinberg P. B., Phys. Rev. Lett., 75
(1995) 3598.
[26] Hori M., Sótér A., Barna D., Dax A., Hayano R., Friedreich S., Juhász B., Pask
T., Widmann E., Horváth D., Venturelli L. and Zurlo N., Nature, 475 (2011) 484.
[27] Werth G., Alonso J., Beier T., Blaum K., Djekic S., Haffner H., Hermanspahn
N., Quint W., Stahl S., Verdu J., Valenzuela T. and Vogel M., Int. J. Mass
Spectrom., 251 (2006) 152.
[28] Sturm S., Kohler F., Zatorski J., Wagner A., Harman Z., Werth G., Quint W.,
Keitel C. H. and Blaum K., Nature, 506 (2014) 467.
[29] Van Dyck R. S., Jr., Farnham D. L., Zafonte S. L. and Schwinberg P. B., Trapped
Charged Particles and Fundamental Physics, edited by Dubin D. H. E. and Schneider
D., AIP Conf. Proc., 457 (1999) 101.
[30] Gabrielse G., Int. J. Mass Spectrom., 251 (2006) 273.
[31] Dehmelt H., Proc. Natl. Acad. Sci. USA, 83 (1986) 2291.
[32] Biraben F., Bouchendira R., Cladé P., Guellati-Khélifa S., Julien L. and Nez F.,
Proceedings of the International School of Physics “Enrico Fermi”, Course 185, Metrology
and Physical Constants, edited by E. Bava, M. Kühne and A. M. Rossi (Ios Press,
Amsterdam; SIF, Bologna) 2013, pp. 447-468.
[33] Hall J. L., Bordé C. J. and Uehara K., Phys. Rev. Lett., 37 (1976) 1339.
[34] Weiss D. S., Young B. C. and Chu S., Appl. Phys. B, 59 (1994) 217 and references
therein.
[35] Bordé C. J., Phys. Lett. A, 140 (1989) 10.
[36] Witch A., Hensley J. M., Sarajlic E. and Chu S., Phys. Scr., T102 (2002) 82.
[37] Chiow S.-W., Herrmann S., Chu S. and Müller H., Phys. Rev. Lett., 103 (2009)
050402.
[38] Chiow S.-W., Kovachy T., Chien H.-C. and Kasevich M. A., Phys. Rev. Lett., 107
(2011) 130403.
[39] Lan S. Y., Kuan P. C., Estey B., English D., Brown J. M., Hohensee M. A.,
Müller H., Science, 339 (2013) 554.
[40] Cladé P., de Mirandes E., Cadoret M., Guellati-Khélifa S., Schwob C., Nez F.,
Julien L. and Biraben F., Phys. Rev. Lett., 96 (2006) 033001.
[41] Cladé P., de Mirandes E., Cadoret M., Guellati-Khélifa S., Schwob C., Nez F.,
Julien L. and Biraben F., Phys. Rev. A., 74 (2006) 052109.
[42] Cadoret M., de Mirandes E., Cladé P., Guellati-Khélifa S., Schwob C., Nez F.,
[43] Bouchendira R., Cladé P., Guellati-Khélifa S., Nez F. and Biraben F., Phys.
Rev. Lett., 106 (2011) 080801.
[44] Battesti R., Cladé P., Guellati-Khélifa S., Schwob C., Grémaud B., Nez F.,
[45] Andia M., Wodey E., Biraben F., Cladé P. and Guellati-Khélifa S., J. Opt. Soc.
Am. B, 32 (2015) 1038.
[46] Cladé P., Guellati-Khélifa S., Nez F. and Biraben F., Phys. Rev. Lett., 102 (2009)
240402.
[47] Cladé P., Plisson T., Guellati-Khélifa S., Nez F. and Biraben F., Eur. Phys. J.
D, 59 (2010) 349.
[48] Jannin R., Cladé P. and Guellati-Khélifa S., Phys. Rev. A, 92 (2015) 013616.
[49] Aoyama T., Hayakawa M., Kinoshita T. and Nio M., Phys. Rev. D, 91 (2015) 033006.
[50] Kinoshita T., in Lepton dipole moments (World Scientific) 2010 and references therein.
[51] Van Dyck R. S., Schwinberg P. B. and Dehmelt H. G., Phys. Rev. Lett., 59 (1987) 26.
[52] Hanneke D., Fogwell Hoogerheide S. and Gabrielse G., Phys. Rev. A., 83 (2011)
052122.
[53] Hanneke D., Fogwell S. and Gabrielse G., Phys. Rev. Lett., 100 (2008) 120801.
[54] Odom B., Hanneke D., D’Urso B. and Gabrielse G., Phys. Rev. Lett., 97 (2006)
030801.
[55] Yerokhin V. A., Berseneva E., Harman Z., Tupitsyn I. I. and Keitel C. H., Phys.
Rev. Lett., 116 (2016) 100801.
[56] Yerokhin V. A., Berseneva E., Harman Z., Tupitsyn I. I. and Keitel C. H., Phys.
Rev. A, 94 (2016) 022502.
[57] Cladé P., Biraben F., Julien L., Nez F. and Guellati-Khélifa S., Metrologia, 53
(2016) A75.
DOI 10.3254/978-1-61499-818-1-395
Cooling, trapping and manipulation of atoms

and Bose-Einstein condensates: Applications
to metrology(∗ )
K. Helmerson and W. D. Phillips
Atomic Physics Division, Physics Laboratory, National Institute of Standards and Technology
Gaithersburg, MD 20899, USA
Summary. — The lecture notes that follow are meant to provide a written ac-
companiment to the lectures given by one of the authors (WDP) in July 2016 at
the “Metrology: from physics fundamentals to quality of life” Enrico Fermi Interna-
tional School of Physics in Varenna. The notes were originally written for an earlier
Metrology school, but still provide a good summary of the material presented in
2016.
The notes review the experimental and theoretical ideas behind the production of
ultracold gases of neutral atoms, the physics of such gases, and the manipulation and
use of ultracold atoms in metrological applications. Light exerts forces on atoms, both
radiation pressure and dipole forces. These optical forces can cool gases of atoms to
temperatures below a microkelvin, and can trap atoms for long periods of time. Laser
cooling also makes atomic gases cold enough to be trapped in magnetic traps when the
atoms have an intrinsic magnetic moment. Trapped atoms can be further cooled by
395
396 K. Helmerson and W. D. Phillips
evaporative cooling to reach quantum degeneracy, producing Bose-Einstein condensates

or degenerate Fermi gases. Light forces can also coherently manipulate atoms, realizing
elements of atom optics. Diffraction processes of atoms from optical standing waves act as
mirrors and beamsplitters for atoms. Atom optics can include non-linear optics in which
the collisional interaction between atoms can produce phenomena like four-wave-mixing
and dark solitons.
Cold atoms and the techniques that manipulate them have a number of metrological
applications. The most important application, and the one that motivated much of the
early work on cold atoms, is to atomic clocks. Thermal motion limits the performance of
typical atomic clocks through Doppler shifts, relativistic time dilation effects, and lim-
ited observation times. The best current primary frequency standards, cesium atomic
clocks, now use laser-cooled atoms in a “fountain” mode, where atoms are launched ver-
tically through a microwave cavity tuned to the cesium hyperfine frequency, then return
through the same cavity after falling back under the influence of gravity. Such clocks
achieve accuracy on the order of parts in 1016 . Optical clocks using laser-cooled atoms
or ions, trapped in optical lattices or in ion traps, and operating on clock transitions
with optical frequencies, achieve even higher accuracy, at the level of parts in 1018 , al-
though these are not primary standards. Other metrological applications of cold atoms
include atom interferometric applications. Atom interferometers can accurately measure
the local acceleration of gravity near the surface of the Earth and are beginning to be
used for measuring the Newtonian gravitational constant. They can also measure the
fundamental constant h/m for the atom being used, which can be converted with high
accuracy to h/m for the electron. That constant in turn can be used to determine the
fine-structure constant.
Since the notes that follow were written, a number of important advances in metrol-
ogy with cold atoms have appeared in the literature. The references listed as “Recent
literature” at the end of the paper represent a selection (admittedly, quite incomplete)
of some of those advances.
1. – Introduction
The thermal motion of atoms often imposes limitations on the precision and accuracy
of measurements involving those atoms. Atomic beam cesium clocks are one outstanding
example, where the typical thermal velocity of the cesium atoms, on the order of 200 m/s,
limits the observation time to a few milliseconds in a typical apparatus. This finite ob-
servation time results in a linewidth for the microwave resonance whose frequency defines
the unit of time. Other motional effects include first- and second-order Doppler effects
whose presence limits the performance of such clocks. Because conventional refrigera-
tion would result in the condensation of cesium or other atoms into a solid, researchers
sought other means of reducing the thermal motion, a quest that led to laser cooling
and electromagnetic manipulation of neutral atoms. Similar techniques have been devel-
oped for trapped ions and used in metrological applications, but these are not treated
here.
Cooling, trapping and manipulation of atoms etc. 397
These notes are based on a series of lectures given at the Enrico Fermi Summer School
on Recent Advances in Metrology and Fundamental Constants held in Varenna in July
and August of 2006. The notes include a general discussion of the mechanical effects of
light; laser cooling and trapping techniques for neutral atoms; experiments with Bose-
Einstein condensed sodium and rubidium atoms at the National Institute of Standards
and Technology (NIST) in Gaithersburg, Maryland; and applications of these techniques
to metrology.
The topics to be covered in the general discussion of the mechanical effects of light
and laser cooling and trapping techniques include: radiative forces, both spontaneous
and dipole; various topics related to trapping neutral atoms, including magnetic traps,
both static and time-averaged orbiting potentials (TOP); laser dipole traps and far-off
resonant traps (FORTs); radiation pressure traps and the optical Earnshaw theorem;
magneto-optical traps (MOTs); mechanisms limiting the loading of MOTs and tech-
niques to increase the densities, including dark spot MOTs; Doppler cooling and the
Doppler cooling limit; deceleration and cooling of atomic beams; optical molasses and
the discovery of sub-Doppler laser cooling; and new laser cooling mechanisms, including
polarization gradient and Sisyphus heating.
These topics have been selected as representing some of the important developments in
the interaction of electromagnetic fields and atoms which have enabled the development
of new metrological tools as well as the creation and manipulation of Bose-Einstein
condensates. More details on these and related topics can be found in a number of notes,
references, and review articles [1-13].
The later sections of these notes discuss metrology applications of cold atoms, par-
ticularly clocks and atom interferometers, as well as Bose-Einstein condensation (BEC)
and its potential applications to metrology.
Before beginning the discussion of the mechanical effect of atom-light interactions,
we define, for convenience, some frequently used notation and symbols, with frequencies
expressed in radians/second. The reader is cautioned that other authors may have used
other conventions:
δ = ωlaser − ωatom is the detuning of the laser frequency from the natural resonant
frequency of the atom.
Γ = τ −1 is the decay rate of the population in the excited state, the inverse of the
natural lifetime, the linewidth of the transition.
k = 2π/λ is the laser photon wave vector.
kB is Boltzmann’s constant.
Ω is the on-resonance Rabi frequency for a laser field, the precession frequency of the
Bloch vector representing the 2-level atom when δ = 0.
I/I0 = 2Ω2 /Γ2 is the normalized intensity of the laser.
M is the mass of the atom.
vrec = h̄k/M is the recoil velocity of an atom upon emission or absorption of a single
photon.
2
Erec = M vrec /2 = h̄2 k 2 /2M is the kinetic energy of an atom having velocity vrec .
2. – Radiative forces
Light comes in quantized packets of energy and momentum called photons. The trans-
fer of energy and momentum between the photon and an atom through either coherent
or incoherent scattering results in a force exerted on the atom. This is the basis for the
mechanical effects of atom-light interactions. Radiative forces or the mechanical effects
of light are generally divided into two categories—the scattering force and the dipole
force.
.
2 1. The scattering force. – The scattering force, also called the spontaneous or ra-
diation pressure force, is the force exerted on an atom by the incoherent scattering of
photons. More precisely it is the force on an atom corresponding to absorption of a pho-
ton followed by spontaneous emission. The photon absorbed transfers h̄k of momentum
to the atom. The spontaneous emission of the photon is symmetrically distributed and
so the momentum transfer due to spontaneous emission, averaged over many absorption-
emission events, is zero. The average force on a two-level atom moving with velocity v
in a plane wave of wave vector k and detuning δ is
Γ I/I0
(1) F(v) = h̄k .
2 1 + I/I0 + [2(δ − k · v)/Γ]2
This force is the maximum scattering rate Γ/2, times the resonance Lorentzian times the
photon momentum, i.e. the rate of absorbing photon momentum. The detuning δ − k · v
accounts for the Doppler shift, and the force is large when |δ−k·v| ≤ Γ. The spontaneous
force is limited by the rate at which spontaneous emissions can occur. These occur at a
rate Γ for excited atoms, whose maximum fractional population is 1/2.
For real, multilevel atoms, the situation can be more complicated. A common occur-
rence, typical of alkali atoms, is that the ground state is split by the hyperfine interaction
into two states separated in frequency by many times the optical linewidth Γ. An atom
excited by a laser from one of these hyperfine levels to an optically excited state may de-
cay by spontaneous emission to the other hyperfine level. Transitions from this level are
then so far out of resonance that effectively no further absorption occurs and no force is
applied to the atom. While various schemes involving selection rules and polarization of
the light may be used to avoid this problem of optical pumping, the most straightforward
method is to apply a second laser frequency, tuned to resonance between the “wrong”
hyperfine state and the optically excited state. This “repumper” keeps the atom out
of the wrong ground state and allows the atom to effectively feel the force of the laser
acting on the main transition.
Equation (1) is only valid if the force can be meaningfully averaged over many
absorption-emission events. If a single event changes the atomic velocity so much that
the resonance condition is not satisfied, such an average is not possible. This imposes a
validity condition h̄k 2 /M Γ on eq. (1). This condition is well satisfied for most atoms
of interest in laser cooling. For example, M Γ/h̄k 2 is 1200 for cesium laser cooled on its
resonance transition at 852 nm and is 200 for sodium cooled at 589 nm.
.
2 2. The dipole force. – The dipole force, also called the gradient force or stimulated
force, is the force exerted on an atom due to coherent redistribution of photons. The
dipole force can be considered as arising from stimulated Raman events—the absorption
and stimulated emission of photons. (Note that the absorption and stimulated emission
cannot be thought of as successive and independent events; their correlation is central
to the proper understanding of the force.)
We can also understand the dipole force in analogy to a driven, classical oscillator.
A harmonically bound charge driven by an oscillating electric field E has an oscillating
dipole moment μ which is in phase with the driving field when driven below resonance,
and out of phase when driven above resonance. The energy of interaction between the
dipole and field is W = −μ · E. Below resonance the energy is negative and the oscillator
will be drawn toward a more intense field, while above resonance it will be drawn to the
weaker part of the driving field.
A particularly powerful and intuitive way of describing the dipole force is in
the dressed-atom picture. This has been treated in detail by Dalibard and Cohen-
Tannoudji [14] and we present the basic idea here. Consider a 2-level atom with
ground state |g and excited state |e. Separately consider a single mode of the ra-
diation field, close to resonance with the atomic transition, having energy levels labeled
. . . , |n − 1, |n, |n + 1, . . . according to the number of photons in the mode. The energy
levels of these two distinct systems are shown in fig. 1a for the case where the photon
energy ωL is greater than the difference in energy between the atomic states, ωA . This is
the “bare” basis in which we consider the atom’s energy levels and those of the photon
field separately. If we now consider the atom and laser field together as a single system
we have the dressed basis (atom dressed by laser photons) of fig. 1b. If there is no inter-
a) b) c) I
uncoupled coupled
|1 , n〉
| n+1〉 | g , n+ 1 〉
| e , n〉
|2 , n〉
x
δ δ >0
|1 , n − 1 〉 |g, n +
| n〉 | g , n〉 (δ 2 + Ω 2)1/2
| e〉 | e , n −1〉 (δ 2 + Ω 2)1/2 δ
|2 , n − 1 〉 |e, n〉
ωA ωL
δ <0
|1 , n − 2 〉 |e, n〉
| g〉 | n −1〉 | g , n −1〉
atom field | e , n −2〉 |g, n
|2 , n − 2 〉
atom + field
Fig. 1. – a) Energy levels of a 2-level atom and a laser field in the bare basis, b) dressed basis
with the atom uncoupled and coupled to the field. c) Top: the intensity profile of the laser field.
Bottom: the dressed energy levels as a function of position for this intensity profile, for laser
detuning above and below resonance. The sizes of the black dots indicate the relative population
of the dressed states.
action between the atom and the laser field (as, for example, when the laser field does
not spatially overlap the atom’s position) the dressed level energies are simply the sum
of the atom and field energies. This gives a ladder where each rung is a pair of nearly
degenerate energy levels, since an atom in the ground state with n photons in the laser
field has nearly the same energy as an atom in the excited state with n − 1 photons
in the field (with the energy difference being equal to the laser detuning). When the
atom interacts with the field, the dressed levels (the eigenstates of the full Hamiltonian)
become superpositions of ground and excited states and of the numbers of photons in
the field. By convention, the higher of the two levels in a rung is called |1, and the
rungs are labeled by the photon number n associated with the excited state in the un-
coupled basis [15]. According to the general rule that interacting energy levels repel each
other by an amount depending on the coupling, the spacing between the √ dressed levels
increases from the detuning δ to the “effective Rabi frequency” Ωeff = δ 2 + Ω2 as the
interaction turns on. Because each of the dressed levels has both ground and excited
state character, an atom can make spontaneous emission transitions between the rungs
of the ladder. These transitions establish an equilibrium population between the two
types of levels [14].
Now consider a light field whose intensity varies in space, such as a focused laser
beam with a Gaussian intensity profile, or a plane standing wave. The Rabi frequency
seen by an atom now varies in space, so the energy of the dressed levels also varies.
The dipole force arises from this variation in energy and the relative populations of the
dressed levels. Figure 1c illustrates the idea. For laser detuning above resonance (δ > 0)
the upper of the two dressed levels is the one that connects to the ground state in the
limit of small interaction. This upper level always has the largest population, and its
potential tends to repel the atom from the region of most intense light. For δ < 0 it is
the lower level that connects to the ground state and that has the higher population; its
potential attracts the atom to the higher intensity. The actual dipole force is the average
force, weighted by population, for the two potentials. The dipole force is derivable from
a potential [14, 16-19], which can be written as
h̄δ I/I0
(2) U= log 1 + ,
2 1 + (2δ/Γ)2
where the spatial dependence of the potential comes in through the dependence of the
intensity I on position.
.
2 3. Doppler cooling. – Doppler cooling results from near resonant radiation pressure
acting on an atom and thus it is a particularly instructive example of an application
of the scattering force. To see how cooling works we shall consider the one-dimensional
motion of an atom in counterpropagating plane waves. If the laser beams are tuned below
the atomic resonance, an atom moving in the direction opposite to one of the beams will,
because of the Doppler shift, see that beam shifted closer to its resonance frequency.
At the same time it sees the other laser beam, propagating in the same direction as its
velocity, shifted further away from resonance. The atom absorbs more photons from the
beam propagating opposite to its velocity and thus slows down. From eq. (1), the forces
due to the positive- and negative-going waves, along the direction of propagation, are
Γ I/I0
(3) F± (v) = ±h̄k .
2 1 + I/I0 + [2(δ ∓ kv)/Γ]2
If we assume that I/I0 1, so that the intensity is low enough for us to add the forces
from the two waves independently, and that kv Γ, the total force is given by
4h̄k 2 I/I0 2δ
(4) Ftot = F+ + F− = v = −αv.
[1 + (2δ/Γ)2 ]2 Γ
For δ < 0, α is positive and the force damps the velocity at a rate v̇/v = −α/M . The
largest damping force is obtained for δ ∼
= −Γ/2.
Atomic motion in a strong standing wave [14, 16, 17, 20] is beyond the scope of this
treatment, however a simple approximation for small velocity, ignoring interference of the
two beams and stimulated redistribution of photons between the beams [21], shows that
the friction coefficient α maximizes at about α = h̄k 2 /4 for I/I0 = 1 in each beam and
2δ/Γ = −1. Under these conditions the velocity damping time v/v̇ for sodium, cooled
on the 589 nm resonance line, would be 13 μs.
So far we have only considered the average force on an atom in a light field. We
should remember that the force arises from discrete photon scatterings, and so must
fluctuate about its average. The fluctuations can be thought of as arising from two
sources: fluctuations in the number of photons absorbed in a given time and fluctuations
in the direction of the spontaneously emitted photons. Both of these effects arise because
of the randomness of spontaneous emission.
The fluctuations represent a random walk of the atomic momentum, with each ran-
dom walk step being of magnitude h̄k. For simplicity we will assume a fictitious one-
dimensional situation where photons are emitted as well as absorbed along a single axis.
Each scattering event represents two random walk steps, one from the absorption, which
could be from either of the counterpropagating beams, and one from the spontaneous
emission, which can be in either direction along the axis. The mean square momentum
of the atom increases linearly with the number of scattering events (random walk steps),
with a rate
d 2
(5) p = 2Rh̄2 k 2 ,
dt
where R is the scattering rate; the factor of 2 comes from the two steps per scattering.
We define a momentum diffusion coefficient
d 2
(6) 2Dp = p ,
dt
so that Dp /M is the rate of increase in kinetic energy, the heating rate. The damping
force decreases the kinetic energy as F · v = −αv 2 , the cooling rate. At equilibrium we
set the sum of the heating and cooling rates to zero, finding
Dp
(7) = M v 2 = kB T.
α
Here we have replaced v 2 with its mean value and used the equipartition theorem to
identify kB T /2 as the mean kinetic energy M v 2 /2 in the single degree of freedom.
Using eq. (4) to get α, eqs. (5) and (6) to get Dp , and remembering that the total
scattering rate from the two laser beams, each of intensity I, is
I/I0
(8) R=Γ ,
1 + (2δ/Γ)2
we find, for low intensity and small velocity:
h̄Γ 1 + (2δ/Γ)2
(9) kB T = .
4 2δ/Γ
This is the Doppler temperature, and we emphasize that it applies to the fictional one-
dimensional case we have constructed. A true 1D experiment, such as cooling an atomic
beam along one axis, would produce a lower temperature depending on the distribution
of scattered photons in 3D. However, it can be shown [21] that in a symmetrical three-
dimensional case, the temperature is also given by eq. (9). The temperature minimizes
at a detuning of δ = −Γ/2, where
h̄Γ
(10) kB TDopp =
2
defines the Doppler cooling limit. To derive the Doppler temperature we assumed that
the velocity was small enough that kv Γ. At the Doppler limit, sodium atoms would
have vr.m.s. = 30 cm/s, corresponding to a temperature of 240 μK and kvr.m.s. = Γ/20, so
the assumption is justified in this case (and most others). We may also ask whether the
velocity distribution corresponds to a temperature. That is, is it Maxwell-Boltzmann?

It can be shown that for sodium, and similar atoms where the recoil energy Erec =
h̄2 k 2 /2M h̄Γ, the velocity distribution is indeed very close to being thermal [21].
What happens if the recoil energy is not small? Then we violate the validity condition
on eq. (1), and the detuning due to the Doppler shift changes significantly with each
emission or absorption. In the limit where the linewidth Γ is small compared to the
recoil energy, we feel intuitively that the cooling limit, rather than being related to Γ, as
in eq. (10), is related to the recoil energy. Indeed, it can be shown [21, 22] that in this
case the lowest temperature attainable (the recoil temperature) is given by
1 h̄2 k 2
(11) kB T = = Erec ,
2 2M
that is, one recoil energy per degree of freedom. While it might seem that this is the
ultimate limit of laser cooling, in fact it is possible to break the recoil limit under certain
circumstances where the interaction with the light is turned off as the atomic velocity
becomes small. One way this may be achieved is by velocity selective coherent popula-
tion trapping in multilevel atoms [23, 24]. Atoms are optically pumped into a coherent
superposition of both internal states and center-of-mass velocities, a superposition that
cannot absorb the laser light. Another way is velocity-space optical pumping [25, 26].
Atoms are cooled on a transition with a narrow linewidth, but with a distribution of laser
frequencies such that the excitation rate for zero velocity atoms is small or vanishing.
This has been accomplished by use the use of pulsed, two-photon Raman transitions [27].
The damping force of eq. (4) is similar to the viscous force on an object moving in
a fluid. Because of this, the configuration of pairs of counterpropagating laser beams is
often called “optical molasses”. The viscosity is so high that a velocity corresponding to
the Doppler cooling limit is damped out and randomized while the atom travels only a
few tens of micrometers, much smaller than the size of the molasses, which is typically
a centimeter. Thus the atoms executes a Brownian-like motion with a short mean free
path, moving diffusively rather than ballistically [28,29]. The evidence for this is the long
residence time of atoms in optical molasses. Atoms require several seconds to diffuse out
of a typical optical molasses [29, 30], whereas, moving ballistically, atoms cooled to the
Doppler cooling limit would traverse a region the size of a typical molasses in a few tens
of milliseconds.
3. – Deceleration and cooling of an atomic beam
Experimentally, atoms must first be at reasonably slow velocities (kv ≤ Γ) before

Doppler cooling can be effective. The two general ways this has been accomplished are
by laser deceleration of an atomic beam [28, 31] and by collection of slow atoms from a
thermal gas [32, 33]. Loading from a thermal gas has the advantage of allowing a more
compact and simpler apparatus, while the beam deceleration technique usually allows
lower background pressure and faster production of slow atoms.
Deceleration of an atomic beam is usually accomplished by directing a near resonant
laser beam so as to oppose the atomic beam. The atoms absorb photons at a rate
determined by the intensity of the laser beam, the detuning from resonance and the atoms
velocity. For each photon absorbed, the atomic velocity changes by vrec in the direction
of the laser propagation. The spontaneously emitted photons are emitted randomly in
a pattern that is symmetric on reflection through the atom, so there is no net average
change in the atomic velocity due to these emissions. If the absorption is followed by
stimulated emission into the same direction as the incident laser beam (we assume the
laser beam is a plane wave), there is no net momentum transfer from the absorption-
emission process. Only absorption followed by spontaneous emission contributes to the
average force, which is given by the rate of scattering photons times the momentum of a
photon. For a two-level atom this force is given by eq. (1). At high intensity this force
saturates to the value h̄kΓ/2.
The acceleration of an atom due to the saturated radiation pressure force is amax =
h̄kΓ/2M = vrec Γ/2, which can be quite large. For sodium with λ = 2π/k = 589 nm,
1/Γ = 16 ns and M = 23 a.m.u., vrec ≈ 3 cm/s and amax ≈ 106 m/s2 . For cesium,
with λ = 2π/k = 852 nm, 1/Γ = 30 ns and M = 133 a.m.u., vrec ≈ 3.5 mm/s and
amax ≈ 6 × 104 m/s2 . This acceleration would stop in 50 cm, a thermal, 1000 m/s Na
atom scattering ≈ 33000 photons in 1 ms and in 80 cm, a thermal, 300 m/s Cs atom
scattering ≈ 84000 photons in 5 ms.
Implicit in eq. (1) is one of the major impediments to effective deceleration of an
atomic beam using a counterpropagating
laser beam. The force acting on the atom is
large if |2(δ −k·v)| ≤ Γ 1 + I/I0 . Atoms much outside this resonant-velocity range will
experience little deceleration, and atoms initially within this range will be decelerated out
of it. This process results in a cooling or velocity compression of a portion of the atomic
beam’s velocity distribution, and was first observed by Andreev et al. [34]. Atoms initially
at the resonant velocity decelerate out of resonance. Other atoms with nearby velocities
will also decelerate, those with larger velocities first decelerating into resonance, then to
slower velocities out of resonance, while initially slower atoms decelerate to still lower ve-
locities. The atoms will “pile up” at a velocity somewhat lower than the resonant velocity.
Both deceleration and cooling occur because a range of velocities around the resonant
velocity are compressed into a narrower range at lower velocity. However, only a small
portion of the total velocity distribution has been decelerated by only a small amount.
There are a number of possible solutions to this problem, some of which have been
discussed in ref. [28]. These include Zeeman tuning [28] where a spatially varying mag-
netic field compensates the changing Doppler shift as the atoms decelerate, so as to keep
the atoms near resonance; white-light deceleration [35] where a range of laser frequencies
ensures that some light is resonant with the atoms, regardless of their velocity (within
the range to be decelerated); diffuse-light deceleration [36] where light impinges on the
atoms from all angles so that, with the Doppler shift, some of the light is resonant with
each velocity; Stark cooling [37] where a spatially varying electric field is used to Stark
shift atoms and keep them near resonance as they Doppler shift due to deceleration; in-
tense standing wave deceleration [38] where the linewidth is sufficiently power broadened
to capture a large velocity distribution for laser deceleration; and “chirp cooling” [39]
in which the frequency of the laser is swept up, or chirped, in time such that the laser
stays in resonance with atoms that have been decelerated, and they continue to absorb
photons and decelerate. Zeeman tuning and chirp cooling were among the earliest atomic
beam slowing techniques demonstrated and they continue to be the most widely used
techniques to date.
.
3 1. Chirp cooling. – In chirp cooling, the frequency of the laser is swept up, or chirped,
in time [39]. Because of the chirp, atoms that have been decelerated by the laser stay
in resonance, continue to absorb photons, and continue to decelerate. Furthermore, the
chirp brings the laser into resonance with additional atoms having lower velocities than
the original group around the velocity initially resonant with the laser.
In order to analyze this process, let us consider atoms having positive velocities near
some velocity V opposed by a laser beam propagating in the negative direction. We

express any atomic velocity as v = V + v . The acceleration of atoms having velocity
V (v = 0) is a = F (V )/M , where F (V ) is given by eq. (1). Therefore, we write
V (t) = V (0) + at. Also we let the detuning vary as δ(t) = δ − kV (t). That is, we
chirp the laser frequency so as to stay a constant detuning δ from resonance with atoms
having the decelerating velocity V (t). Now we transform to a frame decelerating with
V (t). In this frame the atomic velocity is v and the laser detuning is Doppler shifted to
δ . The force on an atom in this frame, for small v , is
I (2δ /Γ)v
(12) F (v ) = 2h̄k 2 .
I0 1 + I/I0 + (2δ /Γ)2 2
The term multiplying v is minus the friction coefficient α. When δ < 0, the force opposes
the velocity v and tends to damp all velocities to zero in the decelerating frame, which is
V (t) in the laboratory frame. Maximum damping occurs for I/I0 = 2 and 2δ /Γ = −1.
The final velocity to which the atoms are decelerated is determined in practice by the
final frequency to which the laser is chirped. The result of chirp cooling an atomic beam
is that all of the atoms in the initial distribution below the velocity resonant with the
laser at the beginning of its chirp are decelerated.
The first definitive experiment showing such chirp cooling is in ref. [40], with decel-
eration to zero velocity first achieved in ref. [41]. The analysis given above is similar to
that given in ref. [42]. The robust character of this sort of cooling is evident. Atoms
within a range of velocities around V (t) are damped (in velocity) toward V (t). Lower
velocities, not initially close to V (t), come within range as the laser chirp brings V (t)
into coincidence with them. If the laser intensity changes during the time an atom is
being decelerated (because, for example, the laser beam is not collimated), the atoms will
continue to decelerate according to the chosen chirp rate, but with a different effective
detuning δ . The chosen chirp rate, however, must be consistent with the achievable
deceleration for the given I/I0 . That is, the chirp rate must satisfy
h̄k 2 Γ I/I0
(13) δ̇ = ka = .
2M 1 + I/I0 + [2δ /Γ]2
This means that δ̇ has an allowable upper limit of kamax . We have noted that for the
velocities to be damped in the decelerating frame we must have δ < 0 and it is easy
to show that the conditions for best damping lead to a deceleration half as large as the
maximum.
.
3 2. Zeeman tuning. – In the Zeeman tuning technique, a spatially varying magnetic
field is used to keep the frequency of an atom resonant with a counterpropagating laser
beam, as the atom slows down by scattering photons from this laser. Because the atoms
are slowed down and the spread of the velocity distribution is narrowed, this process is
often referred to as “Zeeman cooling.” Figure 2 illustrates the general idea of this scheme.
Fig. 2. – Upper: Schematic representation of a Zeeman slower. Lower: Variation of the axial
field with position.
The atomic beam source directs atoms, which have a range of velocities, along the axis
(z-direction) of a tapered solenoid. This magnet has more windings at its entrance end,
near the source, so the field is higher at that end. The laser is tuned so that, given the
field-induced Zeeman shift and the velocity-induced Doppler shift of the atomic transition
frequency, atoms with velocity v0 are resonant with the laser when they reach the point
where the field is maximum. Those atoms then absorb light and begin to slow down. As
their velocity changes, their Doppler shift changes, but is compensated by the change in
the Zeeman shift as the atoms move to a point where the field is weaker. At this point,
atoms with initial velocities slightly lower than v0 come into resonance and begin to slow
down. The process continues with the initially fast atoms decelerating and staying in
resonance while initially slower atoms come into resonance and begin to be slowed as
they move further down the solenoid. Eventually all the atoms with velocities lower than
v0 are brought to a final velocity that depends on the details of the magnetic field and
laser detuning. The magnetic field profile of the tapered solenoid is
2az
(14) B(z) = B0 1−
v02
with 0 ≤ 2az ≤ v02 . B0 is the magnetic field producing a Zeeman shift equal to the
Doppler shift for atoms with velocity v0 and a ≤ amax is the deceleration rate.
The first experiment on the deceleration of atoms using the Zeeman technique is
in ref. [43]. Subsequently, neutral sodium atoms were stopped in ref. [44]. Figure 3
shows the velocity distribution resulting from Zeeman cooling: a large fraction of the
initial distribution had been swept down into a narrow final velocity group. The velocity
distribution after deceleration was measured in a detection region some distance from the
exit end of the solenoid using a separate detection laser. We were able to determine the
velocity distribution in the atomic beam by scanning the frequency of the detection laser
and observing the fluorescence from atoms having the correct velocity to be resonant.
density of atoms (arb. units)
velocity (m/s)
0 400 800 1200 1400 1600 1800
Fig. 3. – Velocity distribution before (dashed) and after (solid) Zeeman cooling. The arrow
indicates the highest velocity resonant with the slowing laser. (The extra bump at 1700 m/s is
from F = 1 atoms, which are optically pumped into F = 2 during the cooling process.)
.
3 3. Optical pumping. – For alkali atoms, whose electronic ground states are split
by the hyperfine interaction, deceleration of an atomic beam may be impeded by the
.
optical pumping problem described above in subsect. 2 1. For chirp cooling, the usual
way of addressing this problem is, as with optical molasses, to use a repumper that
excites atoms out of the “wrong” hyperfine state so that they may again be excited by
the cooling laser. In Zeeman tuning the large Zeeman shifts from the high magnetic field,
the use of a circularly polarized decelerating laser, and the nature of the matrix elements
for radiative transitions in the magnetic field, all work together to make the probability
of deleterious optical pumping very small, so that a repumper is not usually needed.
4. – Traps for neutral atoms
Trapping of atoms usually refers to their confinement by the application of external

fields rather than by the use of a material container. In contrast with the trapping of
ions by electric and magnetic fields, the trapping forces that can be applied to neutral
atoms are relatively weak. Ions have a charge on which an electromagnetic field can exert
a large Coulomb or Lorentz force. Neutral atoms, however, may be acted upon through
their permanent magnetic dipole moments or induced electric dipole moments, allowing
generally smaller forces to be applied. The strongest traps for neutral atoms have energy
depths of only a few kelvin, while ion traps can typically hold room temperature ions,
and have trapped particles with energies of a few thousand electron volts.
Another possible difficulty with traps for neutral atoms is that the trapping potentials
represent changes in the internal energy of the atoms. That is, the positions of and, in
general, the spacings between energy levels are changed by the trapping fields. This
means that, in particular, high accuracy spectroscopy of trapped atoms is problematic.
Neutral atoms have the advantage that the lack of space charge effects means that
one can generally trap larger numbers and densities of neutral atoms than of ions. Fur-
thermore, some applications demand that one work with neutral atoms (as for example
in Bose condensation of an atomic gas). Fortunately, even rather weak forces are capable
of trapping atoms that have been laser cooled, and many different kinds of neutral atom
traps have been demonstrated. Among these are:
i) Magneto-static traps, first demonstrated in 1985 [45], rely on the force exerted by
a gradient magnetic field on the permanent magnetic dipole moment of an atom
such as a ground-state alkali.
ii) Laser dipole traps, first demonstrated in 1986 [46], use the dipole force that results
from the gradient of the energy of the oscillating dipole moment induced on an
atom in an inhomogeneous laser field.
iii) Radiation pressure traps use the scattering force of eq. (1), but are not stable in 3D
for two-level atoms and for laser intensities constant in time. The magneto-optical
trap (MOT), using multilevel atoms and an inhomogeneous magnetic field, was the
first radiation pressure trap to be demonstrated [47].
iv) Magneto-dynamic traps use the micromotion driven by oscillating magnetic field
gradients to allow trapping of high-field–seeking states not stably trapped in static
magnetic fields. Such a trap, first demonstrated in 1991 [48], is analogous to the
radio-frequency Paul trap for ions.
v) Microwave traps are low-frequency, spontaneous-emission-free analogs of laser
dipole traps. Such a trap was first demonstrated near a magnetic resonance transi-
tion [49]. Similarly, traps based on rf dressed states have recently been realized [50].
vi) Electrostatic traps are similar to laser dipole traps but rely on the interaction
between a gradient, DC electric field and the polarization induced in the atom by
this field to generate a trapping force. Electrostatic trapping and deceleration of
molecules has recently been demonstrated [51].

vii) Gravito-optical traps, which combine optical dipole forces with gravity to produce
stable trapping [52], are only one example of hybrid traps that combine different
types of forces to achieve trapping of atoms.
viii) TOP traps (for Time-averaged Orbiting Potential) [53] are an important modi-
fication of one type of magnetostatic trap. While time dependent, they are not
dynamic in the sense of atomic micromotion being essential.
These notes will not treat each of these kinds of traps in detail, nor attempt to give
more complete references about them. We shall, however, discuss the radiation pressure,
laser dipole and magnetic traps in some more detail below.
. .
4 1. Dipole force traps. – The dipole force discussed in subsect. 2 2 can be used to
trap atoms. A single, focused laser beam, tuned below resonance is the simplest dipole
trap and was first proposed by Ashkin in 1978 [54]. As an example, consider sodium
atoms interacting on their strongest transition (I0 = 6 mW/cm2 ) with a modest power,
10 mW Gaussian laser beam focused to a 1/e2 radius of 10 μm. This gives I/I0 ∼ = 106 at
the focus. For the detuning maximizing U , we find Umax /kB ∼ = 100 mK. For a 1 W beam
we find Umax /kB ∼ = 1 K. Such traps can easily confine laser cooled atoms. Gas atoms at
temperature above those achieved by laser cooling will not be easily trapped.
The dipole potential is a conservative potential, so it does not have any dissipative
mechanism associated with it. It was not until the demonstration of laser cooling in
optical molasses [29] that such a trap could be loaded and finally realized [46]. One
of the difficulties involved in such trapping is that although the trap is tuned below
resonance (the proper sign of the detuning to achieve cooling) the cooling provided by
the trapping light does not reduce the thermal energy below the trap depth. Auxiliary
cooling [55] is required, but even this is difficult because the inhomogeneous light shifts
induced by the trapping laser interfere with the cooling process. Dalibard et al. [56, 57]
proposed a solution in which the trapping and cooling are alternated in time, and this
procedure was used for the first dipole trap [46].
Another difficulty with a single focus dipole trap is that the radiation pressure force
pushes the atoms away from the focus, while the dipole force is attracting them to
it. While more complicated, counterpropagating beam geometries can avoid this diffi-
culty [54, 58, 59], one can, at the expense of reduced trap depth, solve the problem by
detuning the laser [46]. According to eq. (1), the destabilizing radiation pressure force
varies as 1/δ 2 (for sufficiently large δ), while the dipole trapping force obtained from
the dipole potential given by eq. (2) varies as 1/δ. Thus, for large enough detuning, the
radiation pressure will be negligible compared to the dipole trapping force.
Large detuning has other advantages, particularly when coupled with multi-level laser
cooling. When the detuning is large enough that the trap depth is comparable to the
natural linewidth, the thermal energy of laser cooled multilevel atoms can still be consid-
erably less than this depth, which is comparable to the Doppler cooling limit of eq. (10).
Furthermore, the optimum detuning for good multilevel laser cooling is at least several
times the linewidth, the inhomogeneous light shifts should not much affect the cooling.
Such a far-off-resonance-trap (FORT) has been demonstrated to work without the need
to alternate cooling and trapping phases [60, 61].
Another advantage [62] of such a FORT is that for sufficiently large detuning the
population of the trapped atoms is almost entirely in the ground state. The trap is then
nearly free of heating due to spontaneous emission [63] and of many of the collisional
perturbations involving excited atoms. FORTs have been used to hold atoms for collision
experiments [64], to trap atoms for evaporative cooling [65], and to confine Bose-Einstein
condensates [66].
Variations of the dipole force trap include using the evanescent wave created by to-
tal internal reflection of light (detuned blue of resonance) to act as a mirror to reflect
atoms [52, 67], crossing two red-detuned laser beams at their foci to achieve a strong
gradient in all directions [65], and crossing sheets of blue-detuned light to form a box
bounded by light (with confinement in the vertical direction sometimes provided by grav-
ity) [68, 69].
.
4 2. Radiation pressure force traps. – Radiation pressure force traps use the sponta-
neous or scattering force to confine atoms. Unlike the dipole force, the force generated
by absorption of a photon followed by spontaneous emission does not depend on the
gradient of the intensity. Hence larger volume traps can be created using the scattering
force compared to the dipole force for a given flux of incident photons. A simple radiation
force trap, shown in fig. 4, can be made in 1D using two counterpropagating focused laser
beams with separated foci [54]. Midway between the foci the radiation pressure from the
two beams is balanced and the net force on an atom is zero. As the atom moves away
from this equilibrium point toward one of the foci it is pushed back by the higher inten-
sity near that focus. In contrast to the dipole force trap, radiation pressure trap depths
on the order of a kelvin can be achieved with a modest, near saturation intensity, i.e.,
a few milliwatts per square centimeter [70]. While the trap of fig. 4 produces radiation
pressure force trapping along only one axis, the dipole force can provide trapping along
the two orthogonal axes [54], a configuration first demonstrated in ref. [59].
It is tempting to extend the trap idea of fig. 4 to three dimensions, but it can be shown
that such an extension is not possible when the radiation pressure force is proportional
to the photon flux from the laser beams. The impossibility of such trapping is related to
the fact that the divergence of the Poynting vector is zero, and has been called the optical
Earnshaw theorem [71] by analogy with the theorem from electrostatics forbidding stable
trapping of a test charge in a charge-free region.
In spite of the optical Earnshaw theorem, which applies to 2-level atoms in a weak,
static laser field, where the scattering force is proportional to the Poynting vector, it is
possible to create a 3D radiation force trap by making use of such features as saturation,
multiple levels, optical pumping and Zeeman shifts, which make the scattering force
not proportional to the Poynting vector. The most successful radiation pressure trap
Fig. 4. – A one-dimensional radiation pressure force trap formed from two counterpropagating,
focussed laser beams with separated foci. The length of each of the pairs of arrows indicates the
magnitude of the force from the respective laser beams on an atom at the indicated positions.
B
a)
σ+ σ−
I
b) c)
m = −1 E
−1 0 1 m=0
δ
m = +1
σ− σ+
σ+ σ−
0
z
Fig. 5. – a) Magnetic field and laser configuration for a 1D MOT. b) Transition scheme. c) Energy
levels and transitions in the spatially varying magnetic field. The designation of the m-state is
with respect to a space-fixed axis, as is the laser polarization.
that circumvents the optical Earnshaw theorem is the magneto-optical trap or MOT.
Conceived by Dalibard [72] and demonstrated in an MIT-Bell Labs collaboration [47], its
principle is illustrated in fig. 5 for 1D operation and a simple J = 0 → J = 1 transition.
A pair of current-carrying coils with opposing currents creates a quadrupole magnetic
field that is zero at the origin and whose vector value is proportional to the displacement
from the origin. The simple J = 0 → J = 1 transition gives us a Zeeman shifted
transition frequency with a non-degenerate ground state. (This non-degeneracy leads to
Doppler rather than sub-Doppler cooling.) Two circularly polarized laser beams with
opposite helicity counterpropagate along the coils’ axis. The σ ± beam excites atoms to
the m = ±1 excited states, respectively. Thus, for a red-detuned laser frequency (δ < 0)
atoms displaced in the positive direction will experience a Zeeman shift that brings the
m = −1 state into resonance with the laser frequency, and the σ − laser beam that excites
this state is the one that pushes it back toward the origin. Similarly, an atom displaced
in the negative direction is pushed back by the σ + beam. In addition to the restoring
force, there is also the usual Doppler-cooling damping force.
The force on an atom with velocity v and position z can be obtained from eqs. (3)
and (4) by replacing the effective detuning δ ∓ kv with δ ∓ (kv + βz), where βz is the
magnitude of the Zeeman frequency shift at position z:

4h̄kI/I0 2δ
(15) F (v, z) ∼
= 2 (kv + βz),
1 + (2δ/Γ)2 Γ
where in the second expression we are in the limit of low velocity, magnetic field and
laser intensity. For negative detuning the above force represents a damped harmonic
oscillator. Typical operating conditions for a MOT might be I/I0 = 1, 2δ/Γ = −1 and
β = 2 π × 10 MHz/cm. This would lead to an oscillation frequency of about 1 kHz for
Na, but with strong overdamping.
While we have considered only the 1D case for a simple transition, the MOT was first
demonstrated in 3D, on an atom with a degenerate ground state (Na). The theory in
3D has been worked out in detail for the J = 0 → J = 1 transition [73]. For transitions
allowing sub-Doppler cooling (see sect. 5 below), some insights have been gained by
studying the forces in 1D on a moving atom in a magnetic field [74-77]. Experiments have
shown that, with a degenerate ground state, sub-Doppler temperatures are achieved in a
MOT [76] along with larger trapping and damping than predicted by the J = 0 → J = 1
theory.
The MOT has become an important tool in the study of cold atoms. It is routinely
used to capture atoms that have been slowed by chirp cooling or Zeeman tuning. A
particularly useful feature of the MOT is that it can capture atoms from an uncooled,
thermal atomic vapor [32, 33]. This often allows considerable simplification of the appa-
ratus compared to one where an atomic beam is first decelerated.
A MOT can concentrate atoms to the point that collisions [78,79] and radiation pres-
sure exerted by the atoms’ fluorescence [80] are factors limiting the density. A major
advance in reducing such problems is the “dark spot” MOT [81] in which atoms are, for
the most part, optically pumped into a state from which the excitation rate is consid-
erably smaller than in a conventional MOT. In practice this is accomplished, for atoms
with ground-state hyperfine structure, by pumping the atoms into one of the hyperfine
states. Normally laser cooling and trapping of such atoms is performed by applying a
separate laser frequency to excite each of the ground hyperfine states to the electronically
excited state, ensuring atoms are not pumped into a state from which they cannot be
excited. In a dark spot MOT, one of the laser frequencies (the re-pumper) is eliminated
from the central region of the trap. Atoms then accumulate in the hyperfine state that
would have been excited by the missing frequency. These atoms are rarely excited (only
by off-resonant light or indirectly scattered resonant light) so problems due to excited
atoms are greatly reduced, even though the atoms are still cooled and trapped. Such
techniques have achieved atomic densities near 1012 cm−3 .
.
4 3. Magneto-static traps. – Both laser dipole and radiation pressure traps involve
optical excitation of the atoms being trapped. Although in principle a sufficiently intense
laser dipole trap can be tuned far enough off resonance for the excitation rate to be small,
it is difficult in practice to make it truly negligible. As a result, optical trapping generally
results in some heating of the atoms due to the random nature of spontaneous emission,
and cooling is required to keep the atoms in equilibrium. Furthermore, whenever the
atoms are excited they are much more likely to undergo collisions that will heat them
and eject them from the trap. Magnetic traps do not suffer this problem and so can
be used to store atoms for long periods of time without the need for additional cooling,
and without suffering much collisional loss. On the other hand, such traps only work for
atoms having a magnetic dipole moment and only for those states of such atoms whose
Zeeman energy increases in increasing magnetic field.
The idea of magnetic trapping was an extension of the magnetic focusing of atomic
beams [82] and was discussed by Paul during the 1950s [83]. The first published proposals
for magnetic trapping of neutral atoms were in the 1960s [84-86]. It was not until after
the demonstration of laser cooling of neutral atoms that the first magnetic trapping of
atoms was achieved [45].
The principle of the magnetic trap can be understood by considering an atom with
a Zeeman sub-structure such as that shown in the excited state of fig. 5c. (Note that
magnetic trapping is generally done on ground-state atoms with such Zeeman structure.)
Since the energy varies with magnetic field, an atom feels a force in a magnetic field
gradient. For states whose energy increases with magnetic field (low-field–seeking states),
a trap is formed by the field of the coils in fig. 5a. This quadrupole field is zero on the
axis midway between the coils, and its magnitude increases linearly along any line away
from this central point. Low-field–seeking atoms experience a restoring force towards and
are trapped around this point. (It can be shown [87, 88] that in a current-free region no
magnetic field can have a relative maximum in its magnitude, so that high-field seekers
cannot be trapped.) The depth of a magnetic trap is given by the maximum Zeeman shift
along the easiest escape path. A magnetic field of 1 mT (10 G) gives a typical shift of
14 MHz, equivalent to 670 μK, so laser-cooled atoms are easily trapped by modest fields.
The quadrupole magnetic field was used in the first magnetic trap [45] and is the sim-
plest of magnetic traps. It works for low-field seekers as long as they do not change their
spin orientation with respect to the local field. This means than they must adiabatically
follow the changes in the direction of the field as they orbit in the trap. The condition
for adiabatic following is that
dθ
(16) ωZeeman ,
dt
where θ is the angle of the magnetic field at the atom’s position and ωZeeman is the Zeeman
frequency separating states of different spin orientation. Since laser-cooled atoms move
so slowly, they generally experience small field rotation rates, and their motion is usually
adiabatic. The quadrupole trap, however, has a point where the magnetic field is zero
and the adiabatic condition is impossible to satisfy. Atoms passing sufficiently close to
the trap center will fail to follow adiabatically, change their spin orientation (undergo
Majorana transitions) and be ejected from the trap [45]. This difficulty can be avoided
by a variety of traps that do not have a point of zero magnetic field [89], but such traps
are generally not as “stiff” as a quadrupole trap. That is, the restoring force near the
center of the trap is not as large as for a quadrupole trap. In practice, non-adiabaticity is
usually a problem only for very cold atoms confined very near the center of a quadrupole
trap (which is exactly what occurs on the way to BEC.)
In order to achieve Bose-Einstein condensation through runaway evaporative cooling,
it is desirable to have a “stiff” trap to maintain a high collision or thermalization rate. For
atoms cold enough to Bose condense, the problem of non-adiabatic spin flip transitions
at the zero-field point of a quadrupole trap is quite severe [90, 91]. The TOP trap
provided one solution to this problem by superposing onto the quadrupole field a magnetic
field rotating in the plane of symmetry. The rotating field guarantees that the field is
not zero at the trap center, and produces a rounding of the sharp-cusp, quadrupole,
trapping potential. This time-averaged, orbiting potential (TOP) trap [53], although
time-dependent, is not a magneto-dynamic trap in the sense that the TOP trap does not
rely on the micromotion of the atoms. The atoms respond to the time-averaged potential
at a given point in the trap.
5. – Sub-Doppler laser cooling
The early experiments [29, 92] on optical molasses produced a satisfying agreement
of the observed temperature and spatial diffusion with the predictions of the theory of
.
Doppler cooling as outlined in subsect. 2 3. Some experiments, however, were in disagree-
ment with expectations [93]. Finally, in 1988, careful temperature measurements [94]
showed conclusively that atoms in optical molasses were much colder than the Doppler
cooling limit. The time-of flight (TOF) method used to measure the temperature in
those experiments has become a standard technique. Atoms are collected and cooled in
the optical molasses at the intersection of the molasses laser beams. The molasses beams
are suddenly extinguished and the released atoms fall toward a probe. As the atoms
pass through the probe laser beam they absorb and fluoresce light. This fluorescence is
measured, with time resolution, by a photodetector. The distribution of detected fluores-
cence in time gives the distribution of times of flight from the molasses to the probe, and
the temperature can be determined from that distribution. With the ultralow tempera-
tures and high densities of or near Bose-Einstein condensation, researchers often use the
direct measurement of the spatial expansion of a cloud of atoms (typically obtained by
imaging the spatial profile of the absorption of a resonant or near-resonant laser beam)
to determine the temperature.
.
5 1. Observation of sub-Doppler temperatures. – Figure 6a shows an example of an
experimental TOF signal for sodium atoms cooled in optical molasses [21]. The exper-
imental points correspond reasonably well with the predicted signal for a temperature
of 25 μK, while 250 μK, about the Doppler cooling limit for Na, is completely inconsis-
tent with the experimental points. Furthermore, the dependence of the temperature on
detuning was found to be inconsistent with the theory of Doppler cooling. Figure 6b
(points) shows the measured temperature as a function of laser detuning for Na, where
the linewidth is 10 MHz. The temperature decreases for detunings larger than Γ/2 (until
the laser frequency approaches resonance with another hyperfine state, about 60 MHz to
the red of the chosen resonance). This is in sharp contrast to the prediction of Doppler
cooling theory (fig. 6b, solid line), which has the temperature increasing for detunings
greater than Γ/2. In addition, the temperature was found to be linearly dependent on
laser intensity [95], again in contrast to the predictions of Doppler cooling theory, and
the temperature was found to depend on magnetic field and laser polarization [94, 95].
1000
250 μK 500
a) b)
fluorescence signal
800 25 μK
400
temperature (μK)
600
300
400 200
200 100
0 0
0 20 40 60 80 100 0 10 20 30 40
time (ms) detuning (MHz)
Fig. 6. – a) Experimental time-of-flight distribution (points) for Na atoms released from an

optical molasses. Expected TOF distributions for temperatures of 25 μK and for 250 μK (ap-
proximately the Doppler cooling limit) are shown. b) Measured temperature as a function of
laser detuning for sodium atoms cooled by a 3D optical molasses. The temperature predicted
by Doppler cooling theory is shown by the solid line.
These latter facts, in particular, suggest that the magnetic sublevels of the atom play an
important role.
The observation of sub-Doppler-limit temperatures was quite surprising. Doppler
cooling theory at low intensity was simple and compelling. Furthermore, at least for
1D, there was a complete theory, taking into account the effects of high intensity and
interference between laser beams that were ignored in the treatment presented in the
previous section. The theory had been restricted to 2-level atoms, but it was widely
believed that this restriction was not particularly important. At low intensity the Doppler
temperature depended on the transition linewidth and the detuning, and these were
the same for any of the degenerate Zeeman sublevels in a given alkali hyperfine level.
Nevertheless, the magnetic field and polarization dependence of the sub-Doppler-limit
temperatures pointed to the importance of the Zeeman sublevels.
.
5 2. New cooling mechanisms. – The explanation for the sub-Doppler temperatures
soon came in the form of a new theory of multilevel laser cooling. The key elements of
the new theory [74, 96, 97] were optical pumping among the magnetic sublevels of the
electronic ground level and differential light shifts of the sublevels. While the original
theories treated cases where a spatial gradient of the polarization of the optical field was
important, it was later demonstrated that such multilevel laser cooling was possible even
without polarization gradients [97-103].
Since the theory of multilevel laser cooling is treated extensively in the literature a
detailed description will not be given here. The most important of the multilevel cooling
mechanisms is Sisyphus cooling [19, 74, 104, 105]. Semiclassically (when the atom can
be considered to be well-localized on the scale of an optical wavelength) the simplified
physical picture for Sisyphus cooling can be understood by considering the atom and
laser field situation illustrated in fig. 7.
m = −1/ 2 +1/ 2
a) x z b)
0 λ/ 4 λ/ 2
y
σ− σ+ σ−
Fig. 7. – a) Interfering, counterpropagating beams having orthogonal, linear polarizations create

a polarization gradient. b) The different Zeeman sublevels are shifted differently in light fields
with different polarizations; optical pumping tends to put atomic population on the lowest
energy level, but non-adiabatic motion results in “Sisyphus” cooling.
Figure 7a shows a 1D set of counterpropagating beams with equal intensity and

orthogonal, linear polarizations. The interference of these beams produces a standing
wave whose polarization varies on a sub-wavelength distance scale. At points in space
where the linear polarizations of the two beams are in phase with each other, the resultant
polarization is linear, with an axis that bisects the polarization axes of the two individual
beams. Where the phases are in quadrature, the resultant polarization is circular and at
other places the polarization is elliptical. An atom in such a standing wave experiences
a fortunate combination of light shifts and optical pumping processes.
Because of the differing Clebsch-Gordan coefficients governing the strength of coupling
between the various ground and excited sublevels of the atom, the light shifts of the
different sublevels are different, and they change with polarization (and therefore with
position). Figure 7b shows the sinusoidal variation of the ground-state energy levels
(reflecting the varying light shifts or dipole forces) of a hypothetical Jg = 1/2 → Je = 3/2
atomic system. Now imagine an atom to be at rest at a place where the polarization is
σ − at z = λ/8 in fig. 7a. As the atom absorbs light with negative angular momentum
and radiates back to the ground states, it will eventually be optically pumped into the
mg = −1/2 ground state, and simply cycle between this state and the excited me = −3/2
state. For low enough intensity and large enough detuning we can ignore the time the
atom spends in the excited state and consider only the motion of the atom on the ground
state potential. In the mg = −1/2 state, the atom is in the lower energy level at z = λ/8,
as shown in fig. 7b. As the atom moves, it climbs the potential hill of the mg = −1/2
state, but as it nears the top of the hill at z = 3λ/8, the polarization of the light becomes
σ + and the optical pumping process tends to excite the atom in such a way that it decays
to the mg = +1/2 state. In the mg = +1/2 state, the atom is now again at the bottom of
a hill, and it again must climb, losing kinetic energy, as it moves. The continual climbing
of hills recalls the Greek myth of Sisyphus, so this process, by which the atom rapidly
slows down while passing through the polarization gradient, is called Sisyphus cooling.
In Sisyphus cooling, the radiated photons, in comparison with the absorbed photons,
have an excess energy equal to the light shift, while in Doppler cooling, the energy excess
comes from the Doppler shift.
In contrast to the case for Doppler cooling (see eq. (4)) the friction force is independent
of laser intensity, and proportional to detuning at low (but not too low) intensity and
low velocity, while the momentum diffusion coefficient is proportional to intensity and
independent of detuning. This leads, according to eq. (7), to a temperature that depends
linearly on intensity and inversely on detuning. That is, the temperature is proportional
to the light shift
h̄Ω2
(17) kB T ∝ h̄Δlight shift = ,
4δ
where the expression for the light shift is valid in the limit of low intensity and large
detuning.
A less restricted treatment [106] shows that the friction force is not linear in the atomic
velocity, nor is the momentum diffusion constant independent of velocity. Nevertheless,
the temperature remains approximately linear in the light shift as long as the intensity is
sufficiently above a critical intensity. The lower limit to the temperature obtainable by
Sisyphus cooling is set by this lowest intensity for which the cooling process works [106,
107]. This is the intensity at which the light shift is comparable to the recoil energy, and
it leads to a lower limit for the thermal velocity on the order of a few times the recoil
velocity.
These qualitative features of multi-level laser cooling have been confirmed by exper-
iments on atoms cooled in 3D optical molasses [21, 99, 108]. The experimental results
showed the linear dependence of the temperature on intensity and light shift for all but
the largest intensity at the smallest detuning, outside the limits of validity of the simple
results listed above. The constancy of the lowest temperature, once the detuning is large
enough, is consistent with its depending only on the recoil energy. The temperature
of 2.5 μK obtained for Cs represents an r.m.s. velocity of about three recoil velocities,
similar to the case for the lowest temperatures observed for Na [21, 99] and Rb [109].
6. – Metrology with cold atoms
Since laser cooled atom move much more slowly than ordinary, thermal atoms, metrol-
ogy with such atoms is less subject to errors and uncertainties associated with the mo-
tion. These include the first-order Doppler shift of atomic transition frequencies and the
second-order Doppler shift or relativistic time dilation shift. Furthermore, the slow veloc-
ity means that atoms remain longer in the apparatus, allowing observation of narrower
linewidths for observation-time–limited features.
The wave nature of atoms become more evident as the atoms become slower be-
cause the deBroglie wavelength is inversely proportional to the velocity. This can make
atom interferometry easier with cold atoms, and atom interferometers can be used for
measurements, as optical interferometers are often used in metrology.
This section briefly describes the use of laser-cooled atoms in atomic clocks and in
metrology applications of atom interferometers. General background on cold atom clocks
can be found in ref. [110], and on atom interferometers in ref. [11].
.
6 1. Atomic fountain clocks. – Primary frequency standards, the devices by which
the SI definition of the second is realized, have traditionally been cesium atomic beams
using the Ramsey method of separated oscillatory fields [82]. In this method the atomic
beam is first state-selected so that the atoms are in one (let us say, the lower) of the two
hyperfine states (clock states) whose frequency difference (the clock frequency) defines
the unit of time. The atoms then enter a microwave cavity where a microwave field tuned
close to the clock frequency induces coherent transitions in the atoms. The strength of
the field and the duration of the passage of the atoms through the cavity are adjusted
so that the atoms receive approximately a π/2-pulse, putting the atoms into a coherent
superposition of the two hyperfine states.
After the atoms exit the cavity, they proceed in a nearly field-free region for a time
T to a second, similar cavity, where they receive another π/2-pulse, while in the field-
free region, the atoms, being in a superposition of the two hyperfine states, evolve at
the clock frequency. When they encounter the second cavity, with its second oscillatory
microwave field, the action of that field depends on the difference in the phase evolution
of the atoms and the phase evolution of the microwaves. When the microwave and clock
frequencies are identical, the second pulse completes the transition begun by the first
pulse, and the atoms are all in the upper hyperfine state. If the frequencies are different
enough so that the atom and field are out of phase by π, then the second pulse puts the
atoms back into the lower hyperfine state. The population of the lower (upper) state
varies sinusoidally (cosinusoidally) with the phase difference. The upper state population
varies from maximum to minimum for a difference of a half-cycle between the atom and
microwave frequencies over the time required for the atoms to travel between the two
cavities. This is also the frequency difference corresponding to the full width at half-
maximum (FWHM) of the feature corresponding to the hyperfine resonance. Thus, the
FWHM in hertz is given by Δν = 1/2T . For cesium atoms traveling at about 200 m/s
over a distance of 1 m between the cavities, this means a FWHM of about 100 Hz, a rather
narrow line, but one whose center must be determined to a part in 10−6 in order to achieve
the 10−14 accuracy that is typical of the best laboratories’ primary frequency standards.
Slower (colder) atoms would allow a longer Ramsey time T , a narrower FWHM, and
with it potentially better accuracy. The way in which such colder atoms could be used
to make a better atomic clock was suggested by Zacharias in the early 1950s [82]. In
this scheme, called an atomic fountain clock, state selected atoms are launched vertically
through a microwave cavity, where they experience a π/2-pulse as in the first passage
through a microwave cavity in the traditional beam clock. The atoms continue along
a vertical trajectory, falling back through the cavity, where they experience the second
π/2-pulse. For a trajectory height of 1.2 m above the cavity, the time between cavity
passages is about 1 s, an improvement of more than two orders of magnitude over a
typical laboratory primary reference standard. Furthermore, the fact that the second
passage is with the same cavity as the first passage eliminates some systematic errors
associated with having two different cavities in the traditional design.
The first realization of an atomic fountain clock of the Zacharias design [111] had
only a 5 cm fountain height, which still gave a 200 ms Ramsey time, significantly longer
than any beam clock. (An earlier atomic fountain design [112] did not have the atoms
passing through the cavity and falling back, but instead exposed the atoms to two pulses
of microwaves while they remained in the cavity near the top of their trajectory.) Today
atomic fountains at the LPTF in Paris [113], PTB in Germany [114] and at NIST in the
USA [115] with significantly larger heights achieve accuracy on the order of 10−15 .
Achieving temperatures below the Doppler cooling limit was an important feature in
the success of cesium fountain clocks. If the temperature of laser-cooled cesium atoms
had been as high as the 125 μK predicted by Doppler cooling theory, a cloud of cesium
atoms in a meter-high fountain would spread too much during the fountain time that less
than 1% of the launched atoms would return through the microwave cavity for the second
passage. (The opening in the 9.2 GHz microwave cavity, which allows the passage of the
atoms, cannot be any larger than about 1 cm without causing significant distortion of the
microwave field.) But even at temperatures as low as 1 μK most of the launched atoms
do not return for detection. One might suppose that a modest loss of atoms due to the
thermal spreading at 1 μK could be compensated by simply launching a larger number of
atoms. Unfortunately, collisional shifts of the clock frequency are large enough [116] that
such a procedure would introduce significant difficulties. A cesium density of 106 cm−3
at 1 μK causes a fractional shift on the order of 10−15 [117].
The collisional shift implies that is would be advantageous to further lower at least
the transverse velocity spread of the launched atoms, so as to insure that more of the
launched atoms are detected. One proposed method for achieving temperatures lower
than those usually achieved by laser cooling is Raman cooling [118, 119]. If we consider
the possibility of further improving the clock performance by increasing the Ramsey
time, such cooling becomes even more attractive. We cannot, however, increase the
Ramsey time significantly in an earth-bound atomic fountain clock. A Ramsey time
of even 10 s would require a fountain 120 m high, a daunting engineering prospect that
would also make shielding of magnetic fields quite difficult. For this reason, a number
of groups are considering the use of the “microgravity” environment of earth orbiting
satellites [120, 121].
.
6 2. Atom interferometers. – Optical interferometers have long been used for metrol-
ogy. Indeed, practical length metrology at the highest levels of precision is almost ex-
clusively accomplished by laser interferometry. Over the past two decades atom interfer-
ometry, where the wave nature of atoms is used in a way analogous to that of light, has
developed rapidly to the point where it too is an important metrological tool. In many
respects, this development has been driven or encouraged by the developments in laser
cooling and trapping of atoms. Two important reasons are that the longer deBroglie
wavelengths associated with cold atoms make their interference effects more accessible
and that the smaller velocity spreads available with laser cooling increase the deBroglie
wave coherence of the atoms.
In many respects the traditional Ramsey-resonance atomic clock, and the atomic
fountain clock that is based on it, may be thought of as atom interferometers: the
first interaction with the microwaves splits the atoms, originally state-selected in one
quantum state, into two components, the two “clock-state” hyperfine energy levels. The
second interaction splits or projects each of these states into the same two states. The
components of each state interfere so as to produce the Ramsey pattern, often called the
Ramsey interference pattern. This picture provides an analogy with optical Michelson
or Mach-Zehnder interferometers where a first interaction (with a beamsplitter) divides
the light and a second interaction (with the same or another beamsplitter) divides each
of these in such a way that they interfere with each other at the output.
One clear difference between the Ramsey interferometer and those optical interfer-
ometers is that in the Ramsey case the atoms are not physically separated while in the
optical case they are. The Ramsey interference is (at least in the traditional description)
between internal energy states of the atoms and does not involve the center-of-mass mo-
tion, deBroglie-wave character of the atoms. This apparent difference disappears when
we introduce photon momentum transfer into the interactions.
Consider the interferometer with the following configuration: An atom, initially at
rest and in internal state |1, is split by a π/2 pulse into a superposition of |1 and
|2. The pulse delivers the photon momentum to that component of the wave function
transferred to state |2, which then separates from state |1. A π pulse, applied after a
time T , transforms |1 into |2 and vice versa. The photon momentum associated with the
transition stops the motion of the portion of the wave function that had been in |2 as it
transforms to |1, and gives momentum to the state that had been in |1 as it transforms
to |2. Finally, after an additional time T , when the two paths of the atoms’ wave function
physically overlap, a π/2 pulse projects part of the amplitude of each component of the
wave function into final states |1 and |2 with their appropriate momenta. This example
is a time-domain, atomic analog of a Mach-Zehnder interferometer where the π/2 pulses
act as beamsplitters and the π pulses act as mirrors. An example of a Mach-Zehnder
type, atom interferometer based on stimulated Raman transitions is shown in fig. 13 of
.
subsect. 8 4.
Steve Chu and his colleagues at Stanford [122] have used this kind of interferome-
ter for precision measurement of the gravitational acceleration of atoms in an atomic
fountain geometry. The π and π/2 pulses in those experiments induce Raman transi-
tions between hyperfine states, using two laser beams whose frequencies differ by the
hyperfine separation. The laser beams are counterpragating, so that the momentum of
two optical photons is transferred to the atom when the hyperfine transition is induced.
Because of the different trajectories followed by the falling atoms in the two arms of the
interferometer, they experience different phase shifts, with the difference depending on
the gravitational acceleration and the square of the time (in the same way as does the
distance a body falls due to gravity). In the most recent reports, these experiments have
a resolution better than 10−10 and an accuracy on the order of parts in 109 . These re-
sults, whose accuracy is comparable to the best measurements with falling corner cubes,
are expected to improve with further experiments. Experiments using somewhat differ-
ent interferometer configurations have been used at Stanford to measure h/M [122], and
at MIT [123] and Yale [124] to measure absolute rotations. Bloch oscillations of laser-
cooled atoms in an optical standing wave have been used for a new determination of α,
the fine-structure constant [125]. While not yet competitive, α has also been measured
using atom interferometry with a BEC [126].
7. – Coherent manipulation of Bose-Einstein condensates with light
The creation of Bose-Einstein condensates (BECs) in dilute atomic vapors of Rb [127],

Na [128] and Li [129] is one of the major triumphs of laser cooling and trapping of neutral
atoms. Furthermore, the creation of BECs has renewed interest in the applications of
laser cooling and trapping techniques for atom optics, the manipulation of atoms with
mirrors, beamsplitters and lenses analogous to the manipulation of light.
Until recently, atom optic experiments have used thermal sources of atoms much as
early experiments in optics used lamps. What was lacking was a coherent source of
matter-waves similar to the laser for light. The creation of a Bose-Einstein condensate
(BEC) of a dilute atomic gas has opened up the possibility of realizing a matter-wave
source analogous to the optical laser. The macroscopic occupation of the ground state
of a trap by a BEC is similar to the occupation of a single mode of an optical cavity by
photons. The atoms forming the condensate all occupy the same wave function—both
in terms of their internal and external degrees of freedom.
Atoms from a BEC are nearly the ideal, monochromatic source for atom optics. Many
atom optical elements involve the interaction of the atoms with an optical field and the
associated transfer of the photon momentum to the atoms. Because of the repulsive
atom-atom interaction, which can be described by a mean field, the BEC swells to a size
significantly larger than the ground-state wave function of the harmonic trap confining
the atoms [130]. The spatial extent of the resulting wave function can be several orders
of magnitude larger than the optical wavelength. Hence the momentum width, given
by the Heisenberg uncertainty principle, can be much less than the photons momentum.
Not all experiments will realize this reduced, intrinsic momentum width. The interac-
tion energy may be converted to kinetic energy when the atoms are released from the
trap. Nonetheless, the resulting additional momentum spread, due to the atom-atom
interaction, can still be significantly less than the momentum of a single photon.
A principal focus of research on Bose-Einstein condensates in the Laser Cooling and
Trapping Group of NIST is in atom or matter-wave optics. This includes optically
induced diffraction as beamsplitters and mirrors for the condensate atoms, as well as
applications of diffraction in atom interferometry and the realization of an “atom laser”,
the atomic analogue of the optical laser. These experiments are basically demonstra-
tions of “single atom” phenomena applied to a collection of atoms with a high degree
of first-order coherence. When the interactions between the atoms is considered, col-
lective matter-wave phenomena can be observed which are analogous to effects in non-
linear optics with light. We have observed such effects with matter-waves and they are
described in the subsequent section titled “Nonlinear atom optics with Bose-Einstein
condensates”.
Our atom optics experiments are primarily performed on a BEC of sodium atoms.
We begin with a brief description of our experimental apparatus and our approach for
creating a Bose-Einstein condensate. We do this essentially for two reasons: our strategy
for achieving BEC is somewhat different than other approaches, and it illustrates an
application of many of the techniques of laser cooling and trapping developed over the
last 15 years.
As of this writing, only a couple of experiments [126] have been attempted using
Bose-Einstein condensates for high-precision metrology; however, these experiments are
not yet competitive with laser-cooled atoms, primarily because of systematic errors due
to atom-atom interactions. Nevertheless, the properties of condensates as extremely
low-temperature samples of atoms and as coherent sources of deBroglie wave for atom
interferometers suggests that metrological applications will be realized. For example, a
BEC, adiabatically expanded to a 1 cm diameter in a microgravity environment, could
reach an “effective” temperature on the order of a picokelvin, with atomic velocities on
the order of micrometers per second. Such a sample could be used in an atomic clock
with a Ramsey time on the order of 1000 seconds, three orders of magnitude longer than
the best earthbound atomic clocks [120, 121]. Similarly, a BEC could be used in a space-
borne atom interferometer where the extremely small spreading of the condensate could
enable long observation times and high precision.
.
7 1. The triaxial TOP trap for sodium. – Our Bose-Einstein condensates of sodium
atoms are produced in a time-averaged orbiting potential or TOP trap [53]. Our TOP
trap differs from those in other BEC experiments in two respects. First, all other exper-
iments with TOP traps that have resulted in BEC use rubidium. We are currently the
only group making condensates of sodium in a TOP trap. The lighter mass of sodium
poses a greater technical challenge in the design of the TOP trap compared to rubidium.
Since the oscillation frequency of an atom in a trap is higher for the lighter atom, the
frequency of the rotating bias field of the TOP trap must be correspondingly higher
so that the atom experiences the time-averaged potential. Typically, sub-Doppler laser
cooling can cool a sample of atoms to an energy which is a few times the recoil energy.
Since the recoil energy is inversely proportional to the mass of the atom, the TOP trap
must also be deeper or stronger to contain the lower-mass, laser-cooled atoms. Second,
the geometry of our TOP trap is different from other TOP traps resulting in a totally
anisotropic or triaxial, time-averaged potential.

A time orbiting potential or TOP trap is a magnetic trap consisting of a quadrupole
magnetic field and a constant magnitude rotating bias field. The potential resulting from
a superposition of these two magnetic fields, averaged over a rotation period, is harmonic
for small displacements. The time-averaged value of the minimum magnetic field in the
TOP trap is just the magnitude of the rotating bias field. That is, the rotating bias
field has effectively “plugged” the zero-field region of the quadrupole field. The zero-field
point of the quadrupole has been displaced by the bias field and, in fact, rotates with
the rotating bias field, producing the so-called “circle of death”.
In the NIST-Gaithersburg TOP trap, the bias field rotates in a plane containing
the symmetry axis of the quadrupole field. For small displacements, the time-averaged
magnetic field is
b2q 2
(18) Bt = B0 + x + 2y 2 + 4z 2 ,
4B0
where B0 is the magnitude of the rotating bias field and 2bq is the gradient of the
quadrupole field along the symmetry (z) axis. The spring constants for this TOP mag-
netic trap are in the ratio of 1 : 2 : 4 in the x, y and z direction, respectively. In the
standard configuration for the TOP trap fields, such as in the original trap of JILA [53],
used to create the first BEC of Rb, the bias field rotates in the symmetry plane of the
quadrupole field, and the spring constant in the radial (x, y) direction is a factor of eight
less than the axial direction (z) producing a disk-shaped time-averaged potential. The
magnetic trap given by eq. (18) is triaxial; that is, unlike the JILA TOP trap, it has no
rotational symmetry. In addition, it is closer to spherical than the JILA TOP trap and
better matched for loading from the nearly spherical clouds of laser-cooled atoms from
the MOT.
Although the TOP trap has a number of desirable properties, including some inde-
pendent adjustments of the spring constants in the three principle directions, there are
certain limitations associated with trapping a mixed state such as the F = 1, mF = −1
state of sodium. The most important is the quadratic Zeeman effect which reduces the
“effective” magnetic moment of that state.
Equation (18) shows that in order to achieve the stiffest TOP trap for a given radius of
the “circle of death”, the distance where the magnitude of the quadrupole field equals the
bias field (a distance typically chosen to be larger than the radius of the sample of trapped
atoms), the largest possible quadrupole field should be used. If the strength of the
quadrupole field is increased, then the magnitude of the rotating field must be increased to
keep the radius of the “circle of death” constant. The energy of the F = 1, mF = −1 state
of sodium as a function of magnetic field initially increases linearly (the linear Zeeman
effect), but the slope decreases with magnetic field due to the quadratic Zeeman effect.
Eventually the energy of this state reaches a maximum around 31.5 mT (315 Gauss)
above which it becomes an anti-trapped state. Thus for a fixed radius of the “circle of
death”, the stiffness of the TOP trap no longer increases linearly with the strength of the
quadrupole field, becoming extremely weak for sufficiently large values of the bias field.
.
7 2. BEC of sodium in a TOP trap. – Similar to other BEC experiments, our Bose-
Einstein condensate of alkali atoms is produced by evaporative cooling in a magnetic
trap loaded with laser-cooled and trapped atoms, however, the details of our technique
differ from other groups. More specifically, we load a dark spot MOT [81] from an
effusive source of sodium at 625 K using Zeeman slowing. The slowing laser beam passes
through the trap so that the capture area of the trap subtends a large solid angle of the
flux of slow atoms. In order to minimize the effect of the slowing laser on the trapped
atoms, we use a hybrid slower geometry [131]. A conventional Zeeman slower is used
to slow atoms from about 800 m/s down to about 160 m/s, followed by a short section
of reverse slower [132] to slow atoms from about 160 m/s down to a few m/s, which is
within the capture velocity of the dark spot MOT. To avoid the loss of atoms in the
slowing process due to optical pumping as the atoms pass through the zero-field region
between the conventional and reverse Zeeman slowing magnets, a second laser frequency
is in the slowing beam to pump atoms from the F = 1 hyperfine level to the F = 2
level where they can continue to be slowed. In addition, a dark spot is placed in the
slowing laser beam, creating a shadow in the region of the cloud of atoms in the dark
spot MOT [133]. This further reduces the effect of the slowing laser beam on the trapped
atoms. Typically, we load more than 1010 atoms into the MOT in less than 0.5 seconds.
After loading into the dark spot MOT, the magnetic fields are rapidly switched off
and the atoms are further cooled to ≈ 200 μK by a brief period of dark molasses (there is
a dark spot in the repumper light in the molasses beams), followed by optical pumping of
the entire population into the F = 1 hyperfine level. The magnetic trap is then rapidly
switched on, trapping atoms in the mF = −1 sublevel of the F = 1 ground state. Since
all three sublevels are present in equal populations at the time when the magnetic trap
is turned on, two-thirds of the sample of laser-cooled atoms are necessarily lost in the
transfer. Experimentally, we find that we are able to trap 4 to 5 ×109 sodium atoms in
the magnetic trap.
The atoms are confined in the benign environment of a magnetic trap in order to
be evaporatively cooled [134, 135]. We have developed two strategies for evaporatively
cooling atoms to Bose-Einstein condensation. The first strategy evolved in response
to the quadratic Zeeman effect problem in the mF = −1 state of sodium, discussed
.
in subsect. 7 1. That is, for the large cloud of atoms initially confined in the TOP
trap, the stiffness of the trap could not be increased, while keeping the “circle of death”
well outside the cloud, to sufficiently compress the sample for runaway evaporation.
Instead, we initially compress and evaporatively cool the sample of atoms trapped in a
quadrupole magnetic field. When the sample is sufficiently cold and dense enough, but
before there is significant loss at the zero-field region due to Majorana transitions, we
transfer them to the TOP trap by rapidly switching on the rotating bias field while the
quadrupole field is on, after which further evaporative cooling can proceed in the TOP
trap. Evaporative cooling in the TOP trap can be achieved by removing the higher-energy
atoms to untrapped states with either rf-induced transitions or the circle of death. For
most experiments we use rf-induced transitions because it allows us greater control and
flexibility. By appropriate choice of frequency and power, we can control the final energy
of the atoms we are removing as well as the width of cut made into the sample of atoms.
In addition, the parameters of the rf can be changed rapidly compared to changing values
of magnetic fields for the circle of death. This technique works for a large range of initial
numbers of trapped atoms.
When the initial number of trapped atoms in the MOT is ≥ 109 , we can achieve
BEC with atoms in the TOP trap, directly. We initially lose a large number of atoms
after transfer into the TOP trap because the circle of death cannot be placed sufficiently
outside the cloud of atoms. After this initial loss and the cloud of atoms has rethermalized
to a lower temperature and higher density, runaway evaporation can be achieved by
compressing the sample and removing high-energy atoms with the circle of death. Both
strategies produce approximately the same number of final condensate atoms, about
3 × 106 , at a BEC transition temperature of 1.2 μK.
We probe our samples of atoms using the technique of absorption imaging [127]. The
TOP trap is rapidly shut off and after a variable delay, a short laser pulse optically
pumps the atoms from F = 1 to F = 2, after which another short laser pulse resonant
with the 3S1/2 , F = 2 → 3P3/2 , F = 3 transition is applied to the atoms along the
direction of gravity (the x-direction). The light absorbed from this laser beam is imaged
onto a CCD camera. From this image we extract the transverse spatial dependence of
the optical depth along the direction of the probe beam.
.
7 3. Diffraction of atoms by a standing wave. – When an atomic beam passes through
a periodic optical potential formed by a standing light wave, it diffracts similar to the
diffraction of light by a periodic grating. The diffraction can be divided into two regimes,
normal and Bragg diffraction. Both diffraction processes can be thought of as arising from
the simultaneous absorption of a photon from one laser beam of the optical standing wave,
and stimulated emission of a photon due to the counterpropagating laser beam. (This
is a similar picture for the origin of the optical dipole force and the momentum transfer
resulting from diffraction can be thought of as arising from this force.) This necessarily
means that the momentum transfer to the atomic beam by the optical standing wave is
quantized in units of 2h̄k, twice the momentum associated with a single photon.
In normal diffraction illustrated in fig. 8a, the incident atomic beam non-adiabatically
enters the light field at normal incidence. As there is no difference in frequency between
the two laser beams comprising the standing wave, the exiting atomic beam is symmetri-
cally diffracted with respect to the incident atomic beam. Energy conservation is satisfied
by the spread in energies associated with the non-adiabatic “turn-on” and “turn-off” of
the standing wave. For short interaction times such that the atoms do not move ap-
preciably along the direction of the standing wave, the standing wave potential can be
considered a thin phase grating that modifies the atomic deBroglie wave with a phase
a) b) c) E |e 〉
atomic beam atomic beam

|g 〉
p = 0 +2!k
− 4!k −2!k p = 0 +2!k + 4!k n!δ
P
0 2n!k
Fig. 8. – a) Normal incidence diffraction. b) Bragg diffraction. c) Bragg diffraction as a 2n-

photon Raman transition.
Fig. 9. – Diffraction of a BEC by a short pulse, optical standing wave. a) For low intensities
only first-order diffraction into ±2h̄k momentum states is visible. b) At higher intensities,
higher-order diffraction (±4h̄k and ±6h̄k) is observed.
modulation, which for a square profile laser beam is given by φ(x) = (U0 τ /h̄) cos2 (kx),
where U0 is the maximum depth of the optical potential given by eq. (2) and τ is the
interaction time of the atomic beam with the standing wave. An atom with zero momen-
tum is therefore split by the standing wave into multiple components with transverse
momenta pn = 2nh̄k, (n = 0, ±1, ±2, . . .), with populations Pn = Jn2 (U0 τ /2h̄), where
Jn (x) are Bessel functions of the first kind. Normal incidence diffraction of atoms by
a near resonant optical standing wave was first demonstrated in Pritchard’s group at
MIT [136] in 1983.
In the early experiments demonstrating diffraction of atoms by optical standing
waves [136, 137], a beam of atoms passed through an optical standing wave and the
diffracted beam was detected downstream. In our experiments [138], we start with BEC
at a temperature sufficiently below the transition temperature such that no discernible
thermal fraction is present. We then adiabatically reduce the strength of the confining
potential, which lowers the energy of the condensate by both reducing the mean-field
interactions and increasing the size of the condensate wave function. We then expose the
atoms to a short pulse of the optical standing wave, while they are either still in the TOP
trap, or shortly after releasing them from the trap by rapidly shutting off the magnetic
fields. Hence the condensates are essentially at rest and we expose them to the optical
standing wave temporally. We detect the momentum transferred to the atoms from the
diffraction process by taking an absorption image after a sufficient time delay, such that
the various atomic wave-packets with different momenta have spatially separated.
Figure 9 is an example of normal diffraction of BEC by a short pulse of a weak and
strong optical standing wave. The optical standing wave is applied along the z-axis,
2 ms after the condensate atoms have been released from the adiabatically expanded
trap. For a weak pulse there is only a small phase modulation imposed on the cloud of
atoms by the optical standing wave and only the first diffraction orders with momentum
±2h̄k are observed (fig. 9a). When the pulse intensity is increased by a factor of 5
there is a substantial phase modulation imposed on the released condensate atoms
and higher diffraction orders are observed. In fig. 9b, both second- and third-order
diffraction, corresponding to momentum transfer of ±4h̄k and ±6h̄k, is clearly evident.
The images in fig. 9 were taken 10 ms after the application of the diffraction pulse.
The momentum spread of the undiffracted atoms is approximately 0.06 h̄k and so the
diffracted components are clearly resolved.
.
7 4. Bragg diffraction of atoms. – When the atoms enter and exit the monochromatic
standing wave adiabatically, energy conservation must be explicitly satisfied in the in-
teraction between the atoms and the light field. In this regime, also known as the Bragg
regime, the energy difference of the atom before and after the change of momentum of 2h̄k
must come from the photon field. This is typically accomplished by having the atomic
beam incident on the standing wave at an angle such that the atoms see a differential
Doppler shift between the two counterpropagating laser beams comprising the standing
wave. This geometry is shown schematically in fig. 8b. Under these conditions, Bragg
diffraction can be understood as a stimulated Raman transition between two momentum
states. Figure 8c shows n-th-order Bragg diffraction as a 2n-photon, stimulated Raman
process in which photons are absorbed from one beam and stimulated into the other.
Conservation of energy and momentum requires (n2h̄k)2 /2M = 2nh̄kv sin θ, where M is
the mass of the atom, v is the longitudinal velocity of the atomic beam and θ is the angle
of incidence of the atomic beam on the standing wave. Bragg diffraction of atoms by a
near resonant optical standing wave was also first demonstrated in Pritchard’s group at
MIT [137] in 1988.
Fig. 10. – Bragg diffraction of a BEC: By applying a moving standing wave (whose velocity is
determined by the frequency difference of the two waves comprising the standing wave) we can
Bragg diffract a portion of the condensate into a well-defined momentum state.
In the case where we start with a BEC essentially at rest, two laser frequencies are
needed to satisfy energy conservation in the Bragg diffraction process. The difference in
frequencies results in a moving standing wave, that would correspond to the differential
Doppler shift observed by the atom in the stationary frame of the standing wave. In
our experiments, we create our moving standing wave by having a frequency difference
δ between the two counterpropagating waves that make up the standing wave. In the
presence of this moving standing wave, an atom initially at rest will simultaneously absorb
photons from the higher-frequency laser beam and be stimulated to emit photons into
lower-frequency beam acquiring momentum ±2nh̄k in the process. In order to satisfy
energy conservation, the detuning δ must be chosen such that nδ = n2 4Erec /h̄, where
Erec is the recoil energy.
Figure 10 shows first-, second- and third-order Bragg diffraction of Bose condensed
atoms released from the magnetic trap. When the frequency difference between the
two lasers is 100 kHz, atoms initially at rest can resonantly absorb a photon from the
higher-frequency laser beam and be stimulated to emit a photon into the lower-frequency
beam. The result of this process is a transfer of two units of photon momentum to the
atoms, which then travels ballistically with a velocity of 6 cm/s. Similarly, by setting
the frequency difference of the lasers to −100 kHz, the momentum transfer to the atoms
from the Raman process will be in the opposite direction. Since Bragg diffraction of
the atoms can be thought of as a two-level system (the initial and final momentum
states) coupled by the Raman process, it is possible to transfer all of the atoms to the
final momentum state. We have observed first-order Bragg diffraction of 100% of the
condensate atoms. The amount of transfer was reduced in the images of fig. 10 so that
the location of the condensate atoms, initially at rest, could be easily identified. Second-
and third-order Bragg diffraction was observed when the laser detuning was increased to
200 kHz and 300 kHz, respectively. We have observed up to sixth-order Bragg diffraction
with a momentum transfer of ≈ 12h̄k (corresponding to a velocity of 0.35 m/s).
.
7 5. Raman output coupling: Demonstration of a CW atom laser . – In order to realize
an atom laser from BEC, it is necessary to coherently extract the condensed atoms; that
is, an atom output coupler is needed. The first demonstration of an output coupler
for BEC was reported in 1997 [139], where coherent, rf-induced transitions were used
to change the internal state of the atoms from a trapped state to an untrapped one.
This method, however, did not allow the direction of the output coupled atoms to be
chosen. The extracted atoms fell under the influence of gravity and expanded because
of the intrinsic repulsion of the atoms. We have developed a highly directional method
to optically couple out a variable fraction of a condensate. We use stimulated Raman
transitions between magnetic sublevels to coherently transfer trapped condensate atoms
to an untrapped state while giving them a momentum kick [140].
The Bragg diffraction of atoms [138] discussed earlier involves a stimulated Raman
transition between different momentum states while keeping the atoms in the same mag-
netic sublevel. If the frequency difference between the lasers includes the additional
Zeeman energy between two magnetic sublevels, a simultaneous change in the momen-
Fig. 11. – Raman output coupler. Left: A stimulated Raman transition is used to transfer 2 h̄k
of momenta, and change the magnetic sublevel from the trapped m = −1 to the untrapped
m = 0 state. Right: Series of images demonstrating multiple Raman output coupling of atoms
from BEC en route to demonstrating a continuous stream of coherent atoms. In a)-c), one,
three and seven 6 μs Raman pulses were applied to the condensate, respectively; d) is the result
of the application of 1 μs Raman pulses at the full repetition rate of ≈ 20 kHz imposed by the
frequency of the rotating TOP bias field (140 pulses in 7 ms). The pictures are absorption images
taken after a time-of-flight period.
tum and internal state of the condensate atoms can be achieved. This is illustrated in
fig. 11, where BEC trapped in the F = 1, mF = −1 state is transferred to the F = 1,
mF = 0 state. Two units of photon momentum are transferred in the Raman process,
so the cloud of atoms in the mF = 0 state has a velocity of 6 cm/s with respect to those
atoms in the mF = −1 state (fig. 11a).
We can repeatedly apply the Raman pulses to achieve multiple output coupling of
atoms from a BEC. This is shown in fig. 11a)-d). In order to avoid changes in the
Raman resonance frequency between different magnetic sublevels we synchronized the
application of the Raman pulses to our rotating TOP field. (Our condensate atoms
were displaced by gravity away from the zero of the quadrupole field, such that the
local magnetic field was modulated by the rotating TOP bias field.) Figures 11a)-c) are
optical absorption images of the condensate after one, three and seven Raman pulses,
respectively. For these images, the TOP trap was held on for a 9 ms window during
which time 6 μs Raman pulses were applied at a subharmonic of the rotating TOP bias
frequency. The magnetic fields were then extinguished and the atoms were imaged 1.6 ms
later. In fig. 11d, the TOP trap was held on for a 7 ms window during which time 140
Raman pulses were applied at the 20 kHz frequency of the rotating bias field and the
distribution of atoms was imaged 1.6 ms later. The Raman pulse duration was reduced
to 1 μs in order to couple less atoms out of the condensate during each Raman pulse. In
the time between two Raman pulses each output coupled wavepacket moves only 2.9 μm.
These pulses strongly overlap because this spatial separation of 2.9 μm is much smaller
than the ≈ 50 μm size of the condensate, therefore the output coupled atoms form a
continuous coherent matter-wave.
Our Raman output coupling scheme dramatically reduces the transverse momentum
width of the extracted atoms compared to other methods such as rf output coupling [139].
This dramatic reduction occurs because the output coupled atoms have received a sub-
stantial momentum kick from the Raman process. If the atoms were simply released
from the trap with no momentum transfer, they would undergo a burst of expansion due
to the repulsive interactions with the other condensate atoms. In our output coupling
scheme, however, this additional expansion energy is primarily channeled into the for-
ward direction. The increase in the transverse momentum width due to the interaction
between the atoms is reduced by roughly the ratio of the timescale over which the mean
field repulsion acts on the freely expanding condensate, divided by the characteristic
time it takes the output coupled atoms to leave the still trapped condensate. In our case,
the reduction ratio is about a factor of ≈ 20 which results in a well-collimated beam of
atoms.
One of the important properties of an optical laser is that the coherence length of the
emitted beam of photons is much longer than the size of the cavity. A similar property
for an atom laser would be highly desirable. The output coupled beam of our atom laser
beam is much longer than the characteristic size of the condensate. Since stimulated
Raman transitions are coherent processes, we expect the coherence length of this beam
to be much longer than the size of the condensate. While we have not measured the
intrinsic coherence length of our atoms laser beam, we have shown, in an interference
experiment [141], that successive pulses are fully coherent. The coherence length of an
atom laser beam, produced by using rf to continuously extract atoms from a magnetically
trapped Bose condensate, was measured first by Hänsch’s group in Germany [142].
8. – Nonlinear atom optics with Bose-Einstein condensates
The advent of the laser as an intense, coherent, light source enabled the field of non-
linear optics to flourish. The interaction of light in materials, whose index of refraction
depends on the intensity, has led to effects such as multi-wave mixing of optical fields to
produce coherent light of a new frequency, and optical solitons, pulses of light that prop-
agate without dispersion. Nonlinear optics now plays an important role in many areas of
science and technology. With the experimental realization of Bose-Einstein condensation

(many atoms in a single quantum state) and the matter-wave or atom “laser” (atoms
coherently extracted from a condensate), we now have an intense source of matter-waves
analogous to the source of light from an optical laser. This has led us to the threshold
of a new field of physics: nonlinear atom optics [143].
The analogy between nonlinear optics with lasers and nonlinear atom optics with
Bose-Einstein condensates can be seen in the similarities between the equations that
govern each system. For a condensate of interacting bosons, in a trapping potential V ,
the macroscopic wave function Ψ satisfies a nonlinear Schrödinger equation [130],

∂Ψ h̄2 2
(19) ih̄ = − ∇ + V + g|Ψ|2 Ψ,
∂t 2M
where M is the atomic mass, g describes the strength of the atom-atom interaction (g > 0
for sodium atoms), and |Ψ|2 is proportional to atomic number density.
.
8 1. Four-wave mixing with matter-waves. – The nonlinear term in eq. (19) is similar
to the third-order susceptibility term, χ(3) , in the wave equation for the electric field
describing optical four-wave mixing. We therefore expect that if three coherent matter-
waves are spatially overlapped with the appropriate momenta, a fourth matter-wave will
be produced due to the nonlinear interaction, analogous to optical four-wave mixing. In
contrast to optical four-wave mixing, the nonlinearity in matter-wave four-wave mixing
comes from atom-atom interactions, described by a mean field; there is no need for an
external nonlinear medium.
Using the atoms from a BEC, we have observed such four-wave mixing of matter-
waves. This work is described in detail in ref. [144]. In our four-wave mixing experiment,
we used optically induced Bragg diffraction [138] to create three overlapping wavepackets
with appropriately chosen momenta. As the three wavepackets spatially separate, a
fourth wavepacket, due to the wave-mixing process, is observed (see fig. 12).
The process of four-wave mixing of matter-waves (and also optical waves), can be
thought of as Bragg diffraction off of a matter grating. In this picture, two of the
matter-waves interfere to form a standing matter-wave grating. The third wave can
Bragg diffract off of this grating, giving rise to the fourth wave. An alternative picture of
four-wave mixing is in terms of stimulated emission. In this picture it is helpful to view
the four-wave mixing process in a reference frame where the process looks like degenerate
Fig. 12. – The process of four-wave mixing of matter-waves can always be transformed to a
reference frame where the mixing process is degenerate (all of the waves have the same energy;
left). The nonlinear term describing the mean-field, s-wave interaction of the atoms is responsible
for the four-wave mixing. Atoms from waves 1 and 3 scatter off each other and go off back-to-
back. The scattering process can be stimulated by wave 2, so that it is more likely that one of
the scattering pairs goes into this wave. By momentum conservation, wave 4 is created. The
small cloud of atoms in the image on the right is the fourth wave generated by four-wave mixing
of matter-waves.
four-wave mixing; that is, all of the waves have the same energy.
In four-wave mixing, both energy and momentum (corresponding to phase matching)
must be conserved. Since atoms, unlike photons, cannot be created out of the vacuum,
we have the additional requirement for matter-waves of particle number conservation.
(If we included the rest mass of the atom, particle number conservation is contained in
energy conservation.) Given these three conditions, one can show that the only four-wave
mixing configurations possible with matter-waves are those that can be viewed in some
frame of reference as degenerate four-wave mixing. This is also the geometry of phase
conjugation. Figure 12 shows the four-wave mixing geometry for matter-waves viewed
in the degenerate or phase conjugation frame.
In the picture of four-wave mixing as arising from stimulated emission, atoms in waves
1 and 3 can be considered as undergoing an elastic collision. The scattering process results
in atoms going off back-to-back in order to conserve momentum, but at some arbitrary
angle with respect to the incident direction. (The scattering process is typically s-wave
and the outgoing waves can be considered spherical.) In the presence of wave 2, however,
this scattering process can be stimulated. There is an enhanced probability that one of
the atoms from the collision of waves 1 and 3 will scatter into wave 2. (This probability
is enhanced by the atoms in wave 2.) Because of momentum conservation, the enhanced
scattering of atoms into wave 2 results in an enhanced number of atoms in wave 4. In
this picture, it is obvious that the four-wave mixing process removes atoms from waves 1
and 3 and puts them into waves 2 and 4. This may have some interesting consequences
in terms of quantum correlations between the waves.
.
8 2. Quantum phase engineering. – A three-dimensional image of an arbitrarily com-
plex object can be constructed by sending light, with sufficient spatial coherence, through
the appropriate phase and/or amplitude mask. This is the basic principle behind phys-
ical optics, which includes wave phenomena like diffraction and holography. Diffraction
can be achieved with a periodic phase and/or amplitude mask, while a more compli-
cated mask is needed to construct a complex holographic image. In each case, the mask
modifies the incoming wave and subsequent propagation produces the desired pattern of
light. This idea can be readily adapted to atom optics, especially when the “incoming”
matter-wave is from a highly coherent source such as a Bose-Einstein condensate.
We have developed a technique to optically imprint complex phase patterns onto a

Bose-Einstein condensate in order to create interesting topological states. This technique
is analogous to sending a wave through a thin phase mask. The basic idea is to expose
the condensate atoms to a short pulse of laser light with a spatially varying intensity
pattern. The laser detuning is chosen such that spontaneous emission is negligible. (The
phase mask can also serve as an amplitude mask by tuning closer to resonance, so that
spontaneous emission is significant.) The pulse duration is sufficiently short such that
the atoms do not move an appreciable distance (i.e. the wavelength of light) during the
pulse. This is sometimes referred to as the Raman-Nath regime. During the laser pulse,
the AC Stark effect or optical dipole potential (see eq. (2)) shifts the energy of the atoms
by U (r, t). Hence the effect on the atomic wave function is to “instantaneously” change
its phase. This effect can be represented

by multiplying the wave function by the phase
factor exp[iφ(r)], where φ(r) = − U (r, t) dt/h̄. Since the AC Stark or light shift is
proportional to the intensity of the light, any spatial intensity variation in the light field
will be written onto the BEC wave function as a spatial variation in its phase.
Optically induced phase imprinting is a tool for “quantum phase engineering” of
the wave function to create a wide variety of states. For example, as discussed in the
section on diffraction of the condensate, the application of a short pulse of standing wave
light will imprint a sinusoidal phase onto the condensate. The imprinted wave function
subsequently evolves in momentum states differing by 2h̄k. It should be possible to use
quantum phase engineering to produce collective states of excitation of the interacting
BEC, such as solitons and vortices. The application of a uniform intensity light field to
half of the BEC imprints a relative phase difference between the two halves. This phase
step is expected to give rise to dark solitons (see following section). Such solitons will
propagate with a speed related to the phase difference [145], which can be adjusted by
the intensity of the laser pulse.
It should also be possible to produce one or more vortices by applying a laser pulse
which has a linearly varying, azimuthal, intensity dependence [146]. This will produce a
topological winding of the BEC phase, which if large enough (i.e. 2 π) should produce a
vortex. Numerical solutions to a 3D Gross-Pitaevskii equation [147] show that this is the
case; and also show that such a vortex, although unstable because it is created in a non-
rotating trap, will live for a sufficient time to be observable. Increasing the phase winding
will generate multiple vortices (vortices with more than h̄ of angular momentum are not
stable and will immediately split into multiple vortices each with angular momentum
h̄). Quantum phase engineering can generate arbitrary phase patterns, and perhaps
other interesting quantum states. In this sense, it is a form of atom holography [148].
The technological challenge is mostly one of imaging. Any complicated pattern must be
imaged to the size of the BEC, typically of order 50 μm.
.
8 3. Solitons in a BEC. – Solitons are stable, localized waves that propagate in a
nonlinear medium without spreading. They may be either bright or dark, depending on
the details of the governing nonlinear wave equation. A bright soliton is a peak in the
amplitude while a dark soliton is a notch with a characteristic phase step across it. Equa-
tion (19), which describes the weakly interacting, zero-temperature BEC, also supports
solitons. The solitons propagate without spreading (dispersing) because the nonlinearity
balances the dispersion; for eq. (19) the corresponding terms are the nonlinear interaction
g|Ψ|2 , and the kinetic energy −(h̄2 /2M )∇2 , respectively. Our sodium condensate only
supports dark solitons because the atom-atom interactions are repulsive [145,149] (g > 0).
A distinguishing characteristic of a dark soliton is that its velocity is less than the
Bogoliubov speed of sound [145, 149] v0 = gn/M (n is the unperturbed condensate
density) and they travel opposite to the direction of the phase gradient. The soliton
speed vs can be expressed either in terms of the phase step δ (0 < δ ≤ π), or the soliton
“depth” nd , which is the difference between n and the density at the bottom of the
notch [145, 149]:

vs δ nd
(20) = cos = 1− .
v0 2 n
For δ = π the soliton has zero velocity, zero density at its center, a width on the order
of the healing length [149], and a discontinuous phase step. As δ decreases the velocity
increases, approaching the speed of sound. The solitons are shallower and wider, with
a more gradual phase step. Because a soliton has a characteristic phase step, optically
imprinting a phase step on the BEC wave function should be a way to create a soliton.
.
8 4. Observation of solitons in a BEC. – We modified the phase distribution of the
BEC by employing the technique of quantum phase engineering discussed in an earlier
section. The condensate atoms were exposed to a pulsed, off-resonant laser beam, coaxial
with the absorption probe beam, with a spatial intensity profile such that only half of the
BEC was illuminated. This was accomplished by blocking half of the laser beam with
a razor blade and imaging this razor blade onto the condensate. The intensity pattern
at the condensate, as observed by our absorption imaging system, had a light to dark
(90% to 0%) transition region of 7 μm. The intensity required to imprint a phase of π
was checked with a Mach-Zehnder atom interferometer based on optically induced Bragg
diffraction [150,151]. Our Bragg interferometer [152] differs from previous ones in that we
phase Bragg beam

imprint
A
x direction
Port 1
B
1 ms 1 ms
Port 2
50 mm
Time
Fig. 13. – Space-time diagram of the matter wave interferometer used to measure the spatial
phase distribution imposed on the BEC. Three optically induced Bragg diffraction pulses formed
the interferometer. Each pulse consisted of two counter-propagating laser beams detuned by
about −2 GHz from atomic resonance (so that spontaneous emission is negligible) with their
frequencies differing by 100 kHz. The first pulse had a duration of 8 μs and coherently split the
condensate into two components |A and |B with equal number of atoms. |A remained at
rest and |B received two photon recoils of momentum. When they were completely separated,
we exposed the top half of |A to a phase imprinting pulse of π, which changed the phase
distribution of |A while |B served as a phase reference. 1 ms after the first Bragg pulse, a
second Bragg pulse of 16 μs duration brought |B to rest and imparted two photon momenta
to |A. When they overlapped again, 1 ms later, a third pulse of 8 μs duration converted their
phase differences into density distributions at ports 1 and 2, which appears in the images.
can independently manipulate atoms in the two arms (because of their large separation)
and can resolve the output ports to reveal the spatial distribution of the condensate
phase. When a phase of π was imprinted on one half of the condensate relative to the
other half, the two output ports of the interferometer displayed the complementary halves
of the condensate (see fig. 13).
To observe the creation and propagation of solitons, we measure BEC density dis-
tributions with absorption imaging after imprinting a phase step. Figure 14 shows the
evolution of the condensate after the top half was phase imprinted with φ = 1.5 π, a
phase for which we observe a single deep soliton (the reason for imprinting a phase step
larger than π is discussed below). Immediately after the phase imprint, there is a steep
phase gradient across the middle of the condensate such that this portion has a large
velocity in the +x-direction. This velocity can be understood as arising from the impulse
imparted by the optical dipole force, and results in a positive density disturbance that
travels at or above the speed of sound. A dark notch is left behind, which is a soliton
moving slowly in the −x-direction (opposite to the direction of the applied force).
A striking feature of the images is the curvature of the soliton. This curvature is
due to the 3D geometry of the trapped condensate, and occurs for two reasons. First,
the speed of sound v0 is largest at the trap center where the density is greatest, and
decreases towards the condensate edge. Second, as the soliton moves into regions of
lower condensate density, we find numerically that the density at its center, n − nd ,
approaches zero, δ approaches π, and vs decreases to zero before reaching the edge. This
is because the soliton depth nd rather than its phase offset δ appears to be a conserved
quantity in a non-uniform medium.
A clear indication that the notches seen in fig. 14 are solitons, rather than simply sound
waves, is their subsonic propagation velocity. To determine this velocity, we measure the
distance after propagation between the notch and the position of the imprinted phase step
along the x-direction. Because the position of our condensate varies randomly from shot-
to-shot (presumably due to stray, time-varying fields) we cannot always apply the phase
step at the center. A marker for the location of the initial phase step is the intersection of
the soliton with the condensate edge, because at this point the soliton has zero velocity.
Using images taken 5 ms after the imprint, at which time the soliton has not traveled
far from the BEC center, we obtain a mean soliton velocity of 1.8(4) mm/s. This speed
is significantly less than the mean Bogoliubov speed of sound v0 = 2.8(1) mm/s. From
the propagation of the notch in the numerical solutions (fig. 14, lower images) we obtain
a mean soliton velocity, vs = 1.6 mm/s, in agreement with the experimental value. The
experimental uncertainty is mainly due to the difficulty in determining the position of
the initial phase step.
From the lower image of fig. 14 at 5 ms, we can extract the theoretical density and
phase profile along the x-axis through the center of the condensate. The dark soliton
notch and its phase step are centered at x = −8 μm. This phase step, δ = 0.58 π, is less
than the imprinted phase of 1.5 π. The difference is caused by the mismatch between the
phase imprint and the phase and depth of the soliton solution of the nonlinear Schrödinger
equation (eq. (19)): Our imprinting resolution of 7 μm is larger than the soliton width,
Fig. 14. – Experimental (upper) and theoretical (lower) images of the integrated BEC density
for various times after we imprint a phase of about 1.5 π on the top half of the condensate with
a 1 μs pulse. The measured number of atoms in the condensate was 1.7(3) × 106 , and this value
was used in the calculations. A positive density disturbance moved rapidly in the +x-direction
and a dark soliton moved oppositely at significantly less than the speed of sound. Because the
imaging is destructive, each image shows a different BEC. The width of the images is 70 μm.
which is on the order of the healing length (0.7 μm), and we do not control the amplitude
of the wave function.
In order to improve our measurement of the soliton velocity, we avoid the uncertainty
in the position of the initial phase step by replacing the razor blade mask with a thin slit.
This produces a stripe of light with a Gaussian profile (1/e2 full width ≈ 15 μm). With
this stripe in the center of the condensate, numerical simulations predict the generation
of solitons that propagate symmetrically outwards. We select experimental images with
solitons symmetrically located about the middle of the condensate, and measure the dis-
tance between them. For a small phase imprint of φ ≈ 0.5 π (at Gaussian maximum),
we observe solitons moving at the Bogoliubov speed of sound, within experimental un-
certainty. For a larger phase imprint of φ ≈ 1.5 π, we observe much slower soliton
propagation, in agreement with numerical simulations. An even larger phase imprint
generates many solitons. The results of these experiments on the creation and propaga-
tion of solitons can be found in ref. [152]. Solitons in a BEC have also been observed by
a group in Hannover [153].
.
8 5. Quantum atom optics. – Imagine taking a BEC of N atoms and splitting it in
half, putting one half in one trap and the other half in another trap. Furthermore,
assume (quite reasonably) that each atom has a 50% probability of ending up in one
trap or the other. What would we expect for the distribution of atoms in the two traps?
We√would find that the mean number of atoms in each trap is N/2, with an uncertainty
of N /2 (it is extremely unlikely to get exactly 500 000 heads if a coin is flipped 106
times.) This uncertainty in the number is intrinsic in physics, and is the source of noise
in many experimental situations, such as the fundamental noise (called shot-noise) on a
laser beam. Classically, this noise is unavoidable. Interactions, however, offer the ability
to change the noise characteristics and create non-classical states. For example, nonlinear
optics can be used to create squeezed states of light, where the noise characteristics in one
degree of freedom can be less than the standard quantum limit given by the Heisenberg
uncertainty principle. Squeezing does not violate the uncertainty principle. Instead, the
noise in a degree of freedom that is unimportant (such as the intensity in the measurement
of a laser frequency) is increased, while the noise is reduced in a relevant one (such as
the phase of the laser). The field concerned with the non-classical statistical properties
of light is called quantum optics. By analogy, interactions in a BEC have enabled the
generation of non-classical statistical states of atoms, or quantum atom-optics.
The interactions between the atoms in a BEC can be exploited to generate number
squeezed states; that is, atoms in traps with reduced number fluctuations. One of the
earliest demonstrations of this was an experiment [154] in the group of Mark Kasevich,
then at Yale. In this experiment, they loaded a BEC of 105 rubidium atoms into a
shallow, 1-D optical lattice and then slowly increased in time the depth of the optical
lattice (it took ≈ 200 ms to reach the final value). Initially, the atoms could tunnel
from well to well, but as the depth of the optical lattice was increased, the tunneling
was suppressed. The atom-atom interactions responsible for the nonlinear term in the
Gross-Pitaevskii equation also means that there is an energy cost when two atoms get
close to one another. Hence having many atoms in one well is energetically unfavorable,
so the tendency is for the number of atoms in each well to be equal (see fig. 15, left).
By raising the lattice intensity slowly enough, the system, starting in the lowest possible
energy state (the BEC), could remain in the lowest energy state (by adiabatic following),
and eventually evolve into an equal partitioning of the atoms into the wells of the lattice.
The reduction in number fluctuations could be inferred from the disappearance of
phase coherence from well to well. That is, as the uncertainty in number decreases,
the uncertainty in phase increases due to the number-phase relationship, which is well
known in quantum optics. Suddenly releasing the atoms from the optical lattice results
.
in diffraction peaks in the atomic momentum (see subsect. 7 3) if there is phase coher-
ence across several wells. The disappearance of the diffraction implies a loss of phase
coherence from well to well, which in this experiment was due to the reduction in num-
ber fluctuations. A factor of ≈ 10 suppression in number fluctuations (from ≈ 30 atoms
per well to about 2 or 3 per well) was observed in the experiment. More recently, an
experiment [155] was performed with a BEC in a 3-D optical lattice in which complete
suppression of number fluctuations was observed. This complete disappearance of num-
ber fluctuations is a result of a quantum phase transition, which takes the system into
the so-called Mott insulator phase where there is the same number of atoms (uniform
filling) in each lattice site. The mechanism for this phase transition is the same as the one
responsible for suppression of number fluctuations in Kasevich experiment. The atoms
initially loaded in to the optical lattice from a BEC are delocalized and can move from
well to well by tunneling. As the depth of the wells of the optical lattice is increased
the movement of the atoms becomes constrained and the interactions between the atoms
favors the filling of lattice sites with uniform numbers of atoms. The 3-D nature of the
optical lattice results in a more abrupt transition from delocalized state to Fock state
Fig. 15. – Left: Schematic illustrating the relevant concepts for creating number (Fock) states
in an optical lattice. Atoms can tunnel, at rate J, to adjacent lattice sites, however, there
is an energy cost of U associated with putting two atoms in the same site. V0 is the height
of the optical lattice potential. Right: Images of the atomic distribution (for a fixed time-of-
flight period) after rapidly extinguishing the optical lattice at a particular depth. An initially
delocalized state (a) exhibits diffraction peaks due to the periodic character of the optical lattice.
As the depth of the optical lattice is increased to suppress tunneling (b), the system undergoes
the transition to Mott insulator such that the atoms are in a Fock state and the diffraction
disappears. If the optical potential is lowered (c) to allow tunneling, then the diffraction peaks
reappear.
(uniform filling) than in the case of Kasevich 1-D optical lattice experiment. Similar
experiments have been performed at NIST [156] (see fig. 15, right).
The reduced number fluctuations of Fock states can be exploited to perform Heisen-
berg limited interferometry [157] in which an N -particle Fock state would have an N -fold
increase in the frequency of phase evolution. To exploit this feature, a coherent superpo-
sition of Fock states (a so-called Schrödinger cat state) needs to be generated. One way
to do this is to send two N -particle Fock states on a beamsplitter, simultaneously. An
early example of such an experiment [158] using photons is the Hong-Ou-Mandel (HOM)
effect. In this experiment, two photons produced by frequency down-conversion are sent
to the two input ports of a beamsplitter, one photon in one input port and the other
photon in the other port. If the twin photons arrive simultaneously on the beamsplitter,
than each of the two output ports of the beamsplitter are 2-particle states. Since only
two photons were sent into the beamsplitter and each output port of the beamsplitter
are 2-particle states, the two particles must be in a superposition of two particles out
one port and two particles out the other port, or a 2-particle cat state. This result has
been confirmed in numerous experiments using photons and is often used to study the
non-classical behavior of number states of photons. More recently, the HOM effect has
been observed with Fock states of atoms in an optical lattice where the unit cell of can
be dynamically transformed between a single and double-well structure [159].
Other measurements of the higher-order correlation functions can reveal the non-
classical nature of a multi-particle field. The landmark experiment of Hanbury Brown and
Twiss [160] is essentially a measurement of the 2-particle correlation function of an optical

field. They observed that a thermal field (in their case, starlight) has a two-particle
correlation function that starts at two and decays to zero over a time (the coherence time)
inversely proportional to the spectral bandwidth. This experiment marked the beginning
of the study the statistical properties of light and can arguably represent the start of the
modern field of quantum optics. A similar measurement of the light emitted by a laser
results in a 2-particle correlation function that starts at one and decays to zero after the
coherence time. The same is true for observing three, four, five, etc. number of photons
within this correlation time. This is due to the non-classical statistical properties of the
state of the laser field, which can be represented by a coherent state. A measurement of
the 2-particle correlation function for atoms extracted from a BEC also shows a similar
behavior as a laser [161, 162]. A comparison of the 3-particle correlation at essentially
zero time (measured by looking at three-body loss rates) for a thermal cloud and BEC
differs by a factor of 6 (or 3!) [163], which is also the expected difference between thermal
or classical source and a laser. Hence it seems even more appropriate to call the BEC
a laser-like source of atoms, or atom laser. There are, however, instances where the
statistical properties of atoms have no optical analog. Atoms come in two varieties:
bosons and fermions, while photons only exist as bosons. A recent experiment [164]
measuring the 2-particle correlation function for ultracold clouds of metastable helium
atoms has shown the Hanbury Brown Twiss effect for bosons (4 He) and the so-called
anti-Hanbury Brown Twiss effect for fermions (3 He), because of Pauli blocking.
∗ ∗ ∗
Much of the experimental work described here was carried out in the Laser Cooling
and Trapping Group in the Atomic Physics Division of the National Institute of Standards
and Technology. It is a pleasure to acknowledge the contributions to this work by the
other members of the NIST staff of the Laser Cooling Group, S. Rolston and P. Lett, as
well as former member C. Westbrook. This work would not have been possible without
the contributions of many postdocs, students, and visitors to the group. While not
attempting to be a review of all the work in this field, these notes have drawn extensively,
both explicitly and implicitly, on the work of many groups throughout the world that
have advanced the art and science of laser cooling. We are indebted to all of those
colleagues, both cited and not, who have contributed so much to our understanding
of laser cooling and to the pleasure of working in this field. We recognize that the
references given in these notes are by no means complete and we apologize to those
authors whose work has not been cited. The BEC experiments at NIST described in
this paper notes were carried out by present and past members of the Laser Cooling
Group, Trey Porto, Steve Rolston, Jenni Sebby-Strabley, Marco Anderlini, Patty Lee,
Ben Brown, Ed Hagley, Lu Deng, Mikio Kozuma, Johannes Denschlag, Jesse Simsarian,
Jesse Wen, Yuri Ovchinnikov, Joerg-Helge Muller, Julien Cubizolles, Robert Lutwak, Rob
Thompson, Aephraim Steinberg, Mike Gatzke and Gerhard Birkl. These experiments
have also benefited greatly from direct interactions with our theoretical colleagues Paul
Julienne, Carl Williams, Eite Tiesinga, Marek Trippenbach, Yehuda Band, Marya Doery,
Charles Clark, David Feder, Mark Edwards, Phil Johnson and Keith Burnett.We also
gratefully acknowledge the financial support not only of NIST, but of the U.S. Office of
Naval Research, NASA and NSF.
REFERENCES
[1] Arimondo E., Phillips W. D. and Strumia F. (Editors), Laser Manipulation of Atoms
and Ions, Proceedings of the International School of Physics “Enrico Fermi”, Varenna,
1991, Course CXVIII (North Holland, Amsterdam) 1992.
[2] Dalibard J., Raimond J.-M. and Zinn-Justin J. (Editors), Fundamental Systems in
Quantum Optics (North Holland, Amsterdam) 1992.
[3] Chu S. and Wieman C. (Editors), Feature issue on laser cooling and trapping of atoms,
J. Opt. Soc. Am. B, 6 (1989) 2020.
[4] Meystre P. and Stenholm S. (Editors), Feature issue on mechanical effects of light, J.
Opt. Soc. Am. B, 2 (1985) 1706.
[5] Metcalf H. and van der Straten P., Phys. Rep., 244 (1994) 203.
[6] Kazantsev A. P., Surdutovich G. I. and Yakovlev V. P., Mechanical Action of
Light on Atoms (World Scientific, Singapore) 1990.
[7] Minogin V. G. and Letokhov V. S., Laser Light Pressure on Atoms (Gordon and
Breach, New York) 1987.
[8] Chu S., Rev. Mod. Phys., 70 (1998) 685.
[9] Cohen-Tannoudji C., Rev. Mod. Phys., 70 (1998) 707.
[10] Phillips W. D., Rev. Mod. Phys., 70 (1998) 721.
[11] Special issue on atom optics and interferometry, Appl. Phys. B, 54 (1992) 321.
[12] Special issue on atom optics and interferometry, J. Phys. (Paris), 4 (1994) 1877.
[13] Special issue on atom optics and interferometry, Quantum Semiclass. Optics, 8 (1996)
495.
[14] Dalilbard J. and Cohen-Tannoudji C., J. Opt. Soc. Am. B, 2 (1985) 1707.
[15] Cohen-Tannoudji C., Dupont-Roc J. and Grynberg G., Atom-photon interactions:
Basic Processes and Applications (Wiley, New York) 1992.
[16] Gordon J. P. and Ashkin A., Phys. Rev. A, 21 (1980) 1606.
[17] Cook R. J., Phys. Rev. A, 22 (1980) 1078.
[18] Stenholm S., Rev. Mod. Phys., 58 (1986) 699.
[19] Cohen-Tannoudji C., in Fundamental Systems in Quantum Optics, edited by J.
Dalibard, J.-M. Raimond and J. Zinn-Justin (North Holland, Amsterdam) 1992, p. 1.

[20] Dalibard J., Thèse de doctorat d’état en Sciences Physique, Université de Paris VI
(1986).
[21] Lett P. D., Phillips W. D., Rolston S. L., Tanner C. E., Watts R. N. and
Westbrook C. I., J. Opt. Soc. Am. B, 6 (1989) 2084.
[22] Castin Y., Wallis H. and Dalibard J., J. Opt. Soc. Am. B, 6 (1989) 2046.
[23] Aspect A., Arimondo E., Kaiser R., Vansteenkiste N. and Cohen-Tannoudji
C., J. Opt. Soc. Am. B, 6 (1989) 2112.
[24] Aspect A., Arimondo E., Kaiser R., Vansteenkiste N. and Cohen-Tannoudji
C., Phys. Rev. Lett., 61 (1988) 826.
[25] Pritchard D. E., Helmerson K., Bagnato V. S., Lafyatis G. P. and Martin
A. G., in Laser Spectroscopy, edited by W. Persson and S. Svanberg, Vol. VIII
(Springer-Verlag, Berlin) 1987, p. 68.
[26] Wallis H. and Ertmer W., J. Opt. Soc. Am. B, 6 (1989) 2211.
[27] Kasevich M. and Chu S., Phys. Rev. Lett., 69 (1992) 1741.
[28] Phillips W. D., Prodan J. and Metcalf H., J. Opt. Soc. Am. B, 2 (1985) 1751.
[29] Chu S., Hollberg L., Bjorkholm J., Cable A. and Ashkin A., Phys. Rev. Lett.,
55 (1985) 48.
[30] Hodapp T. W., Gerz C., Furtlelehner C., Westbrook C. I. and Phillips W. D.,
Appl. Phys. B, 60 (1995) 135.
[31] Phillips W. and Metcalf H., Phys. Rev. Lett., 48 (1982) 596.
[32] Cable A., Prentiss M. and Bigelow N. P., Opt. Lett., 15 (1990) 507.
[33] Monroe C., Swann W., Robinson H. and Wieman C., Phys. Rev. Lett., 65 (1990)
1571.
[34] Andreev S., Balykin V., Letokhov V. and Minogin V., Pis’ma Zh. Eksp. Teor.
Fiz., 34 (1981) 463 (JETP Lett., 34 (1981) 442).
[35] Hoffnagle J., Opt. Lett., 13 (1988) 102.
[36] Ketterle W., Martin A., Joffe M. A. and Pritchard D. E., Phys. Rev. Lett., 69
(1992) 2483.
[37] Gaggl R., Windholz L., Umfer C. and Neureiter C., Phys. Rev. A, 49 (1994) 1119.
[38] Prentiss M. and Cable A., Phys. Rev. Lett., 62 (1989) 1354.
[39] Letokhov V. S., Minogin V. G and Pavlik B. D., Opt. Commun., 19 (1976) 72.
[40] Phillips W. D. and Prodan J. V., in Laser-Cooled and Trapped Atoms, edited by
W. D. Phillips, Natl. Bur. Stand. (U.S.) Spec. Publ., 653 (1983) 137; Prog. Quantum
Electron., 8 (1984) 231; in Coherence and Quantum Optics V, edited by L. Mandel and
E. Wolf (Plenum, New York) 1984, p. 15; Phillips W., Prodan J. and Metcalf H.,
in Laser Spectroscopy Laser VI, edited by H. Weber and W. Luthy (Springer-Verlag,
Berlin) 1983, p. 162.
[41] Ertmer W., Blatt R., Hall J. and Zhu M., Phys. Rev. Lett., 54 (1985) 996.
[42] Salomon C. and Dalibard J., C. R. Acad. Sci. Paris, 306 (1988) 1319.
[43] Prodan J., Phillips W. and Metcalf H., Phys. Rev. Lett., 49 (1982) 1149.
[44] Prodan J., Migdall A., Phillips W. D., So I., Metcalf H. and Dalibard J.,
Phys. Rev. Lett., 54 (1985) 992.
[45] Migdall A., Prodan J., Phillips W., Bergeman T. and Metcalf H., Phys. Rev.
Lett., 54 (1985) 2596.
[46] Chu S., Bjorkholm J., Ashkin A. and Cable A., Phys. Rev. Lett., 57 (1986) 314.
[47] Raab E., Prentiss M., Cable A., Chu S. and Pritchard D., Phys. Rev. Lett., 59
(1987) 2631.
[48] Cornell E., Monroe C. and Wieman C., Phys. Rev. Lett., 67 (1991) 3049.
[49] Spreeuw R. J. C., Gerz C., Goldner L., Phillips W. D., Rolston S. L.,
Westbrook C., Reynolds M. W. and Silvera I. F., Phys. Rev. Lett., 72 (1994) 3162.
[50] Hofferberth S., Lesanovsky I., Fischer B., Verdu J. and Schmiedmayer J.,
Nature Phys., 2 (2006) 710.
[51] Bethlem H. L., Berden G., Crompvoets F. M. H., Jongma R. T., van Roij
A. J. A. and Meijer G., Nature, 406 (2000) 491.
[52] Aminoff C. G., Steane A. M., Bouyer P., Desbiolles P., Dalibard J. and
Cohen-Tannoudji C., Phys. Rev. Lett., 71 (1993) 3083.
[53] Petrich W., Anderson M. H., Ensher J. R. and Cornell E. A., Phys. Rev. Lett.,
74 (1995) 3352.
[54] Ashkin A., Phys. Rev. Lett., 40 (1978) 729.
[55] Ashkin A. and Gordon J., Opt. Lett., 4 (1979) 161.
[56] Dalibard J., Reynaud S. and Cohen-Tannoudji C., Opt. Commun., 47 (1983) 395.
[57] Dalibard J., Reynaud S. and Cohen-Tannoudji C., J. Phys. B, 17 (1984) 4577.
[58] Gould P. L., Lett P. D., Phillips W. D., Julienne P. S., Thorsheim H. R.
and Weiner J., in Advances in Laser Science, edited by A. Tam, J. Gole and W.
Stwalley, Vol. III (American Institute of Physics, New York) 1987, p. 295.
[59] Gould P. L., Lett P. D., Julienne P. S., Phillips W. D., Thorsheim H. R. and
Weiner J., Phys. Rev. Lett., 60 (1988) 788.
[60] Helmerson K., Interdisciplinary Laser Conference (Unpublished, Monterey, CA) 1991.
[61] Rolston S. L., Gerz C., Helmerson K., Jessen P. S., Lett P. D., Phillips W. D.,
Spreeuw R. J. and Westbrook C. I., in 1992 Shanghai International Symposium on
Quantum Optics, edited by Yuzhu Wang, Yigui Wang and Zugeng Wang, Proc.
SPIE, Vol. 1726 (1992), p. 205.
[62] Chu S., Bjorkholm J. E., Ashkin A., Gordon J. P. and Hollberg L. W., Opt.
Lett., 11 (1986) 73.
[63] Miller J. D., Cline R. A. and Heinzen D. J., Phys. Rev. A, 47 (1993) R4567.
[64] Miller J. D., Cline R. A. and Heinzen D. J., Phys. Rev. Lett., 71 (1993) 2204.
[65] Adams C. S., Lee H. J., Davidson N., Kasevich M. and Chu S., Phys. Rev. Lett.,
74 (1995) 3577.
[66] Stamper-Kurn D. M., Andrews M. R., Chikkatur A. P., Inouye S., Miesner
H.-J., Stenger J. and Ketterle W., Phys. Rev. Lett., 80 (1998) 2027.
[67] Cook R. and Hill R., Opt. Commun., 43 (1982) 258.
[68] Davidson N., Lee H. J., Adams C. S., Kasevich M. and Chu S., Phys. Rev. Lett.,
74 (1995) 1311.
[69] Lee H.-J., Adams C. S., Davidson N., Young B., Weitz M., Kasevich M. and
Chu S., in Atomic Physics, edited by D. Wineland, C. Wieman and S. Smith, Vol. 14
(AIP Press, New York) 1995, p. 258.
[70] Phillips W. D., in Laser Manipulation of Atoms and Ions, Proceedings of the
International School of Physics, “Enrico Fermi”, Varenna, 1991, Course CXVIII, edited
by E. Arimondo, W. D. Phillips and F. Strumia (North Holland, Amsterdam) 1992,
p. 289.
[71] Ashkin A. and Gordon J., Opt. Lett., 8 (1983) 511.
[72] Dalibard J., personal communication (1986).
[73] Walker T., Laser Physics, 4 (1994) 965.
[74] Dalibard J. and Cohen-Tannoudji C., J. Opt. Soc. Am. B, 6 (1989) 2023.
[75] Walhout M., Dalibard J., Rolston S. L. and Phillips W. D., J. Opt. Soc. Am. B,
9 (1992) 1997.
[76] Steane A. and Foot C., Europhys. Lett., 14 (1991) 231.
[77] Werner J., Wallis H. and Ertmer W., Opt. Commun., 94 (1992) 525.
[78] Prentiss M., Cable A., Bjorkholm J. E., Chu S., Raab E. L. and Pritchard
D. E., Opt. Lett., 13 (1988) 452.

[79] Sesko D., Walker T., Monroe C., Gallagher A. and Wieman C., Phys. Rev.
Lett., 63 (1989) 961.
[80] Walker T., Sesko D. and Wieman C., Phys. Rev. Lett., 64 (1990) 408.
[81] Ketterle W., Davis K. B., Joffe M. A., Martin A. and Pritchard D. E., Phys.
Rev. Lett., 70 (1993) 2253.
[82] Ramsey N. F., Molecular Beams, in International Series of Monographs on Physics,
edited by R. J. Elliott, J. A. Krumhansl, W. Marshall and D. H. Wilkinson
(Oxford University Press, Oxford) 1956, 1985.
[83] Paul W., personal communication (1985).
[84] Heer C. V., Rev. Sci. Instrum., 34 (1963) 532.
[85] Heer C. V., in Quantum Electronics, edited by C. H. Townes (Columbia University
Press, New York) 1960, p. 17.
[86] Vladimirski V. V., Zh. Eksp. Teor. Fiz, 39 (1960) 1062.

[87] Wing W., Prog. Quantum Electron., 8 (1984) 181.
[88] Ketterle W. and Pritchard D. E., Appl. Phys. B, 54 (1992) 403.
[89] Bergeman T., Erez G. and Metcalf H. J., Phys. Rev. A, 35 (1987) 1535.
[90] Davis K. B., Mewes M.-O., Joffe M. A., Andrews M. R. and Ketterle W., Phys.
Rev. Lett., 74 (1995) 5202.
[91] Petrich W., Anderson M. H., Ensher J. R. and Cornell E., in Abstracts of
Contributed Papers, Vol. ICAP XIV, 1994, p. 1M-7; Davis K. B., Mewes M.-O.,
Joffe M. A. and Ketterle W., ibid., p. 1M-3.
[92] Sesko D., Fan C. G. and Wieman C. E., J. Opt. Soc. Am. B, 5 (1988) 1225.
[93] Phillips W. D., Gould P. L. and Lett P. D., in Laser Spectroscopy, edited by W.
Persson and S. Svanberg, Vol. VIII (Springer, Berlin) 1987, p. 64.
[94] Lett P. D., Watts R. N., Westbrook C. I., Phillips W. D., Gould P. L. and
Metcalf H. J., Phys. Rev. Lett., 61 (1988) 169.
[95] Phillips W. D., Westbrook C. I., Lett P. D., Watts R. N., Gould P. L.
and Metcalf H. J., in Atomic Physics, edited by S. Haroche, J. C. Gay and G.
Grynberg, Vol. 11 (World Scientific, Singapore) 1989, p. 633.
[96] Dalibard J., Salomon C., Aspect A., Arimondo E., Kaiser R., Vansteenkiste
N. and Cohen-Tannoudji C., in Atomic Physics, edited by S. Haroche, J. C. Gay
and G. Grynberg, Vol. 11 (World Scientific, Singapore) 1989, p. 199.
[97] Ungar P. J., Weiss D. S., Riis E. and Chu S., J. Opt. Soc. Am. B, 6 (1989) 2058.
[98] Sheehy B., Shang S.-Q., Watts R., Hatamian S. and Metcalf H., J. Opt. Soc.
Am. B, 6 (1989) 2165.
[99] Weiss D. S., Riis E., Shevy Y., Ungar P. J. and Chu S., J. Opt. Soc. Am. B, 6
(1989) 2072.
[100] Sheehy B., Shang S.-Q., van der Straten P., Hatamian S. and Metcalf H.,
Phys. Rev. Lett., 64 (1990) 858.
[101] Nienhuis G., in Laser Manipulation of Atoms and Ions, Proceedings of the International
School of Physics “Enrico Fermi”, Varenna, 1991, Course CXVIII, edited by E.
Arimondo, W. D. Phillips and F. Strumia (North Holland, Amsterdam) 1992, p. 171.
[102] Emile O., Kaiser R., Gerz C., Wallis H., Aspect A. and Cohen-Tannoudji C.,
J. Phys. II, 3 (1993) 1709.
[103] Aspect A., Emile O., Gerz C., Kaiser R., Vansteenkiste N., Wallis H. and
Cohen-Tannoudji C., in Laser Manipulation of Atoms and Ions, Proceedings of the
International School of Physics “Enrico Fermi”, Varenna, 1991, Course CXVIII, edited
by E. Arimondo, W. D. Phillips and F. Strumia (North Holland, Amsterdam)
1992, p. 401.
[104] Cohen-Tannoudji C. and Phillips W. D., Phys. Today, 43 (1990) 33.

[105] Cohen-Tannoudji C., in Laser Manipulation of Atoms and Ions, Proceedings of the
International School of Physics “Enrico Fermi”, Varenna, 1991, Course CXVIII, edited
by E. Arimondo, W. D. Phillips and F. Strumia (North Holland, Amsterdam) 1992,
p. 99.
[106] Castin Y., Dalibard J. and Cohen-Tannoudji C., in Light Induced Kinetic Effects
on Atoms, Ions and Molecules, edited by L. Moi, S. Gozzini, C. Gabbanini, E.
Arimondo and F. Strumia (ETS Editrice, Pisa) 1991, p. 5.
[107] Castin Y., Doctoral Dissertation, Ecole Normale Supérieure (1992).
[108] Salomon C., Dalibard J., Phillips W. D., Clairon A. and Guellati S., Europhys.
Lett., 12 (1990) 683.
[109] Gerz C., Hodapp T. W., Jessen, Jones K. M., Phillips W. D., Westbrook C. I.
and Mølmer K., Europhys. Lett., 21 (1993) 661.
[110] Rolston S. and Phillips W., Proc. IEEE, 79 (1991) 943.

[111] Clairon A., Salomon C., Guellati S. and Phillips W., Europhys. Lett., 16 (1991)
165.
[112] Kasevich M., Riis E., Chu S. and DeVoe R., Phys. Rev. Lett., 63 (1989) 612.
[113] Bize S., Sortais Y., Abgral M., Zhang S., Calonico D., Mandache C,
Lemonde P., Laurent Ph., Santarellie G., Salomon C., Clairon A., Luiten A.
and Tobar M., in Proc. 6th Symp. Freq. Standards and Metrology edited by Gill P.
(World Scientific) 2002, p. 53.
[114] Weyers S., Hübner U., Schröder R., Tamm Chr. and Bauch A., Metrologia, 38
(2001) 343.
[115] Jefferts S. R., Shirley J. H., Parker T. E., Heavner T. P., Meekhof D.
M., Nelson C. W., Levi F., Costanzo G., DeMarchi A., Drullinger R. E.,
Hollberg L., Lee W. D. and Walls F. L., Metrologia, 39 (2002) 321.
[116] Gibble K. and Chu S., Phys. Rev. Lett., 70 (1993) 1771.
[117] Tiesinga E. and Williams C., private communication.
[118] Kasevich M. and Chu S., Phys. Rev. Lett., 69 (1992) 1741.
[119] Davidson N., Lee H. J., Kasevich M. and Chu S., Phys. Rev. Lett., 72 (1994) 3158.
[120] Lemonde P., Laurent Ph., Simon E., Santarelli G., Clairon A., Salomon C.,
Dimarcq N. and Petit P., Trans. Inst. Meas., 48 (1999) 512; see also
www.estec.esa.nl/spaceflight/map/map/atomiclock.htm.
[121] Ashby N., Bull. Am. Phys. Soc., April 2000, Q21.001; see also
www.jpl.nasa.gov/sespd/space/fphysics.html.
[122] Peters A., Chung K., Young B., Hensley J. and Chu S., Philos. Trans. R. Soc.
London, Ser. A, 355 (1997) 2223.
[123] Lenef A., Hammond T. D., Smith E. T., Chapman M. S., Rubinstein R. A. and
Pritchard D. E., Phys. Rev. Lett., 78 (1997) 760.
[124] Gustavson T. L., Bouyer P. and Kasevich M. A., Phys. Rev. Lett., 78 (1997) 2046.
[125] Cladé T. L., De Mirandes E., Cadoret M., Guellati-Khélifa S., Schwob C.
Nez F., Julien L. and Biraben F., Phys. Rev. Lett., 96 (2006) 033001.
[126] Gupta S., Dieckmann K., Hadzibabic Z. and Pritchard D. E., Phys. Rev. Lett.,
89 (2002) 140401.
[127] Anderson M. H., Ensher J. R., Matthews M. R., Wieman C. E. and Cornell
E. A., Science, 269 (1995) 198.
[128] Davis K. B., Mewes M.-O., Andrews M. R., van Druten N. J., Durfee D. S.,
Kurn D. M. and Ketterle W., Phys. Rev. Lett., 75 (1995) 3969.
[129] Sackett C. A., Stoof H. T. C. and Hulet R. G., Phys. Rev. Lett., 80 (1998) 2031;
Bradley C. C., Sackett C. A., Tollett J. J. and Hulet R. G., Phys. Rev. Lett.,
75 (1995) 1687.
[130] Dalfovo F., Giorgini S., Pitaevskii L. P. and Stringari S., Rev. Mod. Phys., 71
(1999) 463.
[131] Witte A., Kisters T., Riehle F. and Helmcke J., J. Opt. Soc. Am. B, 9 (1992) 1030.
[132] Barrett T., Dapore-Schwartz S., Ray M. and Lafyatis G., Phys. Rev. Lett., 67
(1991) 3483.
[133] Miranda S. G., Muniz S. R., Telles G. D., Marcassa L. G., Helmerson K. and
Bagnato V. S., Phys. Rev. A, 59 (1999) 882.
[134] Hess H. F., Phys. Rev. B, 34 (1986) 3476.
[135] Masuhara N., Doyle J. M., Sandberg J. C., Kleppner D., Greytak T. J., Hess
H. F. and Kochanski G. P., Phys. Rev. Lett., 61 (1988) 935.
[136] Moskowitz P. E., Gould P. L., Atlas S. R. and Pritchard D. E., Phys. Rev. Lett.,
51 (1983) 370; Gould P. L., Ruff G. E. and Pritchard D. E., Phys. Rev. Lett., 56
(1986) 827.
[137] Martin P. J., Oldaker B. G., Miklich A. H. and Pritchard D. E., Phys. Rev.
Lett., 60 (1988) 515.
[138] Kozuma M., Deng L., Hagley E. W., Wen J., Lutwak R., Helmerson K.,
Rolston S. L. and Phillips W. D., Phys. Rev. Lett., 82 (1999) 871.
[139] Mewes M.-O., Andrews M. R., Kurn D. M., Durfee D. S., Townsend C. G. and
Ketterle W., Phys. Rev. Lett., 78 (1997) 582.
[140] Hagley E. W., Deng L., Kozuma M., Wen J., Edwards M. A., Helmerson K.,
Rolston S. L. and Phillips W. D., Science, 283 (1999) 1706.
[141] Hagley E. W., Deng L., Kozuma M., Trippenbach M., Band Y. B., Edwards
M. A., Doery M., Julienne P. S., Helmerson K., Rolston S. L. and Phillips
W. D., Phys. Rev. Lett., 83 (1999) 3112.
[142] Esslinger T., Blochr I. and Hänsch T. W., J. Mod. Opt., 47 (2000) 2725.
[143] Lens G., Meystre P. and Wright E. W., Phys. Rev. Lett., 71 (1993) 3271.
[144] Deng L., Hagley E. W., Wen J., Trippenbach M., Band Y. B., Julienne P. S.,
Simsarian J. E., Helmerson K., Rolston S. L. and Phillips W. D., Nature, 398
(1999) 218.
[145] Reinhardt W. P. and Clark C. W., J. Phys. B, 30 (1997) L785.
[146] Dobrek L., Gajda M., Lewenstein M., Sengstock K., Birkl G. and Ertmer W.,
Phys. Rev. A, 60 (1999) R3381.
[147] Feder D. L., Clark C. W. and Schneider B. I., Phys. Rev. Lett., 82 (1996) 4956.
[148] Fujita J., Morinaga M., Kishimoto T., Yasuda M., Matsui S. and Shimizu F.,
Nature, 380 (1996) 691.
[149] Jackson A. D., Kavoulakis G. M. and Pethick C. J., Phys. Rev. A, 58 (1998) 2417.
[150] Torii Y., Suzuki Y., Kozuma M., Kuga T., Deng L. and Hagley E. W., Phys.
Rev. A, 61 (2000) 041602.
[151] Giltner D. M., McGowan R. W. and Lee S. A., Phys. Rev. Lett., 75 (1995) 2638.
[152] Denschlag J., Simsarian J. E., Feder D. L., Clark C. W., Collins L. A.,
Cubizolles J., Deng L., Hagley E. W., Helmerson K., Reinhardt W. P.,
Rolston S. L., Schneider B. I. and Phillips W. D., Science, 287 (2000) 97.
[153] Burger S., Bongs K., Dettmer S., Ertmer W., Sengstock K., Sanpera A.,
Shlyapnikov G. V. and Lewenstein M., Phys. Rev. Lett., 83 (1999) 5198.
[154] Orzel C., Tuchman A. K., Feneslau M. L., Yasuda M. and Kasevich M. A.,
Science, 291 (2001) 2386.
[155] Greiner M., Mandel O., Esslinger T., Hänsch T. W. and Bloch I., Nature, 415
(2002) 39.
[156] Porto J. V., Rolston S., Laburthe-Tolra B., Williams C. J. and Phillips W.
D., Philos. Trans. R. Soc. London, Ser. A, 361 (2003) 1417.
[157] Wineland D. J., Bollinger J. J., Itano W. M., Moore F. L. and Heinzen D. J.,
Phys. Rev. A, 46 (1992) R6797.
[158] Hong C. K., Ou Z. Y. and Mandel L., Phys. Rev. Lett., 59 (1987) 2044.
[159] Sebby-Strabley J., Brown B. L., Anderlini M., Lee P. J., Phillips W. D.,
Porto J. V. and Johnson P. R., Phys. Rev. Lett., 98 (2007) 200405.
[160] Hanbury Brown R. and Twiss R. Q., Nature, 178 (1956) 1046.
¨
[161] Ottl A., Ritter S., Köhl M. and Esslinger T., Phys. Rev. Lett., 95 (2005) 090404.
[162] Schellekens M., Hoppeler R., Perrin A., Viana Gomes J., Boiron D., Aspect A.
and Westbrook C. I., Science, 310 (2005) 5748.
[163] Burt E. A., Ghrist R. W., Myatt C. J., Holland M. J., Cornell E. A. and
Wieman C. E., Phys. Rev. Lett., 97 (1997) 337.
[164] Jeltes T., McNamara J. M., Hogervorst W., Vassen W., Krachmalnicoff V.,
Schellekens M., Perrin A., Chang H., Boiron D., Aspect A. and Westbrook C.
I., Nature, 445 (2007) 402.
Recent literature
Atomic clocks
Chou C. W., Hume D. B., Koelemeij J. C. J., Wineland D. J. and Rosenband

T., “Frequency Comparison of Two High-Accuracy Al+ Optical Clocks”, Phys. Rev.
Lett., 104 (2010) 070802.
Weyers S., Gerginov V., Nemitz N., Li R. and Gibble K., “Distributed cavity
phase frequency shifts of the caesium fountain PTB-CSF2”, Metrologia, 49 (2012) 82.
Heavner Thomas P., Donley Elizabeth A., Levi Filippo, Costanzo Giovanni,
Parker Thomas E., Shirley Jon H., Ashby Neil, Barlow Stephan and Jef-
ferts S. R., “First accuracy evaluation of NIST-F2”, Metrologia, 51 (2014) 174.
Peil Steven, Hanssen James L., Swanson Thomas B., Taylor Jennifer and Ek-
strom Christopher R., “Evaluation of long term performance of continuously running
atomic fountains”, Metrologia, 51 (2014) 263.
Kessler E. M., Komar P., Bishof M., Jiang L., Sorensen A. S., Ye J. and
Lukin M. D., “Heisenberg-Limited Atom Clocks Based on Entangled Qubits”, Phys.
Rev. Lett., 112 (2014) 190403.
Ludlow A., Boyd M., Ye J., Peik E. and Schmidt P. O., “Optical atomic clocks”,
Rev. Mod. Phys., 87 (2015) 637.
Nicholson T. L., Campbell S. L., Hutson R. B., Marti G. E., Bloom B. J.,
McNally R. L., Zhang W., Barrett M. D., Safronova M. S., Strouse G. F.,
Tew W. L. and Ye J., “Systematic evaluation of an atomic clock at 2 × 10−18 total
uncertainty”, Nature Commun., 6 (2015) 6896.
Rovera G. D., Bize S., Chupin B., Guéna J., Laurent Ph., Rosenbusch P.,
Uhrich P. and Abgrall M., “UTC(OP) based on LNE-SYRTE atomic fountain pri-
mary frequency standards”, Metrologia, 53 (2016) S81.
Szymaniec K., Lea S. N., Gibble K., Park S. E., Liu K. and Glowacki P.,
“NPL Cs fountain frequency standards and the quest for the ultimate accuracy”, J.
Phys.: Conf. Ser., 723 (2016) 012003.
Chen J.-S., Brewer S. M., Chou C. W., Wineland D. J., Leibrandt D. R.

and Hume D. B. “Sympathetic Ground State Cooling and Time-Dilation Shifts in
an 27Al+ Optical Clock”, Phys. Rev. Lett., 118 (2017) 053002.
Atomic interferometry
Bertoldi A., Lamporesi G., Cacciapuoti L., de Angelis M., Fattori M., Pe-
telski T., Peters A., Prevedelli M., Stuhler J. and Tino G. M., “Atom in-
terferometry gravity-gradiometer for the determination of the Newtonian gravitational
constant G”, Eur. Phys. J. D, 40 (2006) 271.
Bouchendira R., Cladé P., Guellati-Khelifa S., Nez F. and Biraben F., “New
Determination of the Fine Structure Constant and Test of the Quantum Electrodynam-
ics”, Phys. Rev. Lett., 106 (2011) 080801.
Zhou Min-Kang, Duan Xiao-Chun, Chen Le-Le, Luo Qin, Xu Yao-Yao and Hu
Zhong-Kun, “Micro-Gal level gravity measurements with cold atom interferometry”,
Chin. Phys. B, 24 (2015) 050401.
Cladé P., Biraben F., Julien L., Nez F. and Guellati-Khelifa S., “Precise
determination of the ratio h/m(u): a way to link microscopic mass to the new kilogram”,
Metrologia, 53 (2016) A75.
Cladé P., “Bloch oscillations in atom interferometry”, Riv. Nuovo Cimento, 38 (2015)
173.
Fang B., Dutta I., Gillot P., Savoie D., Lautier J., Cheng B., Garrido
Alzar C. L., Geiger R., Merlet S., Pereira Dos Santos F. and Landragin
A., “Metrology with Atom Interferometry: Inertial Sensors from Laboratory to Field
Applications”, J. Phys.: Conf. Ser., 723 (2016) 012049.
General reference on metrology with cold atoms
Perrin Hélène, “Ultra cold atoms and Bose-Einstein condensation for quantum metrol-
ogy”, Eur. Phys. J. ST, 172 (2009) 37.
DOI 10.3254/978-1-61499-818-1-449
From the mise-en-pratique for mass to the future

of metrology for the SI(∗ )
Summary. — The third lecture in this series provides a summary of the primary
realization of mass in the new SI and concludes with the relationship between the
new Quantum SI and the field of quantum information science. Before describing
how mass will be realized in the new SI, the lecture briefly reviews the problems
with the current SI which includes mass and the fact that the quantum electrical
standards are part of conventional units and not the SI. The lecture provides a
very brief summary of how the kilogram will be realized using the X-ray crystal
diffraction method and then provide a more detailed mise-en-pratique based on the
NIST-4 watt balance as a primary realization. The lecture also provides an overview
of future technologies for primary realization of small mass and force. The lecture
then proceeds to look at the future of metrology by creating a connection between the
Quantum SI and quantum information science. The new Quantum SI will allow us to
test the Standard Model of physics and build sensors and technology good enough
to provide high resolution gravity gradiometers, improved geodesy, and detectors
that may detect dark matter. In concluding, the lecture returns to the value of
embedded sensors and presents a view of the coming second quantum revolution.
449
1. – Problems with the current SI
The two most troublesome issues with the SI as currently defined involve the artifact
kilogram (kg) and the electrical units.
Defining the kg as exactly equal to the mass of the International Prototype Kilogram
(IPK) poses severe difficulties in scaling, because measurement uncertainties increase by
orders of magnitude the farther the desired scale is from the 1 kg range. Moreover, the
world’s standard masses are known to change over time. In 2013, in preparation for the
SI redefinition, the International Bureau of Weights and Measures (BIPM) conducted
an “extraordinary comparison” of its working standards with the IPK. In the 25 years
since the 3rd periodic verification, the unit of mass as maintained by the International
Bureau of Weights and Measures (BIPM) through its working standards was found to
be 0.035 mg too high relative to the IPK [1].
Only one of the SI base units is electrical — the ampere. It is nearly impossible
to realize under the present definition. Yet the volt and the ohm, which are based on
quantum phenomena (the Josephson effect for the volt; the quantum Hall effect for the
ohm), are so metrologically robust that in 1987 the General Committee on Weights and
Measures (CGPM) established conventional electrical units. The net effect is that since
1990 the SI electrical units have deviated slowly from conventional electrical units.
2. – The redefinition: a quantum SI
The impending redefinition of the SI base units, expected in 2018, will eliminate the
inherent problems with the kilogram and the ampere by basing both units on constants
of nature — the Planck constant (h) and the elementary electric charge (e), respectively.
Two other base units will be redefined — the mole by the Avogadro constant and the
kelvin by the Boltzmann constant), and all unit definitions will be expressed in a uniform
fashion. (See Resolution 1 of the 24th [2011] and 25th [2014] CGPM.) CODATA will
provide updated final values for the constants in 2018.
NIST, along with the national metrology institutes (NMIs) of Canada, Germany, and
Japan, are at the center of this initiative, first proposed by three NIST scientists and
two European colleagues [2]. Independently obtained values for the Planck constant now
agree sufficiently to meet the preconditions for redefinition, and dry runs of the new
mise-en-pratique for mass are underway.
3. – Primary realization of the definition of the kilogram
There are two complementary methods for realizing the definition of the kg, and either
can be used. One employs a device called a watt balance, which equates mechanical
quantity of power to the corresponding electrical quantity when the latter is measured in
terms of quantum electrical effects. This method is now used to determine a value for h.
Alternatively, the kg can be realized using X-ray measurements of crystal density in
a near-perfect sphere of Si-28. This method compares a macroscopic mass to the mass of
From the mise-en-pratique for mass to the future of metrology for the SI 451
a single atom. Values are determined by NA h = [M (e)/m(e)]h = M (e)cα2 /2R∞ , where

M (e) is known to 2 parts in 1011 (1.2 parts in 108 in kg [MKS] — due to the Planck
constant), the fine structure constant, α, is known to 2 parts in 1010 , the Rydberg
constant (R∞ ) to 5 parts in 1012 , and the molar Planck constant (NA ) to 4.5 parts
in 1010 .
At present, worldwide, work is underway to realize the definition of the new quantum
kilogram in the United States, Canada, France, Germany, China, Korea, Japan, Switzer-
land, and New Zealand, as well as at the BIPM. Mexico and South Africa have plans for
table-top watt balances.
At present, dissemination of a primary realization of the kg is relatively straightfor-
ward: National prototypes are compared against the IPK, and those prototypes are used
to calibrate national working standards. After the SI redefinition, the primary realization
obtained by watt balance and the Avogadro method will be somewhat more complex [3].
To disseminate the kg using the Avogadro method, the Si-28 sphere must be carefully
maintained, and oxidation of surface layers must be monitored. Relevant techniques
include spectral ellipsometry, X-ray refractometry, photoelectron spectrometry, X-ray
fluorescence, and infrared absorption. In addition, the volume must be carefully tracked,
with the assumption that the crystal spacing is constant over time and the molar mass
will not change [4]. The sphere mass can be measured in vacuum or in air. In air, a
buoyancy correction is required, but the density and volume are well known.
4. – The mise-en-pratique of the kilogram at NIST

.
4 1. Realization. – At NIST, the primary realization of the kilogram will take place
in vacuum using a new watt or Kibble balance, NIST-4 [5]. A watt or Kibble balance
compares an object’s gravitational force to that of a calculable electromotive force. Watt
balances have conventionally been employed to measure increasingly accurate values of
the Planck constant using a known mass traceably calibrated against the IPK. The
impending redefinition of the kilogram, however, will turn that relationship around: By
fixing the value of the Planck constant, the watt balance can then be used to determine
the unknown mass of an artifact.
The measurement is a two-stage operation, beginning with “weighing mode”. In this
stage, the balance is brought into equilibrium: The artifact’s weight (mg) exactly equals
the electromotive force applied by an electromagnet in a permanent magnetic field, the
magnitude of which is the product of the magnetic field strength (B), the length of wire
(L) in the coil windings of the electromagnet, and the current (I) running through the
wire. Thus mg = BLI.
As a practical matter, however, it is difficult to measure BL directly to adequate
uncertainty. That is why a second stage, called “velocity” or “calibration” mode is
required. Using Faraday’s law of induction, voltage induced in a conductor as it travels
through a magnetic field is exactly proportional to the field strength and the velocity.
So if the velocity is held constant, the induced voltage is a sensitive measure of the field
strength.
In velocity mode, the test mass is removed and the applied current through the coil
is shut off. The coil is then moved using a motor through the surrounding field at a
carefully controlled constant velocity monitored by laser interferometry. The resulting
induced voltage is measured.
Again, an uncomplicated formula governs the magnitude of the induced voltage:
V = νBL, where B and L are the very same field strength and wire length as in weigh-
ing mode, and ν is the velocity. When this equation is combined with the one above
for force on the coil, the problematic B and L cancel out, leaving IV = mgν. That is,
electrical power equals mechanical power, both expressed in watts, hence the name of
the balance(1 ). Or, solving for mass, m = IV /gν.
Everything on the right side of that equation (IV = mgv) can be determined to
extraordinary precision: Current and voltage are obtained by using the von Klitzing
constant (h/e2 , known to 1 part in 109 ) and the Josephson constant (h/2e, known to
about 1 part in 1010 ) along with Ohms law. The local value of g is measured by an
absolute gravimeter. Thus by using a series of substitutions and equivalences along
with the definition of von Klitzing and Josephson constant, mass can be measured with
uncertainties in the range of a few parts in 108 .
The NIST-4 watt balance employs a permanent samarium-cobalt (Sm2 Co17 ) magnet
to produce the field surrounding the movable coil. The 91 kg unit, which can be split, is
10 cm high with a 3 cm gap in the center. The flux density is 0.55 T. The coil, made of
alumina former alloy, has three grooves. The main coil contains 945 turns, and the two
gradient coils (above and below the main coil) contain 135 turns each, for a total wire
length of 1271 m. When the coil is moving at 1 mm/s, the induced voltage is 0.69 V.
As of mid-2016, [6] NIST-4 had measured the Planck constant to a total relative
uncertainty (1-sigma) of 34×10−9 and measured values close to those obtained by another
watt-balance and the Avogadro collaboration. The scientists expect to bring that number
down to 20 × 10−9 by June of 2017.
.
4 2. Vacuum-to-air transfer . – The NIST-4 watt balance realizes the kg in vacuum.
But mass must be disseminated through various ensembles of mass standards and work-
ing standards deployed in air. That situation requires an intermediary process. At
NIST, this can occur by a novel mass comparator or magnetic suspension mass compara-
tor. Here we will describe the former method which is being used by other NMIs with
primary realizations.
A measured reference mass will be removed from the watt balance in vacuum using
a wheeled, evacuated mass transport vehicle (MTV), and loaded into the vacuum mass
comparator that can measure masses in either vacuum or air. The test mass from the
Watt balance will be compared to other masses in vacuum. These other masses along with
(1 ) Note that neither IV nor mgν is actually measured directly in either operational mode.
The watt balance thus makes a virtual comparison between electrical and mechanical power.
This is the reason for the original name watt balance but with the passing of Bryan Kibble who
originated this idea the BIPM decide to rename this the Kibble balance.
their paired sorption artifacts will be affected by water vapor and other contaminants
that are adsorbed by the metal surface, as well as buoyancy effects that depend on the air
density. To characterize these properties, these other mass objects to be measured will
be placed in a vacuum balance that employs surface and volume artifacts to determine
sorption and density values.
This system is used to quantify the amount of mass gained by an artifact that has been
calibrated in vacuum after it is removed to air. The technique measures the difference
in mass between two “sorption” artifacts that have the same nominal mass but very
different surface areas. Since most of the mass change in going from vacuum to air is due
to adsorbed water, more water will desorb from the mass that has more surface area. To
provide a reference, the system compares two artifacts: a stack of masses with air space
between each element of the stack, and a solid cylinder. Knowing the ratio of the surface
areas as well as the mass difference measured from several air-vacuum cycles enables the
calculation of the mass loss of a mass artifact that has the same composition and surface
properties.
When placed in vacuum and weighed, the difference in mass decreases, as more water
desorbs from the stack than from the cylinder. Hydrogen plasma cleaning in situ further
reduces the mass difference to nearly nothing. When the artifacts are removed to air
and re-measured, the mass difference increases due to the adsorption of water. The final
value of the mass difference in air takes about two weeks to stabilize.
.
4 3. Storage of artifacts. – NIST will maintain a pool of Pt-Ir and stainless steel
artifacts that are traceable to the NIST-4 watt balance (and hence to the SI). Some will
be stored in vacuum with a chamber pressure about 10−5 Pa, monitored for pressure and
gas composition. Some will be stored in a constant flow of filtered room air monitored
for temperature, pressure, and humidity. In both cases, artifacts will be inserted and
removed by the MTV, which will lock onto the storage enclosures. Only a small set of
masses and their sorption artifacts will typically move from air to vacuum.
The unit of mass disseminated by NIST will be an “ensemble average” of the mass
values of these artifacts, both Pt-Ir and stainless steel. NIST has eight official prototype
kilograms: K4, K20, K79, K85, K92, K102, K104, and K105.
.
4 4. Dissemination. – An alternative and our preferred dissemination path will use
NIST’s magnetic-suspension mass comparator, assuming it meets the required accuracy.
This device enables direct comparison of a known mass in vacuum obtained from the
watt or Kibble balance to an unknown mass in air using magnetic suspension technology.
The upper (vacuum) chamber contains the reference mass and a sophisticated mass
comparator capable of resolving a difference of 10 micrograms out of a kilogram, or 10
parts in 109 . The comparator has been specially modified at NIST to be able to couple
its weighing pan directly to a reference mass or to a combination solenoid and permanent
magnet that hangs just above the floor of the vacuum chamber.
Directly below the vacuum enclosure, separated by an airtight barrier, is an air cham-
ber containing the mass artifact to be calibrated against the mass that has come from
the Watt balance. The mass is mounted on a suspension carriage at the top of which is a
permanent magnet with one exposed pole. That pole is attracted to the opposite polarity
of the magnet/solenoid combination in the vacuum chamber above, and the attraction
force levitates the mass artifact and carriage when the poles are the right distance apart
(typically in the range of 15 mm to 17 mm).
The system is brought into virtual equilibrium by using feedback sensors to carefully
adjust the current in the solenoid. The magnets in each chamber are encased in shielding
so no stray flux disturbs other parts of the system. Earnshaw’s theorem indicates that it
is not possible to achieve stable stationary equilibrium in such a system. That is, there
are no local minima or maxima of the field potential in free space, only saddle points.
A stable equilibrium of the particle cannot exist and there must be an instability in at
least one direction.
But by using a highly sensitive feedback system, this problem can be minimized. For
each load and distance between poles, there is a specific field strength that will produce
near-motionless levitation. To maintain equilibrium in the balance system, the device
uses a computer-implemented feedback controller to adjust the strength of the solenoid’s
magnetic field. The feedback input comes from a Hall-effect sensor on a cantilever placed
near the exposed pole of the permanent magnet in the air chamber. That device tracks
the magnetic field between the poles, which is a function of the distance between the
poles.
The Hall sensor output is sent to the controller, which uses that signal to vary the
current passing through some 2500 windings in the solenoid. This fine-tuning goes on
20 000 times per second, maintaining the levitated artifact at equilibrium.
A recent, two-stage proof-of-concept demonstration compared two masses. When
compared in a mass calibration laboratory, the artifacts showed a difference of
m1 − m2 = 1.01 ± 0.01 mg. The result with the same masses in the magnetic mass
suspension comparator was m1 − m2 = 0.95 ± 0.13 mg. NIST expects that with an
improved mass comparator that has been installed that we should be able to compare
masses between vacuum and air at around 20 parts in 109 .
5. – Small mass and force realization and metrology at NIST

Under the present definition of the kilogram, scaling down to sub-milligram levels
entails a laborious, time-consuming process of producing ever-smaller traceable mass
artifacts, with a significant increase in uncertainty at each stage. This situation will
improve dramatically when watt balances become the accepted method of realizing mass.
But 1 kg watt balances are expensive, complicated, and require highly trained operating
personnel.
Now, however, it appears possible to construct a tabletop watt balance that can realize
mass at the 1 g level with a relative uncertainty of 10−6 , which is competitive with the
current system. A tabletop watt balance with a cost in the range of USD 50 000 could
be used for calibration on the factory floor or at a calibration lab, and could be used to
realize the unit of mass at smaller NMIs.
NIST is exploring this possibility, and built two prototypes at NIST in early 2016.
The work involves a number of challenges and questions, including:
– AC versus DC metrology. Prototypes explore low-frequency sinusoidal motion

measured using conventional multimeters. Could high frequency designs be coupled
directly to low cost thermal converter chips?
– Traceable velocity measurement. One can build a universal interferometer from

catalog components. But could the entire system be replaced with NIST fiber
frequency comb technologies?
– Robust data acquisition and control. NIST began with NI/Labview, then migrated
to NI/Labview FPGA-based data acquisition with trigger pulse. Could dedicated
FPGA be the solution for truly low-cost implementation?
– Electromagnetic design. Which is preferable: a moving coil or moving magnet?

NIST has built two different optimized coil/magnet systems to explore design trade-
offs.
– 1-D Constrained motion to 1 ppm. Which is preferable: a spring or a balance sus-

pension? The mechanical considerations have proven unexpectedly difficult, and
the design is currently undergoing a revision. A paramount concern is how to
minimize off-axis motion. The extent of that motion depends on the magnet de-
sign. NIST hopes to relax the constraints on the magnet, so we anticipate needing
vertical motions of millimeters with horizontal motions of nanometers!
Recent progress in small mass and force metrology promises to reduce uncertainties
by orders of magnitude while providing SI-traceable measurements in new instrument
designs.
One project, called the Electronic Milligram, uses the newly improved NIST Electro-
static Force Balance (EFB) to establish a link between mass artifacts and SI electrical
units [7]. The EFB generates an electrostatic force between the plates of a concentric
cylinder capacitor, the magnitude of which is Fz = 1/2V 2 dC/dz, where V is the poten-
tial difference across the capacitor, C is the capacitance, and z is the axial displacement
of the inner electrode of the capacitor.

To measure a force, the balance is run in two modes. In the first, the capacitance
of the main capacitor is measured as a function of position using a second, auxiliary
concentric cylinder capacitor. In the second mode, the instrument measures the change
in potential across the main capacitor that is needed to maintain a null displacement as
a force is applied to the balance.
Measurements with the new EFB are capable of 1 to 2 orders of magnitude less
uncertainty (k = 1) than a chain of subdivisions of the kilogram. Differences in results
between initial work downs from EFB in vacuum to forces in air are still unexplained,
and further research is underway. EFB measurements take about 2 days, and a work
down from 100 mg takes at least 2 days, with a full work down requiring about a week.
A second project is underway for ultrasensitive detection of small forces (in the range
of AFM probes) and small masses suited to MEMS and NEMS devices. Resonant sensors
in this range are limited by large systematic uncertainties. NIST has developed an
optomechanical sensor for small mass and force metrology that
– is self-calibrating, enabling in situ calibration in cryo/UHV using an integral Fabry-

Perot interferometer for displacement measurement, and an integral fiber mirror
system to deliver traceable laser power for photon momentum exchange force;
– is ultra-sensitive (Mechanical Quality factor ∼ 106 ) and highly stable;
– has well-characterized dynamical properties; and
– allows accurate force measurements in cryogenic ultrahigh-vacuum environments,
where in situ calibration methods are often difficult (e.g., for atomic scale force
measurements using frequency-modulated atomic force microscopy).
The device combines the interferometer with a superluminescent diode to provide

photon momentum force, and a flexure stage for mass and restoring force — all in a unit
that is a few millimeters wide and is capable of femtonewton resolution [8].
6. – The quantum SI, quantum information, and the Standard Model
The impending change in the SI will define each of the base units in terms of a
defining constant, resulting in a “quantum SI”. It is likely that further modifications will
be desirable in the future. For example, the current definition of the second — based
on the microwave frequency of a cesium-133 transition — may eventually be defined in
terms of the Rydberg constant and a hydrogen transition. In any event, that would not
affect the quantum basis of the SI.
Similarly, a redefinition of the candela (or perhaps the lumen) would only change the
defining constant, not the basic structure of the quantum SI.
But ultimately, there are limits to the quantum SI. It is based on the Standard Model
of Particle Physics, which cannot account for numerous phenomena such as dark matter,
dark energy, and the neutrino mass. Thus, at some level of accuracy the quantum SI
will fail or change with time. Clearly, physics needs to explore the limits of the Standard
Model, and metrology needs to begin anticipating the day in which clocks approach
an accuracy of 10−19 , at which point they can become environmental sensors — with
potentially profound implications for the future of metrology.
As the evolution of the SI continues, it is important to recall that even the greatest
minds can be wrong about fundamental things. One notorious example is the 1935 paper
by Einstein, Podolsky, and Rosen [9] arguing that quantum mechanics could not possibly
be a complete description of physical reality. As Wolfgang Pauli observed at the time,
“Einstein has once again expressed himself publicly on quantum mechanics, indeed in
the 15 May issue of Physical Review. As is well known, every time this happens it is a
catastrophe”.
But even the most forceful original proponents of quantum mechanics might not have
been able to foresee the eventual connection between quantum science and quantum
information. One of the earliest suggestions was made by Richard Feynman in his cele-
brated talk in 1959 titled “Plenty of Room at the Bottom” [10]:
“When we get to the very, very small world—say circuits of seven atoms—we have
a lot of new things that would happen that represent completely new opportunities for
design. Atoms on a small scale behave like nothing on a large scale, for they satisfy
the laws of quantum mechanics. So, as we go down and fiddle around with the atoms
down there, we are working with different laws, and we can expect to do different things.
We can manufacture in different ways. We can use, not just circuits, but some system
involving the quantized energy levels, or the interactions of quantized spins, etc.”
Five decades later, we are routinely doing just that: processing information through
the use of quantum-mechanical systems. This effort, collectively known as quantum
information science (QIS), is the fundamental research, applied research, and technol-
ogy development that are associated with the physical and computational limits that
quantum mechanics places on simple quantum information systems and the emergent
behavior stemming from attempts to control or manipulate complex quantum informa-
tion systems.
QIS is important theoretically because it reveals how nature computes, indicates what
is possible, and helps provide insight into quantum many-body problems. It is important
in practical terms because it uncovers the possibilities for metrology. Many of those
possibilities involve generating and manipulating entangled states.
As M. Nielsen and I. Chuang put it in their book Quantum Computation and Quantum
Information [11]:
“Entanglement is a uniquely quantum mechanical resource that plays a key role in
many of the interesting applications of quantum computation and quantum information;
entanglement is iron to the classical world’s bronze age. In recent years, there has
been a tremendous effort to better understand the properties of entanglement considered
a fundamental resource of Nature, of comparable importance to energy, information,
entropy, or any other fundamental resource”.
In 1994, mathematician Peter Shor [12] postulated that, if a quantum computer (ex-
ploiting both superposition and entanglement) could be built, it could factor an N -digit
number in approximately N 3 steps. It was the first quantum algorithm to tackle an
important problem that is classically computationally hard (2 ) — that is, the number of
operations required grows exponentially with the length of the input string.
Shor’s prime-factoring algorithm raises a significant dilemma in which either 1) the
Extended Church-Turing Thesis is false, or 2) textbook quantum mechanics is false, or
3) there is an efficient classical factoring algorithm. Because the three of these appear to
(2 ) This is a problem that belongs in the complexity class NP (nondeterministic polynomial

time) — hard to factor, but with a solution verifiable in polynomial time. It is not in the sub-class
of NP-complete problems — see for e.g. https://en.wikipedia.org/wiki/NP (complexity).
be the foundation of much of modern technology — specifically for 1) computer science,

2) physics, and 3) mathematics and complexity theory — it appears that this is pure
speculation. However, at least one of them is true! This dilemma was pointed out by
Scott Aaronson an MIT (now U. of Texas, Austin) Professor and is one of the reasons we
should care about quantum computing and quantum information science. For those who
understand the history of computation, it is likely that the answer is 1) and the Church-
Turing thesis will once again be modified to allow for efficient solution of problems on a
“Quantum Turing Machine”.
7. – Future metrology and quantum information science
Recent developments and the current pace of research make certain future develop-
ments extremely likely. First, as noted above, science will find more and different ways
in which to test the Standard Model.
Second, as optical atomic clock uncertainties improve an order of magnitude or more
beyond the current milestone of a few parts in 1018 , and the frequency stability of optical
clocks advances (now about 1 × 10−16 at 5 seconds for NIST’s Sr lattice clock versus
∼ 1×10−16 at 106 seconds for the cesium fountain clock NIST-F2), clocks will increasingly
serve as sensors for a range of applications. These include high-precision geodesy, gravity
gradiometry, gravitational wave detection, and investigation of time variation in the fine-
structure constant.
At NIST, one of our most precise clocks is the Al ion clock that uses quantum logic [13].
This clock has also been used to measure time dilation resulting from both special and
general relativity [14]. This has been observed in objects traveling as slowly as 10 m/s,
and at altitude differences of 33 cm, where the difference in frequency between the higher
and lower clocks differed by only a few parts in 1017 .
Another quantum phenomenon that will have a profound effect on further metrology
is entanglement. Extended coherence times and increasingly sophisticated control and
sensing of entangled states are already making unprecedented measurements possible.
For example, a recent research has shown that applying a periodic series of pulses in a
“stroboscopic” protocol can provide insight into three-body interactions [15]. Alternating
between attractive and repulsive pairwise interactions undoes two-body effects, and leaves
the weak three-body interactions for observation. By using this technique one can beat
the standard quantum shot noise limit that scales as 1/N 1/2 , where N is the number of
atoms in the ensemble. Typically such an approach will move toward the Heisenberg limit
that scales as 1/N but in this stroboscopic approach the measurement uncertainty scales
as 1/N 5/2 . In a one-dimensional optical lattice clock of Sr atoms NIST has explored the
quantum many-body dynamics that show the effects of weak interactions [16]. Although
the effects can be observed with extremely fine resolution — e.g., fractional energy-
level fluctuations as small as 1 part in 1016 can produce clearly discernible spectral
features in the range of 500 mHz — it has been extremely difficult to characterize these
interactions.
Entanglement can also be used as a technique to reduce the amount of quantum

noise in various kinds of measurements. In one recent experiment, a population of rubid-
ium atoms in an optical lattice was entangled via spin-squeezing by exposing it to red
light, which was captured and analyzed. The precision with which phase information
could be extracted from the population was increased by a factor of 10 compared to the
unentangled population [17].
Quantum phenomena can also be exploited to maintain the integrity of optical infor-
mation as it moves from place to place on a photonic chip. Researchers recently developed
a two-dimensional lattice of coupled ring resonators — a design that is prone to fabrica-
tion defects, resonator frequency shifts, high loss, high scatter, and low transmittance.
In this case, the novel resonator pattern induced a “synthetic magnetic field”, with pho-
tons behaving like electrons in a topological insulator: immobile in the bulk material, but
propagating along the edges even in the face of defects [18]. What is extraordinary about
this result is that it is analogous to the Quantum Hall Resistance standard that is the
basis of our electrical resistance and this device has the ability to control the direction
that light travels depending on the frequency of the light.
Quantum control over spin correlations is one potential pathway to quantum infor-
mation processing. For example, a two-dimensional array of hundreds of ions — formed
and controlled in a Penning trap — provides a platform for simulation of quantum mag-
netism with a number of spins (more than ∼ 30) that is intractable on a conventional
computer [19]. A new trap design, which increases spin-spin coupling by a factor of 50 is
now fabricated and operational. Already numerous ions have been effectively entangled
and larger couplings (> 1 kHz) should enable simulation of non-equilibrium processes
that cannot be simulated on a conventional computer.
Finally, increased precision and reduced uncertainty in measuring key quantum phe-
nomena may contribute to longstanding research investigations of possible temporal vari-
ation in the fine structure constant (α). Some astronomical observations have produced
evidence that α may have varied on the order of 1 part in 105 over the past billion
years. At NIST, comparison of cesium, mercury ion, and aluminum ion atomic clocks
over five years indicates fractional variation less than 1.6 × 10−17 per year [20]. The
variation of the fine-structure constant along with measurements of things like the elec-
tron dipole moment have the possibility of providing evidence for physics beyond the
Standard Model.
With the increasing precision of our atomic clocks and the use of entanglement
schemes there are now numerous suggestions of new experiments that exploit these pre-
cision measurements and entanglement to search for dark matter. As mentioned earlier
in this lecture, there are limits to the “New SI” or quantum SI. It is clearly based on the
Standard Model of Particle Physics, which does not account for numerous phenomena
such as dark matter, dark energy, and the neutrino mass. Thus, at some level of accu-
racy the quantum SI will fail. However, the quantum SI will serve its broader community
probably with minor changes over the next 100 years.
8. – Ubiquitous embedded sensors
One of the major benefits of the redefinition of the SI is that sensors based on quantum
effects will all become directly traceable to the SI. These sensors are being miniaturized
at a rapid rate and will become cheaper, more accurate, and more robust. At NIST, a
major goal is to develop SI-traceable measurements and physical standards that are
– Deployable in a factory, lab, device, system, home, or anywhere.
– Usable in (usually) small dimensions, with low power consumption, rugged con-
struction, and ease of integration and operation.
– Flexible, providing a range of SI-traceable measurements and standards (often

quantum-based) relevant to the customer’s needs/applications. One, few, or many
measurements of time, electrical and magnetic properties, temperature, force,
length and more may be possible from a single small-form package.
– Manufacturable, enabled by multiple technologies such as micro-combs and stable

lasers, with the potential for production costs commensurate with the applications:
Low cost for broad deployment, or acceptable cost for high-value applications.
These devices will lead to a variety of embedded standards and sensors. A chip-scale
atomic clock has already been commercialized, and many more are in advanced stages
of development, including: chip-scale atomic magnetometers, ultraminiature gyroscopes,
various sensors based on atomic vapor cells, temperature and pressure sensors (such as
NIST’s photonic thermometer/barometer, the Fixed Length Optical Cavity described in
my previous lecture), and near zero-field NMR and MRI for “remote” chemical analysis
made possible by ultrasensitive magnetometers.
We are now witnessing the second quantum revolution, in which technology will ex-
ploit the weird properties of quantum mechanics and the way nature works at the quan-
tum level. Progress has been rapid. In the 20th Century, only atomic clocks utilized
the strange aspects of quantum mechanics to, for example, provide timing for the Global
Positioning System. Today, chip-scale atomic clocks are available and new devices are
being demonstrated.
Sensitivity will continue to improve, with major impact on devices such as mag-
netometers, accelerometers, and gravimeters. And new technologies such as nitrogen-
vacancy centers may serve as qubits for quantum computers and magnetic imaging sys-
tems unimaginable today.
We are engaged in the great exploration that Richard P. Feynman predicted in his
visionary 1981 address titled “Simulating Physics with Computers” where he stated: “If
you want to make a simulation of Nature, you’d better make it quantum mechanical, and
by golly it’s a wonderful problem, because it doesn’t look so easy” [21].
Clearly the 21st Century will be one of the most exciting ever. The redefinition of the
SI will finally do what the founders of the metric system dreamed and will have created
a set of measurement units based on nature that are realizable in the same way from
region to region or country to country. This transformation is remarkable. Even as this

is being done we know that minor changes to the SI will continue with the redefinition of
the second based upon optical clocks probably occurring in the next decade. Eventually,
the second may be redefined in terms of the Rydberg constant (R∞ ), when it can be
measured to the accuracy of our best clocks. The candela will probably eventually be
redefined in terms of counting photons of a certain frequency and we can see how that
will occur. However, we expect that the Standard Model will finally see its limitations
and at some level of precision new cracks will occur in the foundation of the SI. These
cracks will likely produce new metrologists, new technologies, and will certainly produce
new Nobel Laureates in Physics.
REFERENCES
[1] Stock M., Barat P., Davis R. S., Picard A. and Milton M. J. T., Metrologia, 52
(2015) 310.
[2] Mills I. M., Mohr P. J., Quinn T. J., Taylor B. N. and Williams E. R., Metrologia,
42 (2005) 71; 43 (2006) 227.
[3] Bettin H. and Schlamminger S., Metrologia, 53 (2015) A1.
[4] Fujii K., Bettin H., Becker P. et al., Metrologia, 53 (2015) A19.
[5] Robinson I. A. and Schlamminger S., Metrologia, 53 (2015) A46.
[6] Haddad D., Seifert F., Chao L. S. et al., Rev. Sci. Instrum., 87 (2016) 061301.
[7] Shaw G. A., Stirling J., Kramar J. A. et al., Metrologia, 53 (2015) A86.
[8] Melcher J., Stirling J., Cervantes F. G., Pratt J. R. and Shaw G. A., Appl. Phys.
Lett., 105 (2014) 233109.
[9] Einstein A., Podolsky B. and Rosen N., Phys. Rev., 47 (1935) 777.
[10] Feynman R., Caltech Eng. Science, 23 (1960) 22 or http://www.its.caltech.edu/∼
feynman/plenty.html.
[11] Nielsen M. A. and Chuang I. L., Quantum Computation and Quantum Information
(Cambridge University Press) 2000.
[12] Shor P., SIAM J. Comput., 26 (1997) 1484.
[13] Chou C. W., Hume D. B., Koelemeij J. C. J., Wineland D. J. and Rosenband T.,
Phys. Rev. Lett., 104 (2010) 070802.
[14] Chou C. W., Hume D. B., Rosenband T. and Wineland D. J., Science, 329 (2010)
1630.
[15] Mahmud K. W., Tiesinga E. and Johnson P. R., Phys. Rev. A, 90 (2014) 041602(R).
[16] Martin M. J., Bishof M., Swallows M. D. et al., Science, 341 (2013) 632.
[17] Bohnet J. G., Cox K. C., Norcia M. A., Weiner J. M., Chen Z. and Thompson
J. K., Nat. Photon., 8 (2014) 731.
[18] Mittal S., Fan J., Faez S., Migdall A., Taylor J. M. and Hafezi M., Phys. Rev.
Lett., 113 (2014) 087403.
[19] Bohnet J. G., Sawyer B. C., Britton J. W. et al., Science, 352 (2016) 1297.
[20] Rosenband T., Hume D. B., Schmidt P. O. et al., Science, 319 (2008) 1808.
[21] Feynman R., Int. J. Theor. Phys., 21 (1982) 467.
DOI 10.3254/978-1-61499-818-1-463
Precise time scales and navigation systems

Patrizia Tavella
INRIM - Torino, Italy
Summary. — Precise timekeeping and navigation systems have a lot in com-

mon. Both are based on ultraprecise atomic clocks, on the capacity to measure
and synchronise clocks, on the formation of a common reference time traced to the
international standard time. These similarities are described showing how the time
and frequency metrology has found its major application in the navigation systems,
and how the pressing needs of navigation have promoted the research in the time
metrology fields. The paper mostly focus on the clock and navigation equations, the
necessary algorithms, and the definition of time scales used as reference in navigation
and in timekeeping.
1. – Introduction
On Sept 7th, 1612 Galileo wrote a letter to the King of Spain, proposing the use
of the Jupiter satellites for the identification of the position at sea, a huge problem
for navigation at all times. He wrote something like “These stars display conjunctions,
separations, eclipses, and other precise configurations . . . more than 100 a year . . . and
they are so unique, identifiable, and so exact that nobody, of medium intelligence, is not
able to use them to estimate the longitude and the position of the ship based on the
ephemerids I have computed for the next years to come [1].”
463
464 Patrizia Tavella
In 1612 Galileo was not lucky as the idea seemed too futuristic and nobody took
care, but the basic intuition of Galileo is at the fundamentals of the modern satellite
navigation system, as by observing the position of artificial satellites, and measuring their
distance, a user can estimate his position. In Global Navigation Satellite System (GNSS),
as in the Global Positioning System [2], the position is estimated as the intersection
of 3 or more spheres whose centre is in the known position of a satellite and whose
radius is given by the travel time of an electromagnetic signal leaving the satellite at
a certain known time and received by the ground receiver at a certain measured time
instant, multiplied by the velocity of light. Since the light velocity is a large number, it
is sufficient to make an error of 1 nanosecond in the measurement of the travel time to
have at least 30 cm of error in the estimated position. But 1 nanosecond is the ultimate
capability in time measurements for ground metrological laboratories taking care of all
the calibration and stability issues; in space it is easy to make an error of 100 ns which
gives a positioning error of 30 meters and we immediately understand why time metrology
matters in navigation. This paper explains the fundamental of the navigation equations,
showing how time plays an important role not only as a time interval has to be measured,
but also because many timing information is necessary in the system, and all that requires
an understanding of clock modelling, and the use of appropriate timing algorithms. Being
time a fundamental component, a navigation system can also be a mean to disseminate
precise time, and therefore an overview of the international reference time, the Universal
Coordinated Time is also described. After the GPS born in the United States, other
navigation systems were developed. In particular Europe is devoting a large effort in the
development of a GNSS which, to recognize the initial intuition of the Italian physicist,
is named Galileo.
2. – Time inside a Global Navigation Satellite System
In a GNSS the clocks are necessary on board of the satellites, on ground in the monitor
stations, and in the control centre. All these clocks are to be synchronized to a high level
of accuracy. To reach this aim, good clocks are necessary which means stable, reliable,
and space qualified. Also good space-to-ground measurement systems are necessary. In
addition, the clocks are to be estimated and predicted to be able to set appropriately
each clock on board, therefore algorithms are also necessary.

Most of the GNSS are not only positioning systems, but also timing system, dissem-
inating the reference standard time UTC. To this aim it is also necessary to establish a
connection with UTC community comparing the GNSS clocks with a local realization of
UTC indicated by UTC(k), where k indicates a time laboratory.
When we talk about clocks, synchronization systems, algorithms, reference time
scales, agreement with UTC, we usually talk about the typical problems of time and
frequency metrology. In fact usually the issues are the same but the answers are a bit
different as addressing different needs.
The atomic clocks on board are usually based on the Cesium and Rubidium technol-
ogy, whose basic physics can be found in this volume, and in [3]. The on-board clock
Precise time scales and navigation systems 465
measurement system and the different estimation algorithms can be found in [2], the
details on the Galileo clocks in [4].
Dealing with atomic clocks, general relativity becomes normal work. Moving clocks
run slower and clocks on the mountain run faster due to the different gravitational
field. The gravitational effect causes a shift of 10−13 to the relative frequency offset
for each kilometre of height difference. This makes a clock going faster of about 1
microsecond a year for each kilometer of altitude. On GPS and Galileo satellites at
about 20000 km of height the summation of the two relativistic effects results to about
10−10 , i.e. 1 millisecond offset is gained each year, 40 microseconds in a day. This means
that about 10 km error in positioning would be accumulated in one day if the clock were
not appropriated compensated by these relativistic effects.
We will mostly focus our discussion on one single equation which is the fundamental
one of the navigation system.
Let us suppose a user has a receiver placed in an unknown 3-dimensional position
denoted by xrec . Similarly we write the satellite position by xsat . Let then define tsat as
the epoch at which the signal leaves the satellite, and trec the epoch at which it arrives
at the receiver. The distance P between satellite and receiver could be written as
sat
Prec = c [(trec − tsat )] = |xsat − xrec |.
The satellite position and epoch of transmission are known, as transmitted by the same
satellite in the navigation message, the epoch of reception is measured, therefore the
unknowns are the 3 space coordinate xsat and by measuring 3 satellites the system could
be solved. This is true in principle but what a user measures is actually a pseudo-distance,
or pseudo-range, which contains also other unwanted contributions. A most complete
equation could be
(1) sat
Prec = c [(trec −tsat )] = |xsat −xrec |+c [(trec − tref ) − (tsat − tref )] + Irec + Tr +εrec ,
where we find more terms. First it is to be taken into account that the system and the
receiver clocks are not perfectly synchronized, they may have an offset with is indicated
with respect to a certain common reference time indicated by tref . The receiver clock is
usually not a very precise clock and its offset is actually introduced as the 4th unknown
of the system which asks for a 4th satellite to be solved.
The offset (tsat − tref ) between the satellite and the reference time takes into account
the fact that the clock on board can have an offset, and this offset is to be communicated
to the user to be taken into account in the navigation solution. The terms Irec , Tr, and
εrec take into account additional delays due to the ionosphere, the troposphere, and the
receiver hardware.
The most important effect is the ionospheric delay. The ionosphere is due to the solar
radiation, which ionizes the atoms and molecules in the upper atmosphere, producing
a sea of ions and free electrons around the Earth in a layer that can expand from 50
to 1000 km. The effect on the navigation signal is an additional delay that can cause
a position error from a few meters up to 45 meters, depending on the elevation of the
observed satellite.
From eq. (1), we see some terms related to time which deserve additional attention:
1) The reference time tref is the navigation system Ref erence Time to be defined from
the ensemble of space/ground clocks, similarly to any national reference time scale.
For example:
– GPS time is a paper time scale estimated with a Kalman filter and steered
versus UTC(USNO),
– Galileo System Time is a weighted average of the ground clocks steered versus
UTC,
2) The offset (tsat − tref ) is estimated by a complex algorithm using the same pseudo-
range measures and estimating orbits and clocks (Kalman filter in case of GPS [2],
Batch least square in case of Galileo, . . . ),
3) The real time offset (tsat − tref ) transmitted to the user is actually a prediction
based on previous measures.
Any satellite is in fact transmitting information on its position and time, this information
comes from the ground control segment and is estimated with previous measures. What
is uploaded to the satellite cannot be anything else than a prediction to be transmitted
in the navigation message to the users. The predicted satellite clock offsets (and orbits)
need to be valid for a certain period in the future (GPS needs one day validity, Galileo
plans 100 minutes validity) till the satellite is again in view of the uploading station to
receive fresh estimates.
The definition of an ensemble time, the estimation of the clock offset, and its prediction
call for typical time and frequency algorithms used in the metrological laboratory. See
for example [5] and [6].
3. – Time from a Global Navigation Satellite System

Even if we may not pay attention, our GNSS receiver is also estimating the offset of
the receiver clock with respect to the system reference time as the 4th unknown of the
navigation solution, therefore we can know the offset (tsat − tref ). The reference time of
the GNSS is not a standard time; what we may want to measure, if we are interested
in timing, is the offset of our receiver clock with respect to the international standard
time, i.e. the Universal Coordinated Time, UTC. The GNSS helps us by transmitting
an additional information in the navigation message which is a prediction of the offset of
the system reference time versus UTC. Again what is transmitted is an estimate based
on previous measures, maybe not really referenced to the international UTC but to a
local approximation called UTC(k). The prediction gets old and is affected by a certain
number of errors, let us indicate by UTCSIS the UTC obtainable by the Signals-In-Space
(SIS). The navigation message contains information on the offset (UTCSIS − tref ), from
the navigation solution we get (trec − tref ), therefore we can get
(UTCSIS − trec ) = (UTCSIS − tref ) − (trec − tref )
by a simple subtraction, with all terms taken at the same epoch. Usually the timing user
has a special clock or time scale connected to the receiver and the receiver time trec can
be a national time scale indicated by UTC(k) and it is immediate to estimate the offset
of this time scale versus UTC as disseminated by the navigation system.
In case of GPS the reference time of the system is called GPStime and what is trans-
mitted in the navigation message is a prediction of UTC(USNO) − GPStime, where
UTC(USNO) is the local approximation of UTC as realised by the US Naval Observa-
tory. The Galileo System Time is called GST and in the navigation message a prediction
with respect to the international UTC, i.e. a prediction of UTC-GST, is transmitted.
From this basic measurement many time transfer techniques and processing algorithms
have been developed. See for example [7-11].
4. – The universal coordinated time and the definition of the seconds in the
centuries
We have seen that navigation systems need time and that time can be obtained
from navigation systems. We will now focus on the precise time scales nowadays used
as references and how the definition of the time scale and of its unit, the second, has
evolved in time. This description is partially based on [12].
For centuries timekeeping was based on the observation of the rotating Earth, day
and nights were the reference “ticks” and all the other clocks were set in agreement to
this reference. In 1967 the perspective was completely changed linking the definition of
the second to an atom, the Cesium atom, and with the atomic “swinging” we were able
to observe and measures the irregularities of the Earth rotation.
.
4 1. Universal time. – We have no written documents defining the rotation of the
Earth as measurement standard, but it is know that for centuries the observation of day
and night has been used as time standard. It was known that during the year the time
necessary for one spin of the Earth was changing and to take into account such a variation
the “mean solar day” was considered and that was deemed constant and universal. The
unit of time measurement, the second, was obtained as the faction of 1/86400 of the
rotational day (Mean Solar Time). Then in 1884 the Greenwich meridian was chosen as
the reference one, and in 1925 the International Astronomical Union fixed the beginning
of the mean solar day at h. 00 (no more at noon) and defined the Universal Time (UT)
as obtained from the Mean Solar Day.
Around 1850 some astronomers began suspecting some irregularities in the rotation of
the Earth as the observation of the Polar star presented some anomalies. The Polar star
is in fact a fixed star in the direction of the Earth spinning axis; if the apparent position
Fig. 1. – The movement of the North pole as observed and reported by IERS http://www.iers.
org.
of the Polar star changes, then a possible explanation is a change in the axis of Earth
rotation. This effect is now known as Polar Motion and it is precisely measured by the
network of astronomical and geodetic laboratories in the world. Information on the Polar
Motion and the current status of Earth orientation are found on the International Earth
Rotation and Reference Systems Service (IERS), web site http://www.iers.org. Fig-
ure 1 for example reports the Polar motion in the year 2010, the units are milliarcseconds
(mas). 1 milliarcsecond corresponds to 4.848 × 10−9 radians and an arc of 1mas from the
center of the Earth at distance equal to the polar radius of about 6356 km corresponds to
about 3 cm, therefore the reported movement covers about 6 meters in both directions.
An important fact is that the Polar motion can be measured but is not predictable,
therefore it can be taken into consideration in the definition of the time scales as an
observed effect, but we cannot state any mathematical law able to predict and compensate
the Polar motion forever.
A time scale called UT1 was defined as the Universal Time corrected by Polar motion.
UT1 was then the official name of what was still called Greenwich Mean Time (GMT).
Universal Time is the correct naming, but unbelievably GMT still resists today despite
one century of new naming agreement.
The Polar motion was just the first of a series of irregularities in the Earth rotation
that today are well know and precisely measured. A second effect was discovered around
the year 1936 during the Second World War by A. Scheibe in Berlin, and N. Stoyko in
Fig. 2. – Several irregularities in the Earth rotation as observed by IERS (http://www.iers.

org).
Paris. With the first crystal clocks they were able to measure a variation in the duration of
the day of about 1.2 ms, being the day shorter in the summer of the Northern Hemisphere.
A time scale called UT2 was defined as the Universal Time scale corrected by Polar
motion and seasonal variations.
Similar periodic fluctuations were discovered due to different reasons, again the IERS
offers a variety of information. See for example figs. 2 and 3, reporting the variation in
the duration of the day in milliseconds.
An important effect is known as “Secular slowing down” related to the slow slowing
down of the Earth rotation rate due to the loss of energy for friction. Figure 4 reports
the Length Of Day exceeding 86400 seconds in the last centuries.
Fig. 3. – Duration of the days exceeding 86400 seconds in the year 2010 (http://www.iers.org).
Fig. 4. – Duration of the solar day (only the excess from 86400 seconds) [13].
All this evidences gave the perception that the Earth rotation was not a good standard
of time and metrologists started to discuss a new definition.
.
4 2. Ephemeris time. – What happened in the years 1956-1960 is not really trans-
parent. A recent interesting book by T. Quinn is going back to the original documents
trying to reconstruct the story [14].
In 1955 the first atomic clock was ready in the National Physical Laboratory in UK
and physicist were convinced of the unprecedented stability of such a device, the definition
of the time standard could have been based on the Cesium atom.
Astronomers were not happy to lose the time measurement supremacy. The discussion
was intense and the International Committee for Weights and Measures forced to a deci-
sion. The leading idea was that if not the “rotation”, surely the “revolution” of the Earth
around the Sun was constant; measuring the longitude of the Sun and using the equation
of the apparent Sun orbit, a more stable and uniform time scale was to be obtained.
This new time scale was called Ephemeris Time, with a start fixed on 0 hour UT of
January 0, 1900.
The time unit was defined as the Ephemeris Second corresponding to the fraction
1/31 556 925.9747 of the tropical year on day January 0, 1900.
The definition of the Ephemeris Time was never “universally” accepted, the realization
of the second from this definition requires years of celestial observations and cumbersome
computations. The Ephemeris Second definition lasted just 7 years that in the world of
international metrology is really nothing. The new standard was just defined and people
started debating and arguing on a new more universal and practical definition. It was
just in 1967 that the definition changed again.
But from the definition of Ephemeris second one can notice a very important issue:
the Ephemeris second is not just a certain fraction of the tropical year, it is that fraction
of that duration of the year as estimated on day January 0, 1900. The duration was
actually a mean value estimated over the observations of about one century centred on
the year 1900. As any new definition of a measurement unit has to be in agreement
with the previous one, the definition of the Ephemeris second was in agreement with the
duration of the rotational second in 1900.
But due to the secular slowing down, the definition of the Ephemeris second agreed
in 1960, referring to the rotational second of the year 1900, was already shorter than the
fraction 1/86400 of the Mean Solar Day in 1960. This will have an important impact
also on the subsequent definition of the atomic second.
.
4 3. Atomic time. – It was at the Thirteenth General Conference for Weights and
Measures, in which the new definition of the second was adopted, based on the Cesium
atom in 1967.
The Atomic Second was defined as 9 192 631 770 periods of the radiation correspond-
ing to the transition between the two hyperfine levels of the ground state of the Cs 133
atom.
After the second, the time scale was defined as the International Atomic Time (de-
fined in 1971) as a time scale accumulating atomic seconds starting from a conventional
“beginning” set on 0 hours UT of January 1st, 1958.
As we have seen any new definition keeps coherency with the previous one, therefore
the duration of the atomic second was defined in agreement with the duration of the
Ephemeris second, which was in agreement with the Universal second of the year 1900,
therefore shorter than 1/86400 of the Mean Solar Day in 1967.
Hence, by definition, the atomic second is shorter than the current Rotational Second
of the Universal Time. This means that 86400 atomic seconds cover a duration which
is a bit shorter than the rotation of the Earth today; atomic clocks reach faster noon,
when the Earth is not yet at “noon” position.
This is an important heritage of the definition of the Ephemeris second and the choice
to be the duration of the year 1900. Due to that and then to the addition of the Earth
slowing down, today we have to face a discrepancy between the time given by atomic
clocks and the time obtained by the observation of the Earth rotation. This discrepancy
led to the definition of a trade-off time scale called Coordinated Universal Time (UTC).
The definition of atomic time was then imbedded in the frame of the theory of general
relativity thanks to the work of the astronomers [15].
.
4 4. UTC: the trade-off . – At the epoch of TAI definition, still some users needed to
know the angular position of the Earth mostly for navigation with traditional means. It
was thus necessary to transmit the relationship between the Universal Time UT1 (rota-
tional) and the Atomic Time. This discussion took place at the International Telecom-
munication Union (ITU) in charge of coordinating the radio broadcasted signal and it
was decided to recommend as international reference time scale a compromise between
TAI and UT. The new time scale was named Coordinated Universal Time (UTC). At
the beginning UTC had some time and frequency steps with respect to TAI to maintain
the agreement with UT, but in 1970 it was decided [16] to define UTC as a time scale
beating with atomic second (constant duration of the time unit), but time tagging in
agreement with UT1. Therefore UTC is a discontinuous time scale with leap seconds
added or removed such that the agreement between UTC and UT1 is maintained at the
level of 1 second as:
TAI − UTC = n seconds, n = 0, ±1, ±2, . . . ,

|UT1 − UTC| < 0.9 s.
The evolution of the difference between UTC and TAI is reported in fig. 5.
The realization and dissemination of the international standard UTC is task of the
Bureau International des Poids et Measures (BIPM) located in Sèvres, close to Paris,
and it is obtained as an ensemble time scale with all the atomic clocks maintained in
the different national laboratories ensuring accuracy by appropriate calibration versus
primary frequency standards. Details may be found in [17] and the references quoted
there.
Today the UTC is the basis for civil, legal, and scientific uses.
The international UTC is available in deferred time, every month, to allow the BIPM
to collect and check all the clock data; local real-time approximations are generated by
national metrological lab and are named UTC(k), where k is the name of the laboratory
or of the country. Recently BIPM started the computation of Rapid UTC available every
week (http://www.bipm.org/en/scientific/tfg/rapid-utc.html).
Fig. 5. – The Universal Coordinated Time and the leap seconds.
.
4 5. The future of leap second . – Despite the large acceptance and universality of
UTC, recommended by many international standardisation bodies, the practice of leap
seconds has caused some perplexity in the last decade. Several international organisations
created working groups to evaluate the need of leap seconds today. Every year, at the
ITU meetings, the issue is debated and support was asked to the Consultative Committee
for Time and Frequency. In 2015 ITU General Assembly postponed the decision to next
Assembly in 2022 promoting the work of different international organizations to better
define time scales and the role of each organization.
The issue is related to the difficulties of using a discontinuous time scale with irregular
insertions of leap second (whenever necessary) in case of automated systems. Telecom-
munication organizations, navigation systems, computer networks, for example, in some
cases decided to use a surrogate of UTC for their datation purposes to avoid adding leap
seconds. The most important issues came from navigation system, first of all the Global
Positioning System (GPS), whose internal time scale was set in agreement with UTC at
the beginning of the GPS days, but then no leap second was added and the GPS time
is nowadays ahead of 18 seconds versus UTC. This would not be an issue as GPS time
is just an internal time scale used by the GPS system to synchronize all the clocks, but
the widely spread use of GPS receivers and the different options usually available to the
users displaying UTC or also GPS time as timing information, originated the confusion.
Moreover, other navigation systems have differently chosen the origin of their time scale
and have now an offset with respect to UTC of a different number of seconds [18]. In-
stead of spreading the use of a uniquely recognized international reference, the navigation
systems are causing a proliferation of time scales.
It appears urgent to take a decision to avoid the worldwide diffusion of several, dif-
ferent, reference time scales easily accessible through GNSS and in disagreement by
seconds [18-20].
5. – Will time scales return in space?
When in 1967 the definition of the time unit was related to a Cesium atom and
the second had to be realized by building atomic machines in physical laboratories, the
astronomers were somehow not happy. Keeping time by the observation of celestial
dynamical systems is still a fascinating mission. In the ’80s enthusiasm was raised by the
discovery of rotating stars emitting radiowaves like a lighthouse: they could be used as
clocks! The pulsars were discovered years before by Susan Jocelyn Bell [21] during her
PhD when dealing with other subjects.
Some types of pulsars, called millisecond pulsars, have a rotation period of the order
of milliseconds and this rotation period is quite stable, therefore they seemed the good
candidate for a new astronomical timekeeping.
Millisecond pulsars are neutron stars with about 20 km as diameter and 1.4 time
the solar mass. They are located in our Galaxy but thousand light years apart; their
rotation is highly stable and every millisecond we see a radio pulse. We have to admit
that the rotation period is slowing down and this deceleration is to be measured with
respect to some other clocks; the pulses have to cross 1016 km of interstellar region, so
the observation can be noisy. Moreover, the Earth is a rotating observatory, therefore
basing timekeeping on pulsar observation may not really be trouble-free, but pulsars
allowed an indirect prove of gravitation waves [22], and are able to give some insights on
the long term instabilities of atomic clocks [23]. Nowadays after decades of observation
of millisecond pulsars their use in timing is becoming quite fascinating [24] and the
construction of the Square Kilometre Array will allow much more pulsar discovery.
The idea that each pulsar has its own rotation period, and therefore its own individual
signal, stimulates people interested in extraterrestrial space navigation. In fact we can
imagine that far from the Earth, far from the GPS, a space probe could estimate its
position by comparing arrival times of different pulsar signals with respect to expected
pulse arrival times using information from a pulsar database. By the observation of the
configuration of an ensemble of pulsars, a user could estimate its position, going back
to the initial idea of Galileo: “These stars display conjunctions, separations, eclipses,
and other precise configurations . . . that nobody, of medium intelligence, is not able to
use them to estimate . . . the position”. The European Space Agency supported such a
study [25].
∗ ∗ ∗
This lecture was kindly supported by the IEEE UFFC Distinguished Lecturer Award
(2015-2016) https://ieee-uffc.org/reports-from-distinguished-lecturers/.
REFERENCES
[1] Le opere di Galileo Galilei, Vol. V (Nazionale, Firenze) 1895.

[2] Parkinson B. W. and Spilker J. J. Jr., “Global Positioning System: Theory and
applications” (American Institute of Aeronautics and Astronautics, Washington) 1996.
[3] Levi F., “Microwave frequency standards”, in Proceedings of the International School of
Physics “Enrico Fermi” Course 185 “Metrology and Physical Constants”, edited by Bava
E., Kuehne M. and Rossi A. (IOS Press Amsterdam, SIF Bologna) 2013, pp. 471-494.
[4] Waller P., Gonzalez F., Binda S., Sesia I., Hidalgo I., Tobias G. and Tavella
P., “The In-orbit Performances of GIOVE Clock”, IEEE Trans. Ultrason. Ferroelect.
Frequency Control, 57 (2010) 738.
[5] Tavella P., this volume, p. 477.
[6] Tavella P., “Statistical and mathematical tools for atomic clocks”, Metrologia, 45 (2008)
S183.
[7] Lewandowski W., Azoubib J. and Klepczynski W. J., “GPS: primary tool for time
transfer”, Proc. IEEE Vol. 87, Issue: 1, Jan. 1999.
[8] Larson K. M. and Levine J., “Carrier-phase time transfer”, IEEE Trans. Ultrason.
Ferroelec. Frequency Control, 46 (1999) 1001.
[9] Defraigne P., “Precise GNSS time and frequency transfer”, in Proceedings IEEE on
International Frequency Control Symposium and European Frequency and Time Forum,
Besancon, France (2009).
[10] Cerretto G., Tavella P., Perucca A., Mozo A., Piriz R. and Romay M., “Network
Time and Frequency Transfer with GNSS Receivers located in Time Laboratories”, IEEE
Trans. Ultrason. Ferroelect. Frequency Control, 57 (2010).
[11] Cerretto G., Tavella P., Lahaye F., Mireault Y. and Rovera D., “Nearly Real-
Time Comparison and Monitoring of Time Scales With Precise Point Positioning Using
NRCan Ultra-Rapid Products”, IEEE Trans. Ultrason. Ferroelect. Frequency Control, 59
(2012) 545.
[12] Tavella P., “Time scales”, Proceedings of the International School of Physics “Enrico
Fermi” Course 185 “Metrology and Physical Constants”, edited by Bava E., Kuehne M.
and Rossi A. (IOS Press Amsterdam, SIF Bologna) 2013, pp. 495-506.
[13] Mannucci R. and Cordara F., “Misurare il tempo e la frequenza” (il Rostro) 1997.
[14] Quinn T., “From artefacts to atoms” (Oxford University Press) 2011.
[15] International Astronomical Union, Resolutions from the XXIVth General Assembly
Manchester, UK, 2000 http://www.iau.org/static/resolutions/IAU2000 French.pdf.
[16] Rec. ITU-R TF.460-6 1, RECOMMENDATION ITU-R TF.460-6, “Standard-frequency
and time-signal emissions”, http://www.itu.int/dms pubrec/itu-r/rec/tf/R-REC-TF.
460-6-200202-I!!PDF-E.pdf.
[17] BIPM Time Section web page at http://www.bipm.org/en/scientific/tai/.
[18] Arias E. F. and Lewandowski W. (Editors), “Modern Applications of Timescales”,
Metrologia, Vol. 48, n. 4, August 2011.
[19] ITU/BIPM Workshop, “Future of the International Time Scale” Geneva, 19-20 September
2013 http://www.itu.int/ITU-R/index.asp?category=conferences&rlink=itu-bipm-
workshop-13&lang=en, http://www.itu.int/net/pressoffice/press releases/2012/
03.aspx.
[20] Bureau International des Poinds et Mesures press releases, 2011, http://www.bipm.org/
utils/en/pdf/Press Release UTC 13October.pdf, http://www.bipm.org/utils/en/
pdf/Press Release UTC 21st century.pdf.
[21] Bell Susan Jocelyn, The Measurement of radio source diameters using a diffraction
method, University of Cambridge (1968) http://ethos.bl.uk/OrderDetails.do?uin=uk.
bl.ethos.449485.
[22] Hulse R. A. and Taylor J. H., “Discovery of a pulsar in a binary system”, Astrophys.
J., 195 (1975) pt. 2, p. L51.
[23] Petit G. and Tavella P., “Pulsar and Time Scales”, Astron. Astrophys., 308 (1996)
290.
[24] Hobbs G. et al., “Development of a pulsar-based time-scale”, Mon. Not. R. Astron. Soc.,
427 (2012) 2780 DOI: 10.1111/j.1365-2966.2012.21946.x.
[25] Josep Sala et al., https://www.esa.int/gsp/ACT/doc/EVENTS/
InnovativeSystemWorkshop2/ACT-ART-Bridge2Space-Sala.pdf.
DOI 10.3254/978-1-61499-818-1-477
Statistical tools for time varying quantities

Patrizia Tavella
INRIM - Torino, Italy
Summary. — The analysis of time series of measures and the important instabil-
ities that they may reveal is presented, describing some of the techniques that can
find useful application in the different fields of metrology.
1. – Introduction
Repeating measurements in different epochs always reveals something. It can be a

slow linear drift as in the case of the International Prototype of the Kilogram [1], or
the presence of an unexpected gravitational wave from a binary black hole merger [2].
Viceversa, it can be useful to demonstrate that the characteristics of a standard are
stationary and not degrading with time. What we can often observe is the following:
– nonstationarity
– aging
– slow drift
– abrupt change
– failure.
477
y(t)
y3
y2
y1
t1 t1 + t +3
t
Fig. 1. – The evolution in time of the measurand y(t).
Changes may be what we are looking for, or may lead to unpleasant discovery, or they
may reveal new physics. For sure we need to evaluate how physical quantities evolve in
time when we need
– to estimate the behaviour of a physical quantity and its possible impact on complex
systems
– to predict that behaviour
– to control that behaviour.
This is particularly useful in many fields of science and engineering, any time we need, for
example, to assess noise, identify outliers, simulate the system behaviour, carry out long-
term analysis, filter noise, evaluate the uncertainty propagation, identify drift, validate
performances, reconstruct missing data, fix calibration intervals, etc.
In metrology we are used to evaluate the uncertainty of each measure, and we have
international standards as guidance [3], but there are cases in which the instability in
time is more impacting then the uncertainty on the single measure. Figure 1 represents
an example showing a certain physical quantity y(t) varying in time according to a
certain behaviour. The measurement process is able to precisely estimate the mean
value ȳi over an observation interval τ , but if the same measurement was repeated in
another observation period τ , the measure would not have the same value, due to the
natural variation of the measurand. The typical example is the frequency delivered by
a frequency standard. In this paper many examples will be taken from the Time and
Frequency metrology world, but this is just an example, other cases apply.
When we deal with such physical quantities and we want to produce and sell a ref-
erence measurement standard or measuring device, or we want to use the standard as
primary reference for calibrating other devices, or carry out precise measure, or use the
standard or the measurement device in a complex system, as space, health, environment
applications, we need to find a suitable tool to estimate instability and time varying
behaviours.
Statistical tools for time varying quantities 479
2. – Estimating time varying behaviour
The typical analysis with time varying quantities [4] is based on time series of measures
in which we usually identify:
1) Deterministic components like polynomials, exponentials, or other slopes that can
be processed by least square estimation techniques (batch, recursive, etc.)
2) Stochastic components may also be present and sometime dominant and in this
case we need to use appropriate mathematical tools.
We will mostly concentrate on this second stochastic case, considering that the deter-
ministic component can easily be estimated and, in case, subtracted. We consider thus
a physical quantity y(t) that at each instant is a random variable and the time evolution
of a random variable is often referred to as noise. The estimation of noise can be carried
out in the frequency domain estimating the power spectrum, which means decomposing
the signal in a summation of elementary sinusoids with different power, or we can process
the time series of measures by time domain techniques which evaluate the dispersion of
the data. The typical measure of dispersion is the variance written as
N
i=1 (ȳi − μ)2
σ 2 (τ ) = ,
N
where we have N measures representing the mean values ȳi of the physical quantity y(t)
measured over an observation interval τ . The average value of the N measures is indicated
by μ. If the “noise” is stationary, the estimation of the variance converges as N grows.
We may have cases in which for longer observation interval τ , the variability di-
minishes. In this case usually the variance decreases proportionally to the observation
interval τ and we say the noise is “white”:
1
σ 2 (τ ) ∝ .
τ
Figure 2 represents an example of this behaviour.
There are other cases in which the noise is not strictly stationary, for example it can
be a random walk, as illustrated in fig. 3, and the estimation of the variance does not
converge and depends on N , the number of samples (for any observation interval τ ).
In this case we can use filtered variances whose estimation converges and that can
appropriately represent the noise. In Time and Frequency metrology a “two sample
variance”, also called “Allan Variance”, plays an important role, which comes for an idea
of Allan and Barnes [5-7]. This variance is based only on N = 2 samples, and this means
considering the variability of the increments of the measurand y(t)
1 !
2
σy2 (τ ) = (ȳt+τ − ȳt ) .
2
Each value ȳt is averaged over an observation interval τ . What is very useful of the
Allan variance is that not only it gives an indication of the instability over a certain
Fig. 2. – Evolution in time of the quantity y(t) affected by white noise and the average value
over short (pink), or long term (red).
Fig. 3. – Evolution in time of a quantity y(t) affected by random walk.
observation internal τ , but when we change τ we can observe the Allan variance increasing
or decreasing according to a particular law and this is what gives us an indication of the
type of noise.
We usually use the drawing reported in fig. 4 where, with the Allan variance or the
spectral analysis, the noise can be identified and quantified. The example reports phase
and frequency noise assuming the analysis is applied to frequency measures. We identify
white, flicker, and random walk behaviors which are typical of atomic clocks.
Fig. 4. – The noise analysis in terms of power spectrum (left) or Allan variance (right) applied
to the typical noises of the atomic frequency standards.
Fig. 5. – Example of typical random noises.
3. – Stochastic processes to model time varying quantities
In the analysis of time series of physical quantities, the most common encountered
random behaviours, or noises, are white, random walk, and integrated random walk.
Examples of this random behaviour are shown in fig. 5. Also flicker noise is often en-
countered, but its mathematical model is more complex and we will not treat it here.
Even if the behaviour is random, in some case it is possible to infer some general laws
and this can be done, for example, by the use of stochastic processes, mathematical tools
which allow to estimate many properties of the stochastic behaviour.
While in case of a deterministic behaviour we can easily predict the future trend and
the uncertainty is only due to the limited knowledge of the polynomial coefficients, or to
the limited knowledge of the physics regulating the process, as we may not know perfectly
the mathematical law like the degree of the polynomial, or the complete mathematical
expression. These limitations can be improved by a better knowledge of the system or
more accurate measures.
In case of random behaviour, the physical quantity at hand cannot be exactly pre-
dicted as at any instant there is a random unpredictable contribution which is added,
but we can have some information on the overall behaviour of these types of quantities.
We may say for example which will be the expected position at a certain time, and this
position will be given by a random variable with known distribution. We will be able to
say, for example, that the position of the quantity will be within a certain threshold with
95% probability. Our knowledge of random behaviour will be a probabilistic one, giving
us a certain degree of confidence on the predicted values.
In this case, even if we improve the measurement and the uncertainty, and are able
to better understand the physics of the system, we would never get rid of the random
component.
The model here presented is usually introduced to represent the time and frequency
deviation of an atomic clock but it can be of more general interest. Let us suppose
that the our velocity y(t) when walking can be instantaneously measured and has the
behaviour as reported in fig. 6(left). The space we cover and our final position x(t) will be
given by the integral of the velocity and in this case it will be as depicted in fig. 6(right).
Fig. 6. – Example of white process y(t) and its integral x(t).
If we want to write an equation representing the velocity y(t), we would write
y(t) = y0 + ξ(t) with ξ(t) : N (0, σ 2 ),
where y0 is the mean value and ξ(t) a random variable with Normal Gaussian distribution
with zero mean value and variance equal to σ 2 . It means that at any instant ξ(t) is a
random number whose distribution is always the same. This is a white noise.
If we want to write the equation of the position x(t), we shall integrate and write
t
x(t) = x0 + y0 (t − t0 ) + ξ(s)ds.
t0
We recognize a straight line due to the constant mean velocity, while the integral of the
random part is something that cannot be dealt with the traditional Riemann integral.
To this aim we need the theory of stochastic process [8] and the appropriate mathematics
which has been defined to treat the integral of random process. The integral of a white
process is a Wiener process W (t). Sometimes in metrology we call it random walk or
Brownian motion. These expressions are not rigorous in mathematics, but they give an
idea of what happens. Let us think to the integral of a white process. This is usually
referred to as the drunk walk. Let us suppose the drunk can walk each step east or west.
Each step can be in a random direction with equal probability. The accumulation of
random steps brings the drunk sometimes far away east, sometimes far away west, most
of the time it remain around the initial position. If we had 10000 drunks all together and
we let them go, after a while they will be a bit dispersed, but after one hour or several
hours they will be much more dispersed. We cannot say where the individual drunk
went, but we can say that on the whole the population of drunk people will disperse
with a probabilistic law. In case the steps are Normally distributed N (0, σ 2 ), we can
demonstrate that the position at the epoch t can be described by a Normal Gaussian
distribution N (0, σ 2 t), whose variance is equal to σ 2 t (fig. 7). The variance is proportional
to time, the standard deviation is proportional to the square root of time. This is the
fundamental property of a Wiener process.
Fig. 7. – Example of realisations of the Wiener process and their distribution at epoch t.
To be more precise, we should write the equation of motion as
dx(t)
y(t) = = y0 + ξ(t) with ξ(t)dt = σ dW (t),
dt
dx(t) = y0 dt + σ dW (t),
whose solution would be

t t
x(t) = y0 dt + σ dW (t),
t0 t0
a straight line given by the mean velocity plus the Wiener process due to the white noise
on the velocity.
We can make our model more complex saying that the mean velocity is no more a
constant but a linearly accelerated one, plus the addition of another Wiener process. We
would have
⎧
⎨dx1 (t) = x2 (t)dt + σ1 dW1 (t),

⎩dx (t) = a dt + σ dW (t),
2 2 2
where now we indicate the position with x1 (t), part of the velocity with x2 (t), the accel-
eration with a, and (W1 (t), W2 (t)) the two Wiener processes acting on the position and
the velocity, respectively.
The complete solution is written as
⎧ t
⎪ 2
⎨x (t) = x + y t + a t + σ W (t) + σ
1 0 0 1 1 2 W2 (s)ds,
2 o
⎪
⎩x2 (t) = y0 + at + σ2 W2 (t),
with the initial conditions

⎧
⎨x1 (0) = x0 ,
⎩x (0) = y .
2 0
Concerning the deterministic part, we recognize the parabolic behaviour of the position
due to the accelerated motion. The solution can also be written in an iterative form
which is more convenient for example for simulations. It becomes
⎧ tk+1
⎪
⎨X (t τ2
1 k+1 ) = X1 (tk ) + X2 (tk )τ + a + σ1 W1,k (τ ) + σ2 W2 (s)ds,
2 tk
⎪
⎩X (t
2 k+1 ) = X2 (tk ) + aτ + σ2 W2,k (τ ),
⎧
⎨X1 (0) = x0 ,
⎩X (0) = y ,
2 0
where τ = tk+1 − tk . The important issue is that the stochastic parts in X1 and X2
are correlated and this is to be taken appropriately into account for an exact simulation.
Details on this model, applied to atomic clocks, can be found in [9-14].
Other stochastic processes are studied and can be useful as model for the time varying
physical quantity. An example is the Ornstein-Uhlenbeck (O.U.) process Ut which is the
solution of the following stochastic differential equation:
1
dUt = − Ut dt + σ dWt .
τ
We can interpret the O.U. process as a Wiener process with the addition of a spring in
the origin which is not allowing the process to go too far from the origin. The plots of
fig. 8 compare different realizations of a Wiener process (top, red), white noise (bottom,
red), with an O.U. process (blue). We can see that the Wiener is going further than the
O.U. (top), while the O.U. shows a bit more of memory with respect to pure white noise
(bottom) [15].
From this brief and not rigorous description of the stochastic processes, we have learnt
that for many physical quantities of interest in metrology,
1) deterministic and stochastic behaviour can be modeled by stochastic differential

equations (SDE),
2) the exact (or approximate) solution of the SDE allows the dynamic behaviour
estimation, simulation, and prediction,
3) The noises imbedded in the model have usually zero mean value, they do not impact
on the prediction but on the uncertainty of the predicted values (therefore allowing
the evaluation of confidence intervals).
Fig. 8. – Wiener trajectories (top, red), or white trajectories (bottom, red) with respect to an
O.U. process (blue).
4. – The need of preprocessing
When dealing with real data and real measures, we often have to deal with spurious
data, missing observation, or outliers. In some cases these spurious data are heavily
degrading our capacity to estimate the true and genuine behaviour of the quantity of
interest. When dealing for example with indirect measures obtained from a complex
system we often observe that data series may present gaps, long periods of missing
observation, outliers, and jumps (fig. 9).
These anomalies have to be detected and properly removed before the data processing
and analysis to avoid a biased or incorrect estimation of the system behaviour, and to
be able to really identify the true system underneath the apparent degraded series of
measures. To this aim an efficient preprocessing is fundamental to ensure proper data
analysis.
The possibility to appropriately treat these cases is treated for example in [16, 17].
What we mostly consider now is the capacity to detect a true change in the “normal”
behaviour of the quantity we are measuring that can, for many reasons, change, abruptly
or slowly. To the detection of anomaly we will dedicate the next section.
Fig. 9. – Example of “dirty” data as obtained when measuring the time offset on a clock on
board a GNSS satellite [16].
5. – The detection of anomalous behaviour
We have discussed the possibility to write a mathematical model for the evolution
in time of our physical quantity and the model can contain deterministic or stochastic
behaviour. The next question is: will the modelled behaviour last forever? Will the
behaviour be stationary? There are different causes for change: ageing, failures, wearing,
etc. Examples are reported in fig. 10 taken from measures of the frequency deviation of
space clocks on board of satellite systems [18-21].
In case of clocks used in real time for timekeeping or in navigation system, the in-
surgence of similar anomalous behaviour can cause serious issues and it is fundamental
to detect them rapidly and with a high efficient algorithm. The topic is quite hot when
dealing with clocks and many groups are working on the capacity to detect jumps, for
example, on the clock frequency [22-28]. In any application, the detection of anomalous
behaviour is based on:
1) observation of past behaviour,
2) extrapolation versus the future of the expected “normal” behaviour,
3) comparison between predicted and measured trend, setting a threshold on the “al-
lowed” difference.
This steps are illustrated in fig. 11
When the difference Δy = |ys − μ &e | between the extrapolated behaviour μ&e and the
measured one ys (smoothed over a certain interval Nw3 ) is larger than a fixed threshold,
an alarm is raised.
In other fields this problem is called Quickest detection method, or Optimal stop-
ping [29]. In case the underlying process is a Wiener one, exact solutions exist.
Let us assume that the quantity evolution can be modeled by a Wiener process X(t)
and that we observe a trajectory with a drift changing from 0 to μ at some random
time θ. We want to find a stopping time τ as close as possible to the unknown time θ.
Fig. 10. – Example of anomalous behaviour observed on clocks on board a GNSS satellite [20].
Fig. 11. – Measuring, extrapolating, and checking for anomalous behaviour [27].
The example is illustrated in fig. 12. We recognize the case at hand, where we studied
the jump on the frequency affected by white noise, which would give a change in slope
in the time deviation of a clock modeled by a Wiener process. Or, in a more generic
example, a jump in the velocity would give a change in slope of the attained position.
Therefore we are always dealing with the same case, but looking from a different point
of view.
Theory of optimal stopping [29] rewrites the problem in another domain, but study-
ing an ancillary process π and by introducing the following functional that has to be
minimized:

V (π) = inf P (τ < θ) + cE[τ − θ]+ .
τ
We want to stop at an epoch τ close to θ, but not before and not too late. If we stop
before it means we are declaring a jump when it has not yet occurred and we raise a false
alarm. If we stop too late after the jump, we have identified it, but it may be too late to
take any recovery action. We recognize in P (τ < θ) the probability of false alarm, while
E[τ − θ]+ is the average delay in detecting the anomaly, and c a suitable constant.
Fig. 12. – A Wiener process with a drift μ arising at epoch θ.
We want P (τ < θ) and also E[τ − θ]+ to be small

– to avoid false alarms,
– to minimize the detection delay.
Therefore we minimize the functional V (π) by setting the constant c in such a way as
to prioritise the false alarm or the delay minimisation. It can be demonstrated that
the optimal stopping problem can be transformed in a first passage time, becoming an
analytical problem. In fact the functional V (π) may be rewritten as
τ
V (π) = inf E 1 − πτ + c πt dt ,
τ 0

πt = Pπ θ ≤ t|FtX , Pπ (π0 = π) = 1,
where πt is the a posteriori probability process, describing the probability that by epoch
t the process X(t) has changed drift. FtX is the natural filtration of the process X(t)
which means the history of X(t) up to time t for t ≥ 0. Observe that πt takes values in
[0, 1] for all t ≥ 0. The optimal stopping time is given by
τA = inf {t ≥ 0 : πt ≥ A} ,
the epoch τA at which, for the first time, the process π crosses the boundary A. The
original process X(t) has changed drift when the process π is crossing the boundary A.
A theorem demonstrates that the exact solution exists and studying the additional
process π we can optimally estimate the epoch of the insurgence of the new drift μ. The
boundary A depends on the values of the parameters of the problem, namely the drift μ,
the variance σ 2 of the Wiener process, the constant c, and the parameter λ of the expo-
nential distribution that we assume characterizes the jump epochs θ, with E(θ) = 1/λ.
Fig. 13. – The contemporaneous behaviour of the processes X(t) and π. The new drift arises at
epoch θ, and the algorithm stops at τ .
Examples of behaviour of the original process X(t) and process π are reported in
fig. 13. In [28] other examples show that, according to the settings of the problem,
including the desired probability of false alarm, or delay in detecting, the process π
becomes more or less nervous and ready to cross the boundary A.
This method has been recently applied to the study of the clock on board the NASA
New Horizon spacecraft which was launched on Jan 19th, 2006 to meet Pluto on July 14,
2015. The measurement system required an ultra-stable oscillator (Quartz Crystal) with
very high frequency stability. The quartz seemed to change the frequency drift exactly
when the spacecraft was getting close to Pluto [30].
6. – Dynamic noise analysis

The noise affecting the system can change in time and therefore, in certain applica-
tions, we need a dynamical characterization of the noise.
In the frequency domain of power spectra, a Time and Frequency spectral analysis
has been introduced processing the signal not only to estimate which frequencies existed,
but also to estimate when they existed [31].
We can do the same in time domain introducing the Dynamic Allan variance as
an Allan variance estimator sliding on the data. The principle is illustrated in fig. 14.
Instead of estimating the Allan variance on the entire set of data, we use only a portion
of it and we estimate the Allan variance, then we slide our estimation on the following
period of data, and so on. We obtain an Allan variance which depends on the observation
interval τ , as we know, but now also on the epoch t.
The definition of the Dynamic Allan deviation σy (t, τ ) and its characteristics can be
found in [32-35]. The discrete time formulation from the time deviation samples x[n],
Fig. 14. – The estimation of the Allan deviation on a sliding window.
taken at discrete epochs n, is written as
1 1
n+Nw /2−k−1

σy2 [n, k] = E (xN [m + k] − 2xN [m] + xN [m − k])2 ,
2k 2 τ02 Nw − 2k
m=n−Nw /2+k
where Nw is the window length, xN is the time deviation in the window Nw , τ0 is the
sampling time, with τ = kτ0 .
The definition contains the expectation value E taken as an ensemble average over
an infinite number of sequences. The estimator of the Dynamic Allan variance σ &y2 (t, τ ),
has no expectation value E because we have one realization only.
Figure 15 illustrates this mechanism. Let us suppose to have a sequence of data
whose noise is increasing for a certain period of time panel (a). If we evaluate the Allan
deviation of the entire data series, we obtain the curve panel (b) not showing that the
noise is increasing in the central period. If we had at disposal an infinite number of
sequences, we could exactly estimate the Dynamic Allan deviation corresponding to this
case and we would obtain panel (c), clearly showing the increase of noise around epoch
t = 500. Instead with only one sequence, we obtain the Dynamic Allan deviation estimate
reported in panel (d) which shows the statistical fluctuations due to the limited number
of data. The statistical fluctuations are related to the length of the window Nw . To have
more data and less fluctuations, we would need a long window, conversely to be able to
quickly detect a change in the stability, we prefer a short window. The choice of the
window is to be fixed as a suitable trade-off.
The Dynamic Allan deviation can be plotted in a 3D plot also reporting on the walls
the classical Allan deviation, and the signal y(t). In addition, a flat surface can be plotted
together acting as a visive threshold of the obtained stability (fig. 16). The Dynamic
Fig. 15. – The estimation of the Dynamic Allan deviation on a series of noisy data (a). The
corresponding classical Allan deviation (b). The theoretical Dynamic Allan deviation (c), and
its estimate (d).
Fig. 16. – The estimation of the Dynamic Allan deviation with a reference stationary surface.
Allan variance has been used to estimate changes in the stability such as increase of
noise, jumps, and long-term drift insurgence. It has been also used to demonstrate that
the stability of some devices is not degrading with time, and it has also been applied to
cardiology [36]. Codes implementing the estimation of the Dynamic Allan variance are
available [37].
7. – Conclusion
The evaluation of the evolution in time of a time quantity can reveal important
issues. Instabilities may be highly impacting the physical systems and they need to be
appropriately estimated.
To this aim, we have illustrated some mathematical tools that can be used to model,
estimate, predict, and simulate a physical system.
The behaviour of a system may change due to ageing, failures, or wearing. These
changes are to be detected (rapidly), and the model is to be dynamically updated.
Many other mathematical and statistic techniques for signal processing are used and
available in metrology for the treatment of time series that we have not mentioned here.
For example the use of the Kalman filter [38], or of control theory [39], or the generation
of an ensemble time scale [40], and many others, that may be found in the literature.
REFERENCES
[1] Girard G., “The Third Periodic Verification of National Prototypes of the Kilogram
(1988-1992)”, Metrologia, 31 (1994) 317.
[2] Abbott B. et al. (LIGO Scientific Collaboration and Virgo Collaboration),
“Observation of Gravitational Waves from a Binary Black Hole Merger”, Phys. Rev. Lett.,
116 (2016) 061102.
[3] JCGM 100:2008, Evaluation of measurement data — Guide to the expression of
uncertainty in measurement, https://www.bipm.org/utils/common/documents/jcgm/
JCGM 100 2008 E.pdf.
[4] Box G., Jenkins G. and Reinsel G., Time Series Analysis, Wiley Series in Probability
and Statistics (Wiley) 2013, ISBN:9780470272848.

[5] Allan D. W., “Statistics of atomic frequency standards”, Proc. IEEE, 54 (1966) 221.
[6] Allan D. W., “Time and Frequency (Time-Domain) Characterization, Estimation,
and Prediction of Precision Clocks and Oscillators”, IEEE Trans. Ultrason. Ferroelec.
[7] Levine J., Tavella P. and Santarelli G. (Editors), “Special Issue Celebrating the 50th
Anniversary of the Allan Variance”, IEEE Trans. Ultrason. Ferroelec. Frequency Control,
34 (2016) 511.
[8] Arnold L., Stochastic Differential Equations: Theory and Applications (Krieger,
Malabar) 1974, chapts. 4 and 8; Cox D. R. and Miller H. D., The Theory of Stochastic
Processes (Chapman and Hall, London) 1965, pp. 205–10.
[9] Galleani L., Sacerdote L., Tavella P., Zucca C., “A mathematical model for the
atomic clock error”, Metrologia, 40 (2003) S257.
[10] Tavella P. and Zucca C., “The Clock Model and its Relationship with the Allan and
related Variances”, IEEE Trans. UFFC Ultrason. Ferroelec. Frequency Control, 52 (2005)
288.
[11] Panfilo G. and Tavella P., “Atomic Clock Prediction based on stochastic differential
equations”, Metrologia, 45 (2008) S108.
[12] Zucca C. and Tavella P., “A mathematical model for the atomic clock error in case of
jumps”, Metrologia, 52 (2015) 514.
[13] Vernotte F., Delporte J., Brunet M. and Tournier T., “Uncertainties of drift
coefficients and extrapolation errors: application to clock error prediction”, Metrologia, 38
(2001) 42.
[14] Audoin C. and Dimarcq N., “Stochastic models of stable frequency and time sources and
their relationship”, IEEE Trans. Instrum. Meas., 42 (1993) 682.
[15] Bibbona E., Panfilo G. and Tavella P., “The Ornstein-Uhlenbeck Process as a Model
of the Filtered White Noise”, Metrologia, 45 (2008) S117.
[16] Sesia I., Cantoni E., Cernigliaro A., Signorile G., Fantino G. and Tavella P.,
“An Efficient and Configurable Preprocessing Algorithm to Improve Stability Analysis”,
IEEE Trans. Ultrason. Ferroelec. Frequency Control, 63 (2016) 575.
[17] Sesia I. and Tavella P., “Estimating the Allan Variance in Presence of Long Periods of
Missing Data and Outliers”, Metrologia, 45 (2008) S134.
[18] Waller P., Gonzalez F., Binda S., Sesia I., Hidalgo I., Tobias G. and Tavella
P., “The In-orbit Performances of GIOVE Clocks”, IEEE Trans. Ultrason. Ferroelec.
[19] Cernigliaro A., Valloreia S., Galleani L., Fantino G. and Tavella P., “Analysis
on GNSS space clocks performances”, in Proceedings of the Joint European Frequency and
Time Forum & International Frequency Control Symposium (EFTF/IFC), Prague, 2013,
pp 835-837.
[20] Cernigliaro A., Fantino G., Sesia I., Galleani L. and Tavella P.,
“Nonstationarities in space clocks: Investigations on Experimental Data”, in Proceedings
of the 28th European Frequency and Time Forum (EFTF), June 2014, Neuchâtel,
Switzerland, pp. 126–129.
[21] Formichella V., Camparo J. and Tavella P., “Influence of the ac-Stark Shift on GPS
Atomic Clock Timekeeping”, Appl. Phys. Lett., 110 (2017) 043506.
[22] Nunzi E., Galleani L., Tavella P. and Carbone P., “Detection of anomalies in the
behavior of atomic clocks”, IEEE Trans. Instrum. Meas., 56 (2007) 523.
[23] Wang Q., Droz F. and Rochat P., “Robust Clock Ensemble for Time and Frequency
Reference System”, presented at the Joint Conference of the IEEE International Frequency
Control Symposium & European Frequency and Time Forum, Denver, 2015.
[24] Huang X., Gong H. and Ou G., “Detection of weak frequency jumps for GNSS onboard
clocks”, IEEE Trans. Ultrason. Ferroelectr. Frequency Control, 61 (2014) 747.
[25] Galleani L. and Tavella P., “Detection of Atomic Clock Frequency Jumps with the
Kalman Filter”, IEEE Trans. Ultrason. Ferroelectr. Frequency Control, 59 (2012) 504.
[26] Galleani L. and Tavella P., “Detection and identification of atomic clock anomalies”,
Metrologia, 45 (2008) S127.
[27] Galleani L. and Tavella P., “Robust Detection of Fast and Slow Frequency Jumps of
Atomic Clocks”, IEEE Trans. Ultrason. Ferroelectr. Frequency Control, 64 (2017) 475.
[28] Zucca C., Tavella P. and Peskir G., “Detecting atomic clock frequency trends using
an optimal stopping method”, Metrologia, 53 (2016) S89.
[29] Peskir G. and Shiryaev A., Optimal Stopping and Free-Boundary Problems, Lectures
in Mathematics (Birkhäuser, ETH Zürich) 2006; Shiryaev A., Optimal Stopping Rules
(Springer) 1978.
[30] Weaver G. L., Jensen J. R., Zucca C., Tavella P., Formichella V. and Peskir G.,
“Estimation of the dynamics of frequency drift in mature ultra-stable oscillators: a study
based on the in-flight performance from New Horizons”, in Proceedings ION PTTI Precise
Time and Time Interval meeting, Monterey CA, Jan 2016.
[31] Cohen L., Time-frequency analysis (Prentice-Hall) 1995.
[32] Galleani L. and Tavella P., “Dynamic Allan variance”, IEEE Trans. Ultrason.
Ferroelectr. Frequency Control, 56 (2009) 450.
[33] Sesia I., Galleani L. and Tavella P., “Application of the Dynamic Allan Variance for
the Characterization of Space Clock Behavior”, IEEE Trans. Aerospace Electron. Systems,
47 (2011) 884.
[34] Galleani L. and Tavella P., “The Dynamic Allan Variance IV: Characterization of
Atomic Clock Anomalies”, IEEE Trans. Ultrason. Ferroelectr. Frequency Control, 62
(2015) 791.
[35] Galleani L. and Tavella P., “The Dynamic Allan Variance V: Recent Advances in
Dynamic Stability Analysis”, IEEE Trans. Ultrason. Ferroelectr. Frequency Control, 63
(2016) 624.
[36] Hernández-Pérez R., Guzmán-Vargas L., Reyes-Ramı́rez I. and Angulo-Brown
F., “Evolution in time and scales of the stability of heart interbeat rate”, EPL, 92 (2010)
68006.
[37] Free Matlab implementation on http://www.inrim.it/res/tf/allan.shtml CANVAS by
NRL @ https://goby.nrl.navy.mil/canvas/download/ STABLE 32 Users: Upgrade to
version 1.5.
[38] Galleani L. and Tavella P., “Time and the Kalman filter”, IEEE Control Syst. Mag.,
30 (2010) 44.
[39] Farina M., Galleani L., Tavella P. and Bittanti S., “A Control Theory Approach
to Clock Steering Techniques”, IEEE Trans. Ultrason. Ferroelectr. Frequency Control, 57
(2010) 2257.
[40] Tavella P., “Statistical and mathematical tools for atomic clocks”, Metrologia, 45 (2008)
S183.
International School of Physics “Enrico Fermi”
Villa Monastero, Varenna
Course 196
27 June - 6 July 2016
Metrology: From Physics Fundamentals
to Quality of Life
Directors Michael de Podesta

NPL
Massimo Inguscio
Hampton Road
CNR e Università di Firenze
TW11 0LW Teddington
Dipartimento di Fisica e Astronomia
Middlesex, UK
Via Giovanni Sansone 1
tel: +44 020 8943 6476
50019 Sesto Fiorentino (FI), Italy
michael.depodesta@npl.co.uk
tel: +39 055 4572465
inguscio@lens.unifi.it
Bernd Güttler
Martin J. T. Milton PTB
BIPM Pavillon de Breteuil Bundesallee 100
92312 Sèvres, France 38116 Braunschweig, Germany
tel: +33 1 45 07 70 00 tel: + 49 0531 592 3010
martin.milton@bipm.org bernd.guettler@ptb.de
Patrizia Tavella
INRIM Beat Jeckelmann
Strada delle Cacce 91 METAS
10135 Torino, Italy Lindenweg 50
tel: +39 011 3919 235 3003 Bern-Wabern, Switzerland
p.tavella@inrim.it tel: +41 058 387 02 97
beat.jeckelmann@metas.ch
Scientific Secretary
William Phillips
Natascia De Leo
NIST
INRIM
100 Bureau Drive
Strada delle Cacce 91

MD 20899-8930 Gaithersburg
10135 Torino, Italy
Maryland, USA
tel: +39 011 3919 410
tel: +1 301 975 6554
n.deleo@inrim.it
william.phillips@nist.gov
Lecturers
Roberto Battiston Terry J. Quinn
ASI CBE FRS
Via del Politecnico snc 92 rue Brancas
00133 Roma, Italy 92310 Sèvres, France
tel: +39 06 8567 821 tel: +33 1 4623 0656
roberto.battiston@asi.it tjqfrs@gmail.com

495
496 List of participants
Carl Williams Nigel Fox

NIST NPL
100 Bureau Drive Hampton Road
MD 20899-8930 Gaithersburg TW11 0LW Teddington
Maryland, USA Middlesex, UK
tel: +1 301 975 2220 tel: + 44 020 8943 6825
carl.williams@nist.gov nigel.fox@npl.co.uk
Stephen Wise Andy Henson

NIST BIPM Pavillon de Breteuil
100 Bureau Drive 92312 Sèvres, France
MD 20899-8930 Gaithersburg tel: +33 1 45 07 62 72
Maryland, USA andy.henson@bipm.org
tel: +1 301 975 3112
Stephen.wise@nih.gov François Nez
stephenwise@comcast.net LKB-CNRS
Université Pierre et Marie Curie (UPMC)
Case 74, 4 Place Jussieu
75252 Paris, France
Seminar Speakers tel: +33 1 44 27 72 48
francois.nez@spectro.jussieu.fr
Walter Bich francois.nez@lkb.upmc.fr
INRIM
Strada delle Cacce 91 Thomas Udem
10135 Torino, Italy Max-Planck-Institut für Quantenoptik
tel: +39 011 3919 945 Hans-Kopfermann-Str. 1
w.bich@inrim.it D-85748 Garching, Germany
tel: +49 89 32905 257
Giorgio Brida thomas.udem@mpq.mpg.de
INRIM
Strada delle Cacce 91 Diederik Wiersma
10135, Torino, Italy INRIM
tel: +39 011 3919 222 Strada delle Cacce 91
g.brida@inrim.it 10135 Torino, Italy
tel: +39 011 3919 544
Vincent Delatour wiersma@lens.unifi.it
LNE
rue Gaston Boissier Robert I. Wielgosz
75724 Paris Cedex 15, France BIPM Pavillon de Breteuil
tel: +33 1 40 434 075 92312 Sèvres, France
vincent.delatour@lne.fr tel: +33 1 45 07 62 51
rwielgosz@bipm.org
Students
Dorela Mirabela Albu National Institute of Metrology, Romania
Matteo Barbiero INRIM, Italy
Daniel Beach National Research Council, Canada
Filippo Bregolin INRIM, Italy
List of participants 497
Eleonora Cara INRIM, Italy

Leon Chao NIST, USA
Huang Che Huazhong University of Science and Technology, China
Alessandro Cultrera INRIM, Italy
Davide D’Ambrosio INRIM, Italy
Peter Davis NPL, UK
Andreia de Lima Fioravante INMETRO, Brazil
Marco Di Luzio INRIM, Italy
Masoud Dialameh INRIM, Italy
Dinesh Indika Dias Measurement Units, Sri Lanka
Paolo Durandetto INRIM, Italy
Fang Fang National Institute of Metrology, China
Riccardo Ferrero INRIM, Italy
Ilaria Finardi INRIM, Italy
Valerio Formichella INRIM, Italy
Sebastian Galer NPL, UK
Ghislain Granger National Research Council, Canada
Myriam Guillevic METAS, Switzerland
Fabian Heisse Max-Planck-Institute for Nuclear Physics, Germany
Vicentiu Iancu Faculty of Physics, Romania
Dario Imbraguglio INRIM, Italy
Mihaela Ispas National Institute of Metrology, Romania
Irma Kruger National Metrology Institute, South Africa
Alice Lorenzati Politecnico di Torino, Italy
Luisa Mandrile INRIM, Italy
Valentin-Constantin Mihalcea National Institute of Metrology, Romania
Dennis Mkhize National Metrology Institute, South Africa
Michael Moawad NMI, Australia
Edward C. Mulhern NIST, USA
Meirbek Mussatayev RSE “KazlnMetr”, Kazakhstan
Luis Carlos Nobre University of Lisboa, Portugal
Wilson Ombati Nyangau Bureau of Standards, Kenya
Sara Orski NIST, USA

Martin Parker Tallinn University of Technology, Estonia
Elia Perego INRIM, Italy
Enrica Pessana INRIM, Italy
Ben Piper NPL, UK
Faranak Pourdanesh INRIM, Italy
Zachary Reed NIST, USA
Ionut Rosu National Institute of Metrology, Romania
Daniel Sarno NPL, UK
Alejandro Savarin Instituto Nacional de Tecnologı́a Industrial, Argentina
John E. Schiel NIST, USA
498 List of participants
Yljon Seferi General Directorate of Metrology, Albania

Daniel Stokes NPL, UK
Florian Stuker METAS, Switzerland
Shahin Tabandeh INRIM, Italy
Anna Tampellini INRIM, Italy
Mostafa Tarboosh National Institute for Standards, Egypt
Konstantina Vasilatou METAS, Switzerland
Adrian Vicent Claramunt Center for Physical Sciences and Technology, Lithuania
Nicholas Vlajic NIST, USA
Steven Wang National Measurement Institute, Australia
Aaron Whittam, NPL, UK
Jinjie Wu National Institute of Metrology, China
Ping Yang National Institute of Metrology, China
Chien-Chih Yeh Center for Measurement Standards, Taiwan
Yu-shan Yeh Center for Measurement Standards, Taiwan
Observers
Faisal Alqahtani Saudi Standards, Metrology and Quality Org., Saudi Arabia
Kuniaki Amemiya National Metrology Institute, Japan
Marc-Olivier André METAS, Switzerland
Natalie Belsey NPL, UK
Luca Boarino INRIM, Italy
Paul Green NPL, UK
Rasmus Havelund NPL, UK
Moto Kinoshita National Metrology Institute, Japan
Ryunosuke Kuroda National Metrology Institute, Japan
Jürgen Mlynek Humboldt Universität Berlin, Germany
Hideaki Nozato National Metrology Institute, Japan
Takamitsu Otake National Metrology Institute, Japan
Elijah Petersen NIST, USA
Osamu Sato National Metrology Institute, Japan

Hironobu Takahashi National Metrology Institute, Japan
Kevin Thomson National Research Council, Canada
Finito di stampare
nel mese di Dicembre 2017

Metrologia: Dalla Fisica Fondamentale Alla Qualit' A Della Vita

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Metrologia: Dalla Fisica Fondamentale Alla Qualit' A Della Vita

Uploaded by

Copyright:

Available Formats

` ITALIANA DI FISICA

SCUOLA INTERNAZIONALE DI FISICA

27 Giugno – 6 Luglio 2016

Metrologia: dalla ﬁsica

INTERNATIONAL SCHOOL OF PHYSICS

27 June – 6 July 2016

Metrology: From Physics

AMSTERDAM, OXFORD, TOKIO, WASHINGTON DC

ISSN 0074-784X (print)

jointly published and distributed by:

IOS PRESS SOCIETÀ ITALIANA DI FISICA

Distributor in the USA and Canada

Camera di Commercio di Lecco

Istituto Nazionale di Fisica Nucleare (INFN)

Istituto Nazionale di Geoﬁsica e Vulcanologia (INGV)

Museo Storico della Fisica e Centro Studi e Ricerche “Enrico Fermi”

Istituto Nazionale di Ricerca Metrologica (INRIM)

FEI Italia s.r.l.

Cal Power s.r.l.

Bureau International des Poids et Mesures (BIPM)

Production Editor: Marcella Missiroli

Graphic elaboration by Simona Oleandri

Proprietà Letteraria Riservata

Printed in Italy by Labanti e Nanni Industrie Graﬁche S.r.l. - Crespellano (BO)

P. Tavella, M. J. T. Milton and M. Inguscio – Preface . . . . . . . . . . . . p. XIX

Course group shot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XXII

MODULE I. METROLOGY FOR QUALITY OF LIFE

Vincent Delatour, Gustavo Martos, Julie Cabillic, Maryline

3 1. Why commutability matters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

Vincent Delatour, Gustavo Martos, François Gaie-Levrel,

N. P. Fox – SI traceable measurements of the Earth from space to monitor

Bernd Güttler, Detlef Schiel, Axel Pramann and Olaf Rienitz –

R. I. Wielgosz, J. Viallon, E. Flores, P. Moussay, F. Idrees and

2. Methods for gas standard preparation and veriﬁcation . . . . . . . . . . . . . . . . . 91

R. I. Wielgosz, S. Westwood, R. D. Josephs, N. Stoppacher,

MODULE II. FUNDAMENTALS OF METROLOGY

W. Bich – Uncertainty of measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Andy Henson – International recognition of NMI calibration and measure-

Martin J. T. Milton – On the proposed re-deﬁnition of the SI: “for all

T. J. Quinn – The development of units of measurement from the origin of

8. Developments in other base units until 1971 . . . . . . . . . . . . . . . . . . . . . . . . . 237

Th. Udem and F. Riehle – Frequency combs applications and optical

Carl J. Williams – The SI and quantum metrology . . . . . . . . . . . . . . . . . . 301

Carl J. Williams – The future of quantum-based measurements and the SI 309

6. Other atom-based sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

MODULE III. PHYSICAL METROLOGY AND FUNDAMENTAL CONSTANTS

R. Battiston – Space metrology and space-time warping . . . . . . . . . . . . . . 321

Giorgio Brida – Measuring light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

Michael de Podesta – Why redeﬁne the kelvin? . . . . . . . . . . . . . . . . . . . . . 357

H. Fleurbaey, C. Courvoisier, P. Cladé, S. Guellati-Khélifa,

K. Helmerson and W. D. Phillips – Cooling, trapping and manipulation

3. Deceleration and cooling of an atomic beam . . . . . . . . . . . . . . . . . . . . . . . . . 403

Carl J. Williams – From the mise-en-pratique for mass to the future of

1. Problems with the current SI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450

Patrizia Tavella – Precise time scales and navigation systems . . . . . . . . . 463

Patrizia Tavella – Statistical tools for time varying quantities . . . . . . . . . 477

List of participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495

SIF Members are granted free access to these volumes at

For a complete list of published courses see

Module II: Fundamentals of Metrology (30th June to 2nd July 2016)

P. Tavella, M. J. T. Milton and M. Inguscio

Italian Physical Society

Reference methods and commutable reference