P. 1
Kruegel E. The physics of interstellar dust (IoP, 2003)(584s)

Kruegel E. The physics of interstellar dust (IoP, 2003)(584s)

|Views: 3,545|Likes:
Published by Batominovski

More info:

Published by: Batominovski on Mar 11, 2011
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less






  • The dielectric permeability
  • 1.1 Maxwell’s equations
  • 1.1.1 Electric field and magnetic induction
  • 1.1.2 Electric polarization of the medium
  • 1.1.4 The physical meaning of the electric susceptibility χ
  • 1.1.5 Magnetic polarization of the medium
  • 1.1.6 The magnetic susceptibility
  • 1.1.7 Dielectrics and metals
  • 1.1.8 Free charges and polarization charges
  • 1.1.9 The field equations
  • 1.2 Waves in a dielectric medium
  • 1.2.1 The wave equation
  • 1.2.2 The wavenumber
  • 1.2.3 The optical constant or refractive index
  • 1.2.4 Energy dissipation of a grain in a variable field
  • 1.3 The harmonic oscillator
  • 1.3.1 The Lorentz model
  • 1.3.2 Free oscillations
  • 1.3.3 The general solution to the oscillator equation
  • 1.3.4 Dissipation of energy in a forced oscillation
  • 1.3.5 Dissipation of energy in a free oscillation
  • 1.3.6 The plasma frequency
  • 1.3.7 Dispersion relation of the dielectric permeability
  • 1.4 The harmonic oscillator and light
  • 1.4.1 Attenuation and refraction of light
  • 1.4.2 Retarded potentials of a moving charge
  • 1.4.3 Emission of an harmonic oscillator
  • 1.4.4 Radiation of higher order
  • 1.4.5 Radiation damping
  • 1.4.6 The cross section of an harmonic oscillator
  • 1.4.7 The oscillator strength
  • 1.4.8 The natural linewidth
  • 1.5 Waves in a conducting medium
  • 1.5.1 The dielectric permeability of a conductor
  • 1.5.2 Conductivity and the Drude profile
  • 1.5.3 Electromagnetic waves in a plasma with a magnetic field
  • 1.5.4 Group velocity of electromagnetic waves in a plasma
  • 1.6 Polarization through orientation
  • 1.6.1 Polarization in a constant field
  • 1.6.2 Polarization in a time-variable field
  • 1.6.3 Relaxation after switching off the field
  • 1.6.4 The dielectric permeability in Debye relaxation
  • How to evaluate grain cross sections
  • 2.1 Defining cross sections
  • 2.1.1 Cross section for scattering, absorption and extinction
  • 2.1.2 Cross section for radiation pressure
  • 2.1.3 Efficiencies, mass and volume coefficients
  • 2.2 The optical theorem
  • 2.2.1 The intensity of forward scattered light
  • 2.2.2 The refractive index of a dusty medium
  • 2.3 Mie theory for a sphere
  • 2.3.1 The generating function
  • 2.3.2 Separation of variables
  • 2.3.3 Series expansion of waves
  • 2.3.4 Expansion coefficients
  • 2.3.5 Scattered and absorbed power
  • 2.3.6 Absorption and scattering efficiencies
  • 2.4 Polarization and scattering
  • 2.4.1 The amplitude scattering matrix
  • 2.4.2 Angle-dependence of scattering
  • 2.4.3 The polarization ellipse
  • 2.4.4 Stokes parameters
  • 2.4.5 Stokes parameters of scattered light for a sphere
  • 2.5 The Kramers–Kronig relations
  • 2.5.1 Mathematical formulation of the relations
  • 2.5.2 The electric susceptibility and causality
  • 2.5.3 The Kramers–Kronig relation for the dielectric permeability
  • 2.5.4 Extension to metals
  • 2.5.5 Dispersion of the magnetic susceptibility
  • 2.5.6 Three corollaries of the KK relation
  • 2.6 Composite grains
  • 2.6.1 Effective medium theories
  • 2.6.2 Garnett’s mixing rule
  • 2.6.3 The mixing rule of Bruggeman
  • 2.6.4 Composition of grains in protostellar cores
  • 2.6.5 How size, ice and porosity change the absorption coefficient
  • Very small and very big particles
  • 3.1 Tiny spheres
  • 3.1.1 When is a particle in the Rayleigh limit?
  • 3.1.2 Efficiencies of small spheres from Mie theory
  • 3.1.3 A dielectric sphere in a constant electric field
  • 3.1.4 Scattering and absorption in the electrostatic approximation
  • 3.1.5 Polarization and angle-dependent scattering
  • 3.1.6 Small-size effects beyond Mie theory
  • 3.2 A small metallic sphere in a magnetic field
  • 3.2.1 Slowly varying field
  • 3.2.2 The magnetic polarizability
  • 3.2.3 The penetration depth
  • 3.2.4 Limiting values of the magnetic polarizability
  • 3.3 Tiny ellipsoids
  • 3.3.1 Elliptical coordinates
  • 3.3.2 An ellipsoid in a constant electric field
  • 3.3.3 Cross section and shape factor
  • 3.3.4 Randomly oriented ellipsoids
  • 3.3.5 Pancakes and cigars
  • 3.3.6 Rotation about the axis of greatest moment of inertia
  • 3.4 The fields inside a dielectric particle
  • 3.4.1 Internal field and depolarization field
  • 3.4.2 Depolarization field and the distribution of surface charges
  • 3.4.3 The local field at an atom
  • 3.4.4 The Clausius–Mossotti relation
  • 3.5 Very large particles
  • 3.5.1 Babinet’s theorem
  • 3.5.2 Reflection and transmission at a plane surface
  • 3.5.3 Huygens’ principle
  • 3.5.4 Fresnel zones and a check on Huygens’ principle
  • 3.5.5 The reciprocity theorem
  • 3.5.6 Diffraction by a circular hole or a sphere
  • 3.5.7 Diffraction behind a half-plane
  • 3.5.8 Particles of small refractive index
  • 3.5.9 X-ray scattering
  • Case studies of Mie calculus
  • 4.1 Efficiencies of bare spheres
  • 4.1.1 Pure scattering
  • 4.1.2 A weak absorber
  • 4.1.3 A strong absorber
  • 4.1.4 A metal sphere
  • 4.1.5 Efficiency versus cross section and volume coefficient
  • 4.1.6 The atmosphere of the Earth
  • 4.2 Scattering by bare spheres
  • 4.2.1 The scattering diagram
  • 4.2.2 The polarization of scattered light
  • 4.2.3 The intensity of scattered light in a reflection nebula
  • 4.3 Coated spheres
  • 4.4 Surface modes in small grains
  • 4.5 Efficiencies of idealized dielectrics and metals
  • 4.5.2 Dielectric sphere with Debye relaxation
  • 4.5.4 Efficiencies for Drude profiles
  • 4.5.5 Elongated metallic particles
  • Particle statistics
  • 5.1 Boltzmann statistics
  • 5.1.1 The probability of an arbitrary energy distribution
  • 5.1.2 The distribution of maximum probability
  • 5.1.3 Partition function and population of energy cells
  • 5.1.4 The mean energy of harmonic oscillators
  • 5.1.5 The Maxwellian velocity distribution
  • 5.2 Quantum statistics
  • 5.2.2 Bosons and fermions
  • 5.2.3 Bose statistics
  • 5.2.4 Bose statistics for photons
  • 5.2.5 Fermi statistics
  • 5.2.6 Ionization equilibrium and the Saha equation
  • 5.3 Thermodynamics
  • 5.3.1 The ergodic hypothesis
  • 5.3.2 Definition of entropy and temperature
  • 5.3.3 The canonical distribution
  • 5.3.4 Thermodynamic relations for a gas
  • 5.3.5 Equilibrium conditions of the state functions
  • 5.3.6 Specific heat of a gas
  • 5.3.7 The work done by magnetization
  • 5.3.8 Susceptibility and specific heat of magnetic substances
  • 5.4 Blackbody radiation
  • 5.4.1 The Planck function
  • 5.4.2 Low- and high-frequency limit
  • 5.4.3 Wien’s displacement law and the Stefan–Boltzmann law
  • 5.4.4 The Planck function and harmonic oscillators
  • The radiative transition probability
  • 6.1 A charged particle in an electromagnetic field
  • 6.1.1 The classical Hamiltonian
  • 6.1.2 The Hamiltonian of an electron in an electromagnetic field
  • 6.1.3 The Hamilton operator in quantum mechanics
  • 6.1.4 The dipole moment in quantum mechanics
  • 6.1.5 The quantized harmonic oscillator
  • 6.2 Small perturbations
  • 6.2.1 The perturbation energy
  • 6.2.2 The transition probability
  • 6.2.3 Transition probability for a time-variable perturbation
  • 6.3 The Einstein coefficients A and B
  • 6.3.1 Induced and spontaneous transitions
  • 6.3.2 Selection rules and polarization rules
  • 6.3.3 Quantization of the electromagnetic field
  • 6.3.4 Quantum-mechanical derivation of Aand B
  • 6.4 Potential wells and tunneling
  • 6.4.1 Wavefunction of a particle in a constant potential
  • 6.4.2 Potential walls and Fermi energy
  • 6.4.3 Rectangular potential barriers
  • 6.4.4 The double potential well
  • Structure and composition of dust
  • 7.1 Crystal structure
  • 7.1.1 Translational symmetry
  • 7.1.2 Lattice types
  • 7.1.3 The reciprocal lattice
  • 7.2 Binding in crystals
  • 7.2.1 Covalent bonding
  • 7.2.2 Ionic bonding
  • 7.2.3 Metals
  • 7.2.4 van der Waals forces and hydrogen bridges
  • 7.3 Reddening by interstellar grains
  • 7.3.1 Stellar photometry
  • 7.3.2 The interstellar extinction curve
  • 7.3.3 Two-color diagrams
  • 7.3.4 Spectral indices
  • 7.3.5 The mass absorption coefficient
  • 7.4 Carbonaceous grains and silicate grains
  • 7.4.1 Origin of the two major dust constituents
  • 7.4.2 The bonding in carbon
  • 7.4.3 Carbon compounds
  • 7.4.4 Silicates
  • 7.4.5 A standard set of optical constants
  • 7.5 Grain sizes and optical constants
  • 7.5.1 The size distribution
  • 7.5.2 Collisional fragmentation
  • Dust radiation
  • 8.1 Kirchhoff’s law
  • 8.1.1 The emissivity of dust
  • 8.1.2 Thermal emission of grains
  • 8.1.3 Absorption and emission in thermal equilibrium
  • 8.1.4 Equipartition of energy
  • 8.2 The temperature of big grains
  • 8.2.1 The energy equation
  • 8.2.2 Approximate absorption efficiency at infrared wavelengths
  • 8.2.3 Temperature estimates
  • 8.2.4 Relation between grain size and grain temperature
  • 8.2.5 Temperature of dust grains near a star
  • 8.2.6 Dust temperatures from observations
  • 8.3 The emission spectrum of big grains
  • 8.3.1 Constant temperature and low optical depth
  • 8.3.2 Constant temperature and arbitrary optical depth
  • 8.4 Calorific properties of solids
  • 8.4.1 Normal coordinates
  • 8.4.2 Internal energy of a grain
  • 8.4.3 Standing waves in a crystal
  • 8.4.4 The density of vibrational modes in a crystal
  • 8.4.5 Specific heat
  • 8.4.6 Two-dimensional lattices
  • 8.5 Temperature fluctuations of very small grains
  • 8.5.1 The probability density P(T)
  • 8.5.2 The transition matrix
  • 8.5.3 Practical considerations
  • 8.5.4 The stochastic time evolution of grain temperature
  • 8.6 The emission spectrum of very small grains
  • 8.6.1 Small and moderate fluctuations
  • 8.6.2 Strong fluctuations
  • 8.6.3 Temperature fluctuations and flux ratios
  • Dust and its environment
  • 9.1 Grain surfaces
  • 9.1.1 Gas accretion on grains
  • 9.1.2 Physical adsorption and chemisorption
  • 9.1.3 The sticking probability
  • 9.1.5 Tunneling between surface sites
  • 9.1.6 Scanning time
  • 9.2 Grain charge
  • 9.2.1 Charge equilibrium in the absence of a UV radiation field
  • 9.2.2 The photoelectric effect
  • 9.3 Grain motion
  • 9.3.1 Random walk
  • 9.3.2 The drag on a grain subjected to a constant outer force
  • 9.3.3 Brownian motion of a grain
  • 9.3.4 The disorder time
  • 9.3.5 Laminar and turbulent friction
  • 9.3.6 A falling rain drop
  • 9.3.7 The Poynting–Robertson effect
  • 9.4 Grain destruction
  • 9.4.1 Mass balance in the Milky Way
  • 9.4.2 Destruction processes
  • 9.5 Grain formation
  • 9.5.1 Evaporation temperature of dust
  • 9.5.2 Vapor pressure of small grains
  • 9.5.3 Critical saturation
  • 9.5.4 Equations for time-dependent homogeneous nucleation
  • 9.5.5 Equilibrium distribution and steady-state nucleation
  • 9.5.6 Solutions to time-dependent homogeneous nucleation
  • 9.5.7 Similarity relations
  • 10.1 Efficiency of infinite cylinders
  • 10.1.1 Normal incidence and picket fence alignment
  • 10.1.2 Oblique incidence
  • 10.1.3 Rotating cylinders
  • 10.1.4 Absorption efficiency as a function of wavelength
  • 10.2 Linear polarization through extinction
  • 10.2.1 Effective optical depth and degree of polarization p(λ)
  • 10.2.2 The Serkowski curve
  • 10.2.3 Polarization p(λ) of infinite cylinders
  • 10.2.4 Polarization p(λ) of ellipsoids in the Rayleigh limit
  • 10.2.5 Polarization p(λ) of spheroids at optical wavelengths
  • 10.2.6 Polarization and reddening
  • 10.3 Polarized emission
  • 10.3.2 Infrared emission of spheroids
  • 10.3.3 Polarized emission versus polarized extinction
  • 10.4 Circular polarization
  • 10.4.1 The phase shift induced by grains
  • 10.4.2 The wavelength dependence of circular polarization
  • Grain alignment
  • 11.1 Grain rotation
  • 11.1.1 Euler’s equations for a rotating body
  • 11.1.2 Symmetric tops
  • 11.1.3 Atomic magnet in a magnetic field
  • 11.1.4 Rotational Brownian motion
  • 11.1.5 Suprathermal rotation
  • 11.2 Magnetic dissipation
  • 11.2.1 Diamagnetism
  • 11.2.2 Paramagnetism
  • 11.2.3 Ferromagnetism
  • 11.2.4 The magnetization of iron above and below the Curie point
  • 11.2.6 The magnetic susceptibility for spin–lattice relaxation
  • 11.2.7 The magnetic susceptibility in spin–spin relaxation
  • 11.3 Magnetic alignment
  • 11.3.1 A rotating dipole in a magnetic field
  • 11.3.2 Timescales for alignment and disorder
  • 11.3.3 Super-paramagnetism
  • 11.3.4 Ferromagnetic relaxation
  • 11.3.6 Mechanical and magnetic damping
  • 11.4 Non-magnetic alignment
  • 11.4.1 Gas streaming
  • 11.4.2 Anisotropic illumination
  • PAHs and spectral features of dust
  • 12.1 Thermodynamics of PAHs
  • 12.1.1 What are PAHs?
  • 12.1.2 Microcanonic emission of PAHs
  • 12.1.3 The vibrational modes of anthracene
  • 12.1.4 Microcanonic versus thermal level population
  • 12.1.5 Does an ensemble of PAHs have a temperature?
  • 12.2 PAH emission
  • 12.2.1 Photoexcitation of PAHs
  • 12.2.2 Cutoff wavelength for electronic excitation
  • 12.2.3 Photo-destruction and ionization
  • 12.2.4 Cross sections and line profiles of PAHs
  • 12.3 Big grains and ices
  • 12.3.1 The silicate features and the band at 3.4 µm
  • 12.3.2 Icy grain mantles
  • 12.4 An overall dust model
  • 12.4.1 The three dust components
  • 12.4.2 Extinction coefficient in the diffuse medium
  • 12.4.3 Extinction coefficient in protostellar cores
  • Radiative transport
  • 13.1 Basic transfer relations
  • 13.1.1 Radiative intensity and flux
  • 13.1.2 The transfer equation and its formal solution
  • 13.1.3 The brightness temperature
  • 13.1.4 The main-beam-brightness temperature of a telescope
  • 13.2 Spherical clouds
  • 13.2.1 Moment equations for spheres
  • 13.2.2 Frequency averages
  • 13.2.3 Differential equations for the intensity
  • 13.2.4 Integral equations for the intensity
  • 13.2.5 Practical hints
  • 13.3 Passive disks
  • 13.3.1 Radiative transfer in a plane parallel layer
  • 13.3.2 The grazing angle in an inflated disk
  • 13.4 Galactic nuclei
  • 13.4.1 Hot spots in a spherical stellar cluster
  • 13.4.2 Low and high luminosity stars
  • 13.5 Line radiation
  • 13.5.1 Absorption coefficient and absorption profile
  • 13.5.2 The excitation temperature of a line
  • 13.5.3 Radiative transfer in lines
  • Diffuse matter in the Milky Way
  • 14.1 Overview of the Milky Way
  • 14.1.1 Global parameters
  • 14.1.2 The relevance of dust
  • 14.2 Molecular clouds
  • 14.2.1 The CO molecule
  • 14.2.2 Population of levels in CO
  • 14.2.3 Molecular hydrogen
  • 14.2.4 Formation of molecular hydrogen on dust surfaces
  • 14.3 Clouds of atomic hydrogen
  • 14.3.1 General properties of the diffuse gas
  • 14.3.2 The 21 cm line of atomic hydrogen
  • 14.3.3 How the hyperfine levels of atomic hydrogen are excited
  • 14.3.4 Gas density and temperature from the 21 cm line
  • 14.3.5 The deuterium hyperfine line
  • 14.3.6 Electron density and magnetic field in the diffuse gas
  • 14.4 HII regions
  • 14.4.1 Ionization and recombination
  • 14.4.2 Dust–free HII regions
  • 14.4.3 Dusty HII regions
  • 14.4.4 Bremsstrahlung
  • 14.4.5 Recombination lines
  • 14.5 Mass estimates of interstellar clouds
  • 14.5.1 From optically thin CO lines
  • 14.5.2 From the CO luminosity
  • 14.5.3 From dust emission
  • Stars and their formation
  • 15.1 Stars on and beyond the main sequence
  • 15.1.1 Nuclear burning and the creation of elements
  • 15.1.2 The binding energy of an atomic nucleus
  • 15.1.3 Hydrogen burning
  • 15.1.4 The 3α process
  • 15.1.5 Lifetime and luminosity of stars
  • 15.1.6 The initial mass function
  • 15.2 Clouds near gravitational equilibrium
  • 15.2.1 Virialized clouds
  • 15.2.2 Isothermal cloud in pressure equilibrium
  • 15.2.3 Structure and stability of Ebert–Bonnor spheres
  • 15.2.4 Free-fall of a gas ball
  • 15.2.5 The critical mass for gravitational instability
  • 15.2.6 Implications of the Jeans criterion
  • 15.2.7 Magnetic fields and ambipolar diffusion
  • 15.3 Gravitational collapse
  • 15.3.1 The presolar nebula
  • 15.3.2 Hydrodynamic collapse simulations
  • 15.3.3 Similarity solutions of collapse
  • 15.4 Disks
  • 15.4.1 Viscous laminar flows
  • 15.4.2 Dynamical equations of the thin accretion disk
  • 15.4.3 The Kepler disk
  • 15.4.4 Why a star accretes from a disk
  • 15.4.5 The stationary accretion disk
  • 15.4.6 The α-disk
  • 15.4.7 Disk heating by viscosity
  • Emission from young stars
  • 16.1 The earliest stages of star formation
  • 16.1.1 Globules
  • 16.1.2 Isothermal gravitationally-boundclumps
  • 16.2 The collapse phase
  • 16.2.1 The density structure of a protostar
  • 16.2.2 Dust emission from a solar-type protostar
  • 16.2.3 Kinematics of protostellar collapse
  • 16.3 Accretion disks
  • 16.3.1 A flat blackbody disk
  • 16.3.2 A flat non-blackbody disk
  • 16.3.3 Radiative transfer in an inflated disk
  • 16.4 Reflection nebulae
  • 16.5 Cold and warm dust in galaxies
  • 16.6 Starburst nuclei
  • 16.6.1 Repetitive bursts of star formation
  • 16.6.2 Dust emission from starburst nuclei
  • References
  • Index

The Physics of Interstellar Dust

Series in Astronomy and Astrophysics
Series Editors: M Birkinshaw, University of Bristol, UK M Elvis, Harvard–Smithsonian Center for Astrophysics, USA J Silk, University of Oxford, UK The Series in Astronomy and Astrophysics includes books on all aspects of theoretical and experimental astronomy and astrophysics. Books in the series range in level from textbooks and handbooks to more advanced expositions of current research. Other books in the series Dark Sky, Dark Matter J M Overduin and P S Wesson Dust in the Galactic Environment, 2nd Edition D C B Whittet An Introduction to the Science of Cosmology D J Raine and E G Thomas The Origin and Evolution of the Solar System M M Woolfson The Physics of the Interstellar Medium J E Dyson and D A Williams Dust and Chemistry in Astronomy T J Millar and D A Williams (eds) Observational Astrophysics R E White (ed) Stellar Astrophysics R J Tayler (ed) Forthcoming titles Very High Energy Gamma Ray Astronomy T Weekes Numerical Methods in Astrophysics P Bodenheimer, G Laughlin, M Rozyczka and H W Yorker

Series in Astronomy and Astrophysics

The Physics of Interstellar Dust

Endrik Kr¨ gel u
Max-Planck-Institut f¨ r Radioastronomie, u Bonn, Germany

Institute of Physics Publishing Bristol and Philadelphia

c IOP Publishing Ltd 2003 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the publisher. Multiple copying is permitted in accordance with the terms of licences issued by the Copyright Licensing Agency under the terms of its agreement with Universities UK (UUK). British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. ISBN 0 7503 0861 3 Library of Congress Cataloging-in-Publication Data are available

Series Editors: M Birkinshaw, University of Bristol, UK M Elvis, Harvard–Smithsonian Center for Astrophysics, USA J Silk, University of Oxford, UK Commissioning Editor: John Navas Production Editor: Simon Laurenson Production Control: Sarah Plenty Cover Design: Victoria Le Billon Marketing: Nicola Newey and Verity Cooke Published by Institute of Physics Publishing, wholly owned by The Institute of Physics, London Institute of Physics Publishing, Dirac House, Temple Back, Bristol BS1 6BE, UK US Office: Institute of Physics Publishing, The Public Ledger Building, Suite 929, 150 South Independence Mall West, Philadelphia, PA 19106, USA
A Typeset in LTEX 2ε by Text 2 Text, Torquay, Devon Printed in the UK by MPG Books Ltd, Bodmin, Cornwall

F¨ r meine Frau u

This page intentionally left blank


Preface 1 The dielectric permeability 1.1 Maxwell’s equations 1.1.1 Electric field and magnetic induction 1.1.2 Electric polarization of the medium 1.1.3 The dependence of the dielectric permeability on direction and frequency 1.1.4 The physical meaning of the electric susceptibility χ 1.1.5 Magnetic polarization of the medium 1.1.6 The magnetic susceptibility 1.1.7 Dielectrics and metals 1.1.8 Free charges and polarization charges 1.1.9 The field equations 1.2 Waves in a dielectric medium 1.2.1 The wave equation 1.2.2 The wavenumber 1.2.3 The optical constant or refractive index 1.2.4 Energy dissipation of a grain in a variable field 1.3 The harmonic oscillator 1.3.1 The Lorentz model 1.3.2 Free oscillations 1.3.3 The general solution to the oscillator equation 1.3.4 Dissipation of energy in a forced oscillation 1.3.5 Dissipation of energy in a free oscillation 1.3.6 The plasma frequency 1.3.7 Dispersion relation of the dielectric permeability 1.4 The harmonic oscillator and light 1.4.1 Attenuation and refraction of light 1.4.2 Retarded potentials of a moving charge 1.4.3 Emission of an harmonic oscillator 1.4.4 Radiation of higher order 1.4.5 Radiation damping

xix 1 1 1 2 3 4 6 7 7 8 9 10 10 11 12 13 15 15 16 17 18 19 20 20 22 23 24 26 27 28


1.4.6 The cross section of an harmonic oscillator 1.4.7 The oscillator strength 1.4.8 The natural linewidth Waves in a conducting medium 1.5.1 The dielectric permeability of a conductor 1.5.2 Conductivity and the Drude profile 1.5.3 Electromagnetic waves in a plasma with a magnetic field 1.5.4 Group velocity of electromagnetic waves in a plasma Polarization through orientation 1.6.1 Polarization in a constant field 1.6.2 Polarization in a time-variable field 1.6.3 Relaxation after switching off the field 1.6.4 The dielectric permeability in Debye relaxation 29 30 31 32 33 34 36 37 38 38 39 40 41 44 44 44 46 47 47 47 50 51 52 52 54 54 56 57 57 57 58 60 61 62 64 64 66 67 67 68 69 71 72




How to evaluate grain cross sections 2.1 Defining cross sections 2.1.1 Cross section for scattering, absorption and extinction 2.1.2 Cross section for radiation pressure 2.1.3 Efficiencies, mass and volume coefficients 2.2 The optical theorem 2.2.1 The intensity of forward scattered light 2.2.2 The refractive index of a dusty medium 2.3 Mie theory for a sphere 2.3.1 The generating function 2.3.2 Separation of variables 2.3.3 Series expansion of waves 2.3.4 Expansion coefficients 2.3.5 Scattered and absorbed power 2.3.6 Absorption and scattering efficiencies 2.4 Polarization and scattering 2.4.1 The amplitude scattering matrix 2.4.2 Angle-dependence of scattering 2.4.3 The polarization ellipse 2.4.4 Stokes parameters 2.4.5 Stokes parameters of scattered light for a sphere 2.5 The Kramers–Kronig relations 2.5.1 Mathematical formulation of the relations 2.5.2 The electric susceptibility and causality 2.5.3 The Kramers–Kronig relation for the dielectric permeability 2.5.4 Extension to metals 2.5.5 Dispersion of the magnetic susceptibility 2.5.6 Three corollaries of the KK relation 2.6 Composite grains 2.6.1 Effective medium theories

2.6.2 2.6.3 2.6.4 2.6.5 3 Garnett’s mixing rule The mixing rule of Bruggeman Composition of grains in protostellar cores How size, ice and porosity change the absorption coefficient

ix 73 74 74 76

Very small and very big particles 80 3.1 Tiny spheres 80 3.1.1 When is a particle in the Rayleigh limit? 80 3.1.2 Efficiencies of small spheres from Mie theory 81 3.1.3 A dielectric sphere in a constant electric field 82 3.1.4 Scattering and absorption in the electrostatic approximation 84 3.1.5 Polarization and angle-dependent scattering 85 3.1.6 Small-size effects beyond Mie theory 86 3.2 A small metallic sphere in a magnetic field 87 3.2.1 Slowly varying field 87 3.2.2 The magnetic polarizability 89 3.2.3 The penetration depth 89 3.2.4 Limiting values of the magnetic polarizability 90 3.3 Tiny ellipsoids 90 3.3.1 Elliptical coordinates 91 3.3.2 An ellipsoid in a constant electric field 92 3.3.3 Cross section and shape factor 93 3.3.4 Randomly oriented ellipsoids 95 3.3.5 Pancakes and cigars 95 3.3.6 Rotation about the axis of greatest moment of inertia 97 3.4 The fields inside a dielectric particle 99 3.4.1 Internal field and depolarization field 99 3.4.2 Depolarization field and the distribution of surface charges 100 3.4.3 The local field at an atom 101 3.4.4 The Clausius–Mossotti relation 101 3.5 Very large particles 103 3.5.1 Babinet’s theorem 103 3.5.2 Reflection and transmission at a plane surface 104 3.5.3 Huygens’ principle 106 3.5.4 Fresnel zones and a check on Huygens’ principle 109 3.5.5 The reciprocity theorem 111 3.5.6 Diffraction by a circular hole or a sphere 111 3.5.7 Diffraction behind a half-plane 113 3.5.8 Particles of small refractive index 116 3.5.9 X-ray scattering 117

x 4

Case studies of Mie calculus 4.1 Efficiencies of bare spheres 4.1.1 Pure scattering 4.1.2 A weak absorber 4.1.3 A strong absorber 4.1.4 A metal sphere 4.1.5 Efficiency versus cross section and volume coefficient 4.1.6 The atmosphere of the Earth 4.2 Scattering by bare spheres 4.2.1 The scattering diagram 4.2.2 The polarization of scattered light 4.2.3 The intensity of scattered light in a reflection nebula 4.3 Coated spheres 4.4 Surface modes in small grains 4.5 Efficiencies of idealized dielectrics and metals 4.5.1 Dielectric sphere consisting of identical harmonic oscillators 4.5.2 Dielectric sphere with Debye relaxation 4.5.3 Magnetic and electric dipole absorption of small metal spheres 4.5.4 Efficiencies for Drude profiles 4.5.5 Elongated metallic particles Particle statistics 5.1 Boltzmann statistics 5.1.1 The probability of an arbitrary energy distribution 5.1.2 The distribution of maximum probability 5.1.3 Partition function and population of energy cells 5.1.4 The mean energy of harmonic oscillators 5.1.5 The Maxwellian velocity distribution 5.2 Quantum statistics 5.2.1 The unit cell h 3 of the phase space 5.2.2 Bosons and fermions 5.2.3 Bose statistics 5.2.4 Bose statistics for photons 5.2.5 Fermi statistics 5.2.6 Ionization equilibrium and the Saha equation 5.3 Thermodynamics 5.3.1 The ergodic hypothesis 5.3.2 Definition of entropy and temperature 5.3.3 The canonical distribution 5.3.4 Thermodynamic relations for a gas 5.3.5 Equilibrium conditions of the state functions 5.3.6 Specific heat of a gas 119 119 119 120 122 123 123 126 127 127 128 131 132 133 136 136 138 139 141 142 145 145 145 146 147 149 149 151 151 152 154 156 157 158 160 160 162 163 164 166 168


3.4 Spectral indices 7 .2 Selection rules and polarization rules The work done by magnetization 5.4 Potential wells and tunneling 6.1 A charged particle in an electromagnetic field 6.1 Translational symmetry Low.1.2.1 Covalent bonding Stellar photometry 7.2.1 Crystal structure 7.2 Small perturbations 6.1.2 The Hamiltonian of an electron in an electromagnetic field 6.4 6 The radiative transition probability 6.3 The Hamilton operator in quantum mechanics 6.1 The Planck function 5.2 The transition probability 6.3 Rectangular potential barriers 6.2 Binding in crystals 7.2.4 The Planck function and harmonic oscillators xi 168 169 170 170 171 172 173 175 175 175 176 177 179 179 181 181 181 182 183 183 186 186 188 192 192 192 194 198 201 201 201 203 207 207 208 209 211 213 214 214 216 219 220 The Einstein coefficients A and B 6.1 The perturbation energy 6.8 Susceptibility and specific heat of magnetic substances Blackbody radiation 5.3 Wien’s displacement law and the Stefan–Boltzmann law Lattice types 7.1 Induced and spontaneous transitions 6.2 Potential walls and Fermi energy 6.4 The dipole moment in quantum mechanics 6.4.and high-frequency limit 5.1.3 Quantization of the electromagnetic field 6.3 The reciprocal lattice 7.4 Quantum-mechanical derivation of A and B 6.3 Transition probability for a time-variable perturbation 6.1 The classical Hamiltonian 6.1.4 van der Waals forces and hydrogen bridges 7.2 The interstellar extinction curve 7.1 Wavefunction of a particle in a constant potential 6.Contents 5.3 Reddening by interstellar grains 7.3 Two-color diagrams 7.4.5 The quantized harmonic oscillator 6.3 Metals 7.3.4 The double potential well Structure and composition of dust 7.2 Ionic bonding 7.3.

3 Standing waves in a crystal 8.3.1 Small and moderate fluctuations 8.6.4 The stochastic time evolution of grain temperature 8.1 Kirchhoff’s law 8.5 The mass absorption coefficient Carbonaceous grains and silicate grains 7.3 Temperature estimates 8.6 Two-dimensional lattices 8.1 Normal coordinates 8.3 Practical considerations 7.3 The emission spectrum of big grains 8.1 The size distribution 7.5 Temperature fluctuations of very small grains 8.xii Contents The emissivity of dust 8.2 Strong fluctuations 8.1.2 The bonding in carbon 7.5 A standard set of optical constants Grain sizes and optical constants Specific heat 8.3 Absorption and emission in thermal equilibrium 8.4.2 Thermal emission of grains 8.2 The transition matrix 8.5.4 Relation between grain size and grain temperature The energy equation 8.5 8 Dust radiation 8.5.2 Internal energy of a grain 8.1 Constant temperature and low optical depth Collisional fragmentation 222 224 224 225 227 232 233 234 234 236 239 239 239 240 241 242 243 243 243 245 247 248 249 251 251 253 254 254 256 257 258 259 261 262 263 263 265 266 268 268 270 272 7.1.2 The temperature of big grains 8.4.6 Dust temperatures from observations 8.4 The density of vibrational modes in a crystal Equipartition of energy 8.3.1 The probability density P(T ) 8.3 Temperature fluctuations and flux ratios .2.6 The emission spectrum of very small grains 8.1 Origin of the two major dust constituents Temperature of dust grains near a star 8.5.4 Silicates 7.4.2 Approximate absorption efficiency at infrared wavelengths 8.4 Calorific properties of solids 8.2 Constant temperature and arbitrary optical depth 8.6.3 Carbon compounds 7.

1 Grain surfaces Normal incidence and picket fence alignment 10.1 Random walk 9.3 Rotating cylinders 10.6 Solutions to time-dependent homogeneous nucleation 9.4 The disorder time 9.2 Vapor pressure of small grains 9.3 Brownian motion of a grain 9.2 The photoelectric effect 9.4 Absorption efficiency as a function of wavelength 10.3.4 Grain destruction 9.5 Tunneling between surface sites 9.2 Physical adsorption and chemisorption Similarity relations xiii 275 275 275 276 279 281 283 284 285 285 286 289 289 289 292 293 295 296 297 298 298 299 301 301 304 305 307 308 311 316 319 319 319 322 322 325 327 327 329 331 334 337 10 Polarization Polarization p(λ) of spheroids at optical wavelengths .3 Grain motion A falling rain drop 9.2 Grain charge 9.6 Scanning time 9.2.1 Efficiency of infinite cylinders 10.3 Polarization p(λ) of infinite cylinders 10.1 Effective optical depth and degree of polarization p(λ) 10.5 Laminar and turbulent friction 9.4 Polarization p(λ) of ellipsoids in the Rayleigh limit 10.2 Destruction processes 9.5.1 Evaporation temperature of dust 9.2 Linear polarization through extinction 10.5 Equilibrium distribution and steady-state nucleation 9.1 Charge equilibrium in the absence of a UV radiation field 9.2 Oblique incidence Gas accretion on grains 9.2.4 Thermal hopping.3 The sticking probability Equations for time-dependent homogeneous nucleation The Poynting–Robertson effect 9.1 Mass balance in the Milky Way 9.1.3 Critical saturation 9.Contents 9 Dust and its environment evaporation and reactions with activation barrier 9.1.5 Grain formation 9.2 The Serkowski curve 10.2 The drag on a grain subjected to a constant outer force 9.

1 Diamagnetism 355 11.1 What are PAHs? 12.2.2 Anisotropic illumination 375 12 PAHs and spectral features of dust 12.1 Grain rotation 347 11.6 Polarization and reddening 10.2.5 Alignment of angular momentum with the axis of greatest inertia 371 11.6 The magnetic susceptibility for spin–lattice relaxation 360 Magnetic alignment 364 11.4 The magnetization of iron above and below the Curie point 358 11.1 The wavelength dependence of polarized emission for cylinders 10.1.2 Infrared emission of spheroids 10.7 The magnetic susceptibility in spin–spin relaxation 362 The phase shift induced by grains 10.3 Super-paramagnetism 368 11.3.4 Rotational Brownian motion 351 11.5 Does an ensemble of PAHs have a temperature? 377 377 377 378 379 381 382 .3 Polarized emission 10.2 Timescales for alignment and disorder 367 Non-magnetic alignment 373 11.2 Magnetic dissipation 355 11.1.3 Polarized emission versus polarized extinction 10.1.5 Suprathermal rotation 353 11.4.xiv Contents Paramagnetism 355 The vibrational modes of anthracene 12.4 Ferromagnetic relaxation 369 11.1 Gas streaming 373 Mechanical and magnetic damping 372 11.2 The wavelength dependence of circular polarization 338 339 340 340 341 342 343 344 11 Grain alignment 347 11.5 Paramagnetic dissipation: spin–spin and spin–lattice relaxation 359 11.3 Atomic magnet in a magnetic field 351 11.2 Microcanonic emission of PAHs 12.2 Symmetric tops 349 11.1 Euler’s equations for a rotating body 347 11.3 Ferromagnetism 357 Thermodynamics of PAHs 12.3.4 Microcanonic versus thermal level population 12.1 A rotating dipole in a magnetic field 365 11.4 Circular polarization 10.4. The three dust components Hot spots in a spherical stellar cluster PAH emission An overall dust model 12.1.Contents 12.3 Big grains and ices 12.1 Photoexcitation of PAHs 12.4 Cross sections and line profiles of PAHs 12.1 Overview of the Milky Way 14.1 Absorption coefficient and absorption profile 13.2.2 Icy grain mantles Radiative intensity and flux 13.4.3 Extinction coefficient in protostellar cores 13 Radiative transport 13.4.2 The relevance of dust 14.1 Radiative transfer in a plane parallel layer 13.5 Practical hints Cutoff wavelength for electronic excitation 12.3.3 Molecular hydrogen 14.1 The silicate features and the band at 3.2 Low and high luminosity stars Spherical clouds 13.1 Moment equations for spheres 13.2 Frequency averages 13.2 The excitation temperature of a line 13.3 Photo-destruction and ionization 12.4 Integral equations for the intensity 13.3 Differential equations for the intensity 13.1 The CO molecule The transfer equation and its formal solution 13.2 The grazing angle in an inflated disk 13.2 Population of levels in CO 14.4 The main-beam-brightness temperature of a telescope 13.5 Line radiation Basic transfer relations 13.3 Passive disks 13.1 Global parameters 14.4 Galactic nuclei 13.3 The brightness temperature 13.3 Radiative transfer in lines 14 Diffuse matter in the Milky Way 14.2 Molecular clouds Extinction coefficient in the diffuse medium 12.2.4 Formation of molecular hydrogen on dust surfaces xv 384 384 385 386 387 388 389 389 390 392 395 395 396 396 396 398 400 401 402 403 404 405 407 407 409 409 414 415 415 416 418 418 419 420 425 425 425 426 427 428 431 435 435 .5.4 µm 12.

4.4 Disks 15.1 Virialized clouds 15.1 The presolar nebula 15.5 Recombination lines 14.3 Similarity solutions of collapse 15.3.2 The 21 cm line of atomic hydrogen The binding energy of an atomic nucleus 15.3.5 The stationary accretion disk .6 Implications of the Jeans criterion 15.4 Free-fall of a gas ball 15.2.4 Why a star accretes from a disk 15.2.4 The 3α process 15.2 Hydrodynamic collapse simulations 15.1 Nuclear burning and the creation of elements 15.5 The critical mass for gravitational instability 15.1 Viscous laminar flows 15.3 Dusty HII regions 14.2 Dust–free HII regions Gravitational collapse 15.5 The deuterium hyperfine line 14.6 The initial mass function 15.xvi Contents 14.5 Mass estimates of interstellar clouds 14.3 Structure and stability of Ebert–Bonnor spheres 15.4.5 Lifetime and luminosity of stars Bremsstrahlung 14.3 How the hyperfine levels of atomic hydrogen are excited 14.4 HII regions 14.7 Magnetic fields and ambipolar diffusion 15.3 The Kepler disk General properties of the diffuse gas 14.6 Electron density and magnetic field in the diffuse gas 14.3 From dust emission 438 438 439 440 443 444 446 448 448 450 453 455 456 457 457 458 459 461 461 461 463 465 467 469 470 471 471 474 475 479 480 482 484 486 486 487 491 494 494 497 498 499 501 15 Stars and their formation 15.2 Clouds near gravitational equilibrium 15.2 Dynamical equations of the thin accretion disk Stars on and beyond the main sequence 15.1.3 Hydrogen burning Ionization and recombination 14.3 Clouds of atomic hydrogen From the CO luminosity 14.1.1 From optically thin CO lines Gas density and temperature from the 21 cm line Isothermal cloud in pressure equilibrium 15.

2 A flat non-blackbody disk 16.1 Repetitive bursts of star formation 16.1.6 The α-disk 15.3.1 The density structure of a protostar 16.2.5 Cold and warm dust in galaxies 16.Contents 15.6.1 A flat blackbody disk 16.3 Accretion disks 16.2.2 The collapse phase 16.3 Kinematics of protostellar collapse 16.3.2 Dust emission from a solar-type protostar Isothermal gravitationally-bound clumps 16.1.4 Reflection nebulae 16.1 The earliest stages of star formation 16.3 Radiative transfer in an inflated disk Globules 16.6 Starburst nuclei 16.2 Dust emission from starburst nuclei Appendix A Appendix B References Index Mathematical formulae List of symbols xvii 501 503 505 505 505 506 508 508 513 515 518 518 521 522 524 526 531 531 535 539 545 549 552 .7 Disk heating by viscosity 16 Emission from young stars 16.

This page intentionally left blank .

it does not catch our attention as does a variable star. We know we can trust his goods.Preface Dear reader Before you is a compilation of lectures held at the University of Bonn all revolving around interstellar dust and the formation of stars. From lecture notes to print The incentive to turn my scribbled lecture notes into a book was twofold: the desire to reach a larger audience. we esteem a book only at second sight. over a steaming pot of tea or the curly smoke rings of a pipe. a comet or a globular cluster. are disappointingly small and would xix . Instead. He is aggressive. Lecture and written text. In a lecture. Its constituents. the grains. even when covering the same topic. however. Whereas an opinion about a lecture is formed quickly and is not likely to change afterwards. and the wish to hand students a more polished and lasting description of the tools for future work. The speaker stresses ideas and concepts and does not waste time in elaborating lengthy formulae. should not be identical in their contents. formula or paragraph at leisure. A serious writer. but complementary: they are two independent didactical challenges. his arguments are compelling and what he says sounds exciting which prevents us from slamming the door in his face. most of all through thoroughness. although he himself may be a bit boring. Dust only screens the light at optical wavelengths. there is opportunity to contemplate a figure. can convince only by more subtle tones. He is like the unobtrusive shopkeeper whom we have been visiting for years. the student should be able to follow from beginning to end. The topic The central theme of this book is cosmic dust. Not every chapter has to be grasped at first reading. Unless we use special equipment. more sophisticated than binoculars. A good lecturer may be likened to a salesman at the front door. Its relevance for astronomy and for the evolution of the cosmos is not obvious.

But there is more to it. it seems justified to consider astronomy a branch of Space Project Management. when one recalls how astronomical satellites have widened our outlook on the universe. Another reason to concentrate on matters where there is consensus and to avoid being specific or touching upon controversial topics lies in the very nature of the dust itself. physically founded and adapted to the particular problem. giving and taking. or should. nor for fear that excitement might be detrimental to the heart. It is also aimed at PhD students who are starting research and have come across interstellar dust in one of its many manifestations. However. As many of the basic facts are scattered over the literature and are sometimes hard to dig up. not because a book is never fresh. we believe that man is made of this very dust and that. it is in intimate contact with the rest of the world. like pulsars or white dwarfs. The addressee This text was conceived for students who have received an elementary but comprehensive introduction to physics—this is usually the case after two years of university studies—and who have taken a general course in astronomy. and when one thinks of the progress achieved by new telescopes. When an astronomical number is certain. be stretched. astronomy appears as a subfield of Telescope Engineering or Receiver Development. Instead. or come up with a new number. why all numbers about interstellar dust are vague and insecure.xx Preface barely be visible under a microscope. by itself. Only being acquainted with the old results can one fully enjoy the new. and this is best exemplified by the influence it has on star formation and on the appearance of young stars and galaxies. I admit that it contains hardly any exciting new results. say. one should know the physical and technical pillars upon which its derivation rests. The total dust mass in the Milky Way is negligible compared to that of the stars or even the interstellar gas. Hardly any two dust grains in the universe are alike and this immense diversity explains. Astronomy is a branch of physics This is a provocative statement and may arouse indignation. But when the number is ill determined. the mass of a planet or the distance to a star. to a large degree. Instead. the goal was to supply the student with those basic facts about small solid particles that passed the test of time. one can happily apply it in further work without worrying how it was derived. this book will also be of service to astronomers in general. Only then can one estimate how far it may. a selected compilation was thought to be useful. Hopefully. Interstellar dust is not an isolated component of the universe. which could be removed and everything else would stay the same. is reason to study it in depth. It was new- . As if I had forgotten how the great discoveries of the past 30 years have come about: As a result of revolutionary technologies and grand enterprises! Indeed.

out of sheer pleasure for the subject. German would have been beneficial to the reader. too. how many more are required to arrive at it? What would quantum mechanics be if it were reduced to equations .Preface xxi technology instruments that have allowed us to peep into hitherto hidden realms. As English has become so pervasive in our daily scientific work. I consider it a trifle that no sentence came easy. it did me good learning some more of a foreign language while composing the text. and each path offers its specific outlook. A serious argument against writing in a tongue one has not fully mastered is that style and clarity are akin because improving the style usually means improving the thought. But the strongest point in a plea to retain in science one’s mother tongue in all its might. Despite all this. A formula lacks cross relations and does not sufficiently take into account the analogous character of what it asserts. astronomy would continue to thrive for a decade or two. For example. without polemizing against higher resolution or higher sensitivity. Words are needed. If words are needed to explain a formula. After all. More important still is the obligation to preserve and develop one’s language as an inherited gift and an attribute of culture of no less import than the collection of national wines. nothing else. So I freely acknowledge the priority of management and technology over physics. The obvious needs no words but there are lamentable aspects about using the modern lingua franca. Pondering a problem in different languages means approaching it on different paths. Nor do I mind that one suspects behind simple phrases a simple mind. Even ADM (Advanced Data Manipulation) may be more important to astronomy than physics in view of the gigantic quantities of data that have to be crunched and transformed into convincing numbers and pictures. It attempts to summarize the major physical topics with direct application to interstellar grains and wishes to encourage students to try the physical approach to an astronomical problem. the majority of astronomers. even for three. A poignant style enhances the understanding. helps memorize and carries the reader over difficult stretches. tend to forget technical terms in our mother tongue or do not update them and this has the deplorable consequence that we speak and write about our favourite subject clumsily in two languages: in English and in our own. we. is that each language imprints on the mind its own pattern of thinking. this supposition may be true. The language It is obviously English. this book stresses the link between astronomy and physics. in this respect. parallel to the lingua franca. I solve equations containing time but do not very well know what time is. If one were to reduce the physics in astronomy courses to a minimum (one cannot do entirely without it) and teach instead the fields mentioned earlier. It is erroneous to think that the findings of natural sciences are fully expressed in numbers or formulae. a textbook on physical sciences is not a railway timetable. Ach. Indeed. if one includes Science Marketing.

his words carry weight because they shed light on the topic from a different angle of cognition and reassure the natural scientist in his moments of doubt.xxii Preface without extensive accompanying text? Who would shorten R Feynman’s Lectures on Physics? They are the work of a genius not because of the formulae. to add. Nikolaj Voshchinnikov. the figures. The philosophical headline A long scientific text is frequently preceded. one outstanding merit: L Spitzer’s Diffuse Matter in Space is of dazzling perfection. When I sitting heard the astronomer where he lectured with much applause in the lecture-room. Johannes Schmid-Burgk and Alexandr Tutukov. divide. Aigen Li. Werner Tscharnuter. To whom I am indebted I owe to my colleagues who bore with me. I found numerous aphorisms suitable for such a purpose but the most appropriate headlines for this book came to me as a birthday gift from my daughters. they are the usual ones but because of the way the story is told. Michaela Kraus. Although far from being an expert in the scientific subject itself. precise and powerful language to put all its facets into a fruitful perspective. helped with their expertise and advice and encouraged me. such as a philosopher or a poet. Richard Porcas. How soon unaccountable I became tired and sick. Two books served as guides (Vorbilder) which I tried to follow. And a successful struggle with an astronomical problem also needs a vivid. and measure them. When the proofs. It has been on my desk for decades (and I am still struggling with parts of it today). Armin Kirfel. Till rising and gliding out I wandered off by myself . Malcolm Walmsley and Jan Martin Winters. one might even say embellished. to my mind. Dabbling in poetry and philosophical treatises. I am grateful to those who undertook the pains of critically reading parts of the manuscript: Christian Henkel. without pretending to match them. Antonella Natta. in particular: David Graham. were ranged in columns before me. It is the following verse by the 19th century North-American poet Walt Whitman which they had calligraphically written onto cardboard. Here is what Whitman left us: When I heard the learn’d astronomer. Each has. When I was shown the charts and diagrams. with the eyes of a physicist. M Harwit pioneered in his Astrophysical Concepts to teach astronomy anew. Ralf Siebenmorgen. I wish to follow this custom. by words from an authority outside the field. addressing the student and enlightening the professor.

corroborating. at least. condenses into anger and transforms into serenity. a line comes to mind by F M Dostojevskji from his novel The Demons. Some colleagues strongly disagree and regard with contempt those who let themselves be charmed by such a poem. Nor will I marvel at his prophetic power to foresee and congenially describe the feelings of a backbencher at an astronomical meeting more than a century after his time. being a layman in literature. Reflected or not. I take their objections very serious but find the occasional vehemence of their arguments soothing. like the mystical moist night-air. At the end of this longish foreword. that they are not moved by logic and astronomical data alone. irrationality. I venture in the present context the translation: Hmm well. Stepan Verchovenskji. Looked up in perfect silence at the stars. including the quest of an astronomer to understand the cosmos. and from time to time. The reason for picking this poem as the philosophical headline is that it pays a wise tribute to the irrational. is at the root of any sincere endeavour.Preface In the mystical moist night-air. I will not admire how irresistibly Whitman conjures the lure of the night sky and contrasts it to the unnerving ambition of scholars. So I will not say a word about the magic beat that pervades the poem: How the rhythm starts from impatience. any literary praise of these lines from my side is out of place. At a benefit party. well hmm! Let this be the concluding remark. xxiii Of course. Yours sincerely EK Bonn Easter 2002 . makes an ambitious opening speech which Dostojevskji laconically summarizes by the words ÆÓ Ú× Ý «ØÓ Ò Õ Ó¸ ØÓ Ò ÞÒ Ø ÚØÓÖ× ÔÖ ×ÐÓÚ ˘ After intensive consultations with linguists and psychologists. the aging hero of the narrative.

This page intentionally left blank .

including the absorption. therefore. scattering and emission of light by individual dipoles. and a metal where the electrons are free. We summarize the properties of harmonic oscillators. their names and meaning are not universal. a dielectric medium where the charge distribution in the atomic dipoles is fixed but the dipoles themselves may rotate. are Maxwell’s formulae.Chapter 1 The dielectric permeability We begin by acquainting ourselves with the polarization of matter. we learn how its dielectric permeability changes with frequency. This study is carried out for three scenarios: • • • a dielectric medium where the electron clouds oscillate about the atomic nuclei. We recall the basic equations of electrodynamics and outline how plane waves travel in an infinite non-conducting (dielectric) medium and in a plasma. They can be written in different ways and the symbols.1 Maxwell’s equations At the root of all phenomena of classical electrodynamics. Approximating a solid body by an ensemble of such dipoles (identical harmonic oscillators).1. first define the quantities which describe the electromagnetic field. far from it.1) 1 . 1. 1. we.1 Electric field and magnetic induction A charge q traveling with velocity v in a fixed electric field E and a fixed magnetic field of flux density B experiences a force 1 F=q E+ v×B c (1. The fundamental quantity describing how an interstellar grain responds to an electromagnetic wave is the dielectric permeability which relates the polarization of matter to the applied field. Before we exploit Maxwell’s equations. such as the interaction of light with interstellar dust.

For an electromagnetic wave in vacuum.1. and a magnetic component. we introduce as an additional quantity the displacement D. working perpendicular to v and B. |E| = |B| and.2 The dielectric permeability called the Lorentz force. p (1.4) . Equation (1. therefore. (1. through an electric field E: D = E + 4πP.1 Dielectric permeability and electric susceptibility In an electrically neutral medium. the cross × denotes the vector product.2) The dipole moment per unit volume is called the polarization P (capital letter). the term (v × B)/c in formula (1. V However. V It may be interpreted as the number density N of molecules multiplied by the average electric dipole moment pmol per molecule: P = Npmol . Protons move still more slowly because of their greater inertia. E and P being defined.1. the integral of the local charge density ρ(x) at locus x over any volume V vanishes. a neutral body of volume V may have a dipole moment p (small letter) given by p= V xρ(x) d V. their velocities are small.1) shows what happens mechanically to a charge in an electromagnetic field and we use it to define E and B. (1.2.3) P= . ρ(x) d V = 0.2 Electric polarization of the medium 1. The force F has an electric part. (q/c)v × B. P. When the moving charges are electrons in an atom driven by an electromagnetic wave. qE. 1. A typical value is the velocity of the electron in the ground state of the hydrogen atom for which classically v/c equals the fine structure constant α= e2 c 1 137 1. It adds nothing new and is just another way of expressing the polarization of matter.1) is much smaller than E and usually irrelevant for the motion of the charge. B is also called the magnetic induction. which pulls a positive charge in the direction of E.

ε is called the dielectric permeability or dielectric constant or permittivity. P. Equation (1. The relation between them is described in detail in section 3. they mean more or less the same.8)). For example. the polarization P vanishes. 1. E and P. We have.6) ε−1 . the field in the vicinity of the grain becomes distorted. In the trivial case of a vacuum.6)) and on E (from (1. however.2 The electric polarizability Another quantity we will need is the electric polarizability αe . It no has longer the value E but some other value. the proportionality factors being αe and χ.1. Therefore. P = χEi = αe E. The polarization P (capital letter) depends linearly both on Ei (from (1. although two would suffice. χ bears the name electric susceptibility. In a dielectric medium.Maxwell’s equations 3 The local polarization P and the local average electric field E in the dielectric medium are much weaker than the fields on the atomic level and constitute just a small perturbation. As one recedes from the grain. This differs from the constant outer field E. (1. E = D and ε = 1. altogether. we could replace D in all equations by εE. and the dipole moment and the fields do not have the same direction.3 The dependence of the dielectric permeability on direction and frequency Unless specified otherwise.1. There is also a field Ei inside the grain. (1.9) In the general case. a constant field E induces a dipole moment so that ε > 1 and P = 0. αe and χ are tensors.2. three quantities describing the electric field: D. D and E are proportional to each other and we can write D = εE P = χE with χ= (1.5) (1. If we place a small grain of volume V into a constant electric field E. 1. respectively. it acquires a dipole moment p = αe V E. Ee approaches E asymptotically. (1. we always assume (on a macroscopic level with dimensions much greater than an atom) homogeneity so that the dielectric .7) 4π The proportionality factors ε and χ are material constants.5) is the first of two constitutive relations complementing the set of Maxwell’s formulae.8) Because of the dipole moment. say Ee (e for external).4.

a grain is made up of atoms and is thus not homogeneous. that ε shows dispersion. So the term dielectric constant. The relation (1. It is an essential feature for a discussion of interstellar dust that ε is a function of frequency: ω = 2πν or. D0i is always zero. on a microscopic level. is misleading when taken at its face value. The tensor εik is symmetric. E = E 0 e−iωt . The vector representing the complex field E then rotates in the complex plane clockwise. However.5). This happens only in special cases known as pyroelectricity. For a slowly varying field with small ω.11) . In electrostatics.4 The dielectric permeability permeability ε does not depend on the position in the grain. Crystals of the cubic system have this pleasant property but in interstellar grains we will also encounter anisotropy.1. which is the other frequently used name for ε(ω).5) must then be generalized to Di = i εik E i + D0i . Equation (1. The functional form of ε(ω) is called the dispersion relation and will be used extensively in what follows. which means that ε is independent of the field strength. at most. If the three diagonal elements are equal. ε reduces to a scalar and the grain is said to be isotropic. as depicted in figure 1. in an alternating field. χ1 χ(0) (1. The constant term D0i implies a frozen-in dipole moment even in the absence of an outer field. out of mathematical convenience. it may depend on the direction of the fields because. However. and can always be brought into a diagonal form by the appropriate choice of coordinate system.1. for mono-crystals with a cubic structure (see section 7. the susceptibility χ is close to its static value: χ2 χ2 (0) = 0.1). are written as one complex variable: χ(ω) = χ1 (ω) + i χ2 (ω) ε(ω) = ε1 (ω) + i ε2 (ω). We use scalar notation for the moment and let E 0 be real. D = εE. only one parameter is needed to specify the polarization and χ or ε are real. one needs for each frequency ω two independent numbers which. εik = εki . as one says. 1.10) (1.4 The physical meaning of the electric susceptibility χ We explain the physical meaning of χ in the case of a harmonically and slowly varying field. It then has. three independent components. is linear.

The static value χ1 (0) is positive because the polarization must point in the direction of the electric field. It induces a complex polarization P in the material of length P. therefore. in the medium is presented as a complex quantity of length E. the imaginary part χ2 is positive and determines. The polarization P. Some of the field energy is inevitably drained into the dielectric medium. When damping is weak. which implies dissipational losses. the complex representation is for mathematical convenience. E. the electrons have to overcome internal friction. E and P are not in phase because the electrons always need a little while to adjust. The electric field vector. (1. which rotates at frequency ω. The real part of the complex electric susceptibility is equal to the ratio of the radii of the two circles. P lags behind E by a small angle ϕ. Therefore. χ2 = χ1 ϕ.Maxwell’s equations 5 imaginary P ϕ E real Figure 1. Only the real components of P and E have a physical meaning. If χ2 and t were negative. While adapting to the field. Obviously the real part χ1 determines the maximum polarization during one cycle via the relation Pmax χ1 E 0 .1.12) The interval t must be positive for reasons of causality: sensible people squeal only after they have been pinched. the field would draw energy from the dielectric body. for a given χ1 . the time lag. whereas the imaginary part gives the phase lag. χ1 = P/E. As the field changes. lags behind by some time t ω−1 corresponding to a small angle ϕ = ω t = tan(χ2 /χ1 ) χ2 /χ1 1. decreasing its entropy which contradicts the second law of thermodynamics. Even χ2 = 0 is impossible as it would mean that E and P are absolutely synchronous .

the accompanying field is that of a dipole. The magnetic moment of this moving charge is qvr AI m= = 2c c where A = πr 2 signifies the area of the loop. produces a magnetic moment 1 m= (x × J) d V. again with positive χ2 . M). we could also choose a time dependence eiωt for the field and nothing would change. the magnetization of matter comes. Without the motion of the macroscopic charge. The magnetic field H is defined by H = B − 4πM. it is called the magnetization M. The real part of the susceptibility. although one that must be followed consistently. in the classical picture. (1. the magnetic flux density B as well as the polarizations P and M allow a direct physical interpretation.1 and the susceptibility. We would prefer the pairs (E. would be χ(ω) = χ1 (ω) − i χ2 (ω). however. however.14) For example. we can exclude χ2 ≤ 0 for the same reasons as before. The maximum of P is now given by Pmax = 2 2 χ1 + χ2 E 0 . V (1. may become zero as well as negative. . Of course. P) and (B. If A is small. corresponding to a current density J = ρv. B and M. because the electric field E. a stationary motion of charges with density ρ and velocity v. The sign of i is just a convention. When E = E 0 eiωt . Nevertheless.15) Similar to the situation of the electric field. If the frequency ω is not small but arbitrary. the angle ω t is also not necessarily small.5 Magnetic polarization of the medium In a macroscopic picture. from the atomic currents (electron orbits).6 The dielectric permeability and dissipation is completely absent.10). χ1 .13) 2c When the integral extends over a unit volume.1. 1. we could replace H by B in all equations and retain two field quantities. instead of (1. (1. The magnetic moment per unit volume M may be interpreted as the number density N of molecules in the substance multiplied by the total magnetic moment mmol per molecule: M = Nmmol . the field rotates anticlockwise in figure 1. a charge q traveling with velocity v in a circular orbit of radius r constitutes a current I = qv/2πr . M= m . instead of three. conventions urge us to drag D and H along.

17) χ= 4π so that M = χH. . µ is a little smaller than one. For diamagnetic substances. Metals. which are weakly magnetic substances with |µ| close to one. alternating currents produced by a time-variable electric field are possible. time-varying electric fields and currents are possible. the charges accumulate at its surface and arrange themselves in such a way that the electric field inside vanishes and then there is no internal current. Nevertheless. When this piece of metal is placed in a static electric field. However. • In the interstellar medium. This is a synonym for conductors and. a constant outer magnetic field H induces in a small body of volume V a magnetic dipole moment m = V M = αm V H (1.7 Dielectrics and metals We will be dealing with two kinds of substances for which Maxwell’s equations are often formulated separately: • Dielectrics.8). one must explicitly write χm or χe for the magnetic or electric case. If there is any chance of confusion. If the substance is not ferromagnetic. the magnetic induction and magnetic field are practically the same inside and outside. These are insulators and no constant current can be sustained within them. we define the magnetic susceptibility by µ−1 (1. When a piece of metal is connected at its ends to the poles of a battery. In analogy to (1. µ = µ1 + i µ 2 .18) (1. a linear relation holds as in (1. If B denotes the interstellar magnetic field and M the magnetization of the grain. one finds both dielectric and metallic particles but the latter are probably far from being perfect conductors. 1. In vacuo.Maxwell’s equations 1. µ is large only for ferromagnets.7). B = H and µ = 1. µ is the magnetic permeability and also generally complex.1. For interstellar grains.16) where αm is the magnetic polarizability and M the magnetization within the body. for paramagnetic substances a little bigger.5): B = µH. in this sense. In these currents.6 The magnetic susceptibility 7 The second constitutive equation concerns the magnetic field H and the induction B. one usually writes M = χB. they are the opposite of dielectrics.1. Similar to (1. the charges do not travel far from their equilibrium positions. and χ α. a steady current flows under the influence of an electric field. It also shows dispersion.

(1. The total current density Jtot may also be split into parts similarly to (1. For an electrically neutral body. the only exception is the surface of the particle where charges of one kind accumulate. For example. (1.22) . (1.19). positive and negative. ρfree = 0.803 × 10−10 esu (electrostatic ˙ Jpol = P Jmag = c · rot M. their sum is zero. electrons and protons exactly cancel and ρtot integrated over the body is zero. P = 0.19). have huge gradients. So we usually picture the grain material to be a continuous medium. ˚ On a scale of 1 A or less. Often ρtot is written as the sum of two terms: (1.8 The dielectric permeability 1. When charges are brought in from outside. If the polarization within the particle is uniform.21) The first term on the right-hand side is associated with the motion of the free charges. of course. Jtot = Jfree + Jpol + Jmag . ρfree . In an uncharged body. An electron. It contains protons and electrons and the net charge divided by δV gives the total local charge density ρtot .1 Sign conventions We have to make a remark about the sign convention. its absolute value being e = 4. and polarization charges can only appear on the surface. The first arises from polarization.20) The second term in (1. when the polarization is constant.1.1.8. if the polarization P is spatially non-uniform. ρpol does not vanish. the second with the time-varying polarization of the medium and the third represents the current that gives rise to its magnetization: Jfree = vρfree 1. Consider a microscopic volume δV in some piece of matter.8 Free charges and polarization charges We are generally interested in the electromagnetic field averaged over regions containing many atoms. However. Moving charges constitute a current. Then the separation of charges is inhomogeneous and leads to a charge density ρpol = − div P. div P = 0 and ρfree stands for all charges.19) ρtot = ρpol + ρfree . comprises all other charges besides ρpol . In a dielectric medium. Atoms appear with positive nuclei and negative electrons spinning and moving in complicated orbits and the electric and magnetic fields. the medium becomes structured. the electric field polarizes the atoms and thus produces aligned dipoles. which are smooth on a macroscopic scale. has a negative charge. Within a metal. sometimes we have to dive into the atomic world. However. ρpol is zero because the positive and negative charges of the neighboring dipoles exactly balance.

of course. Going from left to right.30) (1. Our choice of mathematical symbols is summarized in appendix A together with some common relations of vector analysis. rot H = D + c c (1. for instance.31) E denotes the electric and H the magnetic field.24) (1.4 of chapter 3.28) (1.27). D the electric displacement. J the current density and c. the electric field of an isolated positive charge (proton) is directed away from it and repels other positive charges according to (1.23) (1. ˙ (1.Maxwell’s equations 9 units). directed from the minus to the plus charge and anti-parallel to the electric field along the line connecting the two charges. for a neutral dielectric medium: div D = 0 div B = 0 1˙ rot E = − B c 1˙ rot H = D c second. Applying the operator div to (1. at a place where div P > 0. Hence a dielectric grain in a constant field E has a polarization P parallel to E and surface charges as depicted in figure 3. In the case of nonuniform polarization within the grain. the electric field pulls a small charge δq out of a tiny volume δV leaving behind an unbalanced negative charge −δq. for a medium with free charges and currents: div D = 4πρfree div B = 0 1˙ rot E = − B c 4π 1˙ Jfree .1).30) gives the expression for charge conservation: ρfree + div Jfree = 0. 1. With the help of this figure.20). we can explain the minus sign in equation (1. After Maxwell’s equation (1.1.9 The field equations We have now defined all quantities that appear in Maxwell’s equations and we write them down.25) (1.2). the polarization jumps at the left edge of the grain from zero to its value inside. B = ∂B/∂t.27) (1. the velocity of light. The moment p of a dipole created by two opposite but equal charges is. The gradient there is positive and thus a negative charge appears on the surface.26) ˙ A dot above a letter stands for partial time derivative. from (1. . B the magnetic induction. first.29) (1.

For real fields. t) = E0 · ei(k·x−ωt ) H(x. (1. E(x.2. All waves in interstellar space that interact with interstellar matter are planar because the sources from which they arise are very distant.38) S= 4π In a vacuum. which have a sinusoidal time dependence proportional to e−iωt . rot H = − i c rot E = i (1.32) and (1.37) where k is a vector with k2 = ω2 µε/c2 .32) (1. we arrive. its momentary value is c E × H.33) Applying the operator rot to (1.33). c2 (1.6) of vector analysis.34) (1.1. which is the energy per unit time and area. at E+ or E− ω2 µεE = 0 c2 µε ¨ E=0 c2 H+ H− ω2 µεH = 0 c2 µε ¨ H = 0. when E 0 denotes the amplitude of the electric field vector. flux and momentum of the electromagnetic field.36) (1. In an infinite medium. The characteristics of a plane harmonic wave are an eik·x space variation and an e−iωt time variation.35) These are the wave equations which describe the change of the electromagnetic field in space and time.26) become simpler: ωµ H c ωε E.1 The wave equation In harmonic fields.10 The dielectric permeability 1. with the help of formula (A. Maxwell’s equations (1.2 Waves in a dielectric medium We derive from Maxwell’s equations how a plane wave propagates in an infinite medium. t) = H0 · e i(k·x−ωt ) (1.25) and (1. or refractive index and recall a few formulae concerning energy.2.1 Flux and momentum of the electromagnetic field An electromagnetic wave carries energy. the time average is c 2 S = E . (1. is given by the Poynting vector S. The flux. 1.39) 8π 0 . one solution to (1. define the optical constant. 1.34) is a plane harmonic wave.

which are responsible for the dissipation of energy. from (A. it follows from (1. In the corpuscular picture. an individual photon travels at the speed of light. (1.42) and (1. k2 = k · k = k2 + k2 + k2 . must be zero. Its magnitude is then k= ω√ εµ c and the imaginary parts ε2 and µ2 . a momentum hν/c and an angular momentum . The wave also transports momentum. k2 and obeys the relation k2 = k2 − k2 + 2i k1 · k2 = 1 2 ω2 µε. Inserting the field of a plane wave given x y z by (1. c2 (1. 1. through a unit area at a rate S/c. we consider the standard case in which k is real.36) is called the wavenumber and is.33) yields ωµ H = k×E c and after scalar multiplication with k k · E = k · H = 0.2 The wavenumber The vector k = (kx . kz ) in (1.Waves in a dielectric medium 11 When we use complex fields.36).37) into equations (1. As k specifies the direction of wave propagation. 8π (1. This is strictly possible only in a vacuum.44) . The wave is planar and travels with undiminished amplitude at a phase velocity vph = c ω = √ . in the most general case. k y .43) that the vectors of the electric and magnetic field are perpendicular to each other and to k. Any other medium is never fully transparent and an electromagnetic wave always suffers some losses. complex: k = k1 + i k2 with real k1 . the time average of the Poynting vector is a real quantity and.2. is given by S = c Re{E × H∗ }.43) and ωε E = −k × H c (1.42) First.40) H∗ is the complex conjugate of H.32) and (1. k εµ (1. has an energy hν. • (1.33).41) Of course.

H0 = (1. and the electric field vector equals E(x. µ and σ taken together.45) µ In particular. light is absorbed and the wave amplitude suffers damping proportional to e−k2 ·x .12 The dielectric permeability The real amplitude of the magnetic and electric field. If k1 and k2 are not parallel. k2 . (1. where k1 .2. c λ • Next. m(ω) = n(ω) + i k(ω) (1. the electromagnetic properties of matter may be described either by the wavevector (from (1. t) = E0 · e−k2 ·x · ei(k1 ·x−ωt ) . In the common case of a non-magnetic medium. There is a unit vector e parallel to k defining the direction of wave propagation such that k = (k1 + i k2 ) · e.115). When the medium is metallic and has a conductivity σ .46) that m does not contain any new information. The name optical constant is unfortunate as m is not constant but varies with frequency.5). let k be complex. As the wave propagates. where µ=1 . It is a complex dimensionless number. Then ε2 and µ2 are complex.3 The optical constant or refractive index For the propagation of light. rather than (1. It is clear from the definition of the optical constant (1. E 0 and H0. If the vectors k1 and k2 are parallel. in a vacuum vph = c H0 = E 0 k= ω 2π = .46) m is also called the (complex) refractive index.42). The geometrical relations (1. the planes of constant phase and constant amplitude coincide. The phase of the wave is constant in a plane perpendicular to k1 and the amplitude is constant in a plane perpendicular to k2 .47) with real part n and imaginary part k. 1.41)) or by the optical constant m= √ εµ. too. in fact less than all material constants ε. one speaks of an inhomogeneous wave and the surface where the field is constant is not a plane. (1. k2 are the lengths of k1 . are related through ε E0 . one must use the dielectric permeability ε as defined in (1. Such a wave is called homogeneous.43) then lose their obvious interpretation.

52) If the space between the charges is filled with a dielectric.Waves in a dielectric medium the real and imaginary parts of the optical constant follow from ε1 and ε2 . E = E0 e−iωt . which is due to their Coulomb attraction. Therefore. ε1 = n 2 − k 2 ε2 = 2nk. 8π 1 E · D.2. so does the polarization and for small energy changes dw = − 1 V d(P · E) = −V P · dE. we find that the energy density of the dielectric medium contains an extra term 1 EP which accounts for the work 2 needed to produce in the medium a polarization P. 1 n=√ 2 1 k=√ 2 and.48) (1. .50) (1. is U= 1 8π E · E dV where the integration extends over all space. If the outer field slowly oscillates.4 Energy dissipation of a grain in a variable field The total energy of an electrostatic field E in vacuum. 1. It suggests an energy density u= 1 E · E. if we place. 2 (1. u= (1.52) with (1. into a constant field in a vacuum. 2 2 ε1 + ε2 + ε1 2 2 ε1 + ε2 − ε1 13 (1.46) has two solutions and we pick the one with positive n and non-negative k 1 .53) When we compare (1. we use the same letter for the imaginary part of the optical constant (italic type k) and for the wavenumber (Roman type k). This formula follows readily from the potential energy between the charges.49) (1. vice versa. the total field energy changes by an amount w = − 1 V P · E. some small dielectric object of volume V . 8π (1.51) (1.53).54) It becomes smaller because some work is expended on the polarization of the grain. 2 1 In order to abide by the customary nomenclature. produced by a fixed distribution of charges.

After some elementary vector analysis. the dielectric permeability ε is a tensor and symmetric. H = H0 e−iωt . Whereas static electric fields are produced by fixed charges and the total energy of the electric field is found by bringing the charges to infinity (which determines their potential energy). 2 (1. Formally similar expressions but based on different physics. . for the total energy of the magnetostatic field. The bracket .4. one has to evaluate its energy by changing H. . . These currents also exert a force F on a charge but from (1.9).57) are the basic equations for understanding the absorption of radiation by interstellar grains.55) and (1. denotes the time average. which we denote by W .14 The dielectric permeability The time derivative w.18)) to a heat dissipation rate W = 1 V ω Im{αm }|H0|2 . So instead of (1. Therefore. in the case of a magnetic field.56) In complete analogy to the electric field. the real part of the ˙ dielectric polarizability Re{αe } disappears in the product PE confirming that only the imaginary part Im{αe } is responsible for heating the grain. This produces according to (1. U= 1 8π H · B dV which implies an energy density of the magnetic field u= 1 H · B. for an anisotropic substance.25) an electric field E and thus a loss rate E · J. a time-variable magnetic field. static magnetic fields are produced by constant currents. 1. we write more generally Pi = j αi j E j . hold for magnetism. 2 (1.33). one gets.57) (1. The same applies to the polarizability.55) It is positive and proportional to frequency ω and volume V . gives the dissipated power ˙ and its mean is ˙ W = −V PE = 1 V ω Im{αe }|E0 |2 > 0. and we have used P = αe E from (1.1 The symmetry of the polarizability tensor We have already mentioned that.1) do no work because F · v = (q/c)v · (v × B) = 0. 8π (1. leads in a small body of volume V and magnetic polarizability αm (see (1. Because of the mathematical relation (A. We first look for the formula of the magnetic energy density. which makes sense.2.8).

that grows linearly with the displacement from the equilibrium position. But generally we assume that a grain is built up of a system of oscillators. −κ x. and (c) a restoring force. in the second step in the y-direction. say. Then one takes back the component E x and finally the component E y . Putting 2 ω0 = κ me γ = b me we get 2 x + γ x + ω0 x = ¨ ˙ eE . one applies an electric field in the x-direction. H A Lorentz applied the harmonic oscillator model to the motion of an electron in an atom.3.1 The Lorentz model We imagine the following idealized situation: An electron of mass m e and charge e is attached to a spring of force constant κ. ¨ ˙ (1. one finds αxz = αzx and αyz = αzy . the condition W = 0 requires αxy = αyx .The harmonic oscillator 15 The symmetry of the tensor αi j can be shown by computing after (1. there is (a) an inertia term m e x. the particle is unpolarized. A harmonic wave with electric field E = E 0 e−iωt exerts a force F = eE which causes the electron to move.59) . in the x-direction. 1.54) the total energy W expended on polarizing a particle in a cycle consisting of four steps. In the first step.58) On the left-hand side. −b x. Its motion is governed by the equation m e x + b x + κ x = F. ¨ ˙ which is proportional to velocity and leads to damping of the system unless it is powered from outside. we later derive the dispersion relation ε = ε(ω) of the dielectric permeability around a resonance at frequency ω0 . the Lorentz model is quite successful in a quantitative description of many phenomena that occur on the atomic level and reveal themselves in the macroscopic world. although a precise idea of what we mean by it is often missing. (b) a frictional force. Using this concept.3 The harmonic oscillator In the early history of atomic theory. me (1. In a similar way. Usually we have in mind the electrons in an atom but occasionally the oppositely charged atomic nuclei of a crystal or some other kind of dipoles are meant. 1. If the cycle is loss-free. The oscillator concept is very fruitful. At the beginning. Despite its simplicity and the fact that electrons ‘move’ in reality in complicated paths.

3. ω0 = 0). with an exponentially decaying 2 amplitude. when there is no friction (γ = 0) and no perturbation from outside (E = 0). where the electron moves along the x-axis but one may readily generalize this to three dimensions.2 Free oscillations In the simplest case. ωγ is real. The amplitude then subsides exponentially without any oscillations.16 The dielectric permeability The oscillator has only three properties: • • • charge-to-mass ratio e/m e .e. The electron then oscillates at a frequency ωγ somewhat smaller than the natural frequency ω0 . γ 2 > 4ω0 and ωγ is imaginary. we discuss only the one-dimensional oscillator. it is required that B be the complex conjugate of A. 1. i. because then x v/ω can be neglected in comparison to the wavelength c/ω of the field. If there is friction (γ = 0) but no external force. the velocity of an electron in an atom is typically of order v/c ∼ 1%. (1. If friction is strong. It is assumed that the electric field is spatially constant over the displacement x of the electron. This is correct as long as the velocity v of the electron is small. B = A∗ 2 which means Re{A} = Re{B} and Im{A} = − Im{B}. v c.60) and the electron oscillates harmonically forever at its natural frequency ω0 .61) the solution is called a transient.63) . (1. damping constant γ and resonant frequency ω0 (if the electron is unbound.62) For x to be real. when damping is weak. 2 x + γ x + ω0 x = 0 ¨ ˙ (1. If γ 2 < 4ω0 . (1. Indeed. It has the general form x = e−γ t /2 · (Ae−iωγ t + Beiωγ t ) where A and B are complex constants and ωγ = 2 ω0 − γ 2 /4.59) reduces to 2 x + ω0 x = 0 ¨ (1. Critical damping occurs for γ = 2ω0 . In the following.

62) eventually dies out and only (1. (1.3 The general solution to the oscillator equation 17 In the most general case of equation (1. A static field E 0 induces in the oscillator a permanent dipole moment p = ex 0 = e2 E 0 . For the complex ˙ amplitude of the displacement of the electron one finds x0 = ¯ ¯ e E0 2 m e (ω0 − ω2 − i ωγ ) .66) ¯ ¯ Putting E 0 = | E 0 | and x 0 = |x 0 |. So if we assume an harmonic field.64). The electron then oscillates at frequency ω and not ω0 .65) describes the steady-state solution into which any particular solution. .61) one particular solution of (1.3. sometimes for clarity we mark the complexity of a quantity explicitly by a bar. It is now convenient to write the displacement and the electric field as complex variables. for example x = x 0 e−iωt . evolves.64) One obtains the general solution of this inhomogeneous equation by adding to the general solution (1.65) In such a sum. the amplitude x 0 is proportional to γ −1 and can become very large. (1. ¯ ¯ E = E 0 e−iωt . ¯ ¯ (1. the transient (1. there is friction plus an external field that acts on the electron.65) remains. E0 e x + γ x + ω0 x = ¯ ¯ me (1.62) of the associated homogeneous equation (1. the real amplitude of the electron’s displacement becomes x0 = eE 0 2 m e (ω0 − ω2 )2 + ω2 γ 2 .59). 2 m e ω0 One can apply this equation to atoms to obtain a rough estimate for the dipole moment induced by a field E 0 if one chooses for the characteristic frequency ω0 a value such that ω0 is of order of the ionization potential of the atom. satisfying certain initial values for x and x at some earlier time t0 . Equation (1.67) If damping (γ ) is small and the incoming wave vibrates with the natural frequency of the electron (ω = ω0 ). ¯ ¯ E and E 0 are complex and e ¯ −iωt 2 ¨ ˙ ¯ .The harmonic oscillator 1.

4 Dissipation of energy in a forced oscillation The total energy of the oscillator is the sum of kinetic energy T plus potential energy V : 2 ˙ (1. friction results from collisions of the electrons with the lattice and this leads to heating. If we think of a grain as being composed of many oscillators (atoms). Without friction (γ = 0). the phase shift changes continuously and amounts to 90◦ at ω = ω0 .1 The phase shift Equation (1. If γ is small. The electron always lags behind the field by a positive angle = • • • tan−1 (x) tan−1 (x) + π if ω < ω0 if ω > ω0 where x = ωγ 2 ω0 − ω2 . is then interpreted as the collisional frequency. This is not a miracle but reflects the steady-state response of the electron. are out of phase because of ¯ the imaginary term i ωγ in (1. With friction. 2 The total energy declines when there is friction. when the field E was switched on. it approaches 180◦. The damping constant γ . Initially. in the same direction as the electric field.3.66). Because the mechanical power W . the time-averaged dissipation rate is ˙ W = 2 γ e2 E 0 ω2 1 2 γ m e ω2 x 0 = . x and E. 1.3.68) T + V = 1 m e (x 2 + ω0 x 2 ). the electron and the field are synchronous.69) . ˙ With x = −ωx. the lag is small. which is converted in a forced oscillation into heat. Around the resonance frequency. For ω ω0 . 2 2 2m e (ω0 − ω2 )2 + ω2 γ 2 (1. ˙ 2 The terms in the brackets represent the total energy. At low frequencies (ω ω0 ). it moves in the forced oscillation with the frequency ω of the external field. of course. the ¯ complex variables for position and field. Then the electron moves opposite to the direction in which it is being pushed by the external force eE. equals force multiplied by velocity: W = Fx ˙ multiplication of formula (1.58) with x gives ˙ W = Fx = ˙ d dt 1 ˙2 2 mex 2 + 1 m e ω0 x 2 + γ m e x 2 . the time derivative of the total energy almost vanishes and only the term γ m e x 2 remains. which has the dimension s−1 .3.65) tells us that although the electron has a natural frequency ω0 .18 The dielectric permeability 1. the acceleration vector of the electron pointed.

however. 2 The Fourier transform of x(t) in equation (1.70). f (ω) = or. (1. For the amplitudes in the Fourier expansion of the free oscillator of (1.3.73) can be regarded as an infinite expansion of the motion of the electron. the same letter as for the electric field) at time t = 0 is 2 2 E 0 = 1 m e ω0 x 0 2 and afterwards it falls due to dissipational losses like E = E 0 e−γ t .72) f (ω)e−iωt dω. (1. by definition. i (ω0 − ω) + γ /2 (1. The integral (1.71) The energy of the system drops by a factor e in a time τ = 1/γ or after ω0 /2πγ cycles. 1. the initial loss rate is 2 2 W = 1 γ m e ω0 x 0 . we should denote the time average by W but for simplicity we write W . we find f (ω) = x 0 0 ∞ e−γ t /2e−i(ω0 −ω)t dt = x0 . x(t) = 1 2π ∞ −∞ ∞ −∞ x(t)eiωt dt (1. while its amplitude gradually declines like e−γ t /2. If the motion of the electron is described by x(t) = x 0 e−γ t /2e−iω0 t 0 for t ≥ 0 for t < 0 (1. Near the resonance frequency (ω ω0 ). x(t). the power W becomes large.73) The two functions x(t) and f (ω) form the Fourier transform pair. and especially when damping is weak.70) is.) • • When the frequency is very high or very low. the heating rate W goes to zero because the velocities are small.5 Dissipation of energy in a free oscillation If the system is not driven by an external force and if damping is small. into harmonic functions e−iωt of amplitude f (ω) and with continuously varying frequencies ω.74) .70) the total energy E (unfortunately.The harmonic oscillator 19 (In fact. in view of the reciprocity relation. the electron swings almost freely near its natural frequency ω0 .

3. they are in the neighboring slab S2 .6 The plasma frequency The plasma frequency is defined by ωp = 4π Ne2 me (1. To prove it. the mean charge density ρ over any macroscopic volume V is zero: positive and negative charges balance.60) shows that the electrons will oscillate around their equilibrium position with the frequency ωp of (1. After this disturbance. Comparison with (1. ωp . which has the same shape and volume as S1 . The inverse. ¨ 4π Ne2 x = m e x. so that the volume of the slab V = x. The loss rate corresponding to each frequency component is then proportional to the square of the amplitude: W (ω) ∝ γ | f (ω)|2 ∝ γ (ω − ω0 )2 + (γ /2)2 . As we assume weak damping. they all cluster around ω0 .75) −1 where N is the number density of free electrons. A real grain contains many such oscillators ¯ (electrons) that are driven by the incoming wave up and down. ¨ The field E = 4π Nex is found by integrating the basic equation 4πρ = div D.7 Dispersion relation of the dielectric permeability Because the restoring force. Equation (1. y. Now consider in the plasma a slab S1 with sides x. gives the relaxation time of the electron density.58) is due to electrostatic attraction between the electron and a proton. 1. in (1.4.75). we may consider the oscillating electron as an alternating dipole of strength xe. only frequencies with a substantial amplitude f (ω) are relevant. If they swing in .74) is the basis for the Lorentz profile to be discussed in section 1. The electric field E arising from the charge separation as a result of shifting electrons from S1 to S2 exerts a force F on the displaced electrons and accelerates them: F = Ne x · E = N xm e x.3. −κ x. and the net charge density (ions plus electrons) in S2 is no longer zero but equal to Ne.20 The dielectric permeability In the decomposition of x(t) into harmonics. z and y · z = 1. immobile positive ions and unbound light electrons. we apply a perturbation to a plasma in equilibrium which consists of heavy. 1. Prior to the disturbance. All electrons inside S1 are suddenly displaced along the x-axis by the infinitesimal length x. therefore.

75). phase and their volume density equals N.77). If we use the plasma frequency of (1.2.2 and ωp = ω0 . The induced dipole moment per . equivalently.6).77) Figure 1. change with frequency. from (1. we get ε = ε1 + i ε2 = 1 + 2 2 ωp (ω0 − ω2 ) 2 (ω0 − ω2 )2 + γ 2 ω2 +i 2 ωp γ ω 2 (ω0 − ω2 )2 + γ 2 ω2 . Also of interest are the limiting values in very rapidly and very slowly changing fields.The harmonic oscillator 21 Figure 1. The vertical line helps us to locate the maxima with respect to the resonance frequency.66). where N is the number density of free electrons. (1. • In a constant field E.48) and (1.49). figure 1. the dipole moment per unit volume becomes.2 presents an example of the dispersion relation ε(ω) calculated with γ /ω0 = 0. Despite this particular choice. Equation (1. The dispersion relation specifies how the dielectric permeability ε = ε1 + iε2 or.2 also gives the dispersion relation for n and k.76) represents the so called dispersion relation of the complex dielectric permeability and specifies how ε varies with frequency.76) where the equals sign on the far right comes from (1. ¯ ε−1 ¯ E Ne2 ¯ = E P = Nxe = ¯ 2 − ω2 − i ωγ m e ω0 4π (1. The bar designates complex quantities. the optical constant m = n + ik. all quantities are real. As we can compute m from ε after (1. here for a harmonic oscillator after (1. it shows the characteristic behavior of the dielectric permeability at a resonance.

The imaginary part ε2 goes to zero far away from the resonance on either side but always remains positive. there is very little polarization as the electrons cannot follow the field due to their inertia and ε can be approximated by ε =1− 2 ωp ω2 +i 2 ωp γ ω3 . n goes towards a constant value and k falls to zero. ε1 (ω0 ) = 1 and 2 ε2 (ω0 ) = ωp /γ ω0 . compute how an oscillator is damped by its own radiation and evaluate the cross section for absorption and scattering of a single oscillator. (1. ε =1+ +i 2 ωp γ 4 ω0 ω. At very high frequencies. ε1 increases from its static value as ω nears ω0 . . It may become zero or negative.22 The dielectric permeability unit volume and the static permeability are: P= Ne2 E 2 m e ω0 2 ωp 2 ω0 ε(0) = 1 + • In a slowly varying field (ω ω0 ).79) • ε is then an essentially real quantity asymptotically approaching unity. whereas ε2 becomes proportional to frequency ω times the damping constant γ and is thus small. and the material becomes transparent because there are no dissipational losses (k. ε2 → 0). the concept of a continuous medium breaks down and modifications are necessary. as displayed in figure 1. 1.2. 2 ωp 2 ω0 . Correspondingly. We derive the emission of an accelerated charge. But note in figure 1. k and ε1 with respect to ω0 . Around the resonance frequency. At high frequencies (ω ω0 ) and far from the resonance. Refraction disappears (n → 1).4 The harmonic oscillator and light We continue to discuss the optical constant from the viewpoint that matter is made of harmonic oscillators.5).2 the significant shifts of the extrema in n. At exactly the resonance frequency. (1.78) • • • The real part ε1 approaches the electrostatic value ε(0). then sharply drops and rises again. ε2 is quite symmetric and has a prominent peak. This is a necessary condition for the fulfilment of the Kramers–Kronig relations (see section 2. when the wavelength is reduced to the size of an atom.

The real part of the dielectric constant. n. For light traveling from focus F1 of an ellipse to focus F2 via the reflection point M on the circumference of the ellipse. the travel time does not change. In a resonance. may take up positive and negative values. the total path length F1 MF2 is indifferent to variations around M. The index n determines the phase velocity vph of the wave. light traveling in a medium of varying n chooses. from among all possible paths.3. ε2 or k can never be negative as this would imply amplification of light. So the chosen route might be a local minimum or a local maximum (see figure 1. In an idealized loss-free medium ε2 and k vanish. c vph = .The harmonic oscillator and light 23 T M C F1 F2 Figure 1. the electrons transfer kinetic energy in collisions to the lattice.46). According to Fermat’s principle. In the harmonic oscillator model. 1. the phase velocity in glass is different from that in vacuum is contained in the concept of superposition which . the quickest. n n is responsible for the phenomenon of refraction.3). If the reflection point M lies on the tangent T. which is impossible for a dust grain in thermodynamic equilibrium. The loss rate depends foremost on the imaginary part of the dielectric constant ε2 . however.4.2). the path is a maximum. ε2 has its maximum value close to but not coinciding with the natural frequency ω0 (see figure 1. is always positive. Fermat’s principle formally explains how spectacles and binoculars work and why a light ray changes its direction when it passes from a medium with index n 1 to one with n 2 .1 Attenuation and refraction of light The attenuation of light is due to the dissipation of energy. The physical reason that. from (1. Without damping one finds. to first order. More precisely. which itself is proportional to the frictional parameter γ . for example. The real part of the optical constant. if it is on the curve C which has a curvature radius smaller than the ellipse at M. when infinitesimally varying the actual path. ε1 .44) and (1. the actual path is a minimum.

80) So for small attenuation. n is nearly constant at low frequencies and equal to 2 2 1 + ωp /2ω0 .77).131)). induces an electric field and so on.80) n(ω) 1+ 2πe2 me Nj j ω2 j − ω2 .e. from (1. reaches its maximum shortward of ω0 . The electric field of the wave accelerates the electrons in the glass and they start to radiate themselves. When n rises with ω.e. for example. However. One gets exactly the same phase shift if one neglects the fields from the electrons altogether and assumes the wave has traveled in the sheet at a reduced phase velocity c/n. If a substance has several resonances arising from different oscillators with number density N j and natural frequency ω j . one has instead of (1. it all depends. one speaks of normal dispersion. for historical reasons. in turn.4. of course. A wave package. When one computes E a (it is not too difficult). for this purpose. a plane wave that encounters a flat. for glass in the visible range: blue light entering a prism is bent more than red light. from the interference of the incoming wave with waves generated by the excited oscillators.48) and (1. according to the principle of superposition. To calculate the emission. we first summarize what the electric and magnetic fields produced by arbitrarily moving charges .2). This is true.24 The dielectric permeability states that the field at any given point is the sum of the fields from all oscillators anywhere in the world. one finds that it has a phase shift relative to the incident wave. when the damping is weak and γ ω0 . if dn/dω < 0. This does not violate special relativity as it is not possible to transmit information with a monochromatic wave. to the left of ω0 in figure 1. the expression for n reduces. In terms of actual frequencies. Consider. the phase velocity exceeds the velocity of light in vacuum. i. (1. to the wave if there were no sheet. n rises. The field E a far ahead of the sheet is.e. thin sheet of glass oriented perpendicular to the direction of wave propagation. which is composed of waves of different frequencies and is capable of carrying a message. an electromagnetic wave is emitted. 1. in the end. dips afterwards below one and goes asymptotically to unity at high frequencies.2. As one approaches the resonance (see figure 1. cannot travel faster than c (see (1. When n is less than one. For a medium that is almost transparent. to n 1+ 2 ωp 2 2(ω0 − ω2 ) . in a subtle way.2 Retarded potentials of a moving charge An oscillating electron represents a current that is surrounded by a time-variable magnetic field which. i. on the value of ω0 . the sum of the fields created by all electrons plus the field of the incident wave. So the phase velocity c/n follows. i. the dispersion is called anomalous.

86) and (1. t) = A(x2 . We start with Maxwell’s equations in the so-called microscopic form where all contributions to charge and current are included. there exists a scalar potential φ with 1˙ E + A = −∇φ. ρtot + div Jtot = 0.27)–(1. c (1. t ) dV |x2 − x1 | J(x1 . Because div B = 0. Then φ and A obey the relations A− 1 ¨ 4π J A= − 2 c c 1 ¨ φ − 2 φ = − 4πρ. 1˙ div A + φ = 0 c (1.89) The potentials A and φ are called retarded as they refer to the present time t but are determined by the configuration at an earlier epoch t .19)–(1.85) (1.88) (1. The delay corresponds to the time it takes light to travel from x1 to x2 : t =t− |x2 − x1 | . the potentials at the point x2 of the observer are (see any textbook on electrodynamics): φ(x2 .The harmonic oscillator and light 25 of density ρ look like.87) are equivalent to Maxwell’s equations. t) = ρ(x1 . c These potentials are gauged by imposing on them the Lorentz condition. c c c (1. there exists a vector potential A such that B = rot A (1.22). c (1. Again there is charge conservation.84) which leaves the fields E and B untouched. t ) d V. If the charges are localized around the center of the coordinate sytem and if their position is denoted by x1 . not only the ‘free’ parts: div B = 0 div E = 4πρtot 1˙ 4π 1˙ rot B = E + rot E = − B Jtot . they become wave equations and their right-hand sides vanish.81) (1. We now ˙ drop the suffix ‘tot’. In vacuum.83) ˙ and because rot(E + A/c) = 0. c |x2 − x1 | (1.90) .82) They follow immediately from the equation set (1.86) (1.30) with the help of (1.87) (1.

(1.26 The dielectric permeability If the size d ∼ |x1 | of the region over which the charges are spread is small compared with the distance r = |x2 | to the observer.92) This expression applies. Moreover. d λ there will be no phase shift among the waves emitted from different parts of the source and the emission is coherent. we may put r |x2 −x1 | and take r out from under the integrals. where v is the non-relativistic charge velocity.84) is negligible because φ represents the potential of the charge (see the comment after equation (6. If there are many charges qi shaking at velocity vi . Far away from the charge. ∇φ in (1. so we can write t and A(x2 .92) yields A(x2 . |x2 −x1 | |x2 |−x1 ·e. for example. t − r/c) d V. (1. t) = 1 ˙ p(x1 . t) = 1 cr t − r/c J(x1 . We do not discuss the near field here. to atoms where a charge is oscillating at a frequency ν and the linear dimension of the system d ∼ ν −1 v = λv/c. where e is the unit vector pointing from the atom to the observer. so t t − (r − x1 · e)/c. t − r/c) cr (1. equation (1. If there is only one small charge q of space density ρ(x) that is oscillating about an opposite charge at rest. the wave is planar and H =e×E E = H × e. the result is the same with p = qi xi .10)) and the electric field of a charge .91) If the d is also small compared with the wavelength. The vector potential A follows from a volume integral over the current density J at the earlier epoch t = t − r/c.4. d r (astronomers are always at a safe distance from the action).3 Emission of an harmonic oscillator Under such simplifications. 1. Far away.93) because the current density is J = ρv and the wiggling charge q constitutes a dipole p = qx1 whose derivative is given by the integral over the current density. one obtains the dipole field.

98) is also evaluated at the time t = t − r/c. 1. cr cr 6c r (1.96) dW = 4πc3 The emission is zero in the direction of the motion of the charge and maximum perpendicular to it.89) for small x1 · e.4 Radiation of higher order More generally. (1. 3) and this Q is used in (1. The vector x has the components x i and the length x.38). Again. one finds the Poynting vector ˙ S = (c/4π)E × H from E = −A/c and H = e × E.The harmonic oscillator and light 27 falls off like 1/r 2 . Q 2 . The vector potential A in formula (1. we can compute the flux d W carried in a solid angle d = sin θ dθ dφ into the direction e which forms an angle θ with the dipole moment p.4.98). This flux is given by the Poynting vector of (1. a traceless tensor. Having determined the fields E and H via the dipole moment p. so ¨ |p|2 sin2 θ d .91) and expands J(x1. Q = (Q 1 . equation (1.97).97) If the dipole oscillates harmonically proportional to cos ωt.99) the electric quadrupole moment.93) is modified and at first order one has to add two correction terms: A= 1 1 ¨ 1 ˙ ˙ p + m × e + 2 Q. 3c3 (1. From Q i j one defines the vector Q. t ) in (1.94) (1.98) m is the magnetic dipole moment of equation (1. Q 3 ) with Q i = j Qi j e j (i = 1. 2. . the timeaveraged power equals half the maximum value of W in (1. Integration over all directions yields for the total momentary power radiated by a dipole: W = 2 ¨ |p|2 . We thus have E = −A/c and H= ω2 1 ¨ p×e= 2 e×p 2r c c r ω2 E = 2 (e × p) × e c r (1. δi j denotes the Kronecker symbol. whereas the field of a light source falls off more slowly like ˙ 1/r .95) where we have assumed that the dipole varies harmonically at frequency ω. if one does not make the simplification t = t − r/c but sticks to equation (1.13) and Qi j = ρ(x)[3x i x j − x 2 δi j ] d V (1.

1. for instance. 3m e c3 Frad has to be added in the equation of motion of the harmonic oscillator. ω0 x + γ u + u − τ u = me τ= It contains time derivatives in x up to third order but.28 The dielectric permeability Even when the electric dipole moment vanishes.4. the steady-state solution is again x = x 0 e−iωt ¯ ¯ with .27 × 10−24 s. Averaged over some time ˙ interval t. associated with Q and m. Although theory (quantum electrodynamics) and experiment agree extremely well. ¨ 3c3 So the force becomes ¨ Frad = m e τ u 2e2 = 6.97)) is equal to the work that the force Frad has done on the electron: − t Frad u dt = 2e2 3c3 u 2 dt. we write the equation of motion in the form ˙ F + Frad = m e u. so (1. there is currently no strict. self-consistent description of the feedback. When a force F accelerates a free and otherwise undamped electron. ˙ t When we integrate the right-hand side by parts and assume a periodic motion where u u = 0 at the beginning and at the end of the time interval t. We describe the feedback in classical terms here. The additional force Frad accounts for the retardation caused by the radiative loss and u = x is the velocity. the two atoms in ¨ ¨ the hydrogen molecule H2 . Frad acts oppositely to F. there may still be electric ¨ ˙ quadrupole or magnetic dipole radiation. the time derivatives p and m are zero so that only the electric quadrupole radiation remains. we assume that the emitted power due to the radiative deceleration (see (1.5 Radiation damping An accelerating electron radiates and the emitted light reacts on the electron because of the conservation of energy and momentum. because u = −ω2 u for a ¨ harmonic oscillation.59) becomes eE 0 −iωt 2 ˙ ¨ e . we find ˙ Frad − t 2e2 2 u u dt = 0. respectively but they are several orders of magnitude weaker. One can easily show that if all particles have the same charge-to-mass ratio as.

from (1.4. 2 m e ω0 − ω2 − i ω(γ + τ ω2 ) 29 This is the same expression for x 0 as in (1.65 × 10−25 cm2 (1.104) . the power W sca = 2 ω4 2 4 2 2 e4 E 0 ω p0 = 3 2 3c3 3c m 2 (ω0 − ω2 )2 + ω2 γ 2 e The term (1. m e c2 (1.82 × 10−13 cm. γrad = 1.6.The harmonic oscillator and light (a bar denotes complexity) and the complex amplitude becomes x0 = ¯ ¯ e E0 1 . becomes σ sca (ω) = where W sca ω4 = σT 2 S (ω0 − ω2 )2 + ω2 γ 2 (1.101) ¯ where we put p0 = | p0|.1 Scattering An oscillating electron with dipole moment ¯ p = p0e−iωt = e x 0 e−iωt ¯ ¯ scatters.100) 3m e c3 is the damping constant due to radiation alone and follows from equating the 2 time-averaged losses 1 γrad m e ω2 x 0 of the harmonic oscillator after (1. Note that we always assume the radiative losses over one cycle to be small in comparison to the energy of the oscillator.103) 3 is the Thomson scattering cross section of a single electron. r0 is the classical electron radius and follows from equalling the rest mass energy m e c2 to the electrostatic energy e2 /r0 : σT = r0 = e2 = 2. 2e2 2 ω (1.6 The cross section of an harmonic oscillator 1.66) if one replaces there the ¯ dissipation constant γ by γ + τ ω2 = γ + γrad .4. So its cross section σ sca .102) 2 8πr0 = 6.69) to the 2 2 time-averaged radiated power p0 ω4 /3c3 of (1. it is frequency independent.38).97).97).67) and (1. defined as the scattered power 2 divided by the incident flux S = (c/4π)E 0 of (1.

4. and at low and high frequencies:  4   σT ω  if ω ω0   ω0 sca 2 (1. We note that ∞ 0 γ (ω − ω0 )2 + (γ /2)2 dω = 2 ∞ −2ω0 /γ dx 1 + x2 2π (γ ω0 ). Near the resonance.74). (1. we get from (1. When the frequency is high. As we have seen in the Fourier analysis of the motion of a free oscillator after (1. 1. the electron is essentially free and the cross section constant and equal to σT . we have to set γ = γrad after (1. .107) A frequency dependence according to the middle line of (1.106) over frequency yields the total cross section for absorption. 1.107).105) results in a Lorentz profile.4. σ sca falls with the fourth power of ω (Rayleigh scattering).6.107) or (1. When the frequency is small. at the root of such a profile is the exponential decline e−γ t /2 of the amplitude of the electron. It has the characteristic feature that the intensity over an emission or absorption line changes proportional to [(ω−ω0 )2 +(γ /2)2 ]−1 . (1.30 The dielectric permeability If there is only radiative damping.2 Absorption In a similar way.105) σ (ω) = σT ω0   if ω ω0  4 (ω − ω )2 + (γ /2)2   0 σT if ω ω0 .102) we find the following approximations for the scattering cross section at the resonance. we have put 2 ω2 − ω0 2ω0 (ω − ω0 ). From equation (1.108) σtot = σ (ω) dω = mec The integrand has significant values only around ω0 . so to evaluate σtot we could also use the approximate second formula of (1.106) (ω) = πe2  −4 2  4γ ω0 ω  γ m e c  (ω − ω0 )2 + (γ /2)2  4γ ω−2 (1.69) for the absorption cross section of the harmonic oscillator σ abs (ω) = with the approximations σ abs 4πe2 ω2 γ W abs = 2 S m e c (ω0 − ω2 )2 + ω2 γ 2 if ω if ω if ω ω0 ω0 ω0 . 2π 2 e2 .100).7 The oscillator strength Integrating σ (ω) after (1.

Because all relevant frequencies call x 0 2 for emission are close to ω0 . its energy decays exponentially like e−γrad (ω0 )t (see (1. .The harmonic oscillator and light When the cross section σ is a function of ν = ω/2π.67). λ = 2πc ω 2 ω0 (1.4). In quantum mechanics.3). in analogy to the classical formula.110) 2r0 2 ω 3c 0 = 4π ˚ r0 = 1. all intrinsic properties of the oscillator have canceled out.100)). It does not depend on the frequency of the transition.71) and (1. B j i and A j i (see section 6. 1. One may 2 (ω) the resonance curve (see figure 1.8 The natural linewidth If the oscillator is not driven by an external field but swings freely near the resonance frequency ω0 . we have ω2 − ω0 2ω0 (ω − ω0 ) and the full width 2 ω of the resonance curve x 0 (ω) taken at half maximum becomes ω = γrad = or.3). f j i is related to the Einstein coefficients for induced and spontaneous emission. A j i specifies the rate at which a system in level j spontaneously decays to a lower level i (see section 6.109) where f j i is called oscillator strength. at a frequency ω is 2 proportional to x 0 (ω). mec 31 The true quantum mechanical cross section for a downward transition j → i integrated over frequency is often written as σtot = πe2 f ji mec (1.18 × 10−4 A. 3 (1. It is of order unity for strong lines. the square of the elongation given in (1. I (ω). otherwise smaller. through πe2 hν c2 f ji = B ji = A ji mec c 8πν 2 where hν is the energy difference between level j and i .111) This is the natural linewidth as derived from classical physics. The probability of finding the system in state j decreases with time like e−A j i t . we get σtot = σ (ν) dν = πe2 .4. the Einstein coefficient A j i is the counterpart to the radiative damping constant γrad of classical electrodynamics. The intensity of emission. expressed in wavelength.

the energy E j of state j is fuzzy because of the uncertainty principle. (More accurately. Therefore. Here t A−1 is the average ji time that the atom stays in level j . in quantum mechanics. in the presence of a magnetic field. We show that.100). 1. one has to sum the A coefficients of all possible downward transitions starting in j and something similar must be done for the lower level i .5 Waves in a conducting medium We modify the expression for the dielectric permeability or the complex wavenumber for the case that some of the electrons in the material are not bound to individual atoms but are free so that they can conduct a current. The emission thus cannot be strictly monochromatic but has ji a frequency spectrum centered at ω0 . the length of the wave train is c A−1 . The emitted intensity of a classical oscillator with a resonance at λ0 that is damped only radiatively with a damping constant γrad after (1. Furthermore. the ˚ width of the line is 1.32 The dielectric permeability Figure 1. E t ≥ . when calculating ω.4. the optical constant n of a plasma depends for circularly polarized plane waves on the sense of rotation. unless it is the ground state. As the pulse travels with the velocity of light. The medium is then a plasma and the dispersion relation is known as the Drude profile. This leads . Irrespective of λ0 . E j is undetermined by an amount E = A j i resulting in a frequency uncertainty ω = A j i .18 × 10−4 A in wavelengths or γrad in frequency.) The time A−1 that the atom resides in state j also gives the duration of the ji emitted pulse.

however. We treat conductors because some interstellar grains have metallic properties and because the interstellar medium is a plasma. (1. c . we get for a homogeneous medium with the help of (1.113) We may look at (1.114) 4πσ . Formally. 1˙ rot E = − B c rot H = 1˙ 4π D+ J.113) reduces to rot H = 4πσ E c (1.1 The dielectric permeability of a conductor A conductor contains free charges that can support a constant current. as we know from everyday experience with electricity. Ohm’s law asserts a proportionality between the electric field E and the current density J (omitting the suffix ‘free’). When a conductor is placed in a static field. We also derive the group velocity of a wave package in a plasma. we can retain the field equation ε˙ rot H = E.30) which are appropriate for conductors.112) σ is called the conductivity and is usually defined for a quasi-stationary electric field. An electric field may exist in a metal. (1. 1. in a vacuum. a static electric field is. more precisely when |ε| Then equation (1. c c When we insert harmonic fields into these formulae. ω which describes the magnetic field produced by a quasi-stationary current.Waves in a conducting medium 33 to the phenomenon of Faraday rotation.25). In a dielectric medium. J = σ E. σ = 0.29) and (1.113) as an expansion of rot H into powers of ω. either because it varies rapidly and the charges do not have time to reach their equilibrium positions or it is produced by a changing magnetic field according to (1. possible.5. no current can flow because the charges arrange themselves on its surface in such a way that the electric field inside it cancels out.112) ωµ H c ω 4πσ rot H = −i ε+i c ω rot E = i E. The term with the conductivity σ dominates at low frequencies. of course. Let us start with the Maxwell equations (1.

Then the restoring force.49) the optical constants n and k are. −κ x. plus a term relating to the conductivity: 4πσ . The electrons are then free.115) ε = εd + i ω The complex wavenumber of (1.3 for a plasma.123) for an exact expression) ε(ω) = i 4πσ .48) and (1. This happens around ω 1017 s−1 .117) The difference between dielectrics and metals vanishes when the electromagnetic field changes so rapidly that the electrons make elongations that are small compared to the atomic radius. If there are N free electrons per cm3 .5).59). m e (γ − i ω) The conductivity follows from the current density J = σ E = Nev. the dielectric permeability of a metal is approximately (see (1.34 The dielectric permeability as formulated in (1.118) . 1. (1. (1.2 Conductivity and the Drude profile We repeat the analysis of the harmonic oscillator in section 1. therefore. also large and roughly equal: n k→∞ for ω → 0. much greater than one (|ε| has a singularity at ω = 0. is in a conducting medium k2 = ω2 µ 4πσ εd + i c2 ω . which appears in the wave equations (1. in (1. From (1.5.115) via m = εµ (see (1. also for a conducting medium. written here as εd to indicate that it refers to a dielectric. At low frequencies. they experience no restoring force after an elongation and. ε must then denote the sum of the dielectric permeability of (1.33) for a dielectric medium.46)).58) is negligible and all electrons are essentially free. With the same ansatz x = x 0 e−iωt as before.116) √ The optical constant m follows from ε of (1. ω 1) and ε is then purely imaginary (ε1 = 0). ω0 = 0 in (1.41). we find the velocity of a free electron in an harmonic electric field: v= ¯ e ¯ E. However. (1. σ = 2 ωp Ne2 = m e (γ − i ω) 4π(γ − i ω) (1. ¯ ¯ therefore.34).

48). k2 = ω2 c2 1− 2 ωp ω2 c2 1− 2 ωp γ 2 + ω2 +i 2 ωp γ ω γ 2 + ω2 . µ = 1. If the frequency is high or damping weak (γ ω). and substitute.122) In this case. A wave with ω ≤ ωp cannot penetrate into the medium and is totally reflected. σ (ω) = σ1 (ω) + i σ2 (ω) = • 2 ωp γ ω +i .121) ω2 .118) into (1. for the conductivity σ . ω γ 2 + ω2 (1. The dielectric permeability ε(ω) corresponding to (1. For ω = ωp . n becomes zero. (1. σ becomes frequency independent. 4π γ 2 + ω2 4π γ 2 + ω2 2 ωp (1. the conductivity is purely imaginary and inversely proportional to frequency.60)) are illustrations.6 × 10 . when ω is greater than the plasma frequency.75). The conductivity depends on the density of free electrons through ωp and on the collision time through γ . When we put εd = 0. So generally speaking.119) At long wavelengths (ω γ ). here are two examples: • Copper is a pure metal with a free electron density N 8 × 1022 cm−3 .123) To get some feeling for the numbers associated with the conductivity. 16 s−1 and a damping constant γ a plasma frequency ωp = 1.121) is known as the Drude profile (see (1. There is then no dissipation of energy in the current.116). we find for the wavenumber of a plasma k2 = Without damping. from (1.Waves in a conducting medium 35 where ωp = 4π Ne2 /m e is the plasma frequency of (1. It is then an essentially real quantity approaching the direct-current limit σ (ω = 0) = 2 ωp 4πγ (1.41)): ε(ω) = 1 − 2 ωp γ 2 + ω2 +i 2 ωp γ . so the phase velocity is greater than the velocity of light. the expression from (1. (1. The bending of short-wavelength radio waves in the ionosphere or the reflection on metals (see (3. vph > c.120) • which appears in Ohm’s law. smaller than one. the real part n of the optical constant is. the conductivity is a complex function and depends on ω.

0)E 0 e−iωt = We. The trajectory r(t) = (x. x= ¨ e eB y ˙ Ex + me mec y= ¨ e eB x. −i. which is found in interstellar space. B) oriented in the z-direction. Therefore copper cables are ideal for carrying electricity. The presence of the magnetic field B modifies the orbit. Graphite is a reasonable conductor only when the electrons move in the basal plane. 2 The solution to this differential equation is r± = with the cyclotron frequency ωcyc = eB . Graphite.124) that r± = ¨ e eB r± . 0) E 0 e−iωt .124) The variable magnetic field of the wave is neglected. For example.3 Electromagnetic waves in a plasma with a magnetic field In the presence of a magnetic field. 0)(t) of an electron follows from integrating the equation of motion. It follows from (1.36 The dielectric permeability 3 × 1013 s−1 . ¨ F = m e r.5.4 × 1020 cm−3 12 s−1 . wave propagation is generally complicated. E = (1.1). so the direct-current conductivity becomes σ and γ 5 × 10 7 × 1015 s−1 . ˙ Ey − me mec (1. We consider the fairly simple situation when a plane wave with electric vector E = (E x . 0) travels without attenuation through a plasma parallel to a constant magnetic field B = (0. therefore. it can be regarded as a superposition of two oppositely circularly polarized waves. put r± = x ± i y E± = E x ± i E y and interpret the (x. 0) + (1.125) e/m e E± ω2 ± ωωcyc A circularly polarized wave makes an electron rotate. it is two orders of magnitude smaller than for copper. has N 1. • 1. i. mec (1. E y . It has a high direct-current conductivity σ ∼ 7 × 1017 s−1 . If the wave is linearly polarized. and E + and E − as the electric vectors of two circularly polarized waves associated with the electron trajectories r+ (t) and r− (t). if the electric vector stays along the x-axis. ˙ E± ∓ i me mec 1 (1. y. where F is the Lorentz force of (1. 0. y)-plane as a complex plane. 0. and the modification is different for clockwise .

A(k) exp i k x − dω t dk dk = C · u x .127) gives u(x.Waves in a conducting medium 37 and anti-clockwise rotation. ω ∓ ωcyc Without damping.115) with εd = 1. We therefore develop the dispersion relation ω = ω(k) around k0 .0 with x =x− dω t. t) = exp i dω k0 − ω0 t dk =C ∞ −∞ ∞ −∞ A(k)eikx dk.127) in which the amplitude A(k) has a sharp peak at some wavenumber k = k0 . ω(k) = ω0 + dω (k − k0 ). At the time t = 0. 1. n2 = 1 − ± 2 ωp ω2 1± ωcyc ω −1 . 0) = Inserting (1. If there are N electrons per unit volume.128) Here ω0 = ω(k0 ) and the derivative dω/dk is evaluated at k = k0 .4 Group velocity of electromagnetic waves in a plasma A one-dimensional wave package is a superposition of monochromatic waves (see (1. their motion ˙ implies a current J± = σ± E ± = Ner± .128) into (1. dk (1.5. This effect is called Faraday rotation. where the conductivity σ± = i Ne2 /m e . (1. therefore. dk (1.36)): u(x. This leads to a change in the direction of the polarization vector of the linearly polarized wave.129) . the optical constant n 2 = ε1 . the wave package has the form u(x. The dielectric permeability ε is given in (1. t) = ∞ −∞ A(k) · ei(kx−ωt ) dk (1.126) We conclude that the two circularly polarized waves travel with different phase velocities vph = c/n ± .

If f (θ ) dθ denotes the number of atoms that have angles between θ . cos θ = f (θ ) cos θ d f (θ ) d (1. we consider molecules with an intrinsic dipole moment and a rigid charge distribution that are allowed to rotate.134) The average value of cos θ follows from integrating f (θ ) over all directions. However.44)).131) For the product of group and phase velocity vph = ω/k (see (1. dk (1.122). (1. θ + dθ . (1.132) 1. In thermal equilibrium at temperature T . provided the molecules are sufficiently ‘round’ so that they can turn and are not hooked too strongly to their neighbors.6.133) where θ is the angle between the vectors p and E. the analysis which we develop here also applies to the alignment of atomic magnets and we will use the results later in chapter 11 on grain alignment.1 Polarization in a constant field An atom or molecule with a dipole moment p in an electric field E has a potential energy U that depends on orientation. vph · vg = c2 .6 Polarization through orientation The polarization which we have modeled with the help of the harmonic oscillator of section 1. the wave package travels with the group velocity vg = dω . . the distribution of orientations is expressed by the Boltzmann factor e−U/ kT . U = −p · E = − p E cos θ (1. as the factor C is purely imaginary and irrelevant. The potential energy is at its minimum when the dipole moment lines up with the field and θ = 0.130) According to (1. then f (θ ) = f 0 e−U/ kT = f 0 e p E cos θ/ kT . (1. . 1. This process is principally relevant to gases and liquids and only marginally to solids.3 is due to the deformation of atoms in the sense that their internal charges are shifted relative to each other. In this section.38 The dielectric permeability Therefore. The atoms are never perfectly aligned because they are tossed around by the motions of their neighbors. their lengths being p and E.135) . for a plasma without damping vg = dω =c· dk 2 1 − ωp /ω2 ≤ c.

137) with x = pE kT We see that the strength of the net polarization P depends on temperature.140) where ζ is a friction coefficient. there is Brownian motion because the dipoles are jostled about in collisions. so ˙ τ = ζθ (1. all directions are almost equally likely and the average angle is pE cos θ = . It now depends not only on the angle θ between dipole moment and field but also on time. The time enters into f for two reasons.138) f (θ ) = 4π kT f (θ ) is now almost constant. (1. t) instead of f (θ ). If there are N dipoles per unit volume. Second. One just has to replace p E by m B.136) In our particular case.2 Polarization in a time-variable field The distribution function f for the orientation of dipoles in the case of a timevariable field is considerably more complicated. The rotation speed θ is assumed to be proportional to the torque τ . First. where f (θ ) ∝ e p E cos θ/ kT . the polarization P in the direction of E equals P = N p cos θ . 1. Easy to handle and relevant to our applications are weak fields ( p E kT ) for which pE N 1+ cos θ . 3kT An identical formula holds for magnetic dipoles (dipole moment m) in a magnetic field.6. t) d be the number of dipoles whose axes fall into a small solid angle d and make an angle θ with respect to the field E.Polarization through orientation 39 where d is an element of the solid angle. (1.139) ˙ which makes it rotate. We now sketch how one derives an expression for the distribution function leaving aside the mathematical details. the mean cosine is given by cos θ = L(x) where L(x) denotes the Langevin function: L(x) = coth(x) − 1 x x3 = − + ···. x 3 45 (1. Let f (θ. the field exerts on each dipole a torque τ = p E sin θ (1. The directions of the individual . so we have to write f (θ.

This process is called relaxation and governed by the equation 1 ∂ ζ ˙ f = kT sin θ ∂θ sin θ ∂f ∂θ (1. The interval δt is chosen long enough so that the dipole axes manage to escape from d but also sufficiently short so that their directions are modified only by a small angle. we try a distribution function p E0 f (θ.40 The dielectric permeability dipoles vary smoothly with time due to Brownian motion and the alternating electric field. indeed.134). • Altogether. t) = 1 + ϕ(t) cos θ kT .140)) that tries to align the moments with the field. But φ 2 is known: for Brownian motion φ 2 = kT δt/4ζ . The other term field is much easier to calculate.6. The difference in the number of axes that leave and enter the solid angle d is field . The change is split into a contribution due to Brownian motion and one due to the field = Brown + field . t) d over the time interval δt. complete disorder is established. the torque τ is absent and the orientations begin to randomize. The axes are turned at an ˙ angular velocity θ by a variable torque τ (see (1. because δt is small. a solution to (1. To solve (1. This leads to an expression for Brown that depends on f and its spatial derivatives times the mean square angular distance φ 2 which the axis of an individual dipole travels during the time δt. Debye [Deb29] found for the time derivative of the distribution function ∂f 1 ∂ sin θ kT +τf . • To evaluate Brown . one can restrict the discussion to the dipoles in the vicinity of d . 1.134) is.141) ζ f˙ = sin θ ∂θ ∂θ The first term in brackets on the right-hand side of (1.141) for f˙ = 0.141) for τ = 0.3 Relaxation after switching off the field The average orientation in a constant time field is described by the Maxwell– Boltzmann equation (1. It does of course not depend on time. Suppose now that the field is suddenly switched off at time t = 0. We may convince ourselves that (1. (1. the second term due to the rotation of the dipoles by the field.142). Then for t > 0.141) is due to Brownian motion. after a while. so f˙ vanishes.142) which follows from (1. One computes the change = f˙ d δt in the number of dipoles f (θ.

134) is then no longer applicable and needs to be generalized.4 The dielectric permeability in Debye relaxation In the presence of an harmonic field E = E 0 e−iωt . 2 2 1 + ω2 trel 1 + ω2 trel (1.e. For cos θ in (1.138).143) The form of the relaxation function implies exponential decay.141) when the variable field is weak.138) but for the relaxation function ϕ(t) and the factor N/4π which has been dropped. p E0 cos θ. Because P = χ E. it yields A= 1 . again without the factor N/4π. (1. t) = 1 + e−iωt p E0 cos θ . (1. we are back to Maxwell–Boltzmann.146) This function is plotted in figure 1. It is easy to find a solution to Debye’s equation (1. with (1.145) For a static field (ω = 0).144) 2kT 1.123). It should be compared with ε(ω) for a dielectric and a metal.145). t) = 1 + Ae−iωt kT 2 When inserted into (1.6. The Maxwell–Boltzmann formula (1. Substituting f (θ. 1 − i ωtrel So the full expression for the distribution function in a variable weak field is f (θ. the distribution function f depends explicitly on time. we try an ansatz for f in the spirit of (1. i. Naturally. We have here the novelty . t) into (1. At very high frequencies. The system relaxes from statistical alignment in a constant field to random orientation on a time scale ζ trel = . we obtain for the susceptibility χ (or the dielectric permeability ε) in the case of rotational polarization: χ= N p2 1 N p2 ε−1 = = 4π 3kT 1 − i ωtrel 3kT ωtrel 1 +i . i.135) we have to use the distribution function from (1.77) and (1.e.136).Polarization through orientation 41 which is similar to (1.5.141) and terms with E 0 are neglected. We can now compute the polarization P of such a medium according to (1. f becomes constant and is independent of the direction of the field. when p E0 kT. kT 1 − i ωtrel (1.142) yields ϕ(t) = e− 2kT ζ t . f (θ.

that ε depends on temperature.146) and the optical constant m = n + ik for the case that polarization is due to the alignment of molecules with a permanent dipole moment p. which lies in the midst between the hydrogen atoms. Liquid water has. The dielectric permeability ε = ε1 + iε2 after (1. resulting in a permanent dipole moment p = 1. and the center of negative charge. trel 2 × 10−12 s approximately. For small frequencies. k and ε1 − 1 and ε2 . inversely proportional to temperature and otherwise determined by the relaxation time trel . one can .42 The dielectric permeability Figure 1. ˙ ˙ θ θ (1. located near the oxygen atom.147) χ0 is the static value of χ(ω) and refers to a time constant field. here we put 4π N p2 /3kT = 1.5.139) and the force F of Stokes’ law (see (9.41)).148) ˚ For a ∼ 1 A. one may approximate the imaginary part of χ by χ2 = χ0 ωtrel 2 1 + ω2 trel ωχ0 trel χ0 = N p2 3kT . The water molecule has an intrinsic electric dipole moment as the three nuclei in H2 O form a triangle with an obtuse angle of 105◦ . ζ = −1 trel Fa τ ∼ ∼ 6πηa 3. (1. the friction constant ζ follows from the viscosity using the equation for the torque (1. at room temperature.01 g cm−1 s−1 . Shown are n. With the support of experimental data. do not coincide. If the water molecules are approximated by spheres of radius a. The dissipation rate is proportional to ω. Let us consider water as an example of rotational polarization.9 Debye. a viscosity η 0. The associated frequency lies in the microwave region. Therefore the center of positive charge.

at room temperature we have roughly N p2 /3kT ∼ 1. Then the viscosity and the relaxation time jump by a large factor and the frequency at which ε2 has its maximum drops and polarization due to Debye relaxation becomes unimportant. As the density of water is N = 3. the molecules find it hard to adjust their orientation because they cannot rotate freely in a solid.3 × 1022 cm−3 .Polarization through orientation 43 improve the estimate to trel 8 × 10−12 s but the first guess was not bad. When water freezes. .

1. some light is scattered into the direction specified by the angles (θ.1 Defining cross sections 2.1. The flux is the energy carried per unit time through a unit area and given in (1.1. They are complex quantities and Kramers and Kronig discovered a dependence between the real and imaginary parts. The flux from this scattered light. such as ε or µ. φ) as depicted in figure 2. write F0 (2. Section 2. In sections 2. we learn how to compute the scattering and absorption coefficients of particles. φ) = 2 2 Ä(θ.1 Cross section for scattering.5 is concerned with a strange but important property of the material constants that appear in Maxwell’s equations. therefore.2 deals with the optical theorem which relates the intensity of light that is scattered by a particle into exactly the forward direction to its extinction cross section.1) F(θ. we define cross sections. we approximate the material constants of matter that is a mixture of different substances.4. F(θ. When the wave hits the particle. In the final section. 2. which is received at a large distance r from the particle is obviously proportional to F0 /r 2 .3 and 2. Section 2.Chapter 2 How to evaluate grain cross sections In section 2. the scattering cross section is defined as follows. φ). The problem was first solved in the general case for spheres by G Mie [Mie08] and the underlying theory bears his name. we. Consider a plane monochromatic electromagnetic wave at frequency ν and with flux F0 . φ) does not depend on r nor on F0 . k r The function Ä(θ. absorption and extinction For a single particle. the most important quantity describing the interaction between light and interstellar grains.39) as the absolute value of the Poynting vector. φ). We have included in the denominator the wavenumber k= 44 2π λ .

φ) dimensionless.φ) Figure 2. φ). follows from the condition that F0 C sca equals the total energy scattered in all directions per unit time. C sca . The sum of absorption plus scattering is called extinction. φ). In this respect. A= C sca . the wavelength λ is then the natural length unit with which to measure the distance r . • The cross section for scattering of the particle. a particle inevitably absorbs some light. an object with a high albedo scatters a lot of light. C ext (2. Consequently. φ) sin θ with the element of solid angle d = sin θ dθ dφ. (2. A grain scatters light from a plane wave with flux F0 into the direction (θ. It is assumed that the frequency of radiation is not changed in the scattering process.2) • • The scattering cross section C sca has the dimension of an area. the temperature or density of the dust material (which only changes very little anyway).4) • It lies between 0 and 1. it is much simpler to calculate cross sections of grains than of gas atoms. (2. The albedo is defined as the ratio of scattering over extinction. . φ) d = 1 k2 2π π dφ 0 0 dθ Ä(θ.3) C ext = C abs + C sca determines the total amount of energy removed from the impinging beam of light. The extinction cross section.Defining cross sections 45 φ F 0 θ F (θ. All these various cross sections do not generally depend on the radiation field. the scattered flux is F(θ. C sca = 1 k2 4π Ä(θ. to make Ä(θ. The corresponding cross section C abs is defined by the condition that F0 C abs equals the energy absorbed by the particle per unit time. Besides scattering.1. In this direction.

7) As g can be negative. φ) specifies how the intensity of the scattered radiation changes with direction. An isotropic scatterer has f˜ = 1. there is. g = cos θ = • • • 1 2 +1 −1 f (x) x d x. the mean of cos θ over all directions weighted by the phase function f (cos θ ).2 The momentum imparted on a grain by radiation Electromagnetic radiation also exerts a pressure on a grain. the cross section for radiation pressure. (2.1 Phase function and asymmetry factor In equation (2. A photon that is absorbed deposits its full momentum hν/c.1). Again f = 1 for isotropic scattering and the normalization condition is 1= 1 2 +1 −1 f (x) d x.5) When one does not know or does not need the full information contained in f (cos θ ). It is then convenient to have a phase function f which has cos θ as the argument. C rp may be greater than To obtain the momentum transmitted per second to the grain by a flux F. Ä(θ. which is proportional to Ä(θ.2. so the transmitted momentum is F · C rp . In the limit of pure forward scattering. although this case is unusual. g is positive. the grain receives only the fraction (1 − cos θ ).1. for pure backscattering g = −1. C rp . 2.2 Cross section for radiation pressure 2. and thus independent of direction.6) It is easy to verify that g lies between −1 and 1.1). C ext . (2.8) . Therefore. If it is scattered at an angle θ (see figure 2. φ) d 4π = 4π and call it the phase function. only on θ . no dependence on the angle φ. For spheres. When scattering is isotropic. so we put f (cos θ ) = f˜(θ ). This number is the asymmetry factor g. otherwise it is negative. for reasons of symmetry. one sometimes uses just one number to characterize the scattering pattern.1. can be written as C rp = C ext − g · C sca . We form a new function f˜(θ. φ) but normalized so that the integral of f˜ over all directions equals 4π: f˜(θ. we have to divide by the velocity of light c.1.46 How to evaluate grain cross sections 2. g = 1. c (2.2. When there is mainly forward scattering. φ). g = 0. (2.

where a is the grain radius. heating it. 2. or 1 cm3 of space.1 The intensity of forward scattered light Consider a plane electromagnetic wave of wavelength λ propagating in a vacuum in the z-direction with electric field E i = ei(kz−ωt ) . some light is absorbed by the grain. is the cross section per unit mass. some light is absorbed and the rest is scattered into all . It refers either to 1 g of dust or to 1 g of interstellar matter. σgeo = πa 2 .3 Efficiencies. is the ratio of C over the projected geometrical surface area σgeo : C . The optical theorem asserts that the reduction of intensity in the forward direction fully determines the particle’s extinction cross section. again Q ext = Q abs + Q sca Q rp = Q ext − g · Q sca . The latter quantity is some hundred times smaller. Q.11) • • With the exception of spheres.2 The optical theorem When a beam of light falls on a particle.10) (2.9) Q= σgeo For spheres. K .12) For easier writing. and some is scattered. mass and volume coefficients 47 The definition of the cross section of a single particle can be extended to 1 g of interstellar matter or 1 g of dust. (2.The optical theorem 2. σgeo as well as the Cs and Qs change with the direction of the incoming light. scattering. We use the following nomenclature (omitting the dependence on frequency): • • The cross section of a single particle is denoted by the letter C.1. (2. 2. When the wave encounters a grain located at the origin of the coordinate system. we neglect the vector character of the field and assume a unit amplitude. There are efficiencies for absorption. Mass coefficient. Efficiency. It refers either to 1 cm3 in space (which typically contains 10−23 g of interstellar matter and 10−25 g of dust) or to 1 cm3 of dust material with a mass of about 1 g. extinction and radiation pressure. (2. Volume coefficient is the cross section per unit volume and also denoted by the letter K .2.

The disk has a diameter d. the scattered field can be presented as E s = S(θ.14) . At a distance r which is large when measured in wavelength units (kr = 2πr/λ 1). −i kr (2. z directions (θ. is scattered by a particle and falls at a large distance r through a disk of diameter d. φ) dimensionless. the factor −i in the denominator is just a convention.1) tells us that |S(θ. Let us now determine the flux F through a disk of area A far behind the grain. φ) ei(kr−ωt ) . φ). The wavenumber k = 2π/λ is introduced to make S(θ. Light enters from the left. (2. It lies in the (x. 2z 0 The radiation that goes through the disk consists of the incident light (E i ) and the light scattered by the particle (E s ). φ).13) with (2. y = 0. The x-axis is perpendicular to the (y.2. As before. Its center coordinates are (x = 0. z 0 ). y)-plane and is thus oriented perpendicular to the z-axis. z)-plane.13) The information about the amplitude of the scattered wave lies in the complex function S(θ. φ). they have to be added: F= A |E i + E s |2 d A. y) in the disk fulfil the inequalities |x| z0 |y| z0 and their distance to the particle is. the exponential term contains the phase. approximately. r z0 + x 2 + y2 . The two fields interfere and to obtain the flux through the disk. so d 2 ∼ A. conservation of energy requires E s ∝ 1/r .48 How to evaluate grain cross sections y d r z0 Figure 2. φ)|2 corresponds to Ä(θ. comparison of (2. All points (x. Because |E s |2 is proportional to the scattered intensity.

also known as the optical theorem. i.15) and for the sum of the incident and scattered field Ei + Es = Ei 1 − Here we put S(0) = S(θ 0. is fixed by the scattering amplitude in the forward direction alone.17) e 2z0 d x d y = e 2z0 d x = i k −∞ −∞ Of course. π (2. φ) eik(r−z) −i kr 49 (2. But the integral still gives√ more or √ the correct value as long as the disk diameter is much greater less than z 0 λ = 2π z 0 /k. λ d We.18) This is the grand extinction formula. it is required that √ A z0. (2.13) by (2. obtain the flux F = |E i |2 A − 4π Re{S(0)} . we find |E i + E s |2 = |E i |2 − 2|E i |2 Re kz 0 S(0) x 2 + y2 exp i k i 2z 0 . therefore.16) To extract the flux F from (2.18) may not be satisfying. (2.14). which includes absorption plus scattering into all directions. we may neglect terms with z 0 . in this expression for |E i + E s |2 . so the disk as viewed from the grain subtends a very small angle. S(0) x 2 + y2 exp i k i kz 0 2z 0 . φ) because d z 0 . k2 Now without an obstacle.The optical theorem Dividing (2.12). It is baffling because it asserts that C ext . The light that has been removed by the particle determines its cross section for extinction: C ext = λ2 Re{S(0)}. the disk does not really extend to infinity. we have to evaluate the integral (see (A. we get E s = E i S(θ.27)) 2 ∞ ik(x 2 +y 2 ) ∞ ikx 2 2π z 0 . In this way. the flux through the disk would obviously be A|E i |2 and thus greater. −2 Therefore.e. But . The purely mathematical derivation of (2.

obviously C ext must depend on S(0): the extinction cross section of an obstacle specifies how much light is removed from a beam if one observes at a large distance.2. we have to sum over all particles. When we compute the field at point P on the z-axis resulting from interference of all waves scattered by the grains. it receives only the forward scattered light and.17) of the preceding subsection where we considered only one particle. 2. Let a plane wave traveling in the z-direction pass through a slab as depicted in figure 2. We see from (2.3. y)-plane like the one in (2. The slab is of thickness l and uniformly filled with identically scattering grains of number density N. 1 − S(0) 2π (2. We determine the field at point P with coordinates (0.2 serve as a detector and place it far away from the particle.2 The refractive index of a dusty medium One may also assign a refractive index to a dusty medium. like a cloud of grains. The slab is uniformly filled with identically scattering particles.3.20) 1 − S(0) 2π .19) The main contribution to the double integral again comes from a region of area z 0 λ (see figure 2.2 for a definition of z 0 ). z). S(0) contains the information about C ext . This leads to an integral in the (x. If we let the disk of figure 2. therefore. 0.50 How to evaluate grain cross sections y P z l Figure 2. Light passes through a plane-parallel slab of thickness l with one face lying in the (x. no matter how and where it is removed (by absorption or scattering). But there is now another integration necessary in the z-direction extending from 0 to l. y)-plane. Altogether the field at P is: Ei + Es = Ei 1 − N = Ei S(0) i kz l dz 0 1 z ∞ −∞ dx dy e ik(x 2 +y 2 ) 2z λ2 Nl .19) that the total field at P is different from the incident field E i and it is obtained by multiplying E i by the factor λ2 Nl. (2.

Hul57. as one can show by studying the frequency dependence of S(0).21) yields n−1= 2π N Im{S(0)} k3 2π N k = − 3 Re{S(0)}. [Boh83. the series coefficients are determined from the boundary conditions on the particle surface. What is in between can be condensed into one sentence: In Mie theory.23) 1 − i kl(m − 1). Therefore. m = n + i k. A positive k implies some kind of dissipation.22) and (2. and resume the text with the last subsection entitled Absorption and scattering efficiencies. which seems unphysical as no light is absorbed.23) for a cloud filled with identical spheres of size parameter x = 2πa/λ and refractive index m = n + i k. with m and x given. Missing links may be filled in from the special literature (for instance. (2.23) because the extinction coefficient C ext of a single grain does not vanish. If the slab were empty containing only a vacuum. assign to the slab an optical constant (with a bar on top) m = n + i k. Its full derivation is lengthy and we present only the sequence of the main steps. However. . or even cut them out carefully. The reader not interested in the mathematics may skip the next pages. We can. wonder whether n and k obey the Kramers–Kronig relation of section 2. refers to a medium that consists of a vacuum plus uniformly distributed particles. it has the property that if the particles are pure scatterers. eikl(m−1) Setting (2.65). from (2.22) (2. Ker69]).5.18) and (2.20) equal to (2.41). k (2. k is nevertheless positive.21) The refractive index of the slab. one can calculate S(0). like ei(kmz−ωt ) . The way m is defined.3 Mie theory for a sphere Scattering and absorption of light by spheres is a problem of classical electrodynamics. To find the refractive index m = n + i k from (2. 2. they do. therefore. the presence of the grains causes a change in the field at P by a factor 1).Mie theory for a sphere 51 Not only do the grains reduce the intensity of radiation in the forward direction but they also change the phase of the wave because the function S(0) is complex. one finds the scattered electromagnetic field and the field inside the particle by expanding both into an infinite series of independent solutions to the wave equation. This follows from comparing (2. The field within the slab then varies according to (1. When m is close to one (|m − 1| interstellar dust cloud. therefore. it would vary like ei(kz−ωt ) . which is certainly true for any eikl(m−1) . One may.

The vector function N given by N= also obeys the wave equation N + k2 N = 0 and M and N are related through rot N = kM. Let c be an arbitrary constant vector and ψ a solution to the scalar wave equation ψ + k2 ψ = 0 where (2. (2.1 The generating function Consider a spherical particle in vacuum illuminated by a linearly polarized monochromatic plane wave of frequency ν = ω/2π. H1 and outside of it by E2 . 2. E2 = Ei + Es H2 = Hi + Hs . (2.29) .25) The further calculations are greatly simplified by the following relations.26) ω2 µε c2 from (1. All material properties are taken into account by the wavenumber k. (2.32) (2. We denote the field within the particle by E1 .52 How to evaluate grain cross sections 2.116) in the case of conductivity).2 Separation of variables In this way.31) (2.28) (2.27) This is easy to prove either by the standard formulae of vector analysis or component–wise. the problem of finding a solution to the vector wave equation reduces to finding one for the scalar wave equation. H2 .3.3. We start with the vector function M = rot(rψ).24) (2. defined by k2 = M = rot(cψ) is divergence-free (div M = 0) and a solution to the vector equation M + k2 M = 0. Then the vector function M. The field outside is the superposition of the incident and the scattered field (subscript s).41) (or (1.30) 1 rot M k (2. Let Ei and Hi describe the incident field.

e stands for even (associated with cosine terms) and o for odd (sine terms). we can separate θ and φ. which we write as n(n + 1).37) + k2r − n(n + 1) R = 0. where n and m are integer and m ∈ [−n.Mie theory for a sphere 53 M is tangential to the sphere of radius |r| because the scalar product of r and M vanishes: r · M = 0. both sides must equal a constant value.42) (2.36) is satisfied by the Legendre functions of the first m kind Pn (cos θ ). Memn = rot(rψemn ) Momn = rot(rψomn ). In spherical polar coordinates (r. .34) We make an ansatz of separated variables ψ(r.29). n]. θ. the wave equation for ψ reads: 1 ∂ r 2 ∂r r2 ∂ψ ∂r + 1 ∂ 2 sin θ ∂θ r sin θ ∂ψ ∂θ + ∂ 2ψ + k2 ψ = 0. In the same spirit. the spherical Bessel functions of the first ( jn ) and second (yn ) kind jn (ρ) = yn (ρ) = where ρ = kr and n + 1 2 π J 1 (ρ) 2ρ n+ 2 π Y 1 (ρ) 2ρ n+ 2 (2. φ). (2. Arranging (2. as solutions.37) has. In the subindices of ψ. θ.36) (2.40) (2. Nemn and Nomn then follow from (2. (2. The linearly independent solutions to (2.43) . we get. .33) in such a way that its left-hand side depends only on r and its right-hand side only on θ and φ. .27). φ) = R(r ) · T (θ ) · P(φ). . As ψ is the generating function for M. from (2. Altogether we obtain (2.33) r 2 sin2 θ ∂φ 2 1 (2.41) m Pn (cos θ ) · z n (kr ). ±2.39) is half-integer. Because they must be single-valued (P(φ) = P(φ + 2π)) it follows that m = 0.38) (2.35) (2. Equation (2. m ψemn = cos(mφ) · Pn (cos θ ) · z n (kr ) ψomn = sin(mφ) · Here z n may either equal jn or yn . Formula (2. ±1.35) are sin mφ and cos mφ. This leads to the three equations: d2 P + m2 P = 0 dφ 2 1 d sin θ dθ sin θ d dr dT dθ r2 + n(n + 1) − dR dr m2 sin2 θ T =0 (2.

but. of course.46) Superscript (3) denotes that the dependence of the generating function is given (1) by the spherical Hankel function h n (z) = jn (z) + i yn (z) of order n. Note that all coefficients with m = 1 have disappeared.3.44) where the scalar E 0 denotes the amplitude of the incident wave. For the rest of this paragraph. When we integrate rot E = (i ωµ/c)H from (1. internal and scattered waves into the spherical harmonics Memn . the integral vanishes so that the tangential components of E .45) Es = (2. the other is immediately below the boundary within the particle.3 Series expansion of waves In the next step. it behaves like distances (kr n h (1) (kr ) n (−i )n eikr . the latter can be excluded because of its behavior at the origin. The superscript (1) at M and N signifies that the radial dependence of the generating function ψ is given by jn and not by yn . Momn . n=1 (3) (3) (1) (1) (2. consider a small loop in the shape of a rectangle with two long and two very much shorter sides. One long side lies just outside the particle in vacuum and runs parallel to the surface S of the grain.25). the symbol m is reserved for the optical constant.32) over the rectangular loop employing Stokes’ theorem (A.47) The magnetic fields Hi .54 How to evaluate grain cross sections 2.4 Expansion coefficients The expansion coefficients in (2. the not straightforward result is ∞ Ei = n=1 E n Mo1n − i Ne1n (1) (1) with E n = E 0 i n 2n + 1 n(n + 1) (2. 2. For the internal and scattered field one obtains ∞ Ei = n=1 ∞ E n (cn Mo1n − i dn Ne1n ) E n (an Ne1n − bn Mo1n ). one expands the incident.15) and make the loop infinitesimally small. i kr (2. these coordinates are the right ones for the scattered light.46) follow from the boundary conditions of the electromagnetic field at the surface of the grain. Starting with the incident plane wave Ei . Hs follow from the corresponding electric fields by applying the curl after Maxwell’s equation (1.3.45) and (2. At large 2 ). Nemn and Nomn .44) for the simple incident wave is the consequence of using spherical coordinates which are not suited for planar geometry. The awkward expression (2. To derive the latter.

49) where e is the outward directed normal to the surface S. and x= 2πa λ the size parameter. the scattered field disappears as an = bn = 0. when the optical constant m = 1. section 10.113) for metals. Therefore. This set of equations leads to four linear equations for the expansion coefficients an . or to (1. Applying the same procedure to equation (1.19] 2n − 1 jn−1 (z) z 2n − 1 yn−1 (z) yn (z) = − yn−2 (z) + z jn (z) = − jn−2(z) + . m the complex optical constant of the sphere.50) (2.51) In the trivial case. Substituting for E2 . we get the same result for the tangential component of H. mζn (x) · ψn (mx) − ψn (mx) · ζn (x) an = (2.48) (2.33). a its radius. The complex functions ψn (z) = z jn (z) ψn (z) = z jn−1 (z) − n jn (z) ζn (z) = z[ jn (z) + i yn (z)] = zh (1) (z) n ζn (z) = z[ jn−1 (z) + i yn−1 (z)] − n[ jn (z) + i yn (z)] may be calculated from the recurrence relations [Abr70. then an and bn are given by ψn (x) · ψn (mx) − mψn (mx) · ψn (x) ζn (x) · ψn (mx) − mψn (mx) · ζn (x) mψn (x) · ψn (mx) − ψn (mx) · ψn (x) bn = . at all points x ∈ S [E2 (x) − E1 (x)] × e = 0 [H2 (x) − H1 (x)] × e = 0 (2.1. bn . If λ is the wavelength of the incident radiation.Mie theory for a sphere 55 must be continuous at the boundary of the grain. cn and dn of the internal and scattered field (see (2.46)). H2 from (2.45) to (2.25) yields for the components of the fields E 1θ = E iθ + E sθ E 1φ = E iφ + E sφ H1θ = Hiθ + Hsθ H1φ = Hiφ + Hsφ .24) and (2.

Wa . The second integral obviously describes the scattered energy.1.5 Scattered and absorbed power If we imagine the particle to be surrounded by a large spherical and totally transparent surface A.12] sin z sin z cos z j1(z) = 2 − z z z cos z cos z sin z y1 (z) = − 2 − . and the minus sign ensures that Wa is positive. −Wa = A Si = (c/8π) Re{Ei × H∗ } i Si · er d A + A Ss · er d A + A Sext · er d A. The first integral vanishes because the incident field (subscript i) enters and leaves the sphere without modification. Ws = c Re 8π 2π 0 0 π ∗ ∗ E sθ Hsφ − E sφ Hsθ r 2 sin θ dθ dφ. (2.52) Here er is the outward normal of the surface. y0 (z) = − z z z j0 (z) = When |mx| is of order 100 or bigger. (2. The Poynting vector S can be considered to consist of three parts: S = Si + Ss + Sext with Ss = (c/8π) Re{Es × H∗ } s Sext = (c/8π) Re{Ei × H∗ + Es × H∗ } s i and. sections 10. the energy absorbed by the grain. If S is the Poynting vector of the electromagnetic field outside the particle.24) and (2. Wa is determined by the integral Wa = − A S · er d A. 2. one encounters numerical difficulties because the complex Bessel functions contain the term e z which becomes excessively large.11 and 10.53) . therefore. is given by the difference between the flux which enters and which leaves the sphere.56 How to evaluate grain cross sections starting with [Abr70.1.25). S= c Re{E2 × H∗ } 2 8π where E2 or H2 is the sum of the incident and scattered field from (2. A numerically much superior algorithm is printed in appendix A of [Boh83].3.

C abs . n+1 n(n + 1) (2. the sum of absorbed plus scattered energy: Wext = Wa + Ws 2π c = Re 8π 0 π 0 ∗ ∗ ∗ ∗ E iφ Hsθ − E iθ Hsφ − E sθ Hiφ + E sφ Hiθ r 2 sin θ dθ dφ (2.3. According to our definition in section 2.4. n=1 (2.Polarization and scattering 57 The negative of the third integral. The asymmetry factor of (2.1 The amplitude scattering matrix Consider a plane harmonic wave propagating in the z-direction and a particle at the origin of the coordinate system.54) and thus the total energy removed from the beam at a large distance. not fully derived but the missing links concern only the mathematics. define the Stokes parameters and compute the radiation field scattered into a certain direction.56) They must usually be evaluated with the help of a computer.1. where Si is the time-averaged Poynting vector of the incident field. Some light is scattered by the particle into a .50) and (2. bn are given in (2.46).55) (2n + 1) · |an |2 + |bn |2 . one obtains from (2. Wa is related to the absorption coefficient of the particle. Likewise for the scattering and extinction coefficient. 2. A number of results in this section are merely summarized.57) 2. therefore.53) and (2.6) becomes g= 4 x 2 Q sca ∞ n=1 2n + 1 n(n + 2) ∗ ∗ ∗ Re{an an+1 + bn bn+1 } + Re{an bn } . through Wa = Si C abs .6 Absorption and scattering efficiencies As the fields Ei and Es have been evaluated in (2. is.54) after some algebra the following formulae for the efficiencies of extinction and scattering: Q ext = Q sca = 2 x2 2 x2 ∞ (2n + 1) · Re{an + bn } n=1 ∞ (2. which we denote by −Wext .4 Polarization and scattering We introduce the amplitude scattering matrix. The expansion coefficients an .51). 2. Ws = Si C sca and Wext = Si C ext .44)–(2.

(2. The z-axis together with e define what is called the scattering plane. Here we deal with vectors and there is an amplitude scattering matrix consisting of four elements S j . p. θ and φ.4. on the scattering direction.59) 2.4 we can easily figure out that E s = cos θ E i and E i⊥ = E s⊥ . respectively. the amplitude scattering matrix (2. They depend. one parallel. (2. Es = ω2 α (e × Ei ) × e. due to the incident wave is p = αEi . so for a dipole S2 S4 S3 S1 = −i k3 α cos θ 0 0 1 . in the far field where r λ. c 2r It is always transverse to e and depends only on e and p (or Ei ) and not on the direction k from which the incident wave is coming. with only one function S. If its polarizability α is isotropic. the other perpendicular to the scattering plane. missing. At a large distance r from the particle. With the help of figure 2. there is a linear relation between (E s⊥ .58) The factor before the amplitude scattering matrix is the same as in (2.15) where we considered a scalar field.60) . of course.58) also acquires a diagonal structure as the elements S3 and S4 vanish: Es E s⊥ = eik(r−z) −i kr S2 0 0 S1 Ei E i⊥ . A threedimensional model would help visualize the geometrical configuration but a twodimensional drawing is no better than words and. Likewise. the components being E s and E s⊥ . we work out the scattering matrix of a small grain of unit volume. The scattered electric field is given by equation (1. E i ) described by the amplitude scattering matrix. The amplitude of the incident electric field Ei may be decomposed into two vectors. As an example. therefore.95). E s ) and (E i⊥ . Es E s⊥ = eik(r−z) −i kr S2 S4 S3 S1 Ei E i⊥ . its dipole. which is specified by the unit vector e or by two angles.2 Angle-dependence of scattering In the case of a sphere. their lengths are denoted by E i and E i⊥ . (2.58 How to evaluate grain cross sections certain direction given by the unit vector e. one may decompose the scattered electric field Es .

S1 = n 2n + 1 (an πn + bn τn ) n(n + 1) 2n + 1 (an τn + bn πn ) n(n + 1) (2.62) S2 = n with 1 1 Pn (cos θ ) d Pn τn (cos θ ) = . S1 .61) (2. .51). beginning with • For unpolarized incident light (E i = E i⊥ ). S2 depend only on µ = cos θ . The notation S11 comes from equation (2. They can again be expressed with the help of the expansion coefficients an . Alternatively (this scenario is not shown in the figure). The incident wave travels in the direction of the wavenumber k. The scattered wave has the electric field Es . Then Es is also perpendicular to the scattering plane.Polarization and scattering 59 Es p Ei k θ e z r Figure 2.1) the same quantity was denoted Ä. An incident wave with electric field Ei excites a dipole p and is scattered into the direction of the unit vector e.50) and (2.63) The factor before the matrix in (2. Ei and p may be perpendicular to the scattering plane given by the same vectors k and e. sin θ dθ The functions πn and τn are computed from the recurrence relations πn (cos θ ) = πn (µ) = 2n − 1 n µπn−1 − πn−2 n−1 n−1 τn (µ) = nµπn − (n + 1)πn−1 π0 = 0 π1 = 1. in (2. Both Ei and Es lie in the scattering plane which is defined by e and k. bn from (2. the intensity of the radiation scattered into the direction θ is given by S11 (cos θ ) = 1 2 |S1 |2 + |S2 |2 . where θ = 0 denotes the forward direction.60) has been negelected. (2.4.76).

36) and (1. Either from (2.60 How to evaluate grain cross sections When integrated over all directions. in the general case. y)-plane. the real part of the electric vector E rotates at frequency ω and the tip of the vector describes in the (x.61) into the general extinction formula (2. π 0 S11 (cos θ ) sin θ dθ = 1 x 2 Q sca .3 The polarization ellipse Consider a plane harmonic wave of angular frequency ω. Q back = 1. Its series expansion is Q back = 1 x2 ∞ n=1 2 (−1)n (2n + 1) · (an − bn ) .67) For a large and perfectly reflecting sphere. Comparison of the two formulae yields S1 (0) = S2 (0) = 1 2 n (2n + 1) · Re{an + bn }. complex.64) • We now have two ways to determine the extinction cross section C ext = πa 2 Q ext of a grain.65) • We add for completeness the formula for the backscattering efficiency. Re{E} = E1 cos ωt + E2 sin ωt. (2. an ellipse. E0 = E1 + i E2 with real E1 and E2 .4.55) or by inserting S1 (0) = S2 (0) of (2. 2. It is defined by S11 (180◦) (2.18).37)) that travels in the z-direction of a Cartesian coordinate system. (2. The amplitude E0 is.68) .5)) is related to S11 through S11 (cos θ ) = x 2 Q sca f (cos θ ). 4 (2. which is perpendicular to the z-axis. 2 • The angle-dependence of the normalized phase function f (cos θ ) (see (2. wavenumber k and electric field E(x. At any fixed z.66) Q back = 4 x2 and it is a useful quantity in radar measurements or whenever a particle is illuminated by a source located between the particle and the observer. (2. t) = E0 · ei(k·x−ωt ) (see (1.

At a fixed location z. y)-plane. it transmits angular momentum.Polarization and scattering 61 The sense of rotation changes when the plus sign in (2.58) and. When a plane harmonic wave is scattered by a grain. 2. At a fixed location z. therefore. it is always possible to make E0 real. r is the distance from the particle. Q. y)-plane (with equal properties in the x. b.and y-directions) into a circular motion. y)-plane. . when E2 flips direction by an angle π. the ellipse degenerates into a line. E2 are linearly dependent.e.  Us S31 S32 S33 S34 Ui k r Vs S41 S42 S43 S44 Vi The Stokes parameters of the scattered light as well as the matrix elements refer to a particular scattering direction (θ. There are two special cases: • Linear polarization. Circular polarization. Instead of these geometrical quantities (a. For E1 = 0 or E2 = 0. The ellipse is a circle. The matrix Si j contains no more information than the matrix in (2. E1 · E2 = 0. or when E1 . only seven of its 16 elements are independent. • One can combine two linearly polarized waves to obtain circular polarization and two circularly polarized waves to obtain linear polarization. a and b. By adjusting the time t. γ ).68) is altered into a minus sign. |E1 | = |E2 |. φ). the Stokes parameters of the incident (subscript i) and the scattered wave (subscript s) are linearly related through      S11 S12 S13 S14 Ii Is 1  S21 S22 S23 S24   Q i   Qs  (2. This could be the angle γ between the major axis and the x-coordinate. We omit the underlying simple mathematical relations as well as the description of the experimental setup. and perpendicular to each other.58) have four absolute values |S j | and three phase differences between them. The vectors are of equal length. which corresponds to the fact that the four elements S j in the matrix of (2. b. it vanishes when Re{E} switches direction. the electric vector never disappears but rotates retaining its full length.4.69) = 2 2  . U and V . They are equivalent to (a. the electric vector swings up and down along a straight line whereby its length changes.4 Stokes parameters The polarization ellipse is completely determined by the length of its major and minor axes. The field can drive a linear oscillator in the (x. γ ) but have the practical advantage that they can be measured directly. The circularly polarized wave sets a two-dimensional harmonic oscillator in the (x. polarization is usually described by the Stokes parameters I. The magnetic field has the same polarization as the electric field because E and H are in phase and have a constant ratio. plus some specification of its orientation in the (x. i.

69). I (2. dropping the factor 1/k2r 2 in (2. S34 Ui S33 Vi (2.74) 2. The degree of circular polarization is V −1 ≤ ≤ 1. Circular polarization implies Q=U =0 V = ±1.5 Stokes parameters of scattered light for a sphere In the case of a sphere. (2.4. • the sign determines the sense of rotation of the electric vector. an inequality I 2 > Q2 + U 2 + V 2.e.72) I Even when the incident light is unpolarized.75) . (2. instead of the strict equality I 2 = Q 2 + U 2 + V 2 .73) it becomes partially polarized after scattering if S21 . The degree of linear polarization is defined by Q2 + U 2 ≤ 1. the transformation matrix scattered Stokes parameters simplifies to    S11 S12 0 Is 1  S12 S11 0  Qs  = 2 2  Us 0 0 S33 k r Vs 0 0 −S34 (2. one has.71) Q = cos 2γ U = sin 2γ • It can vary between 0 and 1. i. when Q i = Ui = Vi = 0 (2.69) between incident and   Ii 0 0   Qi   . we get Is = S11 Ii Q s = S21 Ii Us = S31 Ii Vs = S41 Ii .70) Unpolarized light has Q=U =V =0 and for partially polarized light. (2. Indeed. • • For unit intensity (I = 1).62 • How to evaluate grain cross sections Only three of the four Stokes parameters are independent as for fully polarized light I 2 = Q2 + U 2 + V 2. S31 or S41 are non-zero. linearly polarized light has V =0 so that Q 2 + U 2 = 1.

S12 is negative and then linear polarization is perpendicular to the scattering plane.8). if the incident light is 100% polarized perpendicular to the scattering plane (Ii = −Q i . eight are non-trivial and they have only four significantly different values: S11 = 1 (|S1 |2 + |S2 |2 ) 2 S12 = S33 = S34 = 2 1 2 (|S2 | 1 ∗ 2 (S2 S1 ∗ i 2 (S1 S2 (2.7 and 4.69) vanishes. specifies the direction of polarization. Then the polarization vector can have any inclination towards the scattering plane and p may also be non-zero in the forward direction. or of S12 .76). in this case.76) − |S1 | ) ∗ + S2 S1 ) ∗ − S2 S1 ).Polarization and scattering 63 Out of the 16 matrix elements. The sign of p. Usually. dropping the factor 1/k2r 2 in (2. also be positive. But S12 can. . none of the matrix elements in (2. we get. In the forward direction. as happens for big spheres (see figures 4. Then polarization is parallel to the scattering plane. when the particles are not spherical (anisotropic) and which is outside the scope of what we calculate here.71). S1 (0) = S2 (0) and p = 0. When one defines the quantity p as the difference of the intensities |S1 |2 and |S2 |2 divided by their sum. • So the scattered light is also 100% polarized parallel to the scattering plane. Ui = Vi = 0). Likewise. • If the incident light is 100% polarized and its electric vector parallel to the scattering plane so that Ii = Q i . via the sign. Is = S11 Ii Q s = S12 Ii Us = Vs = 0.77) • the absolute value | p| is equal to the degree of linear polarization of (2. the scattered light is nevertheless usually polarized. from (2. In the general case.75). Ui = Vi = 0. so is the scattered light and Is = (S11 − S12 )Ii Q s = −Is Us = Vs = 0. although it contains. additional information. We will not compose a new name for p but also call it degree of polarization. • If the incident wave is unpolarized (Q i = Ui = Vi = 0). p= S12 |S1 |2 − |S2 |2 =− |S1 |2 + |S2 |2 S11 (2. Is = (S11 + S12 )Ii Q s = Is Us = Vs = 0. 2 Only three of them are independent because 2 2 2 2 S11 = S12 + S33 + S34 .


How to evaluate grain cross sections

2.5 The Kramers–Kronig relations
The formulae of Kramers and Kronig establish a link between the real and imaginary parts of the material constants, like χ, ε, µ or σ (see section 1.1). Their deduction is very abstract and the result baffling. The relations have some fundamental consequences, however, their practical value is, at present, moderate. 2.5.1 Mathematical formulation of the relations We have to digress briefly into the theory of complex functions. Let f (z) = u(z) + i v(z) be a function of the complex variable z = x + i y. The integral over a path p in the z-plane, described through the parametrization z(t) = x(t) + i y(t) with α ≤ t ≤ β, is defined by f (z) dz =
p β α

f (z(t)) · z (t) dt.


The fundamental theorem of complex functions states that if f (z) is regular in a region G, i.e. it is single-valued and has a derivative, then the integral over any closed path p in G vanishes: f (z) dz = 0.


For the function f (z) we choose f (z) = where g(z) = g1 (z) + ig2 (z) is also a complex function and x 0 a real and positive number. If f (z) is regular in the upper half of the z-plane, the integral along the closed path p, which is depicted in figure 2.5, is zero according to (2.79). The path p runs along the xaxis from left to right, makes a small semicircle of radius δ around x 0 , and returns in a big half–circle of radius R to the starting position; we have divided it into four segments p1 to p4. Now assume that R is very big and that g(z) has the property to vanish for |z| → ∞; the integral along the big half–circle p4 then vanishes too. The small semicircle p2 may be parametrized by z(t) = x 0 − δe−it with 0 ≤ t ≤ π. The integral over p2 gives −i
0 π

g(z) z − x0


g(x 0 − δeit ) dt = −i πg(x 0)

for δ → 0.


The Kramers–Kronig relations



p4 p2


R x

Figure 2.5. To derive the Kramers–Kronig relations, we integrate a function f that is related to the dielectric susceptibility χ, or permeability ε, over a closed path (thick line) in the complex (x, y)-plane. The path consists of four segments, p1 to p4 . The points in this plane are identified with complex frequencies ω. Only the positive x-axis has a physical meaning but equation (2.88) allows us to define χ for any complex argument ω.

We thus have for the whole path p, from (2.79), −


g(x) d x − i πg(x 0) = 0. x − x0


The integral with the bar is the Cauchy principal value defined by −


g(x) d x = lim δ→0 x − x0

x 0 −δ −∞

g(x) dx + x − x0

∞ x 0 +δ

g(x) dx . x − x0


Equation (2.82) holds separately for the real and imaginary parts. Let the function g be symmetric (for real x) such that g(−x) = g ∗ (x) or g1 (−x) = g1 (x) and g2 (−x) = −g2 (x).

Writing the integral in (2.82) over the whole x-axis as the sum of two integrals with limits from −∞ to 0 and from 0 to +∞ and exploiting the symmetry of g, one obtains the Kramers–Kronig relations after small manipulations g1(x 0 ) = 2 ∞ xg2 (x) − dx 2 π 0 x 2 − x0 2x 0 ∞ g1 (x) − d x. g2(x 0 ) = − 2 π 0 x 2 − x0 (2.84) (2.85)


How to evaluate grain cross sections
Putting in (2.84) x 0 = 0 yields the special case (no Cauchy principal value) g1 (0) = 2 π
∞ 0

g2(x) d x. x


In formula (2.83), both integrands within the big brackets go at x 0 to infinity like 1/x at x = 0. Therefore, each of the integrals alone diverges; however, their sum is finite. 2.5.2 The electric susceptibility and causality In the application of equations (2.84)–(2.86) to a physical situation, the complex function g is identified with the electric susceptibility χ of (1.6) and the variable x with the frequency ω. Likewise, we could use the dielectric permeability ε, with the small modification that we have to put g1 = ε1 − 1 and g2 = ε2 . We reiterate that, in the constitutive relation (1.6), P = χ(ω) E the polarization P depends linearly on the electric field E because E is so much weaker than the fields on the atomic level. In the most general case, the vector P at time t is not determined by the present value of E alone but by the whole preceding history. We, therefore, write the linear relation in the form P(t) =
0 ∞

F(τ ) E(t − τ ) dτ.


This equation specifies how P responds to the application of an electric field E. Besides linearity, formula (2.87) also expresses causality because P(t) results from an integration over the past. The function F(τ ) depends on time and on the properties of the medium. It has values substantially different from zero only in the immediate past over an interval τ corresponding to the time scale for polarizing the molecules. At times much larger than τ , the function F(τ ) vanishes because the distant past does not influence the present. When we consider monochromatic electric fields, E(t) = E0 e−iωt , the polarization P = χE has exactly the form of (2.87), if the dielectric permeability is given by χ(ω) =
0 ∞

eiωτ F(τ ) dτ.


Although only a real and positive frequency has a physical meaning, this equation formally extends χ to any complex argument ω of the upper half of figure 2.5; in the lower half where Im{ω} < 0, the integral (2.88) diverges. Such an extension is necessary to perform the integration in (2.79). We summarize here some of the properties of the electric susceptibility of (2.88): • If |ω| → ∞, χ goes to zero. For large and real frequencies, it is physically understandable because the electrons cannot follow the field, so

The Kramers–Kronig relations


the polarization is zero; for large imaginary ω, it is mathematically clear because of the factor e−|ω|τ in (2.88). The function F(τ ) in (2.87) must for physical reasons, and to guarantee causality, be finite over the integration interval [0, ∞], i.e. over the past. As a consequence, the susceptibility χ of (2.88) has no singularities in the upper half of the complex ω-plane and this is a necessary condition for obtaining the Kramers–Kronig relations (see (2.79)). χ fulfils the symmetry relations χ(−ω) = χ ∗ (ω) χ(−ω ) = χ (ω)
∗ ∗

for real ω for complex ω.


The last equation expresses the fact that a real field E produces a real polarization P. When, in (2.80), we replace g(z) by χ(ω), we may convince ourselves that χ(ω) has all the desired mathematical properties to obey the formulae (2.84)– (2.86), if ω0 is a real and positive frequency. 2.5.3 The Kramers–Kronig relation for the dielectric permeability Here is the final result formulated explicitly for the dielectric permeability. For an arbitrary medium, ε1 and ε2 are not completely independent of each other but for any frequency ω0 , ε1 (ω0 ) − 1 = 2 ∞ ωε2 (ω) − dω 2 π 0 ω2 − ω0 2ω0 ∞ ε1 (ω) − dω. ε2 (ω0 ) = − 2 π 0 ω2 − ω0 2 π
∞ 0

(2.90) (2.91)

For the static limit of ε1 at zero frequency, ε1 (0) − 1 = ε2 (ω) dω. ω (2.92)

Similar relations hold for the electric susceptibility χ, the electric polarizability αe or the optical constant m = n + i k. Whenever the vacuum value of the material constant is one (as for ε), the −1 appears on the left-hand side (see (2.90) and (2.92)), when the vacuum value is zero (as for χ), the −1 is missing. 2.5.4 Extension to metals In a metallic medium, the conductivity σ of (1.119) also follows the Kramers– Kronig (KK) relations. We may either put g(ω) = σ (ω) or g(ω) = i 4πσ (ω) ω


How to evaluate grain cross sections

which is the expression that appears in the generalized permeability of (1.115). However, for a metallic medium we have a pole not only at ω0 (see figure 2.5) but also at ω = 0. When, on itegrating the function g(ω)/(ω − ω0 ) along the real frequency axis from −∞ to +∞, we circumvent this additional pole at ω = 0 into the upper half of the complex plane in the same way as ω0 (see (2.81)), we get an additional term −4π 2 σ (0)/ω0 on the left-hand side of equation (2.82) where σ (0) is the direct-current conductivity. When we consider a conducting medium with a dielectric constant ε = εd + i 4πσ = ε1 + i ε2 ω

from (1.115), equation (2.91) for ε2 (ω0 ) has to be replaced by ε2 (ω0 ) = − 2ω0 ∞ ε1 (ω) 4πσ (0) − dω + . 2 π 0 ω2 − ω0 ω0 (2.93)

Equation (2.90) for ε1 (ω0 ) stays in force as it is but the long wavelength limit ε1 (0) must be adapted because ε2 (ω) has a singularity at ω = 0. Going back to the more basic formula (2.84) for ε1 (ω0 ), there now appears a term −
0 ∞

dω ω2
2 − ω0

which is always zero. What remains is the following modification of (2.92): ε1 (0) − 1 = 2 π
∞ 0

ε2 (ω) − 4πσ (0)/ω dω. ω

2.5.5 Dispersion of the magnetic susceptibility The magnetic susceptibility χm is defined in equations (1.16) and (1.17) and connects the field H with the magnetization M through M = χm H. M has, from (1.13) and (1.14) the physical meaning of a volume density of magnetic moments. To speak of a magnetic susceptibility χm makes sense only if, neglecting free charges, in the expression (1.21) for the total current density Jtot the magnetic current dominates over the polarization current: Jmag = c rot M ˙ Jpol = P.

˙ With the help of Maxwell’s equation, rot E = −B/c, we get the order of magnitude estimate ˙ ˙ P = χe E ∼ χe ωE ∼ χe ω2 µH l/c.

The Kramers–Kronig relations


Assuming χe , µ ∼ 1 and a typical length l that must be much greater than Jpol requires frequencies an atomic radius, the condition Jmag ω cχm /l. Therefore χm (ω) becomes constant and real above some critical value ωcr . In practice, magnetic dispersion stops well below optical frequencies, so ωcr 1015 s−1 . Whereas at high frequencies the dielectric permeability ε approaches one, the magnetic permeability µ goes to a value µcr = µ(ωcr ) which may be different from unity. This necessitates the following modification in the KK relations: 2 ∞ ωµ2 (ω) µ1 (ω0 ) − µcr = − dω. 2 π 0 ω2 − ω0 2.5.6 Three corollaries of the KK relation The dependence between ε1 and ε2 Any set of physically possible values ε1 (ω) and ε2 (ω) for any grain material must obey the KK relations (2.90) and (2.91). They thus serve as a check for the internal consistency of data measured, for example, in the laboratory or derived otherwise. It is even sufficient to know one of them over the entire wavelength range, either ε1 (ω) or ε2 (ω), to compute the other. Whereas ε1 (ω) is not restricted at all, ε2 (ω) is associated with the entropy and must be positive everywhere. A data set for ε1 (ω) is wrong if it yields at just one frequency a negative value for ε2 (ω). As an example of a dispersion formula that obeys the KK relations we may take equation (1.77) or (1.121). They apply to the harmonic oscillator or a metal, respectively. That they fulfil the KK relations may be verified from general mathematical considerations for the function ε(ω), which is the smart way; or by doing explicitly the KK integrals for (1.77), that is the hard way; or numerically, which is the brute way. Even the last method requires some delicacy when handling the Cauchy principal value. Numerical integration is inevitable when ε(ω) is available only in tabulated form. When we look at the dispersion relation (1.77) and realize that ε1 and ε2 have the same denominator and contain the same quantities e, m e , γ , ω and ω0 , equations (2.90)–(2.92) which link ε1 with ε2 are no longer perplexing. However, when we discussed the physics associated with the optical constants n and k, which determine the phase velocity and the extinction, the two parameters appeared very distinct and independent so that any general connection between them comes, at first glance, as a surprise. Dust absorption at very long wavelengths In a slowly varying electromagnetic field, a dielectric grain of arbitrary shape and composition absorbs, from (1.55), the power
2 W = 1 V ω Im{αe }E 0 = C abs (ω) · S 2


How to evaluate grain cross sections

2 where V is the volume of the grain, C abs its cross section and S = (c/8π)E 0 the Poynting vector. The particle has to be small, which means that the frequency of the wave must stay below some critical value, say ω1 . We now apply equation (2.86) to the polarizability α of the grain and split the integral into two:

α1 (ω = 0) =

2 π 2 = π

∞ 0 λ1 0

α2 (ω) 2 dω = ω π 2 α2 (λ) dλ + λ π

∞ 0 ∞ λ1

α2 (λ) dλ λ α2 (λ) dλ. λ

Note that we have swapped the integration variable from frequency to wavelength. Because α2 is positive and the integral over α2 /λ in the total interval [0, ∞] finite, the last integral must also be finite. If we make in the range λ > λ1 = 2πc/ω1 the replacement C abs (λ) α2 (λ) = λ 8π 2 V and write C abs = σgeo Q abs , where σgeo is the geometrical cross section, we get the following convergence condition:
∞ 0

Q abs (λ) dλ < ∞.

Therefore, the absorption efficiency of any grain must, at long wavelengths, fall off more steeply than λ−1 . But this last constraint attains practical importance only when we know the threshold wavelength after which it is valid. We could also have derived the result for Q abs (λ) at long wavelengths using the dielectric permeability ε instead of αe but then metals would have required some extra remarks because in conductors ε2 → ∞ for ω → 0, whereas their polarizability αe stays finite (see, for instance, (3.9)). A similar discussion to the one we carried out for αe holds for the magnetic polarizability αm . Total grain volume One may apply the KK relation not only to grain material but also to the interstellar medium as such [Pur69]. Because it is so tenuous, its optical constant m = n + i k is very close to one. A small volume V in the interstellar medium of size d may, therefore, be considered to represent a Rayleigh–Gans particle. This class of grains, for which |m − 1| 1 and d|m − 1|/λ 1, is discussed in section 3.5. Of course, the particle is inhomogeneous as it consists of a mixture of gas, dust and mostly vacuum but for its absorption cross section we may nevertheless use (3.42) C abs (λ) = 2π V 4π V k(λ) = ε2 (λ). λ λ

Composite grains
The static limit (2.92) of the KK relation then gives ε1 (0) − 1 = 2 π
∞ 0


1 ε2 (λ) dλ = 2 λ π V

∞ 0

C abs (λ) dλ.


In a static electric field E, a volume V of interstellar medium with dielectric permeability ε(0) has, from equations (1.3), (1.6) and (2.94), a dipole moment p= E ε(0) − 1 EV = 4π 4π 3
∞ 0

C abs (λ) dλ.

Alternatively, we may express p as resulting from the polarization of the individual grains within V . They have a total volume Vg and at zero frequency a dielectric constant εg (0), so, from (3.8), p= εg (0) − 1 3 Vg E. 4π εg (0) + 2

When we crudely evaluate the long wavelength limit of this ratio from figure 7.19, we obtain a value of order one, εg (0) − 1 ∼ 1. εg (0) + 2 Along a line of sight of length L, the optical depth for absorption is τ (λ) = LC abs . Therefore, the total grain volume Vdust in a column of length L that produces an optical depth τ (λ) and has a cross section of 1 cm2 is Vdust 1 3π 2
∞ 0

τ (λ) dλ.

When we take the standard normalized interstellar extinction curve of figure 7.8 for τ (λ), which refers to a visual extinction AV = 1 mag, we obtain for the previous integral in the interval 0.1–10 µm a value of about 2 × 10−4 . Consequently, the total dust volume in a column of 1 cm2 cross section with AV = 1 mag is Vdust 6 × 10−6 cm−3 . (2.95) Such an estimate is of principal value, although its precision does not allow one to discriminate between grain models.

2.6 Composite grains
Interstellar grains are probably not solid blocks made of one kind of material but are more likely to be inhomogeneous so that the dielectric function within them varies from place to place. There are many ways in which inhomogeneity can come about. For instance, if particles coagulate at some stage during their


How to evaluate grain cross sections

evolution, the result will be a bigger particle with voids inside. The new big grain then has a fluffy structure and for its description, even if the grains before coagulation were homogeneous and chemically identical, at least one additional dielectric function is needed, namely that of vacuum (ε =1). If chemically diverse particles stick together, one gets a heterogenous mixture. We note that purely thermal Brownian motion of grains (see section 9.3) is too small to make encounters between them significant but omnipresent turbulent velocities of 10 m s−1 are sufficient to ensure coagulation in dense clouds. Another way to produce inhomogeneities is to freeze out gas molecules on the surface of grains; this happens in cold clouds. A third possibility is that during grain formation tiny solid subparticles, like PAHs (section 12.1.1) or metal atoms, are built into the bulk material and contaminate it chemically. The chemical composition of interstellar dust is discussed in section 7.4 and in chapter 12. 2.6.1 Effective medium theories The cross section of composite particles can be computed exactly in those few cases where the components are homogeneous and the geometrical structure is simple; examples are spherical shells, cylinders with mantles or coated ellipsoids. For any real composite particle, where different media are intermixed in a most complicated manner, such computations are out of the question. To estimate the optical behavior of composite particles, one has to derive an average dielectric function εav representing the mixture as a whole. Once determined, εav is then used in Mie theory, usually assuming a spherical shape for the total composite grain. The starting point is the constitutive relation D = εav E . (2.96)

Here E is the average internal field and D the average displacement defined as E = 1 V E(x) d V D = 1 V ε(x) E(x) d V. (2.97)

The integration extends over the whole grain volume V . If we envisage the grain to consist of a finite number of homogeneous components (subscript j ), each with its own dielectric function ε j and volume fraction f j , we can write E =

f jEj

D =

f jεjEj.


The E j are averages themselves (from (2.97)) over the subvolume f j V ; we have just dropped the brackets. The constitutive relation (2.96) is thus replaced by ε j f j E j = εav f jEj. (2.99)
j j

Composite grains


We envisage the components to be present in the form of many identical subparticles that are much smaller than the wavelength. When such a subparticle is placed into an extended medium with a spatially constant but time-variable field E , there is a linear relation between the field in the subparticle and the field in the outer medium (see sections 3.1 and 3.3), E j = βE . (2.100)

When one assumes that such a large-scale average field E in the grain exists, one can remove the local fields E j and find the average εav . 2.6.2 Garnett’s mixing rule To exploit (2.99), we first imagine the grain to consist of a matrix (subscript m) containing inclusions (subscript i ). For simplicity, let there be only one kind of inclusions with a total volume fraction f i so that f i + f m = 1. For the constant large-scale field E in (2.100), we take the field in the matrix and obtain εi f i β + εm f m = εav f i β + εav f m . For spherical inclusions, the proportionality factor β in equation (2.100) has the form 3εm β= εi + 2εm which is a generalization of equation (3.7) when the medium surrounding the sphere is not a vacuum but has some permeability εm . We thus arrive at the Garnett formula: 1 + 2 f i (εi − εm )/(εi + 2εm ) . (2.101) εav = εm 1 − fi (εi − εm )/(εi + 2εm ) The expression is evidently not symmetric with respect to the inclusion and matrix. One has to make up one’s mind which component should be regarded as the inclusion that pollutes the matrix. If the concentration of the inclusions, fi , is small, equation (2.101) simplifies to εav = εm 1 + 3 f i εi − εm εi + 2εm .

The Garnett mixing rule is very similar to the Clausius–Mossotti law (see (3.55)). Indeed, the latter follows almost immediately from (2.101). The Clausius–Mossotti law gives the dielectric constant of an inhomogeneous medium that consists of a vacuum matrix (εm = 1) with embedded spherical inclusions, the latter being atoms of polarizability α = (3/4π)(εi − 1)/(εi + 2). The Garnett rule can be extended to an arbitrary number of components. One still has to make the distinction between the matrix (subscript m) and the inclusions (index i = 1, 2, 3, . . .). Putting as before Ei = βi Em


How to evaluate grain cross sections

we get the following generalization of (2.101), εav = f m εm + fm + f i βi ε i f i βi (2.102)

where the sum of the volume fractions is unity, fm +

f i = 1.

Written in the form of (2.102), the restriction that the inclusions are spheres has been lifted. They may be of any shape, for example ellipsoids, for which βi can be calculated as outlined in Section 3.3. One can also take a mean over randomly oriented ellipsoids. 2.6.3 The mixing rule of Bruggeman Next we suppose that none of the components of the grain is special, like the matrix in the Garnett theory. Then the components distinguish themselves only through their permeability and volume fraction; no assumption is made about the average field E . Inserting (2.100) into (2.99) yields the Bruggeman rule: 0=

(ε j − εav ) f j β j


with f j = 1. Contrary to the Garnett rule, this formula is symmetric in all components j . If they consist of spherical entities, 0=


ε j − εav . ε j + 2εav


Thus for n components, εav is determined from a complex polynomial of nth degree. When we imagine the interstellar dust to be a democratic compound of silicate, amorphous carbon, ice and vacuum, the Bruggeman mixing rule is preferred. However, when ice becomes dirty through contamination by tiny impurities (metal atoms or PAHs) that amount to only a small volume fraction f m ), the Garnett rule is more appropriate. ( fi An illustration of how the dielectric permeabilities of two components combine to an avarage εav is shown in figure 2.6 for the two mixing rules and for a very idealized situation. The materials have the optical properties of harmonic oscillators with different resonance frequencies. In the mixture, the resonances are damped, broadened and shifted. 2.6.4 Composition of grains in protostellar cores Likely and astronomically relevant candidates for composite grains are the solid particles in cold and dense protostellar cloud cores (see section 15.3). Such grains

the frosting of molecules in ice layers on the surface of the subparticles precedes the process of coagulation.6.2) and given by the dotted curves (top. by the full line. ice sheets and . They are composed of refractory (resistant to heating) and compact subparticles made of silicates or amorphous carbon. are also expected to be found in the cooler parts of stellar disks or in comets. Both components have equal volume and both averaged permeabilities obey KK’s relations.Composite grains 75 Figure 2. In this picture. the dielectric constants of S1 and S2. be represented by harmonic oscillators (see figure 1. as a whole. compact carbon subparticles. the porous grain. S1 and S2. bottom. probably substantially bigger than normal interstellar grains. The average permeability of a mixture of these two substances calculated from the Bruggeman mixing rule is shown by the broken curve. The subparticles are enshrouded by a thick sheet of ice as a result of molecules that have frozen out. that calculated from the Garnett mixing rule. the grains in protostellar cores have four distinct components: • • • compact silicate subparticles. Let the dielectric permeability of two substances. as do. of course. Altogether. We figure them as fluffy aggregates. The volatile (easy to evaporate) ice balls with their compact cores are loosely bound together and form. imaginary part ε2 ). real part ε1 .

1. Standard gas abundances imply a ratio of volatile-to-refractory mass of Mice = 1 . We adopt a = 0. The available mass of condensable material is determined by the gas phase abundances of C. again because of coagulation. the ice volume is 2.3) is for a fixed wavelength a function of grain radius a. The mass absorption coefficient K abs (defined in subsection 2. and for the specific weight of the refractory subparticles ρref ≈ 2. which is a wavelength where protostellar cores are frequently observed. In the following examples.6. we normalize K abs to its value when the grains are small (a λ). we fix λ to 1 mm. The volume fractions of the components. The possible diversity of such grains is infinite and their structure can be extremely complex. N and O and their hydrides.5 How size. f .76 • How to evaluate grain cross sections vacuum. 2. 2. ice and porosity change the absorption coefficient These composite grains differ in three fundamental aspects from their compact silicate and carbon subparticles: • • • they are bigger because of coagulation. we adopt for the volume ratio of silicate-to-carbon material in the grain f Si = 1. so K abs = K abs (a). . Mref Consequently. it is then insensitive to a.4 . 2 fC compatible with the cosmic abundances in solids. add up to one. To demonstrate the size effect.5 to 5 times bigger than the volume of the refractories.1 µm. To be specific.6. . they contain ices because of frosting and they are porous.5 g cm−3 ρice ≈ 1 g cm−3 . which is a typical radius of an interstellar grain and certainly . 2.5. . . deposition of ice and porosity each affect the absorption coefficient and we now illustrate how. f Si + f C + f ice + f vac = 1.1 Grain size We first investigate the influence of grain size. Coagulation.

7. one has to do proper calculations. The qualitative features stay the same but some details are quite interesting.7 for silicate and carbon spheres. One can create plots like those in figure 2. When a λ.03). For sizes in between. The spikes in figure 2.1 µm) It gives the relative change with respect to ordinary interstellar dust particles and it is plotted in figure 2.1. when abs a λ. The influence of grain growth on the mass absorption coefficient K abs for silicate (Si) and amorphous carbon spheres (aC) at a wavelength of 1 mm.1). the mass absorption abs of carbon grains increases by a factor K abs ∼ 50 when the radius of the coefficient K n particles grows from 0. if all particles in the diffuse interstellar medium had a radius of 1 µm (without changing the total dust mass). the extinction optical . The curves abs show the normalized coefficient K n defined in the text. They reveal an enhancement in the mass absorption coefficient K abs which can be very significant (> 10) and which would strongly boost millimeter dust emission because the latter is proportional to K abs (see section 8.7 for silicates are resonances that disappear when the grains are not all of the same size but have a size distribution. K (0. For example. the normalized coefficient K n ∝ a −1 because Q abs ≈ 1.Composite grains 77 Figure 2.7 for other wavelengths and optical constants. abs smaller than the wavelength (1 mm). However. For example. the normalized coefficient abs K n equals one and is constant because this is the Rayleigh limit. big lumps are not efficacious in blocking light. The normalized coefficient is denoted K n and defined as K abs (a) abs K n (a) = abs .1 µm (a/λ = 10−4 ) to 30 µm (a/λ = 0.

The total grain’s volume is then three and a half times bigger and the mass twice that of the bare silicate core. on the absorption coefficient C abs of a grain that consists of silicate and ice with a mass ratio 1:1. The solid plot refers to a coated sphere with a silicate core and an ice mantle. the imaginary part of the optical constant of ice.01 but uncertain. The influence of kice . here λ = 1 mm. the broken line to a homogeneous sphere where ice and silicate are mixed and the optical constant m av of the mixture is computed after the Bruggeman theory.8. depth at 2.2 µm (K-band) would be almost ten times larger than it really is. 2. Defining the normalized absorption coefficient of a single grain by abs Cn = C abs (core + ice) C abs (core) .5. The relevant parameter for the absorption coefficient C abs is kice . We take a silicate sphere of arbitrary radius a λ and deposit an ice mantle of the same mass on it. such a mass ratio of ice to refractory core is suggested by the cosmic abundances in the case of complete freeze–out. Both curves are based on Mie calculations.6.2 Ice mantles Next we illustrate the influence of ice in the grain material. The ordinate abs Cn gives the increase in the cross section with respect to the bare silicate core.78 How to evaluate grain cross sections Figure 2. again for a wavelength λ = 1 mm. Estimates for kice at this wavelength are around 0. The grain radius a is much smaller than the wavelength. It depends on how much the ice is polluted by impurities. the imaginary part of the optical constant of ice.

04. Note that an ice mantle increases C ice = 0 because the mantle grain is larger than the refractory core and collects more light. The normalized cross abs section Cn is defined in the text. The influence of porosity on the grain cross section. We also abs show in figure 2. We consider. abs we learn from figure 2.8. A porosity parameter f vac between 0. f vac . The mass of the grain is kept constant and does not vary with the vacuum fraction f vac . while the optical constant of the grain. changes with fluffines. Fluffy grains are obviously better absorbers because they are bigger. which may or may not be a reasonable estimate for interstellar conditions.9. in figure 2.01.9. the grain is compact.6. equals zero.5+i0.Composite grains 79 Figure 2.4 and 0.3 Fluffiness A porous grain also has a greater absorption cross section C abs than a compact one of the same mass.8 that an ice mantle enhances C abs by a factor Cn ∼ 3 if abs even when k kice = 0. which is calculated here after Bruggeman.5. The compact particle has m Si = 3. . suggests an increase in the absorption coefficient by a factor of two. 2.8 the value of Cn when the ice is not in a mantle but mixed throughout the grain. C abs (compact grain of same mass) When the volume fraction of vacuum. abs the normalized cross section Cn is defined by abs Cn = C abs (fluffy grain) . silicates with vacuum inclusions.

we presented general solutions of the field equations in the case when a plane wave interacts with a particle with simple geometry.1 Tiny spheres We derive the efficiencies for small spheres of dielectric material. In a few simple configurations. The particle itself is not required to be small.3 and 2. be small. No matter how efficient such a machine is. it is wise to retain some mental independence and bring to one’s mind the physical aspects: computing must not be confused with understanding. which is small compared with the wavelength λ. which occurs at wavelengths where the condition λ a is usually very well fulfilled. 3. They illuminate the problem and are useful for checking the correctness of a computer program. i. a may even be big. This is the basic section for understanding how and why interstellar dust absorbs and scatters light. a. the approximations given here. |m|x 80 1 (3. So the heading of this section is suggestive but not precise. m = n + i k. we have to deliver ourselves to the mercy of a computer.1. 3. there we stressed the mathematical aspects. only the ratio a/λ. one may use for the computation of cross sections.2) .e.Chapter 3 Very small and very big particles In sections 2. when the size parameter 2πa 1 (3. If we additionally stipulate that the product of the size parameter multiplied by optical constant.1 When is a particle in the Rayleigh limit? When a sphere has a radius. With regard to the thermal emission of interstellar grains. Usually.4. analytical solutions are possible and we turn to these in this chapter. when we want to extract numbers from the theory for a specific astronomical application. In fact.1) x= λ the calculation of cross sections becomes easy.

the wave traverses the particle with the phase velocity nxω−1 which is much shorter than the inverse vph = c/n in a time τ circular frequency ω−1 . 45 m2 − 1 m2 + 2 2 + O(x 7 ) Usually non-magnetic materials (µ = 1) are considered. If x is small. this suggests the frequently cited dependences and Q abs ∝ λ−1 Q sca ∝ λ−4 λ λ .Tiny spheres we ensure two things: • • 81 because kx 1. the one with x 6 electric dipole scattering: Q ext = Q sca 6 Re{a1 } x2 4x Im m2 − 1 m2 + 2 = 4x Im ε−1 ε+2 (3. With respect to the wavelength behavior. Then the term with x 3 in the coefficient a1 yields the electric dipole absorption. Because Q abs ∝ x and Q sca ∝ x 4 scattering is negligible at long wavelengths and extinction is reduced to absorption: Q abs Q ext .1.1) and (3. if one understands by the size parameter the ratio of typical dimension over wavelength.51) for Q ext and Q sca into powers of x and retain only terms up to x 5: 2x 3 m 2 − 1 2x 5 (m 2 − 2)(m 2 − 1) 4x 6 a1 = − i + −i 3 m2 + 2 5 9 (m 2 + 2)2 b1 = − i • x5 2 (m − 1) + O(x 7 ). not just spheres.3) (3. the field is only weakly attenuated in the particle.4) 8πa 3ε2 6nk 8πa = λ (n 2 − k 2 + 2)2 + 4n 2 k 2 λ |ε + 2|2 6 |a1 |2 x2 8 4 m2 − 1 x 3 m2 + 2 2 . This concept can be applied to any particle.2 Efficiencies of small spheres from Mie theory A purely mathematical approach to finding simple expressions for the efficiencies of small spheres is to develop the first coefficients a1 .2) hold are said to be in the Rayleigh limit. both Q ext and Q sca approach zero. and because nx 1. 3. Grains for which conditions (3.50) and (2. b1 in the series expansion (2.

the electric field in the grain changes in a quasi-stationary fashion. The basic equations valid everywhere are rot E = 0 div D = 0. m 2 = (n + i k)2 = εµ (k > 0) • equation (3. is for purely magnetic material (µ = 1. When we want to calculate the field in such a configuration. and c1 . there is another approach to obtaining Q sca and Q ext for small particles that gives physical insight. For the potential ϕ of the deformed field we make the ansatz ϕ= −E · r + c1 −c2 E · r E·r r3 outside sphere inside sphere (3. When x 1 and |mx| 1.82 Very small and very big particles for scattering and absorption. we are reduced to an exercise in electrostatics. 3. this leads to the Laplace equation ϕ=0 where ϕ is the potential related to the field through E = −∇ϕ.46). they are true only if m(λ) is more or less constant. ε = 1) replaced by Q abs = λ 3µ2 8πa · . We label the field inside the sphere by Ei and outside it (external) by Ee . the field becomes deformed.2 for a non-magnetic conductor and presents magnetic dipole absorption (see (3. from (1. We now restrict the discussion to a dielectric medium. c2 are constants.5) • The dissipation process refers to magnetic dipole oscillations and is relevant only at frequencies ω < 1012 s−1 . volume V and dielectric constant ε into a constant field E.26)). For a homogeneous medium (ε = constant). Because the optical constant m is. When we place a sphere of radius a. They impose the boundary conditions that the tangential component of the electric field E and the normal component of the displacement D are continuous on the grain surface.1.3). . λ |µ + 2|2 (3. The coefficient b1 is discussed in section 3. However. the potential is simply ϕ0 = −E · r.3 A dielectric sphere in a constant electric field Besides cutting off the series expansion in Mie theory after the first term. Let the sphere be at the center of the coordinate system. which contains only the coefficient a1 .6) where r is the position coordinate of length r = |r|. symmetric in ε and µ. In the absence of the sphere.

10) (3.7) Note that the field is smaller inside the body than outside. therefore. There is then a core (index 1) enveloped by a mantle (index 2) of a different substance.1 A coated sphere in a constant electric field It is not difficult to repeat the previous exercise for a sphere covered by a homogeneous shell.Tiny spheres 83 Equation (3.8)) αe = 3 ε−1 . Continuity of the tangential component of the electric field implies continuity of the tangential derivative of ϕ. Equation (3. outside. 4π ε + 2 (3.6) reflects the expected behavior of the field. Continuity of the normal component of the displacement D yields εc2 = 1 + 2c1 /a 3 .7) gives the internal field.11) 3. Ei = 3 E ε+2 for r ≤ a. If e denotes the unit vector in the direction r. Inside the sphere. This gives c2 = 1 − c1 /a 3 . from which it follows that ϕ is itself continuous (parallel and perpendicular to the surface).6)) P= ε−1 i E 4π the induced dipole moment of the grain is p = PV = a 3 ε−1 E = αe V E ε+2 (3. Ee is the sum of a dipole. (3.3. (3. E i < E. We determine c1 and c2 from the conditions on the grain surface. the outer field Ee is the sum of E plus a dipole field: Edip = so Ee = E − αe V grad 3e(e · p) − p r3 E·r r3 = E + Edip . Ei is constant and parallel to E. Hence we derive ε−1 3 c2 = c1 = a 3 ε+2 ε+2 and. One now has boundary conditions at the interface . Because the polarization is given by (see (1.8) and the electric polarizability αe becomes (see (1.9) For other grain geometries there would be other dependences of αe on ε. which is induced by E and goes to zero at large distances.1. plus the constant field E.

does the dipole moment p. f = a1 a2 3 ≤1 expression (3. Because ω = 0.12) For the dipole moment. (3.1 Scattering The field E is not really static but oscillates proportionally to E 0 e−iωt and so. we can because the variations of the field are assumed to be slow. W must also equal the total power scattered from the incident wave by the particle. emits radiation. therefore.84 Very small and very big particles between the two materials and. the permeability ε(ω) is complex. However. synchronously. If f denotes the volume fraction of the inner sphere relative to the total sphere. which automatically takes care of the time lag between P and E.4. as before.13) The electric polarizability αe for spheres is given by (3. W = 1 ¨ |p|2 = S C sca .9). Its average power integrated over all ¨ directions is W = |p|2 /3c3 and follows from (1.97). The oscillating dipole. which is now the grain as a whole. the polarization of the medium is not that of a static field.4 Scattering and absorption in the electrostatic approximation 3. Instead.9) for the polarizability becomes a bit lengthy: αe = 3 (ε2 − 1)(ε1 + 2ε2) + f (ε1 − ε2 )(1 + 2ε2 ) · . .14) which agrees with equation (3. We thus find C sca = πa 2 Q sca = 24π 3 V 2 ε − 1 λ4 ε+2 2 (3. Can we use it in the case when ω = 0? Yes. just as in the static case. It was obtained in the electrostatic approximation.39). on the outside. This means the electrons always have sufficient time to adjust to the momentary field. therefore C sca = 8π 3 ω c 4 V 2 |αe |2 . 4π (ε2 + 2)(ε1 + 2ε2) + f (2ε2 − 2)(ε1 − ε2 ) 3. we have to insert p = αe V E and for the time2 averaged Poynting vector S = (c/8π)E 0 from (1.1. so the electron configuration is always relaxed.1.4) for the scattering efficiency Q sca . the dielectric permeability must be taken at the actual frequency ω of the outer field. 3c3 (3. Of course.

5 Polarization and angle-dependent scattering For small spheres. 2 2m e (ω0 − ω2 )2 + ω2 γ 2 When we substitute the dielectric permeability of the harmonic oscillator after (1. one in the forward.16) For the electric field E 0 which drives the oscillators. Scattering is symmetrical in θ about π/2 and has two peaks.Tiny spheres 3. we must insert the field Ei inside the grain according to (3. c (3. There are several noteworthy facts: • • The scattering pattern no longer depends on wavelength. The power W absorbed by the oscillators can be expressed in two ways. This gives C abs = 4π ωV Im{α}. V being the volume and N the oscillator density.17) 3.1.7). . as the matrix elements contain only the angle θ . either by the dissipation losses of the harmonic oscillators given in (1. the other in the backward (θ = π) direction.2 Absorption 85 We can find the absorption coefficient of a small dielectric or magnetic grain if we use (1.3) but now we understand the physics: C abs = ε−1 6π V Im λ ε+2 = V ω 9ε2 . 4λ (3. When we do this. and not the outer field E.77).18) θ = 0 gives the forward direction. so W = VN 2 γ E 0 e2 ω2 = SC abs . c |ε + 2|2 (3.69) or via the Poynting vector S multiplied by the absorption cross section C abs .57) and set the absorbed power equal to SC abs .15) But let us be more elaborate and assume that the grain consists of N V harmonic oscillators excited in phase.1. we recover the efficiency for electric dipole absorption of (3.75) reduces further (we drop the factor in front of the matrix):    1 (1 + cos2 θ ) Is 2  Q s   − 1 sin2 θ 2  = Us 0 Vs 0 − 1 sin2 θ 2 1 (1 + cos2 θ ) 2 0 0 0 0 cos θ 0 0   Ii  0   Qi    Ui 0 Vi cos θ (3. the scattering matrix given in (2.55) or (1. we can transform this equation into Vc 2 ε2 E 0 = SC abs .4.

Q i = Ui = Vi = 0). We will consider some quantum effects when we discuss PAHs. the integral in (2. Here we only remark on the influence of the surface in the case of small grains. One way in which the surface affects the optical grain properties can be understood when we interpret the damping constant γ in the motion of an electron (see (1.19) Consequently. φ). If the particle has a diameter a. rather regularly spaced and separated by a distance r0 .6 Small-size effects beyond Mie theory A real grain is not a homogeneous continuum but a crystal built up from atoms. which is based on the classical electrodynamics of a continuous medium. This has consequences. If the incident radiation has unit intensity and is unpolarized (Ii = 1. a ˚ As the lattice constant is of order 2 A. A more general theory is then needed to describe the optical behavior of particles. for their ability to bind to gas atoms or for the specific heat of the grain. the light is completely polarized at a scattering angle of 90◦ . and thus the phase function f (θ. a substantial fraction of atoms is on the surface only when the particle is small. As a result. the ratio of the number of surface atoms Nsurf to all atoms N in the grain is roughly Nsurf N 6 r0 . Mie theory. so the polarization is perpendicular to the scattering plane. is proportional to 1 + cos2 θ . 1 + cos2 θ (3.59)) as a collision frequency with atoms of the crystal. a specific kind of very small carbon grains. The intensity of the scattered light. for example.6) vanishes and the asymmetry factor becomes zero. 3. The degree of linear polarization becomes p= • • sin2 θ . we have Is = 1 (1 + cos2 θ ) 2 Q s = − 1 sin2 θ 2 Us = Vs = 0. g = 0.86 • • Very small and very big particles The matrix element S12 < 0. Whereas atoms inside the particle are surrounded from all sides. the situation is different for those on the surface which have bonds only towards the particle’s interior. fails when the structure of matter or quantum effects become important. an additional term must be added to the damping constant that comes from collisions (reflections) at the surface.1. although scattering by a small sphere is not isotropic. the lattice constant. If the particle is small. Free electrons in a metal move with the Fermi velocity vF = 3 me 3π 2 n e .

say. rot H = 0 div B = 0 and the same boundary conditions on the surface: the tangential component of H and the normal of B = µH are continuous. that the particle is small.54). There are the same types of equations. 4πσ E = rot Hi .114). one estimates that noticeable density n e ˚ changes in γ .A small metallic sphere in a magnetic field 87 which is the threshold speed of fully degenerate (non-relativistic) electrons and follows from the Fermi energy of (6. The charges on the metal surface become polarized in the outer field but the electric field does not penetrate into the particle. For example. c .3)). and thus in the dielectric permeability. in other words. But not quite because we have neglected in our small-size approximation the magnetic field which is also present in a wave. Slowly means that H is spatially uniform over the dimension of the body or. One might. one has to use the x 6 term in the coefficient a1 of equation (3. 3. When one knows Hi . therefore. then p = a 3 E. Mie calculations automatically do. ohmic losses. 3. there is a formal identity with electrostatics. proportionally to e−iωt . rot E = i µω i H c or for a slowly changing field from (1. When we place a particle in a constant magnetic field H. occur for sizes a ≤ 50 A. The presence of E in the conductor implies a current and. think there would be no absorption and just scattering with an efficiency Q sca = 8x 4/3 (because Im{(ε − 1)/(ε + 2)} = 0. in graphite with an electron 1020 cm−3 and γ = 5 × 1012 s−1 . one can derive the induced field E either from the Maxwell equation (1. So there is nothing new in magnetostatics. their time average is σ E2 .2. The magnetic field Hi inside the sphere is then also changing and induces an electric field E.1 Slowly varying field The situation becomes interesting when the magnetic field H is slowly alternating. therefore.25).9) very large.2 A small metallic sphere in a magnetic field The case of a small metallic sphere in an electric field is included in the previous derivation for the dipole moment p by making ε in equations (3. Let us now include it.

To evaluate the somewhat complicated expression Hi = β rot rot( f H). When one works it out. . 4πσ ). The constant β will be adjusted later. the vector potential. Hi + k2 Hi = 0 (3.33)).6) and the easy-to-prove relations div( f H) = H · ∇ f ∇f =e k cos kr − f r ( f H) = −k2 f H. We have used this result before in section 2. A.28)).20) by noting that the scalar function f (r ) = is a spherically symmetric solution of f + k2 f = 0 (see the Laplace operator in (2. c2 We solve (3. as required.26)–(2. The square of the wavenumber is given by (1.20) together with div Hi = 0.88 Very small and very big particles But one cannot use the static solution Hstat because it fulfils div Hstat = 0 and rot Hstat = 0 and thus gives no electric field at all. When metallicity dominates (ω|εd | k2 = i 4πσ ωµ . one obtains for the magnetic field inside the sphere Hi = β f + k2 f r H−β 3f + k2 f r e(e · H).3 (see equations (2. we use formula (A. If we now put Hi = rot A the equations div Hi = 0 and Hi + k2 Hi = 0 are fulfilled.116).34). Because H is constant. we start with the wave equation (1. Instead. defined by the function A = β rot( f H) also fulfils the wave equation A + k2 A = 0. to determine the internal field Hi . sin(kr ) r Here e is a unit vector in the direction of r.

When we divide W by the Poynting vector. + a 2 k2 ak (3. c 1+i δ= √ = k 2πσ ω (3.A small metallic sphere in a magnetic field 3. Suppose the field vector H is parallel to the x-axis and a function of z only. y)-plane marks the surface of the metallic body. Then with µ = 1.21) With αm . the field has the form Hi = H0 e−iωt eikz = H0 e−z/δ ei(−ωt +z/δ) .22) At the boundary of the body. The factors αm and β follow from the boundary conditions on its surface by equating Hi = He for e · H = 0 and e · H = H. therefore. It falls off exponentially in the metal and. the (perturbed) field He outside the sphere. as we assume in this chapter. we do not have to integrate σ E2 over the particle volume. 0. 0) and the (x.20) to a simple one-dimensional situation. H = (Hx (z). We find β = 3/[2k2 f (a)] and αm = − 3 8π 1− 3 3 cot ak .57). we apply the wave equation (3. 3. where a plane wave falls on a metal surface.11).2. where k = c c (3. for which rot He = 0 and div He = 0. we can determine the dissipation rate W in a grain after (1. ∂ 2 Hx + k2 H x = 0 ∂z 2 √ √ i 4πσ ω 2πσ ω = (1 + i ). r3 αm denotes the magnetic polarizability and m = V αm H is the magnetic dipole moment of the sphere of radius a and volume V . H = H0 e−iωt and inside it (z > 0).2. can be written as He = H + V αm 3e(e · H) − H .3 The penetration depth Even when the particle is small compared to the scale on which the outer magnetic field changes. the magnetic field may not fully pervade it.23) .2 The magnetic polarizability 89 In analogy to (3. we get the absorption cross section of a magnetic dipole. To see how far it can penetrate.

• If the penetration depth is large (a in (3.26) If the penetration depth is small (δ a λ). we can determine their scattering and absorption cross section once we have worked out the dipole moment that they .27) αm = − 8π 2a 16π a and C abs = 3 π 2 1/2 a ω .25) A metallic sphere of radius a and conductivity σ has then at frequency ω the cross section for magnetic dipole absorption C abs = • 8π 2 σ 2 5 ω a .21) cancel out and we obtain for the magnetic polarizability in the low frequency limit αm = − 1 105π a δ 4 +i 1 20π a δ 2 .24) where ez is the unit vector in the direction of the z-axis.21) to fifth order in ka. a λ = 2πc/ω.2. (3. the magnetic polarizability becomes 3δ 9 δ 3 1− +i (3. E = (1 − i ) ω Hi × e z 8πσ (3.21) for the magnetic polarizability αm of a sphere simplifies further when the penetration depth δ. the greater δ becomes. we develop the cotangent function √ a 2πσ ω a ka = (1 + i ) = (1 + i ) δ c 1. which we just introduced. the grain is always small compared to the wavelength. is small or large compared to the grain radius. Then the first two terms in the bracket of (3. At any rate. 3. In analogy to the previous discussion. where δ).114). E is parallel to the y-axis.3 Tiny ellipsoids The treatment of a sphere in a constant electric field may be extended to ellipsoids.28) 3. The electric field has the same penetration depth and follows from (1.4 Limiting values of the magnetic polarizability The formula (3. 15c3 (3.90 Very small and very big particles is a characteristic scale for the penetration depth of the magnetic field into the particle: the lower the frequency. 2σ (3.

1 Elliptical coordinates Ellipsoids are naturally handled in elliptical coordinates. (3.Tiny ellipsoids 91 acquire and the strength of their internal field. The essential point is that ξ is constant on the surface of an ellipsoid.3. named ξ.32) . η.31) For u = η or u = ζ .29) is of third order in u. To transform the elliptical into Cartesian coordinates. We will again need the Laplace equation ϕ = 0. with ξ > −c2 − c2 > η > −b2 − b 2 > ζ > −a 2 . 3. (−b 2. An ellipsoid has three principal axes a. c. one has to solve this set for x. Dividing the Laplace operator ϕ by 1 (ξ − η)(ζ − ξ )(η − ζ ). b. They are defined as follows: For any three numbers a.29) constitutes a set of three equations when we put u = ξ. η. we are reduced to a sphere and f (u) has only one root. the equation f (ξ ) = 0 describes an ellipsoid that has the same foci as the ellipsoid y2 z2 x2 + 2 + 2 = 1. ζ . Therefore. (3. −c2 and for reasons of continuity must vanish somewhere in each of the intervals (−a 2 . b. one gets 4 (η − ζ )Rξ ∂ ∂ξ Rξ ∂ϕ ∂ + (ζ − ξ )Rη ∂ξ ∂η Rη ∂ϕ ∂ + (ξ − η)Rζ ∂η ∂ζ Rζ ∂ϕ ∂ζ = 0. −b 2 . this time in elliptical coordinates. We use here the convention a≥b≥c and that a is aligned along the x-axis of a Cartesian coordinate system. one obtains confocal hyperboloids. 2 a b c (3. c with a>b>c>0 the function f (u) = y2 z2 x2 + 2 + 2 −1 a2 + u b +u c +u (3. If a = b = c. −b2 ). x=± (ξ + a 2 )(η + a 2 )(ζ + a 2 ) (b2 − a 2 )(c2 − a 2 ) and corresponding expressions for y and z. −c2 ) and (−c2 . Equation (3. for example. u = η or u = ζ . ζ ) are called elliptical coordinates because for u = ξ . It has poles at −a 2. +∞).30) (ξ. f (u) has three real roots. This gives. y and z.

the perturbation ϕ0 F has the form characteristic of a dipole. ζ ) are related to the length element ds in elliptical coordinates by ds 2 = h 2 dξ 2 + h 2 dη2 + h 2 dζ 2 ξ η ζ with √ (ξ − η)(ξ − ζ ) hξ = 2Rξ √ hη = (η − ζ )(η − ξ ) 2Rη √ hζ = (ζ − ξ )(ζ − η) .3. At great distance r = x 2 + y 2 + z 2 . This allows us to evaluate the integral in (3. equivalently. Let the outer electric field E be directed parallel to the x-axis and thus to the axis a of the ellipsoid. When we insert ϕ = ϕ0 (1 + F) into (3. dξ (3. and F(ξ ) = c1 ∞ ξ dx .33) where ϕ0 = −E · r is the potential of the unperturbed field.32) and take into account that also for the perturbation (ϕ0 F) = 0. c2 is a constant. so the terms a 2 .31) or. (x + a 2 )Rx (3.92 Very small and very big particles The expressions Ru = (u + a 2 )(u + b 2 )(u + c2 ) (u = ξ. we write ϕ= ϕ0 [1 + F(ξ )] c2 ϕ 0 outside ellipsoid inside ellipsoid (3.34): F = constant.6). we arrive at the differential equation 0=F +F Rξ 1 + 2 Rξ ϕ0 =F +F d ln[Rξ (ξ + a 2 )]. by ξ = 0.35) and one obtains the dipole potential 2c1 −3/2 F(ξ ) ξ ∝ r −3 . 3 . In analogy to (3.2 An ellipsoid in a constant electric field 3.3.2. and ϕ0 F is the perturbation evoked by the grain.29) are negligible and r ξ 1/2 . which applies to the interior of the grain. c2 in (3. η. b2 .35) Far away from the grain.34) A prime here and later denotes the derivative with respect to ξ . at least one of the Cartesian coordinates is large. Its boundary is given by (3. The function F depends only on ξ . There are two solutions to (3.1 The dipole potential Imagine an ellipsoidal grain with principal axes a > b > c at the center of a Cartesian coordinate system. 2Rζ 3.

38) 3[1 + L a (ε − 1)] and its polarizability. we know the electric field in and around an ellipsoid. gives c2 = 1 + F(0). when the electric field is parallel to axis a.12) for dipole radiation is applicable with the dipole moment p from (3.3 Cross section and shape factor Because of our experience with spheres. the dipole moment of the ellipsoid becomes ε−1 p= E (3.3. The internal field is constant and parallel to E. This gives C abs = C sca = ε−1 2π V Im λ 1 + L a (ε − 1) 8π 3 V 2 ε−1 3λ4 1 + L a (ε − 1) 2 (3. The normal component of the displacement yields εc2 ϕ0 (0) = ϕ0 (0) · [1 + F(0)] + ϕ0 (0) · F (0) and from there follows c2 = 2a 2 Defining the shape factor La = we find c2 = abc 2 ∞ 0 F (0) . we derive the constants c1 .38).2. In the case of scattering. equation (3.3 The polarizability of an ellipsoid With c1 and c2 being determined. αa = ε−1 .41) . for ellipsoids. so we use (3.Tiny ellipsoids 3.37) With the same arguments as in (3. Continuity of the tangential electric field. C sca and C abs . 4π[1 + L a (ε − 1)] (3.39) 3.40) .2.36) 1 1 + L a (ε − 1) c1 = − 3. ε−1 dx (x + a 2 )Rx abc(ε − 1) .37). or of ϕ.2 The shape factor 93 As in the case of a sphere. we can immediately write down the formulae of the scattering and absorption cross section.3. (3. 2[1 + L a (ε − 1)] (3.3. Absorption is still proportional to grain volume V . Ei = E .16) but now with E 0 = |Ei | from (3.8) for a sphere. 1 + L a (ε − 1) (3. c2 from the continuity conditions on the surface where ξ = 0.

the only thing that changes for the cross section is the shape factor of (3. for example. the cigar in figure 3. in the second it is a broad ellipse and much bigger. the electric field inside and outside are more or less equal. the projected surface is a small circle. Light is traveling in the indicated direction.1. Ei = E. When the small ellipsoid is very transparent (|ε| 1). A cigar rotating about axis b. with a = b = c. We then have to replace L a by Lb = abc 2 ∞ 0 dx (x + b 2 )Rx or Lc = abc 2 ∞ 0 dx (x + c2 )Rx respectively. all Ls are equal to 1 and we recover the formulae of the polarizability and cross 3 sections for spheres. which is perpendicular to the page in this book. Obviously. the shape factors loose their importance.36).1 and let the electric vector E swing parallel to axis c. When the electric field is parallel to axis b or c. We can easily check that the sum over all shape factors is one: La + Lb + Lc = 1 so only two of the three L values are independent. After a quarter of the rotation cycle. In the first case. the cigar has changed from position 1 to position 2. and the polarizability α = (ε−1)/4π = χ. Consider.94 Very small and very big particles a light c c a 1 2 Figure 3. Only in the Rayleigh limit does the cross section C depend solely on the direction of the electric field and not on the direction of wave propagation. The cross section is then the same for light that falls in parallel to axis a or parallel to axis b (which is perpendicular to a and c). The absorption and scattering . and not to axis a as we have assumed so far.

the grand principle of superposition allows us to split the electric field vector E of the wave into components along the orthogonal ellipsoidal axes a. E cos β.4 Randomly oriented ellipsoids When an ellipsoid of fixed orientation in space is illuminated by an electromagnetic wave. b. 3 As the terms on the right-hand side of (3. Here E = |E| and cos2 α + cos2 β + cos2 γ = 1. For an ensemble of particles. Cc denote the cross sections of the ellipsoid when the principal axes a.3. for arbitrary grain orientation the internal field Ei is not parallel to the outer field E. the ellipsoid is called a spheroid.44) Without alignment.Tiny ellipsoids coefficient are then independent of the axial ratios and of orientation. a = b > c. even if the grain material is isotropic. we expect.5 Pancakes and cigars When two of the principal axes are equal.3. • If then the two equally long axes are larger than the third one. c: E = (E cos α.44) and (3.43) 3. Cb . . random rotation and thus random orientation of the grains. in interstellar space.2)) cos2 x = 1 4π 2π π dy 0 0 d x cos2 x sin x = 1 . the body has the shape of a pancake. the total cross section of the grain can be written as C = Ca cos2 α + Cb cos2 β + Cc cos2 γ .45) are true only in the electrostatic approximation.42) (3. c are parallel to the electric vector E of the incoming wave. Ei and E are parallel only when E is directed along one of the principal axes. E cos γ ). all directions are equally likely and the mean of cos2 x over 4π is (see (2. If Ca .44) are independent of each other. 3. 3 3 3 (3. (3. C abs (λ) = 2π V ε2 (λ) λ 8π 3 V 2 C sca (λ) = |ε − 1|2. 3λ4 95 (3. the average cross section for identical ellipsoids under random orientation is equal to the arithmetic mean: C = 1 Ca + 1 Cb + 1 Cc . b. Interestingly.45) Equations (3.

Defining the eccentricity e through c2 (3. It can be found by numerically evaluating the integral in (3.47) and (3.48). More educated terms are oblate and prolate spheroids. besides being nourishing or fragrant.2). and La = g 2 (e) g(e) π − arctan g(e) − 2 2 2 2e with g(e) = 1 − e2 . We will see in chapter 10 that pancakes and cigars. The shape factor for oblate (pancake) and prolate (cigar) spheroids. (3. L b = L c and La = • 1+e 1 − e2 1 ln −1 + 2 e 2e 1−e (3. it resembles a cigar. • Otherwise.36) now permits an analytical solution (figure 3. L a = L b . can explain why and how stellar light is polarized by dust clouds.48) e2 .2.46) e2 = 1 − 2 .36) or by using the analytical expressions (3. if a > b = c. The shape factor of (3.47) pancakes have a = b > c. a • cigars have a > b = c.96 Very small and very big particles Figure 3.

respectively. For example. for pancakes it is axis c. than spheres and the effect increases with eccentricity. When the carbon sheets are stacked parallel to the (x.45). while the other two are spinning. The difference between cigars and pancakes is mild. the effect of particle elongation on the optical depth is more pronounced in the far than in the near infrared. Cav . When we evaluate the Cs assuming a ≥ b ≥ c we find that (mostly) Ca ≥ Cb ≥ Cc in such a way that the increase in Ca more than offsets the decline in Cb or Cc .866. Cc . 3. Any wiggle in m due to a resonance will be reflected in the curves. y)-plane. Figure 3. which implies an eccentricity of 0.19. one has εx = ε y = εz .3 displays. CE⊥rot = 1 2 Ca + Cc .3. Graphite particles are an example. so L b = L c = 1 . Suppose a spheroid rotates about the major axis where the moment of inertia is greatest.Tiny ellipsoids 97 For very long cigars (needles) and very flat pancakes e = 1 and L a = 0. what happens to the optical depth when we replace spheres of a fixed radius by randomly oriented spheroids. are at λ ≥ 100 µm by some 30% better emitters or absorbers than spheres. For cigars. this is axis b or c. over a broad wavelength band. depend on the direction of linear polarization of the incident wave. Overall. although not as a strict rule. that ellipsoids have a higher average cross section. Let the light propagate in a direction perpendicular to the rotation axis and let CE rot CE⊥rot denote the time-averaged cross sections in the case when the electric vector of the incident wave is parallel and perpendicular to the rotation axis. CE rot and CE⊥rot . one can treat optically anisotropic ellipsoids. It is also not difficult to extend the computations of cross sections to coated ellipsoids but for particulars. It turns out. Cb . Because the mean of cos2 x over a half-cycle is cos2 x = 1 π π 0 cos2 x d x = 1 2 we find that the average cross sections. The rotation axis stays fixed in space. identical in shape and of the same volume as the spheres. With fixed eccentricity. from (3. 3 Figure 3. We can understand this because Cav is.6 Rotation about the axis of greatest moment of inertia The cross section of spinning particles changes periodically and one has to take time averages. As a further sophistication. A sphere is the subcase when e = 0 and all shape factors are equal 2 to 1 . the arithmetic mean over Ca . grains with an axial ratio of two. We show results for silicate and carbon particles and we can readily identify the resonances in silicate at 10 and 18 µm by comparison with figure 7. the cross section is still a function of the optical constant m and thus of wavelength λ.1 shows such a situation for a spinning cigar with rotation axis b.

Calculations are done in the electrostatic approximation implying grains much smaller than the wavelength. The cross section of spheres over that of randomly oriented spheroids of the same volume. The grains consist of amorphous carbon (aC. optical constants for the two materials from figure 7.3. which is .98 Very small and very big particles Figure 3.19. The eccentricity of the spheroids defined in (3. bottom). If a pancake rotates about its axis of maximum moment of inertia. top) or silicate (Si. CE rot = Cb < CE⊥rot .46) is indicated.

3. E1 and Ei are such averages over many atoms. It arises from all atomic dipoles and is directed opposite to the polarization P.3 we generalized to ellipsoids. Because E i < E. Writing the internal field Ei in the form Ei = E + E1 defines a new field E1 .51) It is important to make a distinction between the local field at exactly one point and macroscopic averages. (3. the field has tremendous gradients. On a microscopic level. In section 3. and light comes in perpendicular or parallel to c.7) the field Ei inside a dielectric sphere that sits in a time-constant homogeneous outer field E. there are. • For cigars (a > b = c and Cb = Cc ): C = 1 Ca + 3 Cc 4 4 • for pancakes (a = b > c and Ca = Cb ): C= 1 2 C= 1 2 Ca − Cc (3. Atoms are not at random positions but at privileged sites (lattice grid points are loci of minimum potential . To calculate in chapter 10 the degree of polarization produced by grains. For example.7) we find for a sphere E1 = − 4π P. In both cases.4.49) Ca + Cc C = Ca − Cc .4 The fields inside a dielectric particle 3. 3 (3.1 Internal field and depolarization field We determined in equation (3.50) If the light travels parallel to the rotation axis of the spheroid (axis b for cigars. the polarization P of the medium is constant and the field inside smaller than outside. no polarization effects.The fields inside a dielectric particle 99 c. a hundred or so but at least over one unit cell in a crystalline structure. one calls E1 the depolarization field (see figure 3. we need the difference C = CE⊥rot − CE rot . from (3. The effective cross section for incident unpolarized light is C= 1 2 CE rot + CE⊥rot .4). axis c for pancakes). on average. no averaging is necessary and CE⊥rot = Ca CE rot = Cc < CE⊥rot .

Inside the body. 3. it is discontinuous. Let ϕ(r) be the electrostatic potential from all dipoles in the body. a surface charge σ of strength σ =e·P (3. This expression for σ follows when we recall that according to (1. r Now we transform the volume integral into a surface integral employing the relation div( f P) = f div P + P · ∇ f .100 Very small and very big particles energy) and the local field Eloc acting on an atom is usually different from the average field Ei . (3. We. There is a theorem which states that for any body of constant polarization P.2 Depolarization field and the distribution of surface charges A body of constant polarization P has. say at x0 . it does not appear to be so close to x0 . . With f (r) = 1/r and div P = 0. imagine a spherical cavity around x0 of such a size that beyond the cavity the dipole distribution may be regarded as smooth. we have to add to E the fields from all dipoles. ϕ(r) = − (P · ∇r −1 ) d V. the field from the dipoles within the cavity has to be explicitly written as a sum. the divergence of the polarization vector is zero and ρpol = 0 but on its surface. we get ϕ(r) = − 1 P · dS = r σ d S. so generally Eloc = Ei . the depolarization field E1 is identical to the field that arises in vacuum from the distribution of surface charges as given by (3. To find Eloc at a specific locus in the grain. therefore. so Edip + Edip . r The second integral sums up the potentials from all surface charges and their total field is thus equivalent to the field of all atomic dipoles.53). whereas inside it. We prove the theorem by writing ϕ(r) as a volume integral. their distribution is not smooth. at least.20) a non-uniform polarization creates a charge ρpol = − div P. As the dipoles are at discrete lattice points.53) where e is the outward surface normal.52) Eloc = E + outside cavity The field arising from the smooth distribution outside the cavity can be expressed as a volume integral.4. P is discontinuous and a charge appears. on its outside. This is correct because a single dipole p has a potential ϕdip = −p · ∇ 1 .

3 3.51). The sum is zero because of the symmetry of the grid.and y-components of the total field vanish. we find that the surface charge on S2 produces the field (4π/3)P. in an amorphous substance.10)) (Edip )z = p cavity i 3z i2 − ri2 ri5 = p i 2z i2 − x i2 − yi2 ri5 =0 where ri = x i2 + yi2 + z i2 is the distance of dipole i to the origin.3 The local field at an atom 101 The ‘outside’ sum in (3. i. therefore. the dipoles are at positions (x i . gives the depolarization field E1 . z i ). which is the configuration of a parallel–plate condenser. 3 . it is equivalent to the field from the surface charges. yi . for a spherical grain we derive from (3. For such a situation. It remains to include the atoms in the spherical cavity S2 . For a body in the shape of a thin slab with parallel surfaces perpendicular to E. For example. Hence α is called the atomic polarizability. Likewise the x. Comparing with (3. the point x0 is at the origin. Suppose we have a cubic lattice. The inner. S1 .8) which we applied to a grain as a whole.54). there are now two boundaries. one finds Eloc = E − 8π P. only of inverted sign.4 The Clausius–Mossotti relation The local field Eloc produces in each atom of volume V a dipole moment p = αV E loc .8): Eloc = E.7) and (3. P = N V α E loc = N V α E i + 4π P . 3 (3. According to our theorem.52) may be evaluated by an integral. Then the z-component of the total field from all dipoles at x0 is (see (3. the polarization of the matter is. the local field Eloc at an atom depends on the shape of the particle. If there are N atoms per unit volume.54) with the help of (3.4. The outer. in view of (3. their moments are all of strength p and aligned in the z-direction.e.The fields inside a dielectric particle 3. has the same surface charge as a sphere in vacuum of constant polarization P.54) Given the outer field E. However. The formula is analogous to (1. S2 . So we can neglect the influence of the nearest atoms altogether. Eloc = Ei + 4π P. We expect that we may also neglect it if there is no grid order at all.4.

102 Very small and very big particles E Ei E1 S1 S2 Figure 3. The resonance frequency ω0 of the atom lies typically in the ultraviolet. we get the electronic polarizability α(ω) = 2 m e (ω0 e2 . and equate the dipole moment p = ex 0 to α E loc . one can relate the dielectric susceptibility χ of the medium to the polarizability α of the atoms. the value from (1. A field E produces on a dielectric grain a surface charge σ .3. When we take for the amplitude of the electron. S1 and S2 denote surfaces. This is done in the Clausius– Mossotti formula: NV α χ= .55) 1 − 4π N V α 3 The field exerts forces on an atom and distorts the cloud of electrons. it attracts them one way and pushes the heavy nuclei the opposite way.56) It is relevant at optical frequencies because the electrons have little inertia and can swiftly follow the field. The static value of the atomic polarizability in a 2 constant field is α0 = e2 /m e ω0 .6)). one has to take into account the regularly but discontinuously arranged dipoles (small dots) in the vicinity as well as those farther away which can be considered as being distributed smoothly (shaded area).4. (3. According to a theorem of electrostatics. Because P = χ E i (see (1. . x 0 . − ω2 ) (3. σ gives rise to a field E1 that combines with E to form the average internal field Ei . The latter is not necessarily equal to the local field Eloc at a particular atom or unit cell (central larger dot in inner circle) in a crystal. To find Eloc . This can be modeled with the harmonic oscillator of section 1.67) but without damping (γ = 0).

we have to study basic optical principles. the diffraction pattern of the hole arises because each point in the plane of the hole is the origin of a spherical wave. but still much bigger than λ. the diffraction patterns of the hole and the obstacle are identical. on a far-away screen. obstacle into the orifice. Q ext = Q abs + Q sca approaches two. (2) If one places a small. We restrict the discussion to this dark area.5: (1) When a parallel wavefront falls on an orifice.5 Very large particles A particle is defined to be very large when its size is much bigger than the wavelength. the other half by absorption and reflection. a bright spot with a blurred rim. the screen is dark. except for the obstacle itself. 2πa → ∞. Outside the bright spot and the rim. which is much bigger than the wavelength. both diffraction patterns are present simultaneously.Very large particles 103 3. Because the region beyond the blurred rim is then dark. Altogether the obstacle thus removes twice as much light than that which corresponds to its projected geometrical surface: half of it through scattering. depending on the wavelength. the diffraction pattern of the obstacle arises because each point in the plane of the orifice.57) when x = Q ext → 2 λ We illustrate Babinet’s theorem with an experiment carried out in three steps as sketched in figure 3. 3. (3.e. they must have the same intensity but be phase-shifted by 180◦. This important result is called Babinet’s theorem or extinction paradox. it produces. a diffraction pattern appears and light is scattered beyond the blurred rim into the former dark area. For spheres. In the case of the completely uncovered orifice (1). the obstacle must scatter exactly the same amount. According to Huygens’ principle for wave propagation (this result will be shown later). (3) If we cover the orifice with black paper leaving just a hole of the same size and shape as the obstacle.5. . the definition is relative and the same particle may be both small and large. To quantify the diffraction phenomena that occur around very big grains. the patterns from the obstacle and the hole must cancel each other exactly. As the hole scatters all the light falling onto it. i. As in the case of tiny grains. independently of the chemical composition or shape of the particle.1 Babinet’s theorem For any very large particle the extinction efficiency. is the origin of a spherical wave.

1 Normal incidence A ray of light travels in the positive z-direction and hits.41) in the medium on the left √ (z < 0). A diffraction experiment to explain Babinet’s theorem. and k2 = ε2 µ2 ω/c in the medium on the right (z > 0). however. we obtain for .104 Very small and very big particles bright spot 1) dark area 2) obstacle 3) hole Figure 3.2 Reflection and transmission at a plane surface 3. it is always valid for interstellar grains. This quantity is denoted by r and called the reflectance. we know from everyday experience that a brick removes only as much sunlight as falls onto its projected surface and not twice as much. Therefore Q ext = 2 can only be verified at far distances. We wish to evaluate which fraction of the incident flux is reflected. 3. under normal incidence. predominantly forward.2. Scattering at the edge of a large obstacle is. At short distances.5. √ If k1 = ε1 µ1 ω/c is the wavenumber from (1. a large particle as shown in figure

58) (3.Very large particles incident transmitted reflected optical constant m 105 0 z Figure 3. Light falls perpendicular on a plane surface. the reflectivity is high and absorption low.32)). The amplitude of its electric field is weakened . Ei + Er = Et Ei − Er = Et ε 2 µ1 .67).59) To the left-hand side.117)). ε 1 µ2 This immediately allows one to calculate the reflectance r = |E r /E i |2 .17). A pure diamond sparkles at visual wavelengths because k is small and n = 2. The optical constant of the particle then equals m = ε2 and the reflectance is given by r= 1−m 1+m 2 = (n − 1)2 + k 2 . to the right only the transmitted wave (index t). we find. some is reflected and some transmitted. The medium to the right (z > 0) has an optical constant m = n + ik. (n + 1)2 + k 2 (3. so the reflectivity is high (r = 0.6. at z = 0. However. We see from (3. When we express the magnetic field via the rotation of the electric field (see (1. Metal surfaces make good mirrors because they have a large optical constant m and n k (see (1.60) For a large sphere (x = 2πa/λ 1). (3. of standard glass which has n = 1. At the surface of the particle (z = 0).4. The most common case is the one where there is a vacuum on the left and the grain is √ non-magnetic (µ2 = 1). if k 0. the reflectance equals the backscatter efficiency Q back of (2. If the stone in a ring is a fake. the electric field E = E i ei(k1 ·z−ωt ) + E r ei(−k1 ·z−ωt ) E = E t ei(k2 ·z−ωt ) for z < 0 for z > 0. the reflectivity grows with n. we have the incident and the reflected wave (indices i and r). not 17%. the tangential components of the electric and magnetic field are continuous.59) how the light that enters into a large and absorbing particle is attenuated.5. Should n or k be large. it will catch less attention because it reflects only 4% of the light.

any light that enters the grain will. so the intensity It of the transmitted light diminishes like exp(−4πk/λz).62) Using this formula. Q ref lies between 0 and 1 and is related to Q abs through Q abs = 1 − Q ref . One now has to specify in which direction the incident light is polarized which is not necessary under normal incidence.61) Whereas for a non-absorbing medium. The angle of the transmitted beam is given by Snell’s law: sin θt = sin θi . If k > 0.7. require the optical constant of vacuum (m = 1). 3. the oscillations of the electromagnetic field are in phase .5. the transmitted light is removed very quickly and the penetration depth is only a few wavelengths. Unless k is very small. the reflected beam forms an angle θr = θi with the normal. One may wonder about the implications of formula (3. A blackbody can be approximated by a particle with n → 1 and k → 0. If the incident beam is inclined to the normal of the surface element of the particle by some angle θi > 0. 3. in the generalized form of (3. despite the smallness of k.106 Very small and very big particles proportionally to exp(−2πkz/λ).3 Huygens’ principle In figure 3.2.61) m and also θt are complex.62) for an incident angle θi = arcsin(x/a) and a is the grain radius. eventually peter out within it. m (3. therefore.2 Oblique incidence We generalize the reflectance r of (3. By definition. it only scatters the light. (3. only smaller if some fraction of light is reflected off the surface. For incident unpolarized light the reflectance is r= 1 cos θt − m cos θi 2 cos θt + m cos θi 2 + 1 cos θi − m cos θt 2 cos θi + m cos θt 2 . On any spherical surface S around Q. the particle must also have a very large size d such that kd/λ is much greater than one. a point source Q emits isotropically radiation.63) where r (x) is the reflectance after (3. we can determine the limiting value of Q abs and Q sca for very large spheres. under the assumption of ray optics. Per wavelength of penetration. then Q abs = 0 and Q sca = Q ext = 2 according to Babinet’s theorem. But a vacuum is translucent. One can define a reflection efficiency Q ref for spheres by πa 2 Q ref = 0 a 2πr (x) d x (3.60) for a blackbody. So no real substance is a perfect absorber. If the particle is translucent (k = 0).60) to the case of oblique incidence. So the absorption efficiency for large particles can never be greater than one.5. it has zero reflectance and would. m and θt are real. the intensity decreases by a factor exp(−4πk).

7. some region on the surface of its wavefront. To derive them from Huygens’ principle. at time t. The superposition (interference) of all elementary waves of equal radius originating from the surface S fixes the position of the wavefront in the future and thus describes the propagation of the primary wave. if the wavefront is. Interesting examples are the intensity distribution of light behind a straight wire or a slit. each surface element dσ is the source of an elementary (or secondary) spherical wave. oscillate in phase and emit elementary waves which propagate with the phase velocity v = c/n of the respective medium.Very large particles 107 S’ S r0 Q dσ r Figure 3. Then the superposition of the remaining elementary waves leads to diffraction patterns. For instance. Applied to plane waves. a more detailed description is not given. A spherical wavefront traveling from a point source Q at velocity c. The strength of the elementary waves is proportional to the amplitude of the primary field and to the area dσ .7. being greatest radially away from Q and zero towards the rear. or the fuzzy edges of shadows. It also depends on direction. which are spherical waves of very large radius. Outside S . n 1 and n 2 . a time t = r/c later it will be again on a spherical surface but of radius r0 + r . one just has to assume that all surface elements on the plane separating two media of different optical constants. corresponding in size and shape to the projected area of the obstacle. If there is an obstacle in the way of the primary wave. forming a wavefront. Light no longer travels only on straight lines but can bend around corners. According to Huygens’ principle. on the surface S of figure 3. It also explains the laws of refraction and reflection in geometrical optics. Huygens’ principle makes understandable why light propagates along straight lines. . does not create secondary waves. the field of elementary waves is extinguished by interference. The new surface S is the envelope of all elementary waves of radius r .

r .5. φ = 0. The field u must obey the wave equation: (3.3. Spherical waves where u depends only on the distance r = x 2 + y 2 + z 2 so that 1 d2 (r u) + k2 u = 0 r dr 2 are another type of solution.108 Very small and very big particles S r P S’ Figure 3.1 Kirchhoff’s strict formulation of Huygens’ principle Kirchhoff put Huygens’ principle in a mathematically rigorous form. one finds ei(kr−ωt ) . which is encountered in electrostatics and where φ is the electric potential. He was able to determine the electromagnetic field u at any point P inside a region G when u is known on the surface S of the region G (see figure 3. (u v − v u) d V = G S u ∂u ∂v −v ∂n ∂n dσ where the function v is defined as v= eikr . u = ei(k·x−ωt ) .65) describes an outward-going spherical wave.2. To evaluate the field u at point P inside a region G when u is known on its surface S. the time factor e−iωt is again included.14)).65) u= r Equation (3.8).64) u + k2 u = 0. one can use the familiar formalism of electrostatics to solve it.8.64) in section 1. As formula (3. we use Green’s identity (see (A.64) is akin to the Laplace equation. 3. To determine u at a point P inside a bounded region G when u is given on its surface S. By introducing the auxiliary function v = r u. (3. as a solution to (3. We have already discussed plane waves.

that the field from a point source emitting spherical waves can be considered to result from interference of secondary spherical waves. The field u in P at a distance R = Q P is calculated from Huygens’ principle in the form (3.66) 3. u + k2 u = 0 and v + k2 v = 0. a distance R from the point source Q. there is a singularity because r = 0. .7). j = 1. this remaining region is called G.66). It is assumed that u is known on the spherical surface S around Q with radius r0 . 4πu p = − S u eikr ∂u ∂ eikr − ∂n r r ∂n dσ. as had been asserted before (figure 3. (3. of course. The new configuration is depicted in figure 3. i. . . The surface integral in Green’s identity now extends over S and S . . we show. one cuts out from G a small sphere around P of surface S . At P. Because in G. . the field is. At P.4 Fresnel zones and a check on Huygens’ principle To convince ourselves that (3. .66) is indeed a strict form of Huygens’ principle. the wave equation is fulfilled for u and v.Very large particles 109 S n r Q r0 Z3 Z2 Z1 P R Figure 3. To circumvent it. one constructs Fresnel zones Z 1 .5. respectively.e. one eventually arrives at an equation that incorporates Huygens’ principle.9. A point source Q isotropically emits light. Letting the radius of the small sphere around P go to zero. Z 3 .9. The radius r counts the distance from the point P where we want to evaluate the field u. 2. . 3. . the volume integral vanishes and only the surface integral is left. up = eikR R . The volume integral in Green’s identity is then taken only over the remaining region without the small sphere. To carry out the integration. Z 2 . They are the intersection of S with shells around P of inner and outer radius R − r0 + ( j − 1)λ/2 and R − r0 + j λ/2.

67) If the distances are much greater than the wavelength (r. yields u1 = 2eikR . over whose surface one has to integrate. R The contribution u j of the j th zone Z j to the field at P follows from a simple integration over r . and this strengthens our confidence in Huygen’s principle. if all others were covered by some dark material. we only have to consider the surface S. These are segments on S lying between constant radii r and r + λ/2. r) + (i k − r0 ) dσ.110 Very small and very big particles We claim to obtain the same result from superposition of all secondary waves emitted from the spherical surface S. r) . r0 4πu p = −i k eikr0 r0 k−1 = λ/2π). r) with the radius vector r. like the first. r) r where r is the radius vector from P to S and r = |r|. at fixed [1 + cos(n. On the surface S. eikR (−1) j +1 1 + cos(n. R The contributions from the following zones alternate in sign. rr0 (3. r)]. The last zone is the one on the far side of the sphere which.66) of the surface which is at infinity may be neglected. The area of a segment is dσ = 2πrr0 dr. uj = R The first zone alone. divide the surface S into so called Fresnel zones. we get 4πu p = − S eik(r+r0 ) −1 (i k − r −1 ) cos(n. in the limits R − r0 + j λ/2 and R − r0 + ( j − 1)λ/2. for this purpose. comprises now all space outside the sphere of radius r0 . Because eikr ∂ eikr = ∂n r r ik − 1 cos(n. . The contribution to the integral (3. r)] dσ. The surface normal n of a given segment forms an approximately constant angle cos(n. r We evaluate this integral as a sum and. One really obtains eikR /R.66) supposing that we know the field and its normal derivative on S. S eikr [1 + cos(n. It is not difficult to work out the sum u j over all Fresnel zones. The region G surrounding P. So we calculate u p from (3. u = u0 = eikr0 r0 ⇒ ∂u ∂n =− 0 eikr0 r0 ik − 1 r0 . is intersected by the straight line through P and Q. The first two zones together produce almost nothing but darkness.

6 Diffraction by a circular hole or a sphere Equation (3.68) gives the field of an electromagnetic wave that has passed through an orifice. (3. 3.68) with respect to r and r0 implies that when a source at Q produces a field u p at P.5 The reciprocity theorem As very big particles block practically all light that falls onto them. We now apply it to a circular hole of radius a and remember from the . the same source brought to the point P produces the same field at Q.Very large particles 111 P r S r0 θ Q Figure 3.5.10). geometrical optics is valid and the hole deflects light only by small angles.10.5. We compute it from superposition of secondary waves from the orifice S. To first order. interesting phenomena occur only through diffraction at their edges. eik(r+r0 ) varies of course over the hole because it is many wavelengths across. Therefore we approximate equation (3.67) under the assumption k−1 by r. and θ is the angle between Q P and the normal of the plane.68) λ rr0 S 1 Here cos θ 2 [cos(n. 3. Light falls from Q though an orifice S and is observed at P. This is the reciprocity theorem of diffraction theory. The symmetry of formula (3. r) − cos(n. Suppose we have a source Q behind an opaque plane with an orifice S of arbitrary shape and we want to determine the intensity u p at a point P in front of the plane (see figure 3. The distances r and r0 are also almost constant and may therefore stand before the integral. however. r0 )]. r0 i cos θ up = − eik(r+r0 ) dσ.

It is assumed that the hole is small (s R. 0). −R0 ) P = (0. one obtains yp s cos ϕ. r0 is the distance between Q and dσ . y)-plane of the Cartesian coordinate system with origin C. y. The coordinates of the points marked in figure 3. 0) Q = (0. R0 ) and that Q. z p ) C = (0.112 Very small and very big particles y A r0 s . 0. subsection on Babinet’s theorem that a spherical obstacle will produce the same diffraction pattern. The configuration is shown in figure 3. respectively.11. ϕ.10) is close to one and may be omitted. which implies s r + r0 R + (s 2 − 2yyp)/2R. The hole is in the (x. C and P lie roughly on a straight line so that cos θ (see figure 3. r R P Q R0 y P C z Figure 3. As we have to integrate over the hole. 0. The source Q is behind a circular hole (dash-dot line) and illuminates the point P on a screen. z) to the source Q and the observing point P. it is advantageous to use cylindrical coordinates (s. R0 and R are the distances from the center C of the Cartesian coordinate system (x. R R + R0 − . yp . y. z). where s2 = x 2 + y2 Because 2 2 r 0 = R0 + s 2 2 r 2 = x 2 + (yp − y)2 + z p 2 2 R 2 = yp + z p x = s sin ϕ y = s cos ϕ.11 are A = (x. The x-axis is perpendicular to the page of the book. Neglecting terms yp .11. r is the distance between P and an arbitrary surface element dσ located at A. it follows that r0 R0 + s 2 /2R0 and r of order s 2 or higher.

For small x. To evaluate the double integral.70) For example. It may seem puzzling that the intensity of the light which one observes at P and which is equal to |u p |2 rises with the fourth power of the hole radius a. λ 2R Note that the phase lag δ increases quadratically with (y p −y).21). (3. To determine the field at P. in the visual bound. opposite to S. the path T P is approximately equal to R +(y p − y)2 /2R. the diffraction pattern gets smaller and. Therefore. J1 (x) → 1 x. We have introduced the deflection angle α = yp /R. This problem affords a nice graphical solution.5. in the end.69) The Bessel function J1 (x) is tabulated in mathematical encyclopediae. relative to the wave emitted from S. The situation is depicted in figure 3. as much energy as falls onto the hole from the source emerges on the other side. at x = 1. yp − y the wave from T has. 2 The intensity |u p |2 drops to half its maximum value at α = 0 at a deflection angle αhalf 1. the field vectors E j of all elementary waves have to be added up. eik(r+r0 ) dσ = eik(R+R0 ) S 0 a 2π 113 ds s 0 dϕ e−ikαs cos ϕ .24) yield for the double integral a 2π 0 ds s J0 (kαs) = 2πa 2 4π 2 a 4 · 2 λ2r 2r0 J1 (akα) akα 2 and thus |u p |2 = J1 (akα) akα . 2πa (3. Because we assume R. The formulae (A. They have practically all the same length but different directions. although the flux passing through the hole increases only with a 2 . one uses complex integer Bessel functions Jn (z) defined in (A. At P. one has to sum up all elementary waves from the plane in which the wall lies.7 Diffraction behind a half-plane As a final example.22)–(A. energy is conserved.12. 3. This leads to the so called cornu-spiral . αhalf is about 1 arcmin. for a grain with 1 mm diameter.Very large particles Because dσ = s · ds dϕ.617 λ .68) becomes. the integral in (3. But as the hole becomes bigger. a phase lag δ= 2π (y p − y)2 . we derive the intensity pattern on a screen behind a half-plane (wall). this approximation has an error of roughly 10%.

down to the edge of the wall. is represented in figure 3. a position below the edge of the wall. this arrow corresponds to a particular vector E j . opposite to the edge of the wall. the hole was assumed to be much bigger than the wavelength so that sections of its rim can be approximated by a straight line. The asymptotic points of the spiral are A and B.14 also gives roughly the variation of light at the ‘blurred rim’ around the bright spot mentioned in the discussion of Babinet’s theorem (see the top of figure 3. At point P3 .69). But exactly behind a sphere at large distances there is. and the second consists of all E j from points below S.13 would end in the middle of the spiral at the fat point.12.13 by a small arrow. It starts at A and ends at the dot which corresponds to the elementary wave from the rim of the wall. thus the curvature of the spiral increases too. it ends in the lower half of the spiral.13.13 by a large arrow. from (3. which is far up on the screen. If one connects any two neighboring dots in figure 3. of figure 3. Obviously. Light comes in from the left. the large arrow in figure 3. which is the resulting field at P. The sum of all field vectors E j at the point P has two parts: • • the first includes all E j of elementary waves emitted from points above S. Their sum. up to infinity. always a maximum intensity . is blocked by a wall and observed at various points on a screen. the arrow goes from A almost to B. If we observe at position P2 .114 Very small and very big particles screen P 3 light S y T P y p R P 2 P 1 wall Figure 3. A blurred rim appears near P2 .5). Observing at P1 . Figure 3. The following vector E j +1 is rotated by an angle that increases as one approaches the asymptotic points. the intensity at P2 is four times smaller than at P3 .

The normalized intensity around position P2 in figure 3.14. . Figure 3. it falls smoothly to zero which is geometrically clear from figure 3.12. On the bright side of the screen. The squared length of the plotted vector gives the intensity at a point somewhere above P2 in figure 3. the intensity equals 0.Very large particles 115 Figure 3. the intensity oscillates and approaches unity.13.25.42 in √ units λR/π . The separation between the first two maxima in figure 3. Here is an astronomical application of this result which can. On the dark side. because the waves from the circular rim are all in phase on the center line and interfere there constructively. At exactly P2 (y p = 0).13.14 is 1.12. see text. The cornu-spiral.

Therefore. For example. If we imagine the total grain volume V to be divided into vi = V .43). If it occults a quasar. 3 To find the scattering cross section or the scattered intensity in a certain direction. one cannot simply sum the contributions of the subvolumes because there is a phase difference between them which leads to interference. which happens to lie on its path. the absorption efficiency of a Rayleigh–Gans sphere for any x is Q abs = 8x Im{m − 1}. their absorption and scattering cross sections are given by (3.8 Particles of small refractive index A grain that fulfils the conditions that • its optical constant is close to unity. observationally checked. however. each of them absorbs and scatters many tiny subvolumes vi with independently of the others. If the quasar is observed at the radio wavelength λ = 11 cm and if it is much smaller than 1 arcsec and crosses the rim of the moon at a right angle.2 on the optical theorem.54 arcsec s−1 . it may. |m − 1| • 1 or n 1. its intensity will oscillate as shown in figure 3. k 0 (3.42). It is this effect which. even for simple grain geometries (see examples in [Hul57]). .2 s. Inside such a particle.116 Very small and very big particles and has been. (3. which we derived in the Rayleigh limit. Otherwise it may be of arbitrary shape and size.14. Consequently.71) and that it is more or less transparent. tends to reduce C sca (λ) with respect to (3. if the subvolumes are ellipsoids. Of course.43). for large bodies. everything can be extracted from the full Mie theory but not in analytical form. The silvery moon on its heavenly orbit (R = 384 000 km) moves relative to the stars with an angular velocity of 0.72) is called a Rayleigh–Gans particle. d |m − 1| λ 1. the electromagnetic field is only weakly deformed and practically the same as in the incoming wave.42) and (3. be big compared to the wavelength. 3. the absorption cross section C abs of the total grain is also given by formula (3.5. The formalism for how waves originating from different points (subvolumes) are added up is basically the same as that discussed in section 2. therefore. The analytical computation of the scattering cross section or the phase function is. the second intensity maximum follows the first after 5. complicated.

5. therefore. so |ε − 1| 1. The latter is much smaller than for large. .577 215 66. The grains are. The scattered intensity I (θ ) is proportional to I (θ ) ∝ G 2 (2x sin 1 θ )(1 + cos2 θ ) 2 with G(u) = 9π 2u 3 1/2 J3/2 (u) where J3/2 (u) is a half-integer Bessel function. non-transparent spheres which always have Q sca ∼ 1. interstellar dust particles are always big because x = 2πa/λ 1. we obtain the Rayleigh limit Q sca = 32 27 |m − 1|2 x 4 and for x 1.Very large particles 117 Here are some results for spheres: If x is the size parameter and γ = 0.9 X-ray scattering We now apply the results obtained for particles of small refractive index to X˚ rays. |m − 1| 1. say 10 A. The value of the dielectric permeability in this range may be estimated from the high-frequency approximation (1. we expect ε1 to be close to one and ε2 to be small. Q sca = |m − 1|2 ϕ(x) with ϕ(x) = 7 5 sin 4x + 2x 2 − − (1 − cos 4x) 2 4x 16x 2 ∞ cos u 1 du . At a wavelength of. As the frequency ω is large. we get Q sca = 2|m − 1|2 x 2 . 3.79). + − 2 γ + log 4x + u 2x 2 4x For x 1. likewise for the optical constant. very transparent and even satisfy condition (3.72): 2πa |m − 1| λ 1. Euler’s constant. which gives ε1 = 1 − 2 ωp ω2 1 ε2 = 2 ωp γ ω3 1 where ωp is the plasma frequency and γ a damping (not Euler’s) constant.

observe towards strong X-ray sources located behind a dust cloud an X-ray halo of this size. there is forward scattering ˚ over an angle such that θ x ∼ 2. scattering is very much in the forward direction defined by θ = 0. The intensity towards it follows from the fact that the total scattered X-ray flux is smeared out over the halo. the angular dependence of the scattered intensity is given by I (θ ) ∝ G 2 (u) with u = 2x sin 1 θ. 2 When one numerically evaluates the function G 2 (u). . therefore. For grains of 1000 A radius. One can.5 and then stays very small. drops monotonically. Therefore. Consequently. one finds that it has its maximum at u = 0.118 Very small and very big particles For big spheres. this corresponds to θ ∼ 10 arcmin. like a bell-curve. reaches zero at about 4.

Chapter 4 Case studies of Mie calculus We have collected the tools to calculate how an electromagnetic wave interacts with a spherical grain. the formulae presented so far allow us to derive numbers for the cross sections. In this chapter. 4. we may envisage the wavelength λ to be fixed so that x is proportional to grain radius a. therefore. The efficiency depends only on m and on the size parameter x = 2πa/λ. scattering matrix. Figure 4.1.e. we present examples to deepen our understanding. too.1 Pure scattering As a first illustration. the absorption coefficient vanishes1 and Q sca = Q ext . i.1 Efficiencies of bare spheres 4. With the help of a small computer. a dielectric. we understand a dielectric to be the opposite of a metal. goes to zero. 1 Astronomers sometimes call a non-dissipative medium. So a dielectric may well have a truly complex m. where the imaginary part is not zero. Q ext → 0 for x → 0.e. we consider the scattering efficiency of a non-dissipative sphere. phase function. one with a purely real optical constant. 119 .1 demonstrates two features valid for any particle: • When a particle is small. zero. the figures also offer the possibility of checking his computer program against the one used in this text. we naturally expect the cross section C to be small. However. The imaginary part k of the optical constant m = n + i k is. To better interpret the dependence of Q sca on x. which is the cross section over projected area. and so forth. i. To the student who wishes to perform similar calculations. But even the extinction efficiency. Let us apply them. one that does not absorb the incident light.

. bn . when the grain is very large.50).120 Case studies of Mie calculus Figure 4. Then the abscissa of the left frame is proportional to the grain radius. Because in any real material. . m = 1. whereas we have kept it fixed in the figure. this value is appropriate for glass in the visible part of the spectrum. .002 ≤ x −1 ≤ 0.4 times (!) bigger than the projected surface of the particle.2 A weak absorber Figure 4. In figure 4. The first maximum of Q ext is achieved roughly at 8(n − 1).1. bn of (2.46)). in our particular example this occurs at x 4. (2. The extinction efficiency can be quite large. The extinction efficiency Q ext as a function of size parameter x and its inverse x −1 for an optical constant m = 1. Q ext → 2 for x −1 → 0. the extinction efficiency approaches two (see the right-hand frame of figure 4. the optical constant is complex.1). The ripples correspond to small denominators in an . The minima in between stay more or less all around two. Now . In the middle. There are two interesting aspects pertaining to a pure scatterer or weak absorber: • Q ext displays an overall undulating character with broad waves on which semi-regular ripples are superimposed. • 4. the right the range 0. the broad crests of the waves to maxima in the sum an + bn . The phenomenon of ripples and waves can be traced to the behavior of the coefficients an . it is followed by gradually declining maxima around x = 11. 17. . m depends on wavelength. exemplifies the effects of adding impurities to the material. The left box shows the range 0 ≤ x ≤ 20.02.5 + i 0.1 and displays a medium with m = 1.5. it is best to envisage λ as constant.51) and is due to interference in the scattered wave (see(2. The top frame is identical to the left box of figure 4. its maximum value is 4. • In accordance with Babinet’s theorem.05 corresponding to 20 ≤ x ≤ 500.

scattering (dots) and extinction (full line) for three different substances: one is not dissipative. as can be seen by comparing the two bottom panels where k changes from 0.2. The efficiencies for absorption (dashes). Impurities effectively suppress the ripples and they also lead to a decline in the amplitudes of the waves. For example.Efficiencies of bare spheres 121 Figure 4. . the ripples and the waves are irrelevant. the others are weakly absorbing.5 + 0.05. An optical constant m = 1. The damping is sensitive to k. the particle not only scatters the light but also absorbs it. For interstellar grains.02 to 0.05i may be appropriate for astronomical silicate around λ = 6 µm.

when we measure the intensity of a star behind a dust cloud.5. Q abs increases quickly with x at x < 1.122 Case studies of Mie calculus Figure 4. We discern.5 + 0. more importantly.3 we retain the value of n = 1. . This implies strong absorption. The scattering efficiency changes after (3. 4. Q abs Q sca .5 of the previous two plots but choose for the imaginary part of the optical constant k = 0. scattering and extinction.3 A strong absorber In figure 4.4) as Q sca ∝ x 4 • so Q sca falls off more steeply than Q abs (see logarithmic plot of figure 4. interstellar dust displays a range of sizes.5i.1.3). Absorption dominates over scattering. as well as the asymmetry factor g for a strongly absorbing material with m = 1. as predicted by (3. the general features of small-sized grains. The optical constant m(λ) will change somewhat over this bandwidth and.5).3. so the peaks produced by particles of a certain diameter are compensated by minima of slightly smaller or bigger ones. in this particular example. The efficiencies change now very smoothly and there are no signs of ripples or waves left. Because of the strong absorption. the observation is not monochromatic but comprises a certain bandwidth λ. The efficiency for absorption. When x → 0: • • Q abs and x are proportional.

Although isotropic scattering implies g = 0. as defined in equation (3.5. compiled in table 4. When the sphere is big. The other half is scattered at the particle’s rim entirely in the forward direction. as always. a metal has (at low frequencies) a purely imaginary dielectric permeability so that n k goes to infinity (see (1. Surprisingly. • • Figure 4. a small reflectance Q ref = 1 − Q abs .4 illustrates these items in case of large but not extreme metallicity (m = 20 + i 20). The combination of both effects gives g = 1 . of all the light removed. whereas the absorption stays at some low level. the absorption efficiency can exceed unity. the scattering increases and soon takes over. The presentation is different from the preceding figures as we choose for the abscissa the grain radius a.1. When the sphere is big. only the concept of a blackbody is reserved to sizes much larger than the wavelength. In this particular example of m = 1. the calculations are correct.Efficiencies of bare spheres • 123 The asymmetry factor g of (2.4 A metal sphere • • • • Ideally.863 and Q ref = 0. half is scattered isotropically and it alone would yield an asymmetry factor g = 0. Q abs = 0.5 Efficiency versus cross section and volume coefficient In figure 4. and it has a large absorption efficiency and.0 which applies to amorphous carbon at a wavelength λ = 2.7 + i 1. the reverse is not true.6). which describes the mean direction of the scattered light. Nevertheless. the metallic sphere also obeys the relation Q abs Q sca . The logarithmic plots reveal the limiting behavior of small and large particles. the wavelength is kept constant at 2. close to the Babinet limit of two. which alone would produce g = 1. therefore. For small values of |mx|. These features. which happens here when x > 0.2 µm. backward scattering dominates (g < 0). A metallic sphere is. so g is not far from one.1. we compare efficiencies. Unity is the value of Q abs of a blackbody and may thus appear to be an upper limit for any object. 4. Q ext is. 4.63).2µm. But as x grows.5. The asymptotic values for x → ∞ are: g = 0.1. very reflective and Q abs tends to zero. therefore. a very large particle (x 1) • • scatters mostly in the forward direction. cross sections and volume coefficients using an optical constant m = 2. . and not the size parameter x.117)).3 plots only the range up to a size parameter of five but the behavior for larger x is smooth and can be qualitatively extrapolated.137. vanishes. 2 When the sphere has a size comparable or smaller than the wavelength. Figure 4. are independent of the choice of m or λ.9.5 + i 0.918.

4. irrespective of the size distribution of the particles. negligible relative to absorption. of a spherical grain. cross section and mass or volume coefficient on the radius.4).3 but now for a metal for which n and k are large and equal. one may assume the Rayleigh limit to be valid at all wavelengths greater than 10 µm. If m is constant.3) and (3. a. Scattering is. Symbol Efficiency Cross section per grain Mass or volume coefficient Q C K Large radii a λ Q abs = constant Q sca = constant C abs ∝ a 2 C sca ∝ a 2 K abs ∝ a −1 K sca ∝ a −1 Small radii a λ Q abs ∝ a Q abs ∝ a 4 C abs ∝ a 3 C sca ∝ a 6 K abs = constant K sca ∝ a 3 Spheres of small radii • • • The efficiencies depend on wavelength as given by (3. . The asymptotic dependence of efficiciency. Same as figure 4.124 Case studies of Mie calculus Figure 4.1 µm.5. The situation applies approximately to amorphous carbon at millimeter wavelengths (see figure 7. Table 4. Q abs ∝ λ−1 and Q sca ∝ λ−4 .19). abs The mass absorption coefficient K λ is proportional to the grain volume V . therefore. Note that g becomes negative. As interstellar dust has typical sizes of 0.

the net effect is not linear but proportional to the square of the grain volume because of interference of the scattered waves. The optical constant m = 2.7 + i1.2 µm. . wavelength λ = 2. • sca The mass scattering coefficient K λ follows V 2 . Although the scattering process is identical in all subvolumes of a grain. the middle one the cross section per grain C = πa 2 Q.5. and the bottom one the volume coefficient K referring to 1 cm3 of grain volume.Efficiencies of bare spheres 125 Figure 4.0. The top frame gives the absorption and scattering efficiency Q.

The scattered light is polarized and maximum polarization occurs 90◦ away from the Sun (see (3.6 The atmosphere of the Earth The atmosphere of the Earth presents another instructive example. Bλ (Tdust ) is the Planck function at the dust temperature (see (8. V grows by a factor of ∼ 1010. it would appear black. the sky is painted blue. More relevant for the reddening towards the horizon is the removal of blue light through absorption by tiny solid particles.2 Spheres of large radii The efficiencies. reach a finite value with Q ext = Q abs + Q sca = 2. according to table 4. Because Q sca ∝ λ−4 and λblue < λred . the observed flux is directly proportional to the total dust volume. as otherwise interference would cancel out the intensity except in the forward direction. the tiny air molecules scatter the sunlight towards us.5. The size of the water droplets is then still . except for the stars. K .19)). An additional requirement is that the air molecules are randomly moving. The reason that they become visible when they cluster into droplets is the following: The mass scattering coefficient K sca is. However. When on a summer day the solar heating induces convection and a parcel of air rises we see the parcel only at the moment the water condenses.1. a sand storm or a volcano eruption. We can now understand why the daylight sky is blue. or at night time 90◦ away from the moon. 4. the transmitted light becomes redder. During the phase transition from gas to liquid. decrease inversely proportional to the radius. C. We can also explain why we see the water in the air only when it has liquefied into clouds. If there were no atmosphere.1)). a condition that is usually fulfilled. This seems puzzling because upon condensation the water molecules do not drastically change their properties as radiating oscillators. The mass or volume coefficients. Sun and planets. increase without bounds proportional to the projected area of the grain. Provided the infrared photons are not trapped in the cloud. But this does not explain the occasional romantic sunsets.126 • Case studies of Mie calculus The emission coefficient λ abs = K λ Bλ (Tdust) also depends linearly on the grain volume. Q. and not as vapor. proportional to the volume V of a single scatterer (one H2 O molecule or one water droplet). The dependence of absorption on λ is less steep than for scattering but qualitatively similar (Q abs ∝ λ−1 ) and also results in reddening.1.1. As preferentially blue light is removed in scattering. although the H2 O molecules were there before. These may arise from industrial exhausts. The cross sections per grain. 4.

2 Scattering by bare spheres 4. 0).01 represents a dipole pattern which is characteristic for a particle small compared with the wavelength irrespective of the choice of the optical constant m. In the right frame where m = 10 + i 10. see also table 4. All contours are mirror-symmetric with respect to the horizontal arrows. It is symmetric relative to the scattering angle of 90◦ (vertical line in the figure) and therefore has equal maxima in the forward (θ = 0) and backward (θ = 180◦) directions. although g = 0. its value in any other direction is given by the distance between the frame center and the contour line. we are dealing with a reflective material. The change in m relative to the non-dissipative sphere shown on the .Scattering by bare spheres 127 Figure 4.1 The scattering diagram The pattern of how the intensity of scattered light changes with angle is described by the phase function of (2.2. Diagrams of the intensity distribution of scattered light for three sizes for parameter x and two optical constants m. the left frame refers to a non-dissipative sphere with m = 1. Light enters from the left and the grain is at the position (0.6. The pattern is sensitive to the size parameter x = 2πa/λ.2. almost all light goes into a forward cone.6. scattering becomes very forward and for a grain diameter just equal to the wavelength (x 3).5. Each diagram is normalized in such a way that the intensity equals one in the forward direction (θ = 0). The curve labeled x = 0. In these examples. smaller than the wavelength of visible light so that the dipoles in the droplet swing in phase.63) and visualized in figure 4. When x ≥ 1. 4. Note that the scattering is not isotropic.

04 × 10−5 0. whereas in the left-hand panel.7 and 4.2. There is also not much difference for x = 3. The polarization is 100% at a scattering angle of 90◦ . They display the basic features.19 0.53 4. shape and optical constant of the particle. at the intermediate value x = 0. 4.05 2. at x = 1. the scattering is mainly forward. 01.98 × 10−5 0. asymmetry factor and back scatter for spheres whose scattering diagram is displayed in figure 4.451 Q back 3. dashes in figure 4. although curious tiny rear lobes appear.42 left does not affect the curve x = 0.2 The polarization of scattered light For the same optical constants and size parameters as previously.31 × 10−9 2. p vanishes at three intermediate angles. The polarization changes sign and there are several maxima and minima of varying height. For weak absorbers. The behavior of p(θ ) at intermediate sizes (x = 3) is more complicated in both materials.67 × 10−8 2. the observed values • • .5 × 10−9 0.2.0 × 10−8 3.01 1 3 0. p(θ ) hardly changes as x grows from 0.01 1 3 m 1.5 1. we now have preferential backscatter and the asymmetry factor is negative (g = −0.7) is symmetric around 90◦ and has the shape of a bell. m = 10 + i 10.6.77)).5 10 + i10 10 + i10 10 + i10 Q sca 2. k 1. When one measures the linear polarization of scattered light at various locations around the exciting star of a reflection nebula. Here the matrix element S12 switches sign (see (2.15 3. The scattering plane is defined by the unit vectors in the direction from where the light is coming and where it is going. although the details are sensitive to size. figure 4.5 1.8 depict the degree of linear polarization p of light scattered by a sphere (see (2. the polarization p(θ ) is no longer symmetric and attains a maximum value of only 78%.11). figures 4. However.75)) and the direction of polarization with respect to the scattering plane changes from perpendicular to parallel or back. • The polarization curve p(θ ) of tiny and weakly absorbing spheres (x = 0.128 Case studies of Mie calculus Table 4. Tiny metallic particles (x = 0. Scattering efficiency. In our examples. x = 2πa/λ 0.01 up to 1.73 1.20 0.42 2.111 0.08 Asymmetry factor g 1.8) have the same pattern as tiny dielectrics but at x = 1.

The dotted line depicts a very large sphere (x = 1000) with some slight contamination (m = 1.8.5 + i0. the polarization at certain scattering angles goes through zero and becomes negative. The degree of linear polarization p of scattered light as a function of scattering angle for various size parameters x and an optical constant m = 10 + i10. Figure 4. For intermediate sizes (x = 3). the polarization vector flips.5.7. . The degree of linear polarization p of scattered light as a function of scattering angle for various size parameters x and an optical constant m = 1.Scattering by bare spheres 129 Figure 4. At the points where p = 0.003).

it can happen. one observes light from two nebulous lobes separated by a dark lane that is interpreted as a disk inside which sits the central and exciting source named IRS4. the angle of deflection is close to 0◦ or 180◦ . If the particles are small.9. An impressive example of a very regular polarization pattern is presented by the bipolar nebula S106 (figure 4. • • are always an average along the line of sight.9). IRS4 is an infrared source and its radiation is reflected by the dust particles that surround it.130 Case studies of Mie calculus Figure 4. Very big dielectric spheres (x 1000) again show a simple pattern with no sign reversals. although it is an unfamiliar scenario to astronomers. There is 100% polarization in a slightly forward direction (θ 65◦ ).2 µm) a polarization pattern of wonderful circular symmetry (adapted from [Asp90]). There is plenty of visual obscuration towards IRS4 (∼20 mag) but in the near infrared one begins to penetrate most of the foreground and outer disk material. This is a well-known star-forming region. S12 < 0 and the polarization vector is perpendicular to the radial vector pointing towards the star. Polarization is always zero in the forward and backward directions (θ = 0 or 180◦). However. We are viewing the disk from the side. At optical wavelengths. Some light is scattered by dust particles producing in the K -band (2. p(θ ) has more structure. When looking directly towards IRS4. that the radial and polarization vectors are parallel. The infrared source IRS4 (at figure center) in the bipolar nebula S106 illuminates its vicinity. as the figure suggests. Reproduced from [Asp90] with permission of Blackwell Science. For metals of this size.

In the infrared where a λ. therefore. roughly S11 ∝ (a/λ)6 . Fλ = F0λ λ2 3 · S11 (cos θ ). the matrix element S11 (cos θ ) is very sensitive to both grain size and wavelength. scattering occurs at right angles and the light becomes strongly polarized. which is illuminated by a flux F0λ from a nearby star.1) for right angle scattering (cos θ = 0). The term (λ2 /a 3 )S11 (cos θ) in equation (4. It is computed as a function of wavelength for silicate and carbon grains with radii a = 3 × 10−6 . Note that S106 must be optically thin to scattering in the K -band (see table 7.5 g cm−3 .03. Some of the radiation is scattered under an angle cos θ towards the observer who is at a distance D. 0. The flux Fλ that he receives is found from formula (2. we consider 1 g of dust in interstellar space. But at positions a few arcseconds off.Scattering by bare spheres 131 Figure 4.63).1. determines the intensity of the scattered infrared light. 0. φ) by the element S11 of equation (2.3 The intensity of scattered light in a reflection nebula To evaluate the intensity of scattered light.2. and the ensuing degree of linear polarization is small.2) as otherwise multiple scattering would lead to depolarization.1) The expression (λ2 /a 3 )S11 (cos θ ) is delineated in figure 4.10 for a scattering angle θ = 90◦ . The size of the largest grains. consisting of Nd = 3/4πa 3 ρd identical spherical grains of radius a and density ρd = 2.19.3 µm).1). Of .10. The 10 µm feature of silicates is discernible. 4.e. where we replace Ä(θ. Optical constants from figure 7. a = 0. 16π 3 ρd D 2 a 3 (4. 10−5 and 3 × 10−5 cm (i.

An extinction of 0.55 µm for right angle scattering 8 × 105 . The left box in figure 4.80 × 10−15 erg s−1 cm−2 .0 × 1026 g in a column with a cross section of 1 arcsec2 . For spheres.03i ). With the factor wλ = 3640 to convert Jy into that (λ2 /a 3 ) × S11 magnitudes from table 7. we find a surface brightness in the nebula at V of 18. its visual flux at a distance r = 0. for r = 1. As an exercise. with a gas-to-dust mass ratio of 150 and accounting for 10% of helium. If the star is approximated by a blackbody. Elongated grains.23) to a hydrogen column density NH = 2 × 1020 cm−2 . we apply figure 4. Only the appearance of two more boundary conditions at the interface between the two dust materials is new.2.1 mag corresponds after (7. a dust mass of 7. the shape of the particle also plays some role.33 + 0. the dependence of S11 (cos θ ) on the scattering angle is complicated and the curves in figure 4.33 + 0. So the total grain radius is atot = acore + d. In the ultraviolet. only of mantle material (m = 1. isotropic scattering at infrared wavelengths is not a bad assumption.1 mag.03i .7 + i .10 to answer the question: What is the approximate V -band surface brightness of scattered light in a reflection nebula? The nebula is D = 1 kpc away and excited by a main sequence star of spectral type B1 with L = 104 L and T∗ = 2 × 104 K. 4. Such particles may be approximated by two concentric spheres. This gives. We calculate this quantity at a projected distance from the star of r = 0.3 Coated spheres The mathematical formalism of Mie theory presented in section 2. Let the particles have radii a = 10−5 cm.3 can readily be extended to coated spheres. we plot Q abs for coated spheres as a function of acore r= .1 pc is F0.10 at a wavelength of 0.132 Case studies of Mie calculus course.7 + i and m = 1. For a rough estimate. scatter more efficiently than spheres of the same volume. The four curves represent . There is a refractory core of radius acore and optical constant m core surrounded by a volatile shell of thickness d and optical constant m shell. we neglect that the distance of the scattering dust to the star changes along the line of sight. even of arbitrary orientation. atot For r = 0. Then we read off from figure 4.5 mag arcsec−2 .1 pc assuming that the extinction along the line of sight through the nebula equals AV = 0. the grain consists only of core matter with m = 2. In the right box. The relevance of calculating coated particles stems from the observational fact that grains in cold clouds acquire ice mantles.10 may only be used if θ is not too far from 90◦ .11 shows the absorption efficiency Q abs as a function of size parameter x for homogeneous particles with optical constants m = 2.V = 9.

Surface modes in small grains 133 Figure 4.33 + 0. The left frame gives the absorption efficiency as a function of size parameter for homogeneous spheres.1 Small graphite spheres When grains display.5 and 8.7+i is representative of amorphous carbon at λ = 2.4 Surface modes in small grains 4. The right figure demonstrates how the absorption efficiency of a coated grain changes when the core grows at the expense of the mantle while the outer radius atot is fixed. The right frame shows the absorption efficiency of a coated sphere. For easy comparison. in the case of grains which are small compared with the wavelength. The optical constant m = 2. . Dotted vertical lines for reference to right box. As the extinction efficiency is in the Rayleigh limit after (3. the extinction cross section may be large without an accompanying feature in the bulk matter. an extinction peak. x denotes the size parameter of the total grain.03i is applicable to dirty ice at the same wavelength.2 µm and m = 1. Because ε = m 2 . at some wavelength. The phenomenon bears the name surface mode.11. these values are also indicated by dotted lines in the left frame.4. we expect a large value for Q ext at a wavelength where ε+2 0. the reason is generally a resonance in the grain material. However. size parameters x = 2πatot/λ of the total sphere between 0. this happens when the real part n of the optical constant is close to zero and k 2 ≈ 2.0. 4.3) proportional to ε2 /|ε + 2|2 .

the strength of the surface mode as well as its center wavelength. via the shape factor L x . For the Rayleigh approximation to be valid. The variation of ε is.19) but graphite around 2200 A is a candidate.0. 4. To underline the ext is normalized to one at λ−1 = 3 µm−1 . Graphite is optically anisotropic and the effect appears when the electric vector of the incoming wave lies in the basal plane of the stacked carbon sheets.12 illustrates how the peak in Q ext drifts towards smaller wavelengths (greater λ−1 ) with decreasing radius a. They are shown in the upper right frame of figure 4. only at a higher frequency.05).4. must for ellipsoids be replaced by 1 + L x (ε − 1) 0 or similar expressions for the y. which is absent ˚ in the bulk material because particles with a radius greater than 1000 A do not show it.3 Metals versus dielectrics Figure 4. We wish to make two points: . Q ˚ ˚ particles (a ≤ 300 A) clearly display a resonance around 2200 A. only the resonance is here sharper (γ = 0. it follows that the condition ε + 2 0. the particles must have diameters ˚ of 100 A or less.39)). similar to figure 1. We see that a surface mode cannot happen far from a resonance in the bulk material and if it does.0.or z-axis.8). The wavelength ˚ interval stretches from 1670 to 3330 A or from 6 to 3 µm−1 . The upper left box of figure 4. the relevant quantities are.2 Ellipsoids From comparison of the electric polarizability for spherical and ellipsoidal particles (see (3.2.134 Case studies of Mie calculus For the major components of interstellar dust.12 shows ε⊥ = ε1 + i ε2 and the resulting extinction efficiency for spheres of different sizes. But the position of the resonance contains further information on grain size.12. The graphite surface mode is invoked to explain the strong feature around ˚ 2200 A in the interstellar extinction curve (figure 7.13 illustrates the interplay between ε and Q ext in a surface mode for a dielectric consisting of identical harmonic oscillators. however. The form of the particle therefore influences. necessary for the appearance of a surface mode in spheres. In astronomical observations of the extinction curve. There is no further shift for still smaller particles. of course. amorphous carbon and silicate.4. The left-hand side of figure 4. it moves ˚ ˚ from 2280 to 2060 A as the radius shrinks from 300 to 30 A.9) and (3. the dielectric constant is then denoted by ε⊥ . there is no wavelength where the condition of small ε+2 is approximately fulfilled ˚ (see figure 7. not the extinction efficiencies Q ext but the extinction coefficients per cm3 of dust volume. 4. The dependence of K ext on λ−1 is qualitatively similar to that of Q ext but the ˚ discrimination against size is more difficult and not possible below 100 A. Small effect of the surface mode. K ext .

Although the grain is in the Rayleigh limit (|mx| ≤ 0. • • There is a clear offset between the peaks in ε2 and Q ext . The increase for small sizes is due to a surface mode and not to an intrinsic quality of the bulk material. a.3) vanishes when 2 3(ω2 + γ 2 ) − 1 ω2 + γ 2 = 0. not normalized. and the lower right frame the optical constant m = n + ik. the denominator in (3. a strong √ enhancement of Q abs . not small. As γ is usually much smaller than ωp .14). . the extinction efficiency is around one and.12. The upper left frame gives the extinction efficiency for graphite spheres of ˚ various radii. a surface mode. surprisingly. Here the frequency ω and the damping constant γ are in units of the plasma frequency ωp . Not every dielectric material displays a surface mode but for metallic grains the phenomenon is inevitable when damping is weak. i.e. Nevertheless. the upper right frame the volume coefficients. Q → 0 for |mx| → 0 is always correct. The efficiencies are normalized to one at 3 µm−1 .123).1). The lower left frame gives the dielectric constant ε = ε1 + ε2 . appears at ω = ωp / 3 (see figure 4. in the 2200 A wavelength region when the electric vector lies in the basal plane. With the dielectric permeability for metals given in equation (1.Surface modes in small grains 135 Figure 4.

13.1 Dielectric sphere consisting of identical harmonic oscillators The functional form of the dielectric permeability ε of an harmonic oscillator with resonance at frequency ω0 is given by equation (1.5. It is instructive to compute in Mie theory the efficiencies of a particle with such an idealized dielectric constant to see how the complex ε translates into Q abs or Q ext .05 at the frequency ω0 .136 Case studies of Mie calculus Figure 4. The peak in the absorption efficiency is due to a surface mode.77)). because Q ext attains its maximum. ω0 = 1 and γ = 0. For better visibility. the absorption efficieny Q ext can be calculated from the Rayleigh limit (3. 4.2ω0 and a plasma frequency ωp = ω0 . When the grains are large.77).5 Efficiencies of idealized dielectrics and metals 4. Q ext is basically constant over the resonance and does not seem to depend on ε at all. For small spheres. The grain radius a is determined by the condition that the size parameter equals x = ωa/c = 0.3). Figure 4.2 for a damping constant γ = 0. At intermediate sizes. The dotted curve of |ε +2|2 is not always within the box. We plotted ε in figure 1. The dielectric permeability ε of the grain material in the figure is that of an harmonic oscillator (see (1.3) has its minimum. one can discern some correspondence between Q ext and the imaginary part ε2 . When the grain is much smaller than . When the resonance frequency ω0 and the damping constant γ are measured in units of the plasma frequency.15 shows the extinction efficiency Q ext when there is a broad resonance in the bulk material (γ = 0. where the denominator |ε + 2| in (3. as in figure 1.2ω0 .2). Q ext is magnified by a factor of 10.05.

For example. the corresponding signature in the emission or absorption of the grain gets lost if the particle is very large. As figure 4.13. all charges move synchronously in the electromagnetic field.78) into the Rayleigh approximation (3. there is no conflict with Babinet’s theorem.16 shows the same Q ext as in figure 4. as already discussed in figure 4.05ωp . whereas here ε becomes smaller and smaller as the frequency rises.16 labeled big grains (radius a = 10c/ω0) demonstrates that when there is a resonance in the bulk material at some frequency ω0 .15 but over a wider frequency range and on a logarithmic scale. Scattering is then negligible and one would expect Q abs to mimic ε2 rather closely. this is not so. However. size parameter x = ωp a/c = 2πa/λ = 0.13 but for a metal with a dielectric permeability ε from the Drude profile (1. However. When we insert the low-frequency limit of the dielectric permeability ε of an harmonic oscillator after (1.123): γ = 0. The latter states that Q ext → 2 for a/λ → ∞ if the dielectric permeability is kept constant. Another way to swamp a resonance is by high optical depth.14.05.3. this will be discussed in section 8. The curve in figure 4.3) for the . One may wonder why Q ext falls and does not approach two when the particle is much bigger than the wavelength. the wavelength. Figure 4. silicate grains of a size much exceeding 10 µm will not display the famous 10 µm feature.Efficiencies of idealized dielectrics and metals 137 Figure 4.

146). . we get Q abs = 48cπ 2 4 ω0 2 ωp γ 3+ 2 ωp 2 ω0 2 a . Q abs ∝ ν 2 for ν → 0.5.77). a. the particle is big (size parameter x = 10 at frequency ω0 ). The absorption efficiency Q abs is now temperature-dependent and varies with 1/T . This is the reason why one often assumes that optically thin dust emission follows a modified Planck curve ν 2 Bν (T ) in the long wavelength limit.2 Dielectric sphere with Debye relaxation The dielectric permeability for Debye relaxation is given by formula (1. absorption efficiency.. Broken lines represent the extinction efficiency of spheres with the dielectric constant ε of a harmonic oscillator given in equation (1. where ω0 is the resonance frequency.2) The behavior at long wavelengths far away from the resonance is. the case a = c/ω0 represents medium-sized grains (x = 1 at ω0 ) and a = 0.05 at ω0 ). 4.1c/ω0 small ones (x = 0. When a = 10c/ω0 . λ2 (4.2 and ω0 /ωp = 1. here with γ /ω0 = 0.138 Case studies of Mie calculus Figure 4. The full curves represent ε = ε1 + iε2 . this line is enhanced for better visibility by a factor of 5. therefore. Q ext is plotted for three grain radii.15.

it corresponds. in some sense. The inverse of the relaxation time trel is a characteristic frequency denoted −1 by ωrel = trel . We also evaluated . 4.16. 2 αtrel a (3 + α)2 λ2 with α = 4π N p2 3kT (4.5. The efficiencies of figure 4.15. Q abs = 48cπ 2 so again Q abs ∝ ν 2 .57) the average rate W at which a small grain of volume V and polarizability αm oscillating at frequency ω in a time-variable magnetic field of amplitude H0 dissipates energy: 2 W = 1 V ω Im{αm }H0 .15 but now on a logarithmic plot and over a broader frequency range.3) In section 3. For small ω. to the resonance frequency of an harmonic oscillator.Efficiencies of idealized dielectrics and metals 139 Figure 4.3 Magnetic and electric dipole absorption of small metal spheres We derived in equation (1.2 we learnt that a piece of metal in an electromagnetic wave becomes such an oscillating magnetic dipole because circular currents are induced in its interior which are surrounded by a magnetic dipole field. The efficiency Q depends on ω/ωrel and on the grain radius in a way that is qualitatively similar to what is plotted in figure 4.

We thus have all the tools to calculate the cross section due to magnetic dipole absorption.17. λ = 62. We assume that ε has no dielectric contribution and put (see (1. We then calculate the total absorption efficiency Q tot correctly from formula (2. an expression formally identical to (1.55) of Mie theory.2 the magnetic polarizability αm of small metal spheres analytically. the wavelength follows from x. The size parameter x = 10−2 is small and kept constant.8 µm for a = 10−5 cm. In figure 4.115)) ε=i 4πσ ω where σ is the direct-current conductivity of (1. in section 3. The size .120).140 Case studies of Mie calculus Figure 4. There is also magnetic dipole scattering but it is much weaker in the Rayleigh limit and we neglect it here. which also contains electric dipole absorption. we have Q tot = Q mag + Q el .17. But the metal sphere in the electromagnetic wave is also a variable electric dipole because its mobile electrons are shifted about by the electric field. For example. The dissipation rate of the latter is given by (1. we assess the importance of magnetic dipole absorption for pure metal spheres by comparing the corresponding efficiency Q mag with the total Q tot .57). Barring quadrupole terms. Therefore.55). The ratio of magnetic dipole absorption over total absorption for metallic spheres with dielectric permeability ε = i4πσ/ω as a function of the conductivity σ . given the particle radius a.

the relevant quantity |mx| is not necessarily so. as defined in equation (3. although the particle itself is small.6) . we use.17 for a dielectric. however.5) and for the magnetic dipole absorption Q mag = 2 8π 2 ωp a 3 6 Re{b1 } = . As m = n + i k is constant in figure 4. It applies to the low frequency limit where ε i 4πσ/ω and σ is the direct current conductivity. The optical constants n and k are then large and roughly equal. Therefore we can write 32π 2 a Q mag = (4. Obviously. Because the size parameter is constant. 4. At high conductivities.4. the penetration depth of the magnetic field. We consider in figure 4.Efficiencies of idealized dielectrics and metals 141 parameter in figure 4. where the curves are flat.4. the implied frequencies ω lie between 3 × 1012 and 3 × 1014 s−1 . is large compared with the grain size and (see (3.17. we get for the electric dipole absorption (see (3. We now plot Q as a function of frequency (in units of ωp ) in the same manner as in figure 4.123) and keep the grain diameter constant.1ωp ). At low conductivities. x = 2πa/λ = 10−2 . In figure 4.17 is small and kept fixed. Q mag does not depend on radius and is proportional to the wavelength (see (3. • • A large size favours magnetic dipole dissipation. the penetration depth is small. The damping constant is the only free parameter (here γ = 0. being a metal. for ε.23).5. 3 λ the magnetic polarizability αm is taken from (3. 15γ c λ2 x2 (4.21). Q el = • 6 48π 2 cγ a Re{a1 } = 2 x2 ωp λ2 (4.26)) Q mag ∝ a 3 /λ2 • δ a (for small σ ).17 the Q’s for three radii in the likely range of interstellar grains. and in the size parameter x only the grain radius changes. An equivalent formula for the Rayleigh limit of the electric dipole absorption efficiency Q el is not valid everywhere in figure 4.28)): Q mag ∝ λ−1/2 δ a (for large σ ). where the curves in figure 4.3)). the plotted values of Q refer to a fixed frequency.17 still increase. magnetic dipole absorption increases with the conductivity and it can be the dominant absorption process. the Drude profile from (1.4 Efficiencies for Drude profiles We have already computed the efficiency Q of a metal-like sphere in figure 4.18.4) Im{αm }. • At long wavelengths.

1. When the direction .1. 4. note that Q sca Q ext holds only at low frequencies (ω/ωp 1). therefore. We.20 for idealized graphite material (see previous subsection) and a cylinder radius a = 10−5 cm. magnetic dipole absorption is important in the far infrared around 100 µm. • Both decline at low frequencies like ν 2 . Q el ∝ a Figure 4. The permeability ε(ω) is computed from the Drude profile and the particle size (a = 0.1 µm) is small enough for the Rayleigh limit to be valid for both Q mag and Q el .5. An example is plotted in figure 4.19 that in graphitic particles magnetic dipole absorption dominates at wavelengths greater than 100 µm. The right panel shows the extinction efficiencies for three particle radii.1 µm.and high-frequency behavior. which are an extreme version of elongated particles and are treated in section 10.17 suggests that for graphite grains (σ = 7 × 1015 s−1 ) of sizes ∼0.18. The grain radius equals a = c/ωp .19 the calculations of Q mag and Q el for a metal sphere with damping constant and plasma frequency like graphite but without a dielectric contribution.142 Case studies of Mie calculus Figure 4. as indicated. however. The dotted line gives the scattering efficiency for a = c/ωp .5 Elongated metallic particles The absorption by infinitely long cylinders. a. their dependence on radius is different: Q mag ∝ a 3 . The left panel shows the extinction and absorption efficiency of a metal sphere whose dielectric permeability ε follows a Drude profile with a ratio of damping constant over plasma frequency γ /ωp = 0. We learn from figure 4. They are plotted here on a logarithmic ordinate to demonstrate their low. depends on the polarization of the wave. repeat in figure 4. for the same ε.

ωp = 6. of the electric vector is perpendicular to the cylinder axis. we indeed find that Q abs is constant: Q abs = 16πσ a 3c where a is the short major axis.Efficiencies of idealized dielectrics and metals 143 Figure 4. Q is due to electric dipole absorption and can be computed from equation (3. Q ⊥ is due to magnetic dipole absorption and proportional to a 3 . For L x = 0 and ω → 0. The magnetic and electric dipole absorption efficiency of a metal sphere of radius a = 10−5 cm that has the damping constant and plasma frequency of graphite in the basal plane (γ = 5 × 1012 s−1 . the steep λ−2 -decline sets in only when λ is much larger than the big . when the electric vector is parallel to the cylinder axis. At very low frequencies. the corresponding efficiency Q ⊥ is similar to that of a sphere of the same radius. is radically different. at least for ω < ωp . in the long-wavelength limit. The permeability ε is from (1.2. The efficiency does not fall with wavelength but levels off at some high value.7 × 1014 s−1 ). e approaches one and the shape factor L x goes to zero. But if the cigars are very elongated. As we keep stretching the cigar.19.123). All real cigars are finite. The behavior of Q .40). In the long wavelength limit. To understand this behavior. we consider a spheroid with an eccentricity e close to unity so that it resembles an infinitely long cylinder. as displayed in figure 3. and their efficiency is. proportional to aλ−2 .

20. leads to a current and the ohmic losses per unit length are πa 2 σ E 2 .7 × 1014 s−1 corresponding to a wavelength λp = 2. at extremely long wavelengths. A constant field E in the cylinder. ωp = 6.8 µm). major axis of the cigar. the dotted curve presents Q abs for a sphere composed of the same metal and with the same radius. The permeability ε is from (1. The cylinder with Q abs = 16πσ a/3c absorbs per unit length from an electromagnetic wave with Poynting vector S = (c/8π)E 2 the power W = (4/3)a 2σ E 2 . .123). however. The cylinder is purely metallic with damping constant and plasma frequency like graphite (γ = 5 × 1012 s−1 .144 Case studies of Mie calculus Figure 4. outside the observable range. So the infinite cylinder is similar to a wire in which a current flows. i.e. The absorption efficiency of an infinitely long cylinder under normal incidence. For comparison.

Chapter 5 Particle statistics This chapter presents purely physical topics selected for our study of the interstellar medium. . by the following sequence: a1 a2 . a Nn . N1 N2 Nn 145 . . into n ordered cells with mean energy E i so that E1 < E2 < · · · < En .1. basic thermodynamic relations and the radiation of blackbodies. this particular configuration is represented. If the atoms are labeled from a1 to an . To this end. for example. We outline the derivation of this fundamental statistical relation and append a few loosely connected items. . Each cell has a size E i which determines its statistical weight gi = E i /E max . . a N1 +N2 . so gi = 1 i Let each level E i be populated by Ni atoms. a N−Nn +1 . The statistical weight gives the fractional size of a cell. 5. a N1 a N1 +1 .1 Boltzmann statistics For a large number of atoms in thermodynamic equilibrium.1 The probability of an arbitrary energy distribution We first compute the number of ways in which the energy of N identical particles. for example atoms. we divide the energy space. their distribution in energy space is just a function of temperature and described by the Boltzmann equation. They concern the statistical distribution of atoms and photons. . which ranges from zero to some maximum E max . 5. can be distributed and then find among them the most likely distribution. . . .

. . of course. According to elementary probability theory. Because N ln N is constant. which equals one.2) n ln ω N ln N + i=1 Ni ln gi . this sum is (g1 + g2 + · · · + gn ) N . (5. each cell has the same chance that a particular atom will fall into it. 5. Summing the probabilities ω of all possible distributions gives the total probability. When the N atoms are arbitrarily distributed over the available cells.1. . ω= (5. Ni In the equilibrium distribution achieved by nature. . When all cells are of the same size. In view of the binomial theorem. . The probability ω for the particular configuration of the N atoms is. Ni . is called the thermodynamic probability and is a very large number. . N! N N N · g 1 · g2 2 · . But there are many others ways in which the same energy distribution can be realized. · gn n . To handle them. the probability has the maximum possible value.2) becomes ω= N! nN Ni ! . therefore. . this yields for (5.146 Particle statistics The Ni under the horizontal brackets give the number of atoms in the energy bins E i . as it should. Nn ! (5. Then gi = 1/n for all i and the probability ω in equation (5. . one uses Stirling’s formula: √ ϑN (0 ≤ ϑ N ≤ 1). N2 .2) N1 !N2 ! . Here it is assumed that the particles are distinguishable.2) is at the heart of classical statistics.2 The distribution of maximum probability As the numbers Ni in the expression for the probability ω are large. we are looking for an extremum of the function n f (N1 . the following N2 atoms into cell E 2 . Retaining only the first term.1) ways. their faculties are much larger still. . · gn n . . the probability that the first N1 atoms fall into cell E 1 . Nn ) = i=1 Ni · ln gi . smaller than one. ω is.3) ln N! N ln N − N + ln 2π N + N The accuracy of this formula is quite remarkable (see the error estimate ϑN ). and so forth equals N N N g1 1 · g2 2 · . . a configuration where N particles are grouped as shown here can be achieved in = N! N1 !N2 ! . . Nn ! 1 Equation (5.

So for every i . can be found using the method of Lagrangian multiplicators.Boltzmann statistics There are two additional and obvious requirements: • The total number of particles N should be constant and equal to n 147 Ni . in (5. Exploiting Ni = N eliminates α and gives the fractional population Ni = N j gi e−β Ei g j e−β E j . Z= j g j e−β E j equations (5. When we define.5) yield ln = − Ni =− Ni ln gi − β E i − ln Z N = β E + N ln Z − Ni ln gi . (5. Ni = gi eα−1 e−β Ei .3) and (5. (5. In this sum. we get the equation for the differentials d f + α dφ + β dψ = 0 leading to d Ni ln gi − 1 + α − β Ei Ni = 0. (5.1. all brackets must vanish. φ= i=1 Ni − N = 0.5).3 Partition function and population of energy cells We have found Ni /N but we still have to determine β. Ni ln . subject to the conditions expressed by the auxiliary functions φ and ψ.4) The maximum of f . If we denote them by α and β. • The mean particle energy E i is related to the total energy E of all particles through E = N · Ei therefore ψ = N Ei − i=1 n Ni E i = 0.1).5) 5.

6) The expression Z introduced above now becomes Z (T ) = j g j e−E j / kT . T This allows us to express the multiplicator β in terms of the temperature. important deviations from the mean never occur.51)). therefore. The entropy S of the system equals the Boltzmann constant multiplied by the logarithm of the thermodynamic probability of (5.7) This is called the partition function.148 Particle statistics We now plug in some thermodynamics from section 5. ∂S ∂ ln 1 = =k T ∂E V ∂E it follows that ∂β 1 = kβ + k E − kN · T ∂E gi E i e−β Ei ∂β . · ∂E g j e−β E j The last two terms on the right-hand side cancel and thus β= 1 . The fractional abundance of atoms with energy E i is. S = k ln . . practically. Nj g j −(E j −Ei )/kT = e . which depends only on temperature. (5. they are extremely precise and. and the differential of the entropy is (see (5. Ni = N gi e−Ei / kT gi e−Ei / kT = Z g j e−E j / kT (5.8) and the abundance ratio of two levels i and j . Because at constant volume V .1).9) Although the population numbers are statistical averages. kT (5. dS = d E + P dV . Ni gi (5.3.

the average is E = Ni E i = N Ei Ni = N Ei ω ). (see 2 (6. ω The last sum is an infinite series where each term is a factor e−β than the preceding one. we get the classical result that the sum of At high temperatures ( ω potential plus kinetic energy of an oscillator equals kT . At low temperatures ( ω kT ). The energy levels are nondegenerate (gi = 1) and equidistant. E = kT. . .8).25)). 2 5. (5.26) and (5.11) kT ). 1.5 The Maxwellian velocity distribution The probability that an arbitrary gas atom has a velocity in a certain range follows as a corollary from the equilibrium energy distribution (5. the statistical weight gi denotes then the degeneracy of the level. If Ni oscillators have an energy E i . We substitute in this . We compute the partition function of a system of N quantized identical harmonic oscillators in equilibrium at temperature T . equations (5.9) are also valid in quantum mechanics (more general forms are given by (5.1. the average E is close to the zero point energy. E = ( 1 + e− ω/ kT ) ω.7)–(5. therefore. 2. For atoms with internal energy levels E i . Z Z ∂β ∂β Because ln Z = − 1 β ω − ln(1 − e−β 2 E = it follows that ω ω/ kT 1 ω+ 2 e −1 . 1 ∂Z ∂ ln Z e−β Ei =− =− . Z (T ) = 1 − e−β ω 1 smaller (5.Boltzmann statistics 5. E i = (i + 1 ) ω for i = 0.36)). . ∞ Z (T ) = i=0 e−β Ei = e− 2 β 1 ω ∞ i=0 e−iβ ω .1. e− 2 β ω .4 The mean energy of harmonic oscillators 149 With some reinterpretation. With β −1 = kT .10) We can also easily compute the mean oscillator energy E .

13) N(v) dv = N. Changing from momenta p = mv to velocities gives the familiar Maxwellian velocity distribution (figure 5. it is the relative not the absolute value of the statistical weight g that matters. In any arbitrary direction x. we obtain the momentum distribution. When we replace the sum in the partition function Z (T ) of (5. therefore.13)) because near zero velocity.150 Particle statistics formula for E i the kinetic energy of an atom. p2 /2m.or z-direction. v + dv equals N(v) dv = N4πv 2 with 0 ∞ m 2πkT 3/2 e−mv 2 /2kT dv (5.8). the number of atoms with velocities in the interval [vx . All directions of motion are equally likely and the square of the length of the momentum vector is 2 p 2 = p 2 + p 2 + pz .1) which states that if there are N gas atoms altogether. The total number of atoms. 2 (5. (5. the sum over all g need not be one.12) In the ratio Ni /N of (5.14) N(vx ) is symmetric about vx = 0 and has its maximum there (contrary to N(v) in (5. N(vx ) dvx = N m 2πkT 1/2 e−mvx /2kT dvx . vp .14) by integrating from −∞ to ∞. follows from (5.27)–(A.29) in appendix A associated with the bell-curve. v = 8kT . we take g = 4π p2 d p. x y For the size of an energy cell.16) . i. . is the one at the maximum of the curve N(v). vp = • 2kT . except for a normalizing constant.. N. • The most probable velocity. we exploit the integrals (A. m (5.e. There are three kinds of averages of the absolute velocity v. the number of those whose absolute velocities are in the range v . πm (5.7) by an integral. . for the statistical weight. by the 2 Boltzmann factor e−mvx /2kT . N(vx ) includes all particles with any velocity in the y.15) The mean velocity v is relevant for calculating collision rates. To compute them. vx + dvx ] is given.

17).18) 5.17) v2 = m The averages are ordered (see also figure 5. which consists of coordinates and momenta. (5.15)–(5. one divides the phase space. The ordinate gives the probability density N(v) after (5. In quantum statistics.1). (5. into cells. vp < v < v2 . • The mean energy of the atoms is given by the average of v 2 such that 1 3 2 2 m v = 2 kT or 3kT . The real coordinate space was not considered.2. The vertical dotted lines show from left to right the means vp . v and v 2 according to (5.13) with respect to one atom (N = 1).Quantum statistics 151 Figure 5. The Maxwellian velocity distribution for hydrogen atoms at 100 K. we divided the energy or momentum space into arbitrarily small cells. If V is the volume and p the momentum.1.2 Quantum statistics 5. we form cells of size d = V · 4π p2 d p .1 The unit cell h3 of the phase space In deriving the energy distribution for an ensemble of particles after Boltzmann.

x n ) denote the wavefunction of a system of n particles. particles in a box of volume V can only have discrete momenta.2. It states how many unit cells. . This means that when we swap their positions in phase space.31)) is called the statistical weight (see also (5. nothing changes physically. We may. position q and momentum p of a particle can only be determined to an accuracy p and q such that p· q ² . Let particle i have the coordinate x i . This relation results from the wave character possessed by all particles. The quantity g (named d Z in (8. The wavelength of a particle is related to its momentum (after de Broglie) by p = h/λ. then one speaks of a symmetric wavefunction.31) of that section and find that the possible number of eigenfrequencies within the momentum interval p to p + d p in a box of volume V is V · 4π p 2 d p .19) h3 The eigenfrequencies are the only allowed states in which a particle can exist and correspond to standing waves. each of size h 3 . which we treat in section 8. for example. . The total energy of the system is then n E= i=1 Ei . Therefore. Let (x 1 . not just photons. exploit equation (8. also have two (not three) spin directions. (5. are otherwise identical. besides place and momentum. fit into the phase space element d . the number of quantum states of an electron (spin 1 ) is twice the value given in (5. two photons or two electrons. x 2 . . In quantum mechanics. particles are represented by their wavefunctions.152 Particle statistics where 4π p2d p is the volume of a shell in momentum space with radius p and thickness d p. the spin direction. either itself is not altered. The same is true for 2 photons (spin 1) which.2 Bosons and fermions Particles are described by their position in phase space.19). then the wavefunction is called anti-symmetric. . The situation is formally almost identical to standing waves in a crystal. otherwise they would be destroyed by interference. Because of the wave character. According to the uncertainty principle. therefore.12)). . Any two particles of the same kind. one needs. Note that classically the number of possible (physically different) states is unlimited. So after they have been swapped. the wavefunction ψi and the energy E i . Their identity then implies that | |2 stays the same when any two of the particles are interchanged. g= 5. traveling at the speed of light. which would correspond to a unit cell of size zero. the particles are indistinguishable. To determine the state of a particle with non-zero spin. all of one kind and not interacting.4. or is transferred into − .

x 2 .. . (5. however.. We. like photons.22) Fermi (x 1 . if it is anti-symmetric. are not subject to Pauli’s exclusion principle. . ψ1 (x 2 )ψ2 (x 1 ) · . Bosons.. that two fermions cannot have the same wavefunction. . ψn (x n ) As indicated by the subscripts on . the wavefunction of the whole system can be factorized into the N wavefunctions ψi of the individual particles. ψn (x 1 )  ψ1 (x 2 ) ψ2 (x 2 ) . ψn (x 2 )   . . . protons or neutrons are fermions. x ) = x 2 x are g(x 2 . x n . therefore. is symmetric and yields the correct energy E. like electrons... x n ) = det  . Here P is an operator that performs a permutation of the coordinates x 1 . they are linearly dependent and the determinant in (5. if the wavefunction is symmetric. . • It can readily be checked that the linear combination Bose (x 1 . while particles with integer spin. ψi = ψ j . x n ) as a sum of such terms but with interchanged arguments. So f (x 1 . . This has the fundamental consequence.. x n ) (5.22) vanishes. . . x 2 ) = x 1 x 2 and 2 x are identical functions but f (x .. the sum extends over all N! possible permutations. . It is an experimental fact and is also claimed theoretically that particles with half-integer spin. · ψn (x n ) and so forth.20) = P ˆ Pψ1 (x 1 )ψ2 (x 2 ) · . the only anti-symmetric wavefunction of the whole system is the linear combination formed by the determinant   ψ1 (x 1 ) ψ2 (x 1 ) . they cannot occupy the same quantum state or unit cell. x ) and h(x . 2 what counts is the sequence in which they appear. . For example. x 2 . Note that in a multi-dimensional function it does not matter by which symbols the arguments are presented. x 2 . the particles are called bosons. All coefficients before the products ψ1 ψ1 . . the first term of the sum is ψ1 (x 1 )ψ2 (x 2 ) · .Quantum statistics 153 Because the particles are non-interacting. . When two rows or columns of the matrix (5. are bosons. . . . .22) are equal. i. known as Pauli’s exclusion principle. . . Likewise. ψn equal one (we neglect normalization). where particles 1 and 2 have been exchanged. . .. .e. · ψn (x n ) and the next. . . write (x 1 . . fermions. . ψ1 (x n ) ψ2 (x n ) . . this corresponds to swapping particles. x 2 . . . .21) • ˆ and no other. Each such product is a wavefunction that gives the same total energy E but fails to fulfil the symmetry condition. . · ψn (x n ) (5. x 1 ) = x 1 2 1 2 1 2 1 2 different. . . .

for instance.23) h3 h3 unit cells. . . . .2. N2 . Ni . . one particular distribution is given.1 Counting states Let us assume N identical particles of mass m in a box of volume V and with a total energy E. . when considering the total phase space. g2 .3 Bose statistics 5. If we write behind each unit cell the atoms which it contains. E2 . . .. i..154 Particle statistics 5. First. the atoms a1 . .. We ask in how many ways can one distribute its Ni particles over its gi unit cells (quantum states). E1. . . . there is no limit as to how many atoms can be squeezed into one unit cell. Let us label the unit cells within d by Z 1 . . As permuting the atoms or unit cells does not result in a new quantum state of the subsystem d . . In this sequence. and so forth. a5 are in Z 3 . . gi . through the three sets of values gi = N1 .e. A particular distribution of the system is completely determined by specifying for each cell d its population Ni .. We divide the phase into cells of size d = V 4π pi2 d p. like the one here. . a Ni . energy E i and statistical weight gi (size). . the atoms a4 . Z 2 . a2 .3. say cell i of size d . . Each cell contains Ni particles of energy Ei = and comprises V 4π pi2 d p d = (5. Therefore the total number of mathematically different sequences is gi (gi + Ni − 1)!. . by the sequence Z 1 a1 a2 a3 Z 2 Z 3 a4 a5 Z 4 a6 Z 5 a7 . start with some Z .24) pi2 2m The probability of finding this particular distribution is proportional to the number of ways W in which it can be realized. cell Z 2 is empty. Its Ni atoms constitute a subsystem. (5. . . we consider only one cell. All such sequences of symbols Z l and ak ... because we always first specify the cell and then the atoms it contains. .25) = Ni !(gi − 1)! i . a3 are in Z 1 .2. In boson statistics. . a2 . there are (gi + Ni − 1)! (5. only (gi + Ni − 1)! gi (gi + Ni − 1)! = Ni !gi ! Ni !(gi − 1)! of the sequences are physically different. g1 . Therefore.. Ei . Z gi and the atoms by a1 .

as otherwise the denominator under the integral might vanish. We only mention that 1.27) = g e −1 the occupation number. according to this formula. this leads to ln gi + Ni = α + β Ei . Following the procedure outlined in section 5. Ni 5. we get dn = g dp V 4π p2 = α+β E . We still have to determine the coefficients α and β.6)) Æ β= 1 .28) becomes p2 e−α e−E/ kT . kT p2 d p = N V. ln = i (gi + Ni − 1) ln(gi + Ni − 1) − (gi − 1) ln(gi − 1) − Ni ln Ni and exploits the conditions for the conservation of energy and number of particles.Quantum statistics 155 ways in which the configuration of (5.26) As the g unit cells are. the α is related to the chemical potential µ by α = −µ/kT .24) can be achieved. in the last formula. write dn instead of Ni (dn is the number of particles in the phase space element V 4π p2 d p).2. e −1 h 3 eα+β E − 1 (5. populated by dn particles. which is what nature chooses. from dn = 4π V h3 eα+β E (5. one calls 1 dn = α+β E (5. When eα+β E integrand in (5. which yields e−α = h3 N V (2πmkT )3/2 (5. is found in the same manner as the maximum of the probability ω in (5. E− i Ni E i = 0 N− i Ni = 0.3. One forms the logarithm using the first two terms of the Stirling formula (5.29) . so e−α lies between zero and one. which is the number of particles per unit cell. i. we drop the subscript i .28) The coefficient α is always positive.23) for the statistical weight.e.2 The occupation number in Bose statistics When.1. and insert (5.3). The maximum of .2) for Boltzmann statistics. −1 α follows from the requirement that the number of particles is conserved. Using the same thermodynamic arguments as for Boltzmann statistics gives (see (5.

which have a momentum p = hν/c.4 Bose statistics for photons For photons. however. dn = N 4π p 2 d p 2 e− p /2mkT . may be filled by an arbitrary number of atoms.27). (5.13). Because there . 1 Æequil = ehν/ kT − 1 . there is no such constraint as N− i Ni = 0 the parameter α in (5. the average is fixed and follows from integrating u ν / hν over frequency. their number is not conserved.17) for an evaluation of the integral).31) Formula (5.32) The occupation number equals dn/g and. (2πmkT )3/2 5. in which photons are created and destroyed.26) is zero and dn = g ehν/ kT − 1 = V 8πν 2 dν . A unit volume of gas in thermodynamic equilibrium. the statistical weight is g=2 V 4πν 2 dν . As radiative equilibrium is established through emission and absorption processes.2.31) describes the distribution of photons in equilibrium at temperature T —the so called blackbody radiation. Consequently. therefore. Although we did not explicitly prescribe the total number of photons within a unit volume. The occupation number of photons in equilibrium is. n phot = 0 ∞ uν dν = 8π hν kT hc 3 0 ∞ ex x2 dx −1 (see (A. uν = ν3 8πh .156 Particle statistics and one recovers the Maxwell distribution (5. n phot. c3 (5.30) The factor 2 accounts for the two possible modes of circular polarization. from (5. c3 ehν/ kT − 1 The radiative energy density u ν now follows from u ν dν = hν dn/V . also has a value when the photons are out of equilibrium and no temperature is defined. c3 ehν/ kT − 1 (5.

there may be two particles of opposite spin per unit cell because their quantum states differ. h3 (5.4 and rediscover the basic equation (5. obey Pauli’s exclusion principle.2. The essential difference from Bose statistics is that there are now gi ! (gi − Ni )!Ni ! ways to distribute Ni electrons over the gi unit cells. the counterpart to (5. Now there can only be one particle per unit cell h 3 with a given spin direction. each harbouring Ni electrons. We again divide the phase space into cells of size d = V 4π p2 d p. A unit cell is either empty or filled with one electron.5 Fermi statistics Fermi statistics deals with particles that have an anti-symmetric wavefunction and which. (5.27)) Æ = dn = eα+β1 + 1 .26) for the number of particles dn in the phase space element V 4π p2 d p which contains g unit cells. We will return to blackbody radiation in section 5.36) and for the occupation number (cf.34) However. as before β = 1/kT and the parameter α follows from the condition that the total number of particles . with the same procedure as before.33) uν = c3 Likewise. like electrons.37) In the denominator. therefore. Of course. may define the intensity of radiation. We consider a cell d consisting of gi unit cells. (5. thus in all = i gi ! (gi − Ni )!Ni ! (5.31) from less abstract and less strict arguments. which have a total energy g hν. the occupation number may be used to specify the radiative energy density at a particular frequency.35) different possibilities. In counting states. and not −1. 8πhν 3 . Æ Æ Æ Æ 5. it is now +1.Quantum statistics 157 are dn = g photons in the phase cell d . so the statistical weight is gi = V · 4π pi2 d p . This gives. dn = g eα+β E + 1 (5. E g (5. we repeat the steps which we carried out for Bose statistics.

there are ionizing and recombining processes symbolized by Xr ª Xr+1 + e. The Ns of two neighboring ionization stages are related in thermodynamic equilibrium by the Saha formula (2πm e kT )3/2 g1. the quantum mechanical probabilities (5. Æ class = N! i gi i .29)). dn = N V.r are in their ground state. 5. particles are distiguishable and the probability is.r+1 −χ/ kT N1.35).r is the statistical weight of the ground state of ionization stage r. and χ > 0 is the energy difference between the ground state of ion Xr and Xr+1 .2).6 Ionization equilibrium and the Saha equation In a hot gas. In Fermi statistics.2. Degeneracy is absent for large positive α.38) suggests that the Saha equation can be derived from the .25) or after Fermi as in (5.28)). In classical statistics. and if eα+β E physics. The quantum mechanical way of counting is the right way and takes into account the fact that the phase space is partitioned into unit cells h 3 .r+1 Ne =2 e . may be positive as well as negative (from −∞ to +∞). But now α. N1.r g1.4.35) approach class when the Planck constant h goes to zero and the gi become very large (gi Ni ). Denoting the corresponding ion by Xr . particles are indistinguishable and states are counted either after Bose as in (5.r h3 (5. The very form of (5. in Bose statistics.158 Particle statistics comes out correctly (cf. when the gas is hot and rarefied (see (5. from (5. Ni ! N In accordance with the correspondence principle. however. (5. which is also called the degeneracy parameter. there is not more than one particle (of the same spin) per unit cell. changes When the temperature T goes to zero.25) and (5. of which N1. Suppose that in 1 cm3 there are Ne electrons and Nr particles of species Xr . all particles are in their lowest (zero momentum) state at T = 0. we are in the realm of classical µ is the chemical potential. In quantum statistics.38) g1. the atoms of any element X exist in different stages r of ionization.2). the occupation number discontinuously from one for energies below the Fermi limit E F to zero for E > E F (subsection 6. As in Bose statistics: α = −µ/kT where 1.

so V = 1/Ne . The factor 2 comes from the two electron spin directions.9) N( p) d p g( p) χ + p2 /2m e = exp − N1. −E 1 = χ > 0. altogether Np Ne (2πm e kT )3/2 −χ/ kT = e . p + d p]. we restrict the discussion to pure hydrogen gas consisting of electrons. Let us consider • • the atom with the electron bound in level n=1 as the ground state the proton together with a free electron of momentum p as the excited state. The statistical weight of the neutral atom with the electron in level n is equal to the maximum number of electrons in that level.H h3 . the energy is zero. For simplicity.H .Quantum statistics 159 Boltzmann equation. one gets Np gp (2πm e kT )3/2 −χ/ kT =2 e V .34). Let N( p) d p be the number density of excited particles (one such particle consists of a proton and an electron) where the electron has momentum in the range [ p. Thus in the ground state. which always pertains to one proton. protons and neutral hydrogen atoms.H/gp = 2.H = 2n 2 and g1. must contain exactly one electron. g( p) = gp · ge .H g1. E n is negative. Let g( p) be the statistical weight of the excited state. so gn.39) N1. then according to (5. The energies of the excited states form a continuum because the electrons may have any velocity. N1. we note that the phase space cell of a free electron. (5.H kT . After (5. Because ∞ 0 N( p) d p = Ne = Np . Energies are counted in the following way: for the hydrogen atom in level n. and for the proton–free-electron system it is positive and equals p2 /2m e . If N1.H h3 g1. ge = 2 · 4π p2 d p V / h 3 .H To determine the volume V .H are the corresponding numbers for the hydrogen atoms in the ground state. where χ is the ionization potential of hydrogen. g1. At the ionization threshold. The statistical weight g( p) is the number of quantum states of the proton–free– electron particle and thus the product of the statistical weight of the proton and the electron.

For hydrogen.H is equal to the number NH of all hydrogen atoms and Np Ne (2πm e kT )3/2 −χ/ kT = e . Some of the thermodynamic elements presented here are also indispensable in the discussion of phase transition during the formation and evaporation of grains (see section 9.1) we have to recall the definition of temperature. q f and the . h3n2 Nn = e−En / kT . Atoms in high quantum number n are big. . An atom is then closely surrounded by ions and the overlap of their electric fields disturbs the higher quantum states and lowers the ionization energy.160 Particle statistics When the partition function of the neutral hydrogen atom Z = gi e−Ei / kT is −E 1 / kT . For the computation of the population of energy levels in the smallest grains (see section 12. entropy and number of accessible states. A system of f degrees of freedom is described by the spatial coordinates q1 . . the path of the atoms in phase space contains the full mechanical information. . The Saha equation goes wrong when the mean distance d between ions becomes comparable to the size 2r of an atom. instead of energy space one considers the phase space of the system. .3. 5. their radius is r = a0 n 2 where a0 is the Bohr radius. This happens in stellar interiors where the density is high. The formulae around the specific heat find their application mainly in the theory of magnetic dissipation in section 11. although LTE (local thermodynamic equilibrium) certainly prevails.5). which is often possible and implies that approximated by its first term g1 e almost all atoms reside in the ground state. NH h3 We can also immediately calculate the number Nn of atoms in level n under conditions of thermodynamic equilibrium.1 The ergodic hypothesis Boltzmann’s equilibrium population of the energy levels of a system was derived by determining the energetically most likely distribution. T ∼ 15 × 106 K). although the Saha equation predicts that about half of all atoms are neutral.3 Thermodynamics This section summarizes the results of thermal and statistical physics. In the central region of the Sun (ρ ∼ 102 g cm−3 . 5. then N1. More generally.2.40) Note that −E n is positive and for high levels very small. Ne Np (2πm e kT )3/2 (5. d is less than the Bohr radius and even the ground state of hydrogen cannot exist.

it is spread out in one-dimensional phase space over a region x p = E/ f ω where it thus occupies E/ f ω cells of size . (5. we make a rough estimate.e. . . q f . If there are many degrees of freedom (f 1). i. Each oscillator has on average an energy E/ f . p f . p f (t)). As the system evolves in time t. .26). Let (E) be the total number of accessible states (cells). y. . The number of states. the probability to find the system in a certain accessible cell Z r is. At any instant. f = 3N. therefore. E + δ E]. . because in quantum mechanics. . it is in a narrow range from E to E + δ E. E cannot be defined with absolute precision in a finite time. E + δ E] becomes d Ns δ E. E f ω f . Usually one needs three coordinates (x. . so if there are N particles. The total energy E is. If one divides the phase space into small cells Z of equal size. with energies from [E. . In equilibrium. independent of time. it describes a trajectory parametrized by (q1 (t). rather. the system may be represented by a point in the 2 f dimensional phase space with the coordinates q1 . . therefore. Ns = Hence the number of cells. Their energy levels are equally spaced in multiples of ω. As a ρ(E) = . After equation (6. p1 . dE (E) = Ns (E + δ E) − Ns (E) We denote by d Ns dE the density of states around energy E. of the total system whose energy does not exceed E is. each corresponding to one degree of freedom. the function (E) increases extremely rapidly with the total energy E of the system. Let our system consist of identical quantum oscillators. • The fundamental (ergodic) postulate asserts that when the system is in equilibrium. contrary to (E) which obviously increases with δ E. It does not make sense to consider cells smaller than because one cannot locate a particle more accurately. . . by definition. . none has any preference. . It is independent of the width of the chosen energy interval δ E. p f . .41) (E). z) for one particle. To see how fast (E) rises.Thermodynamics 161 conjugate momenta p1 . All cells (states) Z i that correspond to an energy between E and E + δ E are called accessible. . Let the system be isolated so that it cannot exchange energy with its surroundings. those with energy in the interval [E. constant or. all accessible states are equally likely. The cells may be enumerated. the microstate of the system at a particular moment is determined by the cell in which it is found. Ns . .

where refer to A and A . If one counts the states more properly but neglects powers of E with exponents f − 2 or smaller. be equally low. by itself. one finds E f −1 ρ(E) . There is no contradiction because there is just one cell corresponding to all atoms being clustered in one corner but a multitude corresponding to a very smooth distribution of atoms over the available box volume. (E) depends on the width of the energy interval δ E but because is so large.2 Definition of entropy and temperature In statistical physics. so the joint energy E tot = E + E is constant. P(E) ∝ (E) · (E tot − E). is given by the product (E) (E tot − E). A + A . the starting point is the number of accessible states with energies from [E. S = k ln . the density of states becomes proportional to n f −1 /( f − 1)!. (E) (5. Note that ρ and rise with E f −1 . like the gas content in an empty bottle. The probability P(E) of finding the and whole system in a state where A has the energy E is evidently proportional to this product. The probability of a particular microstate which actually is encountered and where the atoms are very evenly distributed must. The total system. respectively.43).44) Let us see what these two definitions mean for a bottle of warm lemonade (system A) in an icebox (system A ). one gets ρ(E) f E f −1 /( f ω) f . or 1024. otherwise E and E are arbitrary. (5. This. The probability that all gas atoms in a box are huddled in the upper left-hand corner and the rest of the box is void is extremely low.42) ( f − 1)!( ω) f If the total energy E of the system is n times greater than the natural energy unit ω. After the ergodic principle. The number of accessible states of the whole system. k ∂E ∂E kT (5.3. one can easily show that δ E is irrelevant in (5. .162 Particle statistics first approximation. all accessible cells in a phase space have an equal chance of being populated.43) Of course. The two systems are in thermal contact. subject to the condition that subsystem A has an energy E. The absolute temperature T is then defined by ∂ ln ρ 1 1 ∂S = = . one will never encounter such a configuration. is a huge number but because f appears as an exponent. it determines the entropy. heat can flow between them but the external parameters (like volume) are fixed. 5. the number of degrees of freedom of the atoms is ∼ 1024. therefore. is so large that even its logarithm ln is of order f . i. When we evaluate for something from everyday life. E + δ E]. is isolated (we do not have an electric icebox).e.

only the joint energy E tot of the combined system A + A . The probabity Pr for A being in state r is proportional to the number of states of the total system A + A under that condition. The probability P(E) to find system A in the energy range δ E around E becomes P(E) = (E) e−β Er e−β Es s (5.45) The sum in the denominator represents the partition function Z (T ). as A is fixed to state r . The heat bath has then the energy E = E tot − Er and (E tot − Er ) accessible states. 5. −β E s se (5. or the entropy S + S . Because E tot Er .46) . starting from some arbitrary microstate. the dependence of the number of accessible states on E. the value where it is not zero determines equilibrium. The constant C follows from the condition s Ps = 1 in which the sum includes all possible states s of A. To find system A at an energy E = E eq is totally unlikely. The energy of A is not fixed.3. the lemonade is cool. Because of the extreme When (E) rises. of the whole system has its maximum follows from ∂ ln P = 0. Note that nothing has been said about how long it takes. The probability P(E) is almost a δ-function. and obtains the canonical Pr = e−β Er . to arrive at equilibrium. ln where β = ∂ ln distribution /∂ E (E tot − Er ) = ln (E tot ) − β Er = (kT )−1 (see (5.44)). Pr = C (E tot − Er ). irrespective of their energy. therefore. on E . or of probability P(E) must have a very sharp maximum at some value E eq .Thermodynamics 163 (E tot − E) must fall. one can develop ln around E tot.3 The canonical distribution Let a system A be in thermal contact with a much larger heat bath A of temperature T . The energy E where P. ∂E This equation immediately gives ∂S ∂E = E=E eq ∂S ∂E E =E eq which means that in equilibrium the temperature of the icebox and lemonade are equal. Suppose now that A is in a definite state r of energy Er .

we then symbolically write Y = y. besides having an energy between E and E + δ E. and we denote their total number by (E.49).46) is in agreement with the Boltzmann distribution (5.9). y). For example.9): As the size of system A has not entered the derivation of P(E). the expression is also correct when A is a microscopic system. If A consists of just one atom with discrete energy levels E j . U = U (T.48) P(y) = ρ(E) 5. or lie in the range [y.47) Likewise the probability P can be expressed by the ratio of the densities of states.50) . the probability P of finding the system in the desired configuration Y = y is P(y) = (E. some parameter Y may have to take up a certain precise value y. y + δy]. (5. 5. there is only one state for which E = E j . The parameter Y may be anything. ρ(E. for example.1 Constraints on the system Sometimes a macroscopic system A may have to fulfil additional constraints.4 Thermodynamic relations for a gas When one transfers to a system a small amount of heat δ Q and does the work δ A on the system. according to the first law of thermodynamics. the energy of ten selected atoms or the integrated magnetic moment of all particles.3.49) ∂U ∂T dT + V ∂U ∂V dV T ∂U ∂V + p d V. δQ = ∂U ∂T dT + V (5.3. (E) (5. y) . y) . then. In view of the ergodic postulate. so (E) = 1 and one recovers the Boltzmann distribution (5. The states for which Y = y form a subclass of all accessible states. δ A is positive. Thermal contact then just means that A can exchange energy with A . its internal energy U is altered by the amount dU = δ Q + δ A. So when V is decreased. If the infinitesimal work is due to a change of volume d V .164 Particle statistics where (E) is the corresponding number of states accessible to A. δ A = − p d V. T (5. V ) and dU = and because of equation (5. If U depends on temperature and volume.3. Formula (5.

54) When one presses gas through a small hole from one vessel to another. Because d S is a full differential. i.56) defines the chemical potential. In a homogeneous phase. αV. With some elementary calculus manipulations one obtains the full differential dG = where µj = ∂G ∂G ∂G dT + dp + d N j = −S d T + V d p + ∂T ∂p ∂ Nj ∂G ∂ Nj µj dNj j (5. T T (5. its entropy increases by dS = dU − δ A δQ = . V. zero.Thermodynamics 165 The second law of thermodynamics brings in the entropy S. we note that if. (5.53) • To connect a physical meaning to F. α N j ) = αS(U. therefore. V. it is the enthalpy that stays constant.51) The entropy is a state function and the integral δ Q/T in the ( p. • The free energy or Helmholtz potential F = U − T S. V (5. G = G(T. in an isothermal reversible process. by compressing a gas. ∂2S ∂2S = ∂T ∂V ∂V ∂T we get from (5.52) Other common thermodynamic state functions besides U and S are as follows. The enthalpy H = U + pV. N j ). one applies the work δ A on a system.e. ∂U ∂V =T T ∂p ∂T − p. The free enthalpy or Gibbs potential G = U + pV − T S.55) • As convenient variables one usually uses for G temperature.51) with δ A = − p d V . When the heat δ Q is added reversibly to a system at temperature T . (5. (5. for example. the free energy of the system increases by F = δ A. as in the famous Joule–Thomson experiment. N j ) . in which for any α S(αU. V )-plane along any closed path is. pressure and number N j of atoms of the j th gas component.

j = 1) is G = G fl + G gas = µfl Nfl + µgas Ngas . the free enthalpy of a system consisting of a one-component fluid and its vapor (only one kind of molecules.58) (5. the potential φ is at its minimum and the differential φ = 0. Note that all variables without an index refer to system A.49)). The forces derive from a potential φ. System A receives from the environment A the heat δ Q = −T SA which increases its internal energy by U and does the work δ A on the system (see (5. F has then its minimum because Stot of the total system A + A is at its maximum.3. Therefore. In equilibrium. Therefore. In the case of pure compressional work. the differential SA is zero under small variations of the variables (U. we may drop the index A in the entropy and get U − T S − δ A = 0. V.e. the entropy SA attains its maximum. all forces Fi acting on the particles must vanish. F= (U − T S) = 0. δ A = − p V and U −T S + p V = 0. δ Q = T SA = U − δ A. If A is not isolated but in contact with a heat bath A .166 Particle statistics one finds (without proof but it is not difficult) G= j µj Nj. For a mechanical system. . As all variables here refer to system A. so Fi = −∂φ/∂ x i . N j ).57) Equation (5.58) leads in a straightforward way to the following equilibrium conditions: (1) In a system A of constant temperature and volume ( T = V = 0) and in contact with a heat reservoir A . the equilibrium condition for the total system A + A reads: Stot = (SA + SA ) = 0.5 Equilibrium conditions of the state functions Let us recall the conditions under which a system is in equilibrium. (5. i. 5. For an isolated thermodynamic system A of constant energy U and volume V . the differential of the free energy vanishes under small changes of the variables.

With F = F(y) − F(y ).9): Suppose the N particles of the system possess only two levels. As Y varies from y to y (this need not be a small step). the condition F = U − T S = 0. putting G = G(y) − G(y ). (2) At constant temperature and pressure ( T = p = 0). . there is no compressional work and δ A = 0. P(y ) This result follows also from the canonical distribution (5. almost-equilibrium state b. the entropy of the total system changes by T S − U + δA Stot = T where δ A is the (not necessarily infinitesimally small) work done on system A and S = S(y) − S(y ) and U = U (y) − U (y ). P(y) = e− F/ kT . The thermodynamic probability for the two states states is given by (5. the differential of the free enthalpy G = (U + pV − T S) = 0 and G has its minimum. P(y ) ∝ tot(y ) = e Stot(y )/ k for another value y of the parameter Y .59) (3) At constant pressure ( p = 0) and constant entropy ( S = 0. the lower level 1 is populated by N1 and the upper level 2 by N2 atoms with N1 +N2 = N. H = (U + pV ) = 0 and the enthalpy has its minimum. as in (5. the corresponding populations are N1 +1 and N2 −1. Let system A depend on some parameter Y . for the population ratio of states where Y = y and Y =y.Thermodynamics 167 It is reassuring that the condition F = 0 is also in line with the Boltzmann equation (5.43)) tot (y) = e Stot(y)/ k and the probability P(y) of such states is proportional to tot (y). In an analagous manner as for the free energy. a population ratio of states where the parameter Y has values y and y P(y) = e− P(y ) G/ kT . N2 At constant temperature and volume. one derives Stot = − G/T and. Likewise. 1 and 2.9). In an equilibrium state a. where U is the excitation energy of level 2. adiabatic process).45). As the volume is constant. In a nearby. The number of accessible states where Y takes up values between y and y + δy is (see (5. (5. we get.1) and the entropy changes between state a to b by S = k[ln b − ln a] = k ln b a = k ln N1 N2 + 1 k ln N1 . implies N2 /N1 = e− U/ kT .

CP .50). Because at constant pressure (d p = 0) dp = ∂p ∂T dT + V ∂p ∂V dV = T ∂p ∂T dT + V ∂p ∂V T ∂V ∂T dT P we obtain. This follows from the following thought experiment: (5. (5.7 The work done by magnetization Consider a magnetic body of unit volume in a field H and with magnetization M. U = U (T. to another place Y where there is some field H0. (5.60). Cp − Cv = T ∂p ∂T ∂V ∂T = −T P V ∂p ∂T 2 V ∂p ∂V −1 T . are defined by δQ ∂U Cv = . Its internal energy depends on the temperature and on the magnetic field. say. If a field H is needed to produce in the body the magnetization M. (5. Then we pull the body back to X but while doing so. it acquires the magnetization M0 . near a coil through which runs an electric current.6 Specific heat of a gas The specific heat at constant pressure.62) 5. at each position.3.168 Particle statistics 5. P (5.61) It follows immediately that δ Q = Cv ∂T ∂p d p + Cp V ∂T ∂V d V. where there is no magnetic field.60) Cp = dT P ∂T V Let us determine Cp − Cv . and at constant volume.3. H ). to change the magnetization by d M requires the infinitesimal work δ A = H d M. we fix the magnetization M0 (that is why it is a thought experiment). using (5.52) and (5.63) body H Y coil x X We move an unmagnetized body of unit volume along the x-axis from a position X. M(d H /d x) as follows from (11.16) for . The body is attracted by the coil and in the field H0. C V . The force acting on the body is.

formula (5.68) T where we have exploited.64) M In complete analogy to (5.65) T ∂T ∂M d M + CM H d H. after the second equals sign. therefore. So indeed δ A = H d M and.69) . We compute χad from (5.67) As long as magnetic saturation is excluded. T (5. To obtain the relations for the specific heat at constant magnetic field H and at constant magnetization M. 5. M (5.8 Susceptibility and specific heat of magnetic substances The previously derived formulae for CP and C V now come in handy. CH (5. ∂U ∂T CM = H δQ dT (5. The last equality is obvious when one identifies the integrals with the area under the curves M(H ) and H (M).62). δ Q = T d S = dU − H d M. Therefore χad = CM χT .Thermodynamics 169 the potential energy of a dipole. we write CM − CH = T and δ Q = CH ∂M ∂T ∂H ∂T = −T M H ∂M ∂T ∂T ∂H 2 H ∂M ∂H −1 (5.66) The adiabatic and isothermal (or static) susceptibility are defined as χad = ∂M ∂H and S χT = ∂M ∂H . defined by CH = we just have to interchange M↔p and H ↔ V.65). The total mechanical work A done on the system over the whole path from X to Y and back becomes A=− 0 H0 M dH + 0 H0 M0 d H = 0 M0 H d M. one also has χT = M/H .61) and (5. χad = − CM CH ∂T ∂H ∂M ∂T = H M CM CH ∂M ∂H (5.66) by putting δ Q = 0 (S = constant).3.

This expression. hν c2 e kT − 1 (5.73) In view of the exponential factor. (5. for example. It starts at the origin (0. Bν (T ) = 2h ν 3 . Bν (T ) is usually very sensitive to both temperature and frequency. i. Bν (T ) is related to the monochromatic radiative energy density u ν in the enclosure by 4π 8πh ν3 Bν (T ) = 3 · hν . Bλ (T ) = 2hc2 λ5 (e kT λ − 1) hc . the Planck function may be referred to wavelength and designated Bλ (T ). The Planck function increases monotonically with temperature.74) Fν = π Bν (T ). Written in the frequency scale.2 depicts the plot of the curve y= ex x3 −1 if T2 > T1 . (5.170 Particle statistics 5.70) The intensity is taken per Hz and has the units erg cm−2 s−1 ster−1 Hz−1 . is approximately applicable to stellar atmospheres when T is the effective surface temperature.4.e.4 Blackbody radiation 5. Because of dν = −(c/λ2 ) dλ. T ) of frequency and temperature.72) Bλ (T ) dλ = −Bν (T ) dν. 0) with a slope of zero. its unit is then erg cm−2 s−1 ster−1 cm−1 . .822 and asymptotically approaches zero for large x.71) uν = c c e kT − 1 Alternatively. An object that emits at all frequencies with an intensity Bν (T ) is called a blackbody. The two forms are related through (5. from which one can read off the value of the Planck function for any combination (ν. culminates at x max = 2. at any frequency Bν (T2 ) > Bν (T1 ) Figure 5.1 The Planck function The Planck function gives the intensity of radiation in an enclosure in thermal equilibrium at temperature T . The emergent flux Fν from a unit area of its surface into all directions of the half-sphere is (5. It is a universal function depending only on T and frequency ν.

one has to multiply the ordinate y by the factor (2k 3 / h 2 c2 )T 3 to get Bν (T ). In the Rayleigh–Jeans limit.4. where wavelengths are typically between 1 µm and 1 mm and temperatures from 10 to 2000 K. The universal shape of the Planck function.76) As the photon energy hν is much smaller than the thermal energy kT of an oscillator. 5. one can expand the Planck function into powers of x and enter the realm of classical physics where the Planck constant h .2. x 1 and Bν (T ) → 2ν 2 kT. x 1 and Bν (T ) → 2hν 3 − hν e kT .Blackbody radiation 171 Figure 5.2 Low. c2 (5. As the abscissa x is here in the unit hν/kT . x= • In the Wien limit.and high-frequency limit There are two asymptotic approximations to the Planck function depending on the ratio x of photon energy hν over thermal energy kT .75) hν . kT • With respect to dust emission. c2 (5. Wien’s limit is never appropriate.

The dependence of Bν (T ) on frequency and temperature is then no longer exponential. the Planck function Bλ (T ) reaches its maximum at λmax given by ∂ Bλ /∂λ = 0. one finds λmax T = 0. 1. Interstellar grains are certainly not blackbodies but the shape of the spectral energy distribution from a dusty region may at far infrared wavelengths be approximated by ν m Bν (T ) (see (8.206 cm K.1) for the correct expression) if the emission is optically thin and the absorption coefficient has a power-law dependence K ν ∝ ν m . (5.4.17)) B(T ) = 0 ∞ Bν (T ) dν = σ 4 T π (5.78) Note that λmax from (5.77).76 smaller than the corresponding wavelength c/νmax from (5. (5.44 λT (λ in cm) 5. of course.77) In the frequency scale.3 Wien’s displacement law and the Stefan–Boltzmann law In the wavelength scale. the displacement law determines the temperature of a blackbody. If one wants to detect a blackbody with an instrument that has a sensitivity curve Sν .70 × 10−11 Hz−1 K. Integrating the Planck function over frequency. The Rayleigh–Jeans approximation is always good for centimeter radio astronomy. maximum emission is determined by ∂ Bν /∂ν = 0 and occurs at νmax for which T νmax = 1. is a factor 1. so the λmax of radiating dust is shifted to shorter wavelengths with respect to a blackbody emitter of the same temperature. For K ν ∝ ν 2 . The total energy per s over a certain spectral interval is. Maximum emission follows now from ∂(ν m Bν )/∂ν = 0. and sometimes also applicable to dust emission. The term ν m Bν (T ) is sometimes called the modified Planck function. the same for Fλ and Fν . one usually tries to maximize Sν Bν (T ) dν. we obtain (see (A. Therefore λmax T = 0. When λmax is known. However. which refers to the wavelength scale. one must check whether x= is really small compared to 1.78).289 cm K. The wavelength where the flux from a blackbody peaks depends thus on whether one measures the flux per Hz (Fν ) or per cm (Fλ ).172 Particle statistics vanishes.79) .

87) ∞ ∞ .74)): σ = F= Fν dν = σ T 4 .Blackbody radiation where 173 2π 5 k 4 = 5. we can take 4πω0 I (ω0 ) out from under the integral.80) 15c2h 3 is the radiation constant. The integrated absorption cross section is from (1. we wish to understand it well.56 × 10−15 erg cm−3 K−4 .67 × 10−5 erg cm−2 s−1 K−4 (5.4.82) L ∗ = 4πσ R∗ T∗ . If we denote the radiative intensity by I (ω). In equilibrium. (5. mec 0 0 (5. When we substitute for γ the radiative damping constant from (1. derive it once more by representing the atoms as harmonic oscillators.71).106).86) The cross section σ (ω) is taken from (1. 3c2 (5.83) 4σ = 7. The total emergent flux F from a unit area of a blackbody surface into all directions of the half-sphere is given by the Stefan– Boltzmann law (see (5. We.84) 5. and drop the subscript 0 in the frequency.108). According to (1. we find for its bolometric luminosity: 2 4 (5.4 The Planck function and harmonic oscillators Because the Planck function is fundamental for the emission processes of dust and molecules. Consider a harmonic oscillator in an enclosure in thermal equilibrium with its surroundings. 2π 2 e2 .100). it must pick up the same energy from the radiation in the enclosure. c (5.81) Applying this to a star of radius R∗ and effective temperature T∗ . For the total radiative energy density u. Because the resonance curve is 2 very narrow. we get u= with constant a= u ν dν = aT 4 (5.85) where γ is the damping constant. it loses energy through emission at a rate ˙ −E = γ E (5. therefore. remembering that I (ω) = I (ν)/2π. we obtain γE = 4π I (ω)σ (ω) dω = 4π I (ω0 ) σ (ω) dω = 4π I (ω0 ) I (ν) = 2ν 2 E.

86). . E = kT x ex − 1 x= hν . Inserting E = 3 E = 3kT into (5. from (5. Something is wrong with the classical ideas. potential plus kinetic energy. in agreement with the Stefan– Boltzmann law in (5.174 Particle statistics In (5. we obtain a function that is proportional to the fourth power of the temperature.83). the differential d S = d Q/T of the entropy is given by x dS = u d V + V du + (u/3) d V dU + P d V = .87). T T As d S is a full differential. ∂V ∂T Solving for the energy density u.11).86) still holds in quantum mechanics but we have neglected that light comes in packages (photons). does one obtain the classical result. kT The quantum mechanical mean energy E is smaller than kT . The cross section σ (ω) in (5. we get the equation ∂2S 1 du 4 d = = ∂T ∂V T dT 3 dT u T = ∂2S . . u(T ) ∝ T 4 . Because dU = u d V + V du. In the classical picture. 2. We obtain the correct Planck function if we assume equally spaced energy states E i = i hν with i = 0. A gas of volume V and internal energy U (T ) = u(T )V has the gas pressure P = u/3. 1. not just one. Iν (T ) = 2ν 2 kT. The oscillators must therefore be able to move freely and have three degrees of freedom. . . It is also enlightening to apply the first principles of thermodynamics to a photon gas in equilibrium. one recovers the Rayleigh–Jeans formula of (5.e. c2 This formula. Only when 1. i.76) which is valid for blackbody radiation at low frequencies. E = kT is the average total energy per degree of freedom. fails at photon energies hν comparable to or larger than kT because the intensity would rise infinitely with frequency. . however. the integration extends over the solid angle 4π implying that the oscillators absorb radiation from all directions. for which the average E per degree of freedom is not kT but.

It helps to understand quantitatively very diverse topics: how atoms can wander about the grain surface by way of tunneling through a potential barrier.1 The classical Hamiltonian ˙ Consider a system of particles with generalized coordinates qi and velocities qi . how protons can fuse to deuteron in the interior of the Sun. by integrating the second-order Lagrange equations d dt ∂L ∂ qi ˙ = ∂L . 6.1) Alternatively. H= qi pi − L ˙ (6. ∂qi (6. how electrons are energetically arranged in white dwarfs. one finds the motion of the particles from the Lagrange function L =T −V where T is the kinetic energy of the particles and V the potential.1 A charged particle in an electromagnetic field 6. therefore.1. The final section deals with transmission and reflection of a free particle at a potential barrier. or how electronic bands appear in crystallized grains.2) .Chapter 6 The radiative transition probability Almost all astronomical information eventually comes from light which we detect with our telescopes. If velocities are small (non-relativistic).3) 175 ∂L ∂ qi ˙ (6. using the conjugate momenta pi = one constructs the Hamiltonian. a fundamental process and a major goal of this chapter is to explain the concept of induced and spontaneous emission. Photon emission is.

c (6.2 The Hamiltonian of an electron in an electromagnetic field We are interested in the equation of motion of one particle of charge e and mass m moving in an electromagnetic field. Moreover.1. 6. in our applications.7) As usual. 2 c (6. E = T + V = H (qi . we may readily convince ourselves that equation (6.7) yields the Lorentz force 1 F=e E+ v×B .1) with the Lagrangian of (6. We now use Cartesian coordinates (x. in .4) If H does not explicitly depend on time. c ˙ Often the potential V is only a function of coordinates and the force F = m v on a particle is given by the gradient of the potential. However. it is a constant of motion. pi ).5) (6.8) Remembering that the full-time derivative of the x-component of the vector potential is ∂ Ax ∂ Ax ∂ Ax d Ax ˙ = A x + vx + vy + vz dt ∂x ∂y ∂z with similar expressions for A y and A z . The latter is connected to the magnetic field through B = rot A and the electric field is given by 1˙ E = − grad φ − A. φ denotes the scalar potential and A the vector potential. H equals the total energy E of the system. The kinetic and potential energy equal T = 1 mv2 2 e V = eφ − A · v c which leads to the non-relativistic Lagrangian L= 1 2 e mv − eφ + A · v. ∂qi (6. F = − grad V .6) (6. y.176 The radiative transition probability and integrates the first-order equations qi = ˙ ∂H ∂ pi − pi = ˙ ∂H . z).

10) If we think of the motion of an electron in an atom. The new φ is then constant in time and refers only to the static field. we may regard φ as representing the static electric field of the nucleus and.85). This is because of the gauge condition (1. ˙ div A + φ/c = 0. (6.13) (6. the part p2 Hstat = + eφ 2m which describes the electron without an electromagnetic wave. conjugate momentum and coordinates. (6.) The remaining part of the Hamiltonian.1.1 The Schr¨ dinger equation o In quantum mechanics.1.5) by turning the energy E and the Hamiltonian H into operators using the standard prescription for the conversion of energy. i E →i ∂ ∂t pi → pi = −i ˆ ∂ ∂ xi xi → xi = xi .3. z). x 3 ) = (x. V of (6. separate. ˆ (6.4) governing the motion of the particles are replaced by the Schr¨ dinger equation o ∂ ˆ =H . therefore. the formulae (6.9) E = H = T + eφ. pi ) of (6.3 The Hamilton operator in quantum mechanics 6.11) . y. Hwave = 6. from the Hamiltonian.12) ∂t This equation follows from E = H (x i . (In fact. e2 e (p · A + A · p) + A·A 2mc 2mc2 describes the perturbation of the electron by an electromagnetic wave. Using the conjugate momenta pi = the Hamiltonian becomes e 1 p− A 2m c It equals the total energy E of the system.6) is a generalized potential that also depends on velocity and the force follows from Fi = − ∂V d + ∂ xi dt ∂V ∂ xi ˙ where we put (x 1 . (6.A charged particle in an electromagnetic field 177 the case of an electromagnetic field. H= 2 ∂L e = mvi + Ai ∂ xi ˙ c + eφ. x 2 . for slowly moving charges one can always make a transformation of the given potentials to new ones such that div A = 0 for the new A.

H ] = x i H − H x i = i ˆ ˆ ˆ ˆ pi . (6. t) = ψ(x) e−i Et / .18) The eigenfunctions ψn . One has to keep in mind the commutation rules. 2m e c 2m e c2 (6. The time-independent Schr¨ dinger equation of a particle of o energy E in a potential V (x) follows from (6.1. t).14) m For the Hamiltonian of an electron in an atom perturbed by an electromagnetic field we write ˆ H= e2 ˆ ˆ e ˆ ˆ ˆ ˆ (p · A + A · p) + A · A. the wavefunction can be written as (x. In this case.12). 2 2m ψ + E − U (x) ψ = 0. form a complete set so that an arbitrary wavefunction (x. ˆ (6. t) can be expanded into a sum of ψn : (x.16) This follows from (6.3.19) with expansion coefficients an . . in particular [ x i .12) depends generally on space and time.15) (6. The eigenfunction ψ(x) depends only on coordinate x. t) contains the time merely through the factor e−i Et / which cancels out on forming the probability density | |2. 6. each to an eigenvalue E n . = (x.11). Stationary solutions correspond to fixed energy eigenvalues E. p j ] = x i p j − p j x i = i δi j ˆ ˆ ˆ ˆ ˆ ˆ from which one readily deduces ˆˆ [x i .2 Stationary solutions of the Schr¨ dinger equation o The wavefunction in (6. t) = n an ψn (x) · e−i En t / (6.17) (x.178 The radiative transition probability Operators are marked by a hat on top of the letter. (6. omitting the subscript ‘wave’.

21) ψ ∗ p ψk d x = i ω j k µ j k . with the help of equation (6. The oscillator obeys the equation of motion x + ω2 x = 0 ¨ where ω2 = κ/m is the square of the frequency of oscillation.20) the dipole moment with respect to the states described by the eigenfunctions ψ j and ψk . The total energy of the system E =T +V = 1 p2 + mω2 x 2 2m 2 and the time-independent Schr¨ dinger equation is.23) . we get e (6. If we put ω j k = E j − E k and call µ jk = e ψ ∗ (x) x ψk (x) d V j (6. m the particle mass.4 The dipole moment in quantum mechanics ˆ The Hamiltonian operator H is Hermitian. 2 2m ∂ x 2 2 (6.18)) o ∂ 2 ψn 1 + E n − mω2 x 2 ψn = 0. It follows.3).22) This second-order differential equation looks much better in the form u n + (λn − y 2 )u n = 0 where we have introduced the function u n (αx) = ψn (x) with α2 = mω (6. The asterisk denotes the complex conjugate.1. which means ˆ ψ ∗ ( H φ) d x = ˆ ( H ψ)∗ φ d x 179 for any ψ and φ and guarantees real expectation values. therefore.1.15).A charged particle in an electromagnetic field 6. that e m e ψ ∗ pψk d x = i (E j − E k ) j ˆ ψ ∗ xψk d x j ˆ so one can substitute in the left-hand integral for the momentum operator the coordinate operator (but for a constant). j ˆ m 6. and −κ x the restoring force (see section 1.5 The quantized harmonic oscillator We review the solution of the time-independent Schr¨ dinger equation for the o undamped free linear harmonic oscillator. (see (6.

always disappear for any energy E n .3) we find that the mean position and momentum of an oscillator. With the help of (A.25) is above zero. the square of the eigenfunction |ψn (x)|2 for n = 3 and α 2 = 1.24) Hn are the Hermitian polynomials described in appendix A. For high n. An ansatz for u n (y) of the form e−y yields the normalized eigenfunctions of the harmonic oscillator: ψn (x) = Nn e−α 2 x 2 /2 2 /2 Hn (y) Hn (αx) with Nn = α √ .1 displays. the dotted and full curve converge. here the particle is strictly confined to the allowed region.22) are equidistant: E n = (n + 1 ) ω 2 and the lowest level E0 = 1 2 n = 0. ω (6. . . The undulating quantum mechanical probability P(x) = |ψ(x)|2 for finding the particle at locus x when the oscillator is in its third energy level above ground (n = 3). x and p . The energy levels E n of the harmonic oscillator in (6.180 The radiative transition probability Figure 6. 2n n! π (6. So d x p = dt m . as an example. 2. For comparison. 1. . Figure 6.1. we show the corresponding probability of a classical oscillator (dots) of the same energy. and put y = αx and λn = 2E n / ω.

H ∂ ˆ ˆ H + H −i ∂t ψ(x. t = 0) = ψs (x). t) = 0.12) reads: o small perturbation. the matrix element Hnn gives the perturbation ˆ ˆ ˆ energy.28) Inserting this sum into (6. to a system with Hamiltonian H and energy eigenvalues E n . As the eigenfunctions are orthogonal. 2 ω (6.31) For equal subindices ( f = n).26) 6.2 Small perturbations 6. the energy eigenvalues change from E n to E n + Hnn . t) = n an (t) ψn (x) e−i En t / .2.29) the dot over an means time derivative. f (6. multiplication of (6. This means when the Hamiltonian H is replaced by H + H . (6.30) with the matrix coefficients Hfn = ˆ ψ ∗ H ψn d x.19) into eigenfunctions ψn but now with timeˆ dependent coefficients an (t) because of the additional term H which causes the system to change: ψ(x. a ˆ .1 The perturbation energy ˆ Let us introduce.Small perturbations 181 is fulfilled in a trivial way but x 2 and p2 do not vanish.27) gives e−i En t / n ˆ ˙ an H ψn − i an ψn = 0 (6.27) We expand ψ after (6. . from a starting state s to a f inal state f .2 The transition probability We consider a transition s → f . For the product of the uncertainties we have x p= x2 − x 2 p2 − p 2 = En = (n + 1 ) . so that the Schr¨ dinger equation (6. Suppose that our system was at time t = 0 in state s. (6. 6.2.29) with the eigenfunction ψ ∗ and integration over the space f coordinate x yields e−i En t / an (t)H f n − i a f (t)e−i E f t / = 0 ˙ n (6. ψ(x.

182 The radiative transition probability Then all an (0) in (6.37) (ω f s − ω)2 −∞ . Integrated over frequency. equation (6. a f .28) vanish. (6.32) After integrating from the starting value a f (0) = 0 and assuming that H f s is not time-variable.30) the an (t) by their values at t = 0. t ω−1 . so an (0) = δsn where δsn is the Kronecker symbol.3 Transition probability for a time-variable perturbation ˆ When the perturbation varies harmonically with time. (6. If ω f s = E f − Es is the energy difference between the two states. then becomes ˙ i a f = − H f n ei(ω f s −ω)t ˙ and p f s (t) = |a f (t)|2 = 2 2 |H f s |2 1 − cos(ω f s − ω)t . 1 − eiω f s t i .30) yields for the time derivative of the expansion coefficient a f .35) is large compared to the oscillation period.33) a f (t) = H f s iω f s The probability of finding the system at a later time t in the final state f is p f s (t) = |a f (t)|2 = 2 2 |H f s |2 1 − cos ω f s t ω2 s f . instead of H of the expansion coefficient. we can substitute in (6. If the time t in (6. given in (6. If the interaction ˆ operator H is weak and the perturbation time t not too long. i a f = − H f n eiω f s t .29) for the perturbation operator. (ω f s − ω)2 (6. ˙ (6.34) 6. we have to write H e−iωt ˆ .35) One just has to replace ω f s by ω f s − ω and the probability p f s (t) now has its spike when ω ω f s . except around ω = ω f s in the small interval ω from −π/2t to π/2t.32). (6. the function 1 − cos(ω f s − ω)t (6.36) f (ω) = (ω f s − ω)2 resembles the δ-function: it has values close to zero almost everywhere. The time derivative in equation (6. ∞ 1 − cos(ω f s − ω)t dω = πt.2. except for as (0) = 1.

3. The transition probability from a starting state s to a final state f under the influence of a perturbation potential is proportional to a characteristic function given in (6.The Einstein coefficients A and B 183 Figure 6. let Ni (ν) dν = Ni (ν) dν be the number of atoms in state i that can absorb radiation in the frequency interval [ν. (ν) dν = 1.1.1 How A and B are defined We consider radiative transitions in atoms between an upper level j and a lower one i of energies E j and E i such that E j − E i = hν. 6. respectively. The function f (ω) is plotted in figure 6. of course. If one waits long enough.3. one has almost exactly ω = ω f s . line .2 for t = 100.3 The Einstein coefficients A and B 6. The probability p f s (t) in (6.35) is not negligible and transitions from state s to state f occur only when |ω −ω f s |t ≤ π. As atomic energy levels and lines are not infinitely sharp but have some finite width.1 Induced and spontaneous transitions 6. where.2.36) and shown here for t = 100. ν + dν]. Let the atoms have number densities N j and Ni per cm3 . and be bathed in a radiation field of energy density u ν .

. we get the symmetry relation between the Einstein-B coefficients.3.184 The radiative transition probability Radiative transitions between level j and i may be either spontaneous or induced. Ni gi −x = e Nj gj yields A j i + uν B j i = with x = E j − Ei kT gi Bi j u ν e x . B ji c3 As this equation holds for any small T . we are in the Rayleigh–Jeans part of the A j i . we are in the Wien part of the spectrum and A j i −(E j −Ei )/ kT 8πhν 3 −hν/ kT e = e . we can derive the relations between the As and Bs. Setting u= line u ν (ν) dν (6. As then also spectrum and u ν becomes arbitrarily large so that u ν B j i e x → 1. Bi j are atomic quantities and do not depend on the radiation field.1.40) which connects the A coefficients with the B coefficients.9). gj If we fix ν and let T go to infinity. If we assume blackbody radiation (u ν = 4π Bν (T )/c) and thermodynamic equilibrium.2 How A and B are related A j i . It also follows that A ji = 8πhν 3 B ji c3 (6. B j i .39) If we now let the temperature tend to zero. we obtain hν = E j − E i . The Boltzmann formula (5. 6. Consequently. N j u B j i = rate of induced downward transitions and Ni u Bi j = rate of induced upward transitions. This looks trivial but expresses the non-trivial fact that the radiation field stimulates absorption and emission at the frequency corresponding to the energy difference of the levels. B j i (e x − 1) (6.38) the Einstein coefficients A j i and B j i are defined such that per cm3 • • • N j A j i = rate of spontaneous downward transitions. g j B j i = g i Bi j . uν = A ji .

5 Transition probability for general forces Equation (6.4 Debye.1.97)).41) exactly if one puts µ = 1 p0 and ω = 2πν. the Einstein coefficient becomes A j i ∼ 107 s−1 .1.3. which is 14 powers of ten smaller! 6.3.11 Debye.3 The quantum mechanical expression for A 185 The quantum-mechanical correct expression for the Einstein coefficient A j i for spontaneous emission from upper state j to lower state i is A ji = where µ ji = e ψi∗ (x) x ψ j (x) d V (6. Now A j i ∼ 10−7 s−1 .15 × 1011 Hz.3.41). A rigorous derivation of formula (6. (2) To the lowest rotational transition of the CO molecule.41) that.20)). If x = x 0 e−iωt is the time-variable coordinate of the electron and p0 = ex 0 its dipole moment.41) may also be put in the form A ji = c 8π 3 α 3 p h 3 . Note that 2 according to equation (6. A j i increases with the square of the dipole moment and with the third power of the frequency. The dipole moment in this case is fairly weak.4 × 10−18 cgs = 2.4 A classical analogy The essence of formula (6. ˚ corresponding to λ = 5000 A.41) This yields (6.42) is the dipole moment corresponding to the transition (see (6. the dipole moment is then µ = 2. 3c3 64π 4 ν 3 |µ j i |2 3hc3 (6. At an optical frequency ν = 6 × 1014 Hz.4.The Einstein coefficients A and B 6. We see from (6. cum granu salis. Let x 0 be equal to the ˚ atomic radius in the ground state (0.3. ν 1. ψi and ψ j are the eigenfunctions of the energy states.41) requires quantum electrodynamics. then A j i hν = 2 p0 ω 4 .41) can already be grasped using classical arguments by equating the emission rate A j i hν to the average power radiated by a harmonic oscillator (see (1. A j i is high for optical transitions and low at radio wavelengths. To get a feeling for the numbers. 6. and the frequency low.1. the non-relativistic limit is discussed in section 6.5A). µ = 0. we apply the formula: (1) to an electronic transition in the hydrogen atom.


The radiative transition probability

where α = e2 / c is the fine structure constant, = 2x 0 the size of the system, c/l the inverse crossing time by light, p = hν/c the photon momentum and ( p )3 the volume in the phase space necessary for the creation of photons by the atomic oscillator. In other words, the transition probability is the product of the coupling constant of the electromagnetic radiation, α, multiplied by the number of cells in the phase space divided by the crossing time. When the transition probability is expressed in this way, it can also be applied to other basic forces in the universe, besides electromagnetism. The coupling constant is then, of course, different. 6.3.2 Selection rules and polarization rules Formulae (6.41) and (6.42) incorporate the selection rules because when the integral (6.42) is zero for two eigenfunctions ψi and ψ j (which is usually the case), the transition is forbidden. For instance, from that part of the eigenfunctions of the hydrogen atom which is radius-independent, namely Plm (cos θ )eimϕ follow the selection rules l = ±1 m = 0, ±1.

If they are not fulfilled, no emission is possible. However, a perturbation, like an electric field, can make an otherwise forbidden line to become observable because the eigenfunctions are then deformed so that the integral (6.42) no longer vanishes. One can also derive the polarization rules in the presence of a magnetic field by separating the dipole moment µ of (6.41) into its Cartesian components, µx = e ψi∗ xψ j d V µy = e ψi∗ yψ j d V µz = e ψi∗ zψ j d V

and evaluating the expressions. For example, when the field is in the z-direction and µx = µ y = 0 but µz = 0, the radiation is linearly polarized with its electric vector parallel to the applied magnetic field. 6.3.3 Quantization of the electromagnetic field So far we have learnt that radiative transitions occur at a rate given by the Einstein coefficients. Now we want to understand why they occur and how the As and Bs are computed. For this end, we have to quantize the electromagnetic field and we sketch how this can be done. The spectrum of standing waves in a box Imagine a big box of volume V and length L with an electromagnetic field inside, consisting of standing waves. The vector potential A is a function only of z. It

The Einstein coefficients A and B


oscillates in the x-direction (A y = A z = 0) and vanishes at the walls of the box, which are nodal surfaces. We expand A x into a Fourier series:

A x (z, t) =
j =1

q j (t) sin

jπz L


˙ the coefficients q j (t) are time-dependent. With H = rot A and E = −A/c, and the orthogonality of the sine function,
π 0

sin i x · sin j x d x = δij

π 2

the energy of the field in the box becomes E field = 1 8π
2 2 (E x + H y ) d V =

V 16πc2

(q 2 + ω 2 q 2 ) ˙j j j
j =1

where the frequencies are

j πc . L If the box is very big, there is an almost continuous frequency spectrum. ωj = Equation of motion of an elastic string

There is a strong analogy between the field and an elastic string. Suppose a string has a length L, is fixed at its ends and vibrates exactly like the vector potential A x of (6.43). As it vibrates, it bends and becomes distorted. Hooke’s law of elasticity states how strong the internal deformations are: When we pull with a force F at the end of a rectangular block of length and cross section σ , the change in length d / is proportional to the force per unit area, F/σ = Y d / , where the proportionality factor Y is Young’s elasticity module. This law implies that the potential energy stored in the deformation of the volume element goes with (d )2 . If the shape of the bent string is described by the curve y(z), its local deformation d is proportional to the derivative of the curve, d ∝ y (z). So the potential energy of the string that vibrates like the magnetic field A x is U= 1 κ 2
L 0

∂ Ax ∂z


dz =

κπ 2 4L

j 2q 2 j
j =1

where κ is some elastic constant. If the string has a mass density ρ, its kinetic energy equals ∞ L 1 ˙x A2 dz = Lρ T = 1ρ q 2. ˙j 2 4 0
j =1


The radiative transition probability

We can form the Lagrange function L = T − U and find, from (6.1), the equation of a loss-free harmonic oscillator q j + ω2 q j = 0 ¨ ˜j with frequency ωj = ˜ jπ L

κ . ρ

The total energy equals the sum of kinetic plus potential energy of all oscillators, E string = Lρ 4 (q 2 + ω2 q 2 ). ˙j ˜j j
j =1 Mathematical identity between field and string The expressions for the energy of√ string and the field, E string and E field , the become identical when we put c = κ/ρ and m ≡ Lρ = V 4πc2 (6.44)

m may be considered to be the mass of the oscillator. The frequencies of field and string then coincide, ω j = ω j , as well as their energies. The q j s of the string are ˜ coordinates of an harmonic oscillator of frequency ω j . Such an oscillator has, in quantum mechanics, discrete energy levels E n j = ω j (n + 1 ) 2 with n = 0, 1, 2, . . . .

The quantization of the string implies that the electromagnetic field is quantized, too. Naturally, we associate the oscillator of frequency ω j with radiation at that frequency and identify the quantum number n with the number of photons in the box at that frequency, more precisely, with the occupation number , which is related to the radiative energy density u ν through formula (5.33). Because the field is, from (6.43), completely described by the coefficients q j (t), in quantum mechanics one replaces the field A x by the amplitudes q j ; the q j are called normal coordinates.


6.3.4 Quantum-mechanical derivation of A and B Coupling between oscillator and field Emission or absorption must result from the coupling between the oscillator and the electromagnetic field but how? The trick is to perceive the atom and the field as one system consisting of two subsystems: • ˆ the atom described by coordinates (1) with Hamiltonian H1, eigenvalue E 1 ˆ 1ψ1 (1) = E 1 ψ1 (1); and and eigenfunction ψ1 , so that H

The Einstein coefficients A and B


ˆ the electromagnetic field for which we use the subscript 2; here H2 ψ2 (2) = E 2 ψ2 (2).

ˆ ˆ ˆ The total uncoupled system has the Hamiltonian H = H1 + H2, the wavefunction ψ(1, 2) = ψ1 (1) · ψ2 (2) and the energy E 1 + E 2 , ˆ H ψ = (E 1 + E 2 )ψ1 ψ2 . In reality, the two subsystems couple. For the perturbation operator H we use the Hamiltonian derived in (6.16), e ˆ ˆ ˆ H = − p·A mc (6.45)

ˆ ˆ ˆ ˆ ˆ and neglect the quadratic term A2 . Note that p · A commutes with A · p; this ˆ follows from the definition of the momentum operator p in (6.13) and the Lorentz gauge div A = 0 (see (1.85)). The matrix element H f s Let us consider the process of emission or absorption of the atom as a transition of the whole system: atom plus field. The transition probability from the starting state s to the final state f as a function of time is given by (6.35). To compute it, we have to evaluate the matrix element H f s of (6.31) which contains the operator ˆ ˆ ˆ H = −(e/mc)p · A. The integral that appears in the matrix element H f s can be separated into I1 = e mc
∗ ˆ ψ1 f p ψ1s d x


I2 =

∗ ˆ ψ2 f A ψ2s dq

so that H f s = I1 · I2 . For the first, we get from (6.21) |I1 |2 = ω2 s f c2 |µ f s |2 .

To evaluate the second integral I2 , for ψ2 we insert the eigenfunctions of the electromagnetic field and thus of an harmonic oscillator. They read ψ2 j = N j H j , where H j are the Hermitian polynomials of (6.24) and N j their normalization ˆ coefficients. The field operator A is replaced by the normal coordinate q. In this way, we get I2 = N f Ns H f (αq) q Hs (αq)e−α
2q 2

dq =

N f Ns α2

H f (y)y Hs (y)e−y d y

190 =

The radiative transition probability
Ns {δ f,s+1 + 2sδ f,s−1} 2α N f √  +  Ns = s√ 1 if f = s + 1 (emission)   2α N f α √ 2 =  Ns s   if f = s − 1 (absorption). = √ 2α N f α 2


Here we have exploited formula (A.3). In the event of atomic emission, the energy of the atom makes a downward jump ω f s . For the field it is the other other way round, its final state is higher than the initial state. Therefore in (6.46), which refers to the field, atomic emission corresponds to f = s +1 and absorption to f = s − 1. During the transition there is an exchange of energy between the two subsystems by the amount ω f s but the overall energy stays constant. With respect to the parameter α 2 = mω/ from equation (6.23), we have to insert the field mass m after (6.44), so Vω . (6.47) α2 = 4πc2 This gives s +1 2πc2 · (s + 1) = |I2 |2 = 2α 2 ωV and finally |H f s |2 = |I1 |2 · |I2 |2 = The transition probability The transition probability p f s (t) as a function of time from one discrete state to another, s → f , is given in (6.35). Because of the presence of the field, the energy levels of the total system are not discrete, like those of the atom alone, but closely packed, like a continuum. Furthermore, we do not have one final state but many. If ρ(ω) is the density of these states, there are ρ(ω f s ) dω of them in the frequency interval dω around the energy ω f s . The density is equal to the number of unit cells h 3 in phase space, from (5.30), ρ(ω) = V ω2 . π 2 c3 ω2 s f c2 |µ f s |2 · 2πc2 · (s + 1). ωV

To find the total transition probability of our coupled system, we, therefore, have to multiply p f s (t) by ρ(ω f s ) dω and integrate over frequency. The integral is according to (6.37), proportional to πtρ(ω f s ). When we make the time t large, the total transition probability per unit time becomes P= 2π |H |2 ρ(ω f s ). 3 2 fs (6.48)

The Einstein coefficients A and B


We introduced the correction factor 1 to account for the fact that the 3 orientation of the dipole with respect to the field is arbitrary. It only remains to insert the expressions for ρ(ω f s ) and |H f s |2 . When we do this, the volume V of the box luckily disappears. Here are the final results for the transition probability per unit time of an atom in a radiation field: • If exciting radiation is absent (s = 0), the transition probability P equals the Einstein A coefficient of (6.41), P = A ji = 4ω3 |µ j i |2 . 3 c3

It is puzzling that a seemingly unperturbed atom in an eigenstate (above the ground state) with energy E j should emit radiation at all. One might expect it to stay there forever because the only time dependence in the eigenfunction is, from (6.17), an oscillation of the form e−i E j t / . However, there are downward transitions (A j i > 0), even when the atom sits in the dark. The reason is that the perturbation operator (where the field is represented by eigenfunctions of an harmonic oscillator) does not vanish when photons are absent. In a radiation field, the radiative density is, from (5.33), uν = 2 ω3 s. πc3 s + 1 and the induced

When there are many photons flying around, s emission rate is P= 4ω3 |µ j i |2 (s + 1) 3 c3

Bji u ν .

The same expression holds for induced absorption (if the initial and final states of the atom have the same statistical weight). Note that the symmetry of |H f s |2 with respect to s and f accounts for the symmetry of the B coefficients, Bs f = B f s . The transitions described here involve one field photon because they are ˆ ˆ ˆ rooted in the perturbation operator p· A, which is linear in A. When we include the ˆ 2 in the Hamiltonian of (6.16), it follows, again from the much weaker operator A orthogonality of the Hermitian polynomials which we employed for the derivation ˆ of (6.46), that the A2 -term describes processes involving two photons at a time, either in emission or absorption, or one in emission and the other in absorption.


The radiative transition probability

6.4 Potential wells and tunneling
6.4.1 Wavefunction of a particle in a constant potential Consider in one dimension a free particle of mass m and energy E in a potential V (x). To facilitate writing, in this section we often put U (x) = 2m

V (x)





In this notation, the wavefunction of a stationary state satisfies the equation ψ = U (x) − ε ψ. If U is constant, the general solution reads ψ(x) = A1 eiαx + A2 e−iαx with complex Ai . There are two cases: • • ε − U > 0. Then α > 0 and the wave has an oscillatory behavior. It could be represented by a combination of a sine and a cosine. ε − U < 0. Then i α is real and the two solutions correspond to an exponentially growing and declining function. The former is only allowed in finite intervals because the integral ψ ∗ ψ d x must be bounded. α= √ ε−U (6.51) (6.50)

If the potential U (x) is a step function, i.e. if it is constant over intervals [x i , x i+1 ] but makes jumps at the connecting points x i , with x 1 < x 2 < x 3 < . . . , the wavefunction ψ and its derivative ψ must be continuous at these points; otherwise the Schr¨ dinger equation would not be fulfilled. The boundary o conditions of continuity for ψ and ψ at all x i fully determine the wavefunction. To find it is straightforward but often tedious. 6.4.2 Potential walls and Fermi energy If the particle is trapped in an infinitely deep well of length L, as depicted on the left-hand side of figure 6.3, the wavefunction must vanish at the boundaries. It cannot penetrate beyond the walls, not even a bit. The situation is reminiscent of a vibrating string fixed at its ends. If λ = h/ p is the de Broglie wavelength and p the momentum, the condition nλ/2 = L leads to the discrete energies εn = 2m

En =

n2π 2 L2

n = 1, 2, . . .


and sine-like wavefunctions ψn (x) = A sin nπ x . L

Potential wells and tunneling





ε3 ε2 ε1
−L/2 L/2

ε4 ε3 ε2 ε1 x x

Figure 6.3. The left side shows an electron in a square potential with walls of infinite height. The stationary states correspond to energies E n ∝ n 2 (see (6.52)). The right side shows, for comparison, the parabolic potential and the equidistant energy levels of an harmonic oscillator.

There is a zero-point energy (ε > 0 for n = 1) because the particle is spatially localized in the interval from −L/2 to L/2, and so its momentum cannot vanish. The spacing between energy levels increases quadratically with quantum number n. This is very different from an harmonic oscillator which has a parabolic potential and equally spaced energy levels (see figure 6.3). If a system with N electrons is in its lowest state, all levels n ≤ N/2 are filled. The threshold or Fermi energy of the topmost filled level is EF =
2π 2 N 2

8m e L 2


√ and the velocity of those electrons is vF = 2E F /m e . If the walls of the potential well are finite and of height U and the particle energy ε < U , there is some chance, exponentially decreasing with distance, to find it outside the well. The energies of the eigenstates depend then in a more complicated manner on n. In three dimensions, for a cube of length L and infinite potential barrier, the eigenfunctions are ψ(x) = A sin nyπ nz π nx π x sin y sin z L L L


The radiative transition probability

with positive integers n x , n y , n z , and the Fermi energy is EF =

2m e

3π 2 N L3



(6.54) Examples of Fermi energy and Fermi statistics • White dwarfs are burnt-out stars with extreme densities up to 106 g cm−3 . But only the electrons are degenerate, the more massive atomic nuclei are not because their degeneracy parameter is much smaller (see (5.29)). When the Fermi energy is very high as a result of a high density ρ = N/L 3 , the electrons are relativistic and E F = c(N/V )1/3 . If one expresses the Fermi energy as a temperature, TF = • EF k (6.55)

one finds TF ∼ 109 K for ρ ∼ 106 g cm−3 . An even more spectacular case is presented by neutron stars with densities up to 1015 g cm−3 . The neutrons do not disintegrate into protons and electrons, as one would expect, only a few do, because the energy liberated in the decay n → p + e produces only ∼106 eV, which is miles below the Fermi threshold of (6.55) in a fictitious sea of electrons and protons at density 1015 g cm−3 . The conduction electrons in a metal have values TF between 104 and 105 K, so at temperatures T ∼ 100 K, the energy distribution of the conduction electrons given by (5.26) is highly degenerate, only few of them are thermalized. For graphite particles of interstellar dust grains, the Fermi temperature is somewhat lower, TF ∼ 103 K. Observing astronomers use both isotopes of liquid helium to cool their detectors. The 3 He atoms are made of five elementary particles (two protons, one neutron and two electrons), they have half-integer spin and are fermions. The main isotope 4 He, however, contains one more neutron, has zero total angular momentum and follows Bose statistics. The two isotopes show radically different behavior at low temperatures: 4 He becomes a superfluid below 2.2 K, whereas 3 He does not. Superfluidity is explicable by Bose statistics because Bose statistics allows many (even all) atoms to be at the lowest energy state.

6.4.3 Rectangular potential barriers A single barrier Let a particle travel from the left to the right and encounter a potential wall, as in figure 6.4. If its energy ε is greater than U , it can, of course, overcome

Potential wells and tunneling






Figure 6.4. Rectangular potential barrier for a particle coming from the left. Part of the beam is reflected and part goes through the barrier.

the barrier but contrary to classical physics, part of the wave is reflected. Most relevant is the case when ε < U . Then the particle can tunnel through the barrier which is classically forbidden. What happens, as the particle approaches the barrier, obviously depends on time and therefore stationary solutions may not seem possible in this scenario. However, one may interpret the infinite wave i e ( px−Et ) of a particle with definite energy E as a stationary particle beam, and then the time drops out. The wavefunction is constructed piece-wise:   Aeiαx + Be−iαx x<0 ψ(x) = Ceβ x + De−β x 0<x <L  iαx x > L. Fe + Ge−iαx The coefficient G = 0, as there is no wave coming from the right, and √ √ α= ε>0 β = U − ε > 0. Exploiting the boundary conditions and keeping in mind that (e x + e−x )2 /4 = cosh2 x = 1 + sinh2 x, the transmission coefficient T , which is the fraction of the particle beam that penetrates the barrier, becomes T = . √ 4ε(U − ε) + U 2 sinh2 (L U − ε) 4ε(U − ε) (6.56)

When the barrier is high compared to the kinetic energy of the particle (U ε) and sufficiently broad (LU 1/2 > 1), the tunneling probability can be approximated by 16ε −2L √U T = e . (6.57) U


The radiative transition probability

U 1 −c 0 b 2 a x

Figure 6.5. A periodic square potential of a one-dimensional lattice. The length of the period is a, the length of one barrier is c = a − b. In region 1, where 0 ≤ x ≤ b, the potential is zero, in region 2, where b ≤ x ≤ a, it has the value U . Periodic potential Regularly arranged atoms in a solid (crystal) produce a periodic potential. We discuss this important case in one dimension. Let the potential be periodic with period a, as depicted in figure 6.5. The wavefunction ψ of a particle with energy ε < U is then given by ψ(x) = with α= Aeiαx + Be−iαx Ceβ x + De−β x √ ε β= 0<x <b b<x<a √ U − ε.

Two equations for determining A–D are found from the condition that ψ and ψ must be continuous at x = 0. Another two equations follow from a result detailed in section 7.2.3: In a periodic potential, any eigenfunction ψk , with respect to wavenumber k and energy ε = k2 , must have the form ψk = u k eikx where u k is periodic such that u k (x) = u k (x + a). Therefore, u k (b) = u k (−c) and u k (b) = u k (−c), where u k (x) = ψk (x)e−ikx . Altogether, this gives A+B =C+D e ie
−ikb −ikb



i α(A − B) = β(C − D) + Be−iαb = eikc Ce−βc + Deβc



(α − k)−Be−iαb (α + k) = eikc Ce−βc (β − i k)−Deβc (β + i k) .

58).3 is large. however. 11 or 24.58) as a function of particle energy ε. c = 0. c. It is calculated for the periodic potential of figure 6. we denote it by f (ε). only energies for which f (ε) ∈ [−1.6. after some algebra.Potential wells and tunneling 197 Figure 6. Figure 6. If the length L of the well in figure 6. which is not periodic and where the walls are infinite. the right-hand side of (6. have a vanishing determinant. In a periodic potential. 2αβ (6.5.58) is a function of particle energy ε only. for instance. The function f (ε) is plotted in figure 6. the function ε(k) is only aproximately quadratic and there are now jumps at integral values of |ka/π|. In the potential well on the left-hand side of figure 6. The quasi-sinusoidal variation of the right-hand side of equation (6.7 is the basis for explaining why electrons in crystals are arranged in energy bands. the energy ε rises quadratically with the wavenumber.5 with U = 100. those for which A = B = C = D does not equal zero. U of the potential in figure 6.04. Because of the cosine term on the left-hand side of (6.58) For fixed values a. the case when ε 4. ε = k2 . The condition Det = 0. there will be a continuous spectrum as depicted by the continuous line in figure 6.3. a = 2.6 for a specific set of values U. b. . Energies for which | f (ε)| > 1 are forbidden. This is.7. a. leads to cos ka = β 2 − α2 sinh βc · sin αb + cosh βc · cos αb. 1] are permitted and those with | f (ε)| > 1 are forbidden. Non-trivial solutions of this linear system of equations.

The particle is now not localized in either of them and its wavefunction is a symmetric and anti-symmetric combination of ψ+ and ψ− . there is a certain chance of tunneling from one well to the other. ψ+ ψ− ψ1 = √ + √ 2 2 ψ+ ψ− ψ2 = √ − √ . See also figure 6.4.or right-hand well. respectively. b]. where U (x) ≤ ε.1 Splitting of energy levels Next we consider an atom of mass m and energy ε in a potential U (x) consisting of two symmetric adjacent wells with a barrier U0 between them as depicted in figure 6. The function ε(k) has discontinuties at integer values of ka/π. and satisfy the Schr¨ dinger equation (see (6. (6. a = 2 and c = 0. Classical motion is allowed in the intervals [−b.5 calculated for U = 100. 2 2 .18)): ψ+ + ε − U (x) ψ+ = 0 ψ− + ε − U (x) ψ− = 0.6. The curve without discontinuties is explained in the text.4. The energy spectrum ε(k) of particles in a periodic potential as in figure 6.59) If the barrier U0 is finite.4. −a] and [a.4 The double potential well 6. the particle can only be in the left.198 The radiative transition probability Figure 6.8. The corresponding wavefunctions are denoted by ψ− and o ψ+ .04.7. When the barrier U0 is infinitely large. 6.

o ψ1 + ε1 − U (x) ψ1 = 0 ψ2 + ε2 − U (x) ψ2 = 0 (6. ν0 is the time for the particle to go from a to b and back. one finds ψ+ (0) = p0 ψ+ (0) = ν0 m 1 exp − p0 a 0 | p| d x −1 where ν0 is the classical oscillation frequency. Likewise we such that ψ1 find ε2 − ε. The wavefunctions ψ1 and ψ2 also obey the Schr¨ dinger equations. Exploiting ψ1 (0) = ψ+ (0). an energy level ε splits into two with a separation ε given by (6. therefore.8. So the energy difference is determined by the wavefunction ψ+ in the √ ε − U (x) denotes the classically forbidden region at x = 0. and integrate from x = 0 to x = ∞. √ 4mν0 1 a 4mν0 2a 2mV0 ε2 − ε1 = . i. (6.60) by ψ+ . Because of tunneling. If p = momentum and p0 its value at x = 0.59) by ψ1 and the left equation in (6.61) exp − | p| d x exp − −a .e. subtract the two products.60) but with slightly different energies ε1 and ε2 . we multiply the left equation in (6. To obtain the difference ε2 − ε1 . and.Potential wells and tunneling 199 U(x) U0 ε −b −a a b x Figure 6. ψ1 (0) = 0 and assuming that the tunneling probability is small ψ+ for x > 0 gives ε1 − ε = −21/2ψ+ (0)ψ+ (0). and finally get ε2 − ε1 = ψ+ (0) · ψ+ (0). A particle in a double potential well with a barrier U0 between the minima. The probability of finding the particle either in the left or right well is unity and the integral of |ψi |2 over the whole x-axis equals one.61).

although only in the forbidden region but there they couple differently to the potential U (x) resulting in the splitting of the energy ε into two levels ε1 and ε2 .2 Tunneling time To find the approximate time ttun that the particle needs to tunnel from the left to the right well or back.62) we see that with increasing potential barrier V0 . At time t = 0.200 The radiative transition probability The possibility that the particle can tunnel to the adjacent well leads to two wavefunctions. which happens after a tunneling time π . we find √ 2a 2mV0 −1 . however. A broad energy band E. It reaches the left well when (t) = ψ− . the energy splitting E 2 − E 1 becomes smaller and the tunneling time longer. the particle is then not localized anywhere.49)).4. so the particle starts in the right well. From (6.63) ttun ν0 exp . (6.61) and (6. Their space probabilities |ψi |2 are different.61) into (6. we turn to the time-dependent wavefunction (t) = ψ1 e− i E1 t + ψ2 e− i E2 t . Inserting the energy difference of (6.4. 2 Here we have used again E = ( 2 /2m)ε and V = (√ /2m)U (see (6. (6. implies a high mobility. 6. ψ1 and ψ2 .62). A precisely defined energy means that the particle is fixed in one well and does not tunnel. the wavefunction has the value (0) = 2ψ+ .62) ttun = |E 1 − E 2 | This equation expresses the uncertainty principle E · t .

therefore.1. w.5.1.4 to discuss their atomic structure and bonding. Then we make our first serious acquaintance with observations and examine the interstellar extinction curve (section 7. Its interpretation requires a basic knowledge of stellar photometry and the reddening law. In view of the overall cosmic abundance of elements. the atoms are regularly arranged in a lattice which means that the crystal can be built up by periodic repetition of identical cells. c are said to be primitive and generate the primitive cell.1 the way in which atoms are arranged in a solid and in section 7.1 Translational symmetry 7. v.1 Crystal structure 7. The shape of the extinction curve can only be explained if the grains are not of uniform dimension but display a size distribution. b.2 what holds them together. Equipped with a quantitative idea about size and chemical composition of interstellar dust. the environment is exactly the same.Chapter 7 Structure and composition of dust We study in section 7.1. r in the crystal whose difference can be written as r − r = ua + vb + wc (7.1). What kind of distribution and how it can be achieved in grain–grain collisions is sketched in section 7. In a crystal.1 The lattice and the base ˚ Interstellar grains probably contain crystalline domains of sizes 10 –100 A. we present typical and likely grain cross sections based on a reasonable set of optical constants. the extinction curve rules out major dust constituents other than silicate and carbon grains. This is the one with the 201 . b and c such that for any two points r. the vectors a. If any two points r and r with an identical environment are connected through (7.1) with integral u.3). A cell is a parallelepiped defined by three vectors a. proceed in section 7. We. 7.

of the distribution of charge and matter in the cell. the plane by the triplet in round brackets (hkl) where h. even the choice of the primitive cell is not unique. figure 7.e. b.202 Structure and composition of dust smallest possible volume.1. i. k. a description of the three-dimensional cell structure. a point r = xa + yb + zc within the cell is designated by x yz with x. y. i. For a full specification of a crystal. u v w . Among the many possibilities. forces which lead to sharp vibrational resonances observable at infrared wavelengths.1. one uses the following notation: • • If the origin of the coordinate system is at a cell corner. l are the smallest possible integers such that h:k :l = 1 1 1 : : . 0v0 and 00w. 7. Furthermore. intensified. in non-scientific language: crystals are beautiful. The cell structure is called the base.1. and a generally anisotropic response to fields and forces. one needs besides the lattice points ua + vb + wc. axes or planes. For example. whose angles are the least oblique and which best express the symmetry of the lattice. after Miller. The spatial relation between the lattice points and the base is irrelevant. in others it does not. z ≤ 1. the way crystals are polarized in an electromagnetic field or stretched under mechanical stress depends on direction. the existence of a grid always implies spatial anisotropy because in various directions the structures are periodic with different spacings. For 11 example. One then denotes. the periodic arrangement of atoms are manifest by the morphological shape of the body. bonding strengths that are variable with direction: in one or two directions. 2 Suppose a plane intercepts the axes of the crystal coordinate system at the three lattice points u00. There is an infinite number of ways how to construct cells or vector triplets (a. 7.3 gives a two-dimensional illustration.1. the cell center is always at 1 2 2 . by • • • the appearance of long-range. among others. c).3 Miller’s and other indices To find one’s way around a crystal and to identify points.e. a crystal cleaves well. one prefers those cells whose sides are small. So lattice points may or may not coincide with atomic centers. A primitive cell may contain many atoms and have a complicated base but it always contains only one lattice point.2 Physical consequences of crystalline symmetry Geometrically. The physical relevance of crystalline order reveals itself. elementary or unit cells. These are called conventional.

a = b = c. β = 90◦ = γ = 90◦ = γ = 90◦ = 90◦ . the lengths of the vectors a. furthermore. 7. See figure 7. z are integers. one obtains grids of higher symmetry.1 Highest symmetry • For instance. b.Crystal structure 203 Table 7. v.5). 3.1. c are orthogonal and their lengths are equal. γ = 120◦ = γ = 90◦ = γ = 90◦ Lowest symmetry Not shown in figure 7. only seven of them are primitive. or when angles have special values. There are 14 basic kinds of translation lattices (Bravais lattices) which correspond to 14 elementary cells. 2 A direction within the lattice may be specified by an arrow that goes from the origin of the coordinate system to the point r = xa + yb + zc where x. 040 and 002. The cells of the triclinic and rhombohedral system are always primitive. k = v and l = w. .1 Bravais cells and crystal systems In the general case.1. whereas for other crystal systems. 4. this is shown in figure 7. For the cubic crystal system. These 14 lattice types can be grouped into seven crystal systems which are described in table 7. 7. There is an infinite number of such vectors pointing in the same direction.1 and shown in figure 7. orthorhombic. let ua + vb + wc be the one with the smallest integers u.1 for definition of sides and angles. y. 2. w. The monoclinic. like 90◦. primitive cells do not have a cubic shape (see the two-dimensional analog in figure 7. 1. Lengths and angles in the unit cells of the seven crystal systems. c in the conventional cell and the angles between them are arbitrary. Altogether there are 14 lattice types or Bravais cells. b.2. the direction [uvw] is always perpendicular to the plane (hkl) with h = u.2 shows the three lattice types of the cubic family: simple (sc). 5. When lengths or angles are equal. Triclinic: Monoclinic: Orthorhombic: Tetragonal: Hexagonal: Rhombohedral: Cubic: a a a a a a a =b=c =b=c =b=c =b=c =b=c =b=c =b=c α α α α α α α =β =γ =β =β =β =β =β =γ = 90◦ . The cubic system has the highest symmetry: a. The polarizability is then isotropic.2 Lattice types 7. The direction is then denoted by the triplet [uvw] in square brackets. body centred (bcc) or f ace centred (fcc) of which only the sc cell is primitive. 6. Figure 7. without centring.1 Not shown in figure 7.1. it depends on direction. tetragonal and cubic system are further split into lattice types that are either primitive or centered. The plane (200) is parallel to (100) but cuts the a-axis at 1 a. The bcc cell is two times bigger than the primitive cell and the fcc cell four times. Among them.1. A plane that is parallel to b and c and cuts only the a-axis at a distance a from the coordinate center is identified by (100). In the cubic system (see later).2. the plane (436) runs through the points 300.

b. for the fcc lattice the face-centring atoms are shown only on the three faces directed towards the observer. To avoid confusion.2. A linear translation by one of the vectors a.1. c) is denoted α.2 Microscopic and macroscopic symmetry Under a symmetry operation.2.1. b). a) and γ = (a. we write α = (b. symbolically. sc bcc fcc Figure 7. See table 7. If. c is one such operation but it becomes ˚ apparent only on an atomic scale because the shifts are of order 1 A and thus . The angle between the vector pair (b. we define likewise β = (c.1.204 Structure and composition of dust c c a triclinic c b a b a b orthorhombic tetragonal c a cubic c b b a hexagonal Figure 7. a crystal lattice is mapped onto itself. There are seven crystal systems of which five are shown here. 7. c). body-centered cubic (bcc) and face-centered cubic (fcc). The three lattices types of the cubic crystal system: simple cubic (sc).

other angles are possible. Macroscopically. besides the trivial 360◦ . • In each of these symmetry operations. the result of the microscopic structure.3). γ = 90◦ ). hexagonal (a = b. including two further symmetry elements (screw axis and glide plane). • • Inversion at a center. These 32 classes completely define the morphological (macroscopic) appearance of crystals.4). of course. The translational symmetry is revealed from X-ray images and is described by the Bravais lattices. there are symmetry operations that also are evident on a macroscopic scale. Now there are only five (translational) Bravais lattices and 10 crystal classes. like a gem in an ornament or the grains in the salt shaker. respectively). although the latter is not explicitly used in the derivation of the 32 crystal classes. γ = 90◦ . figure 7. 90◦ . All crystals can be divided into 32 classes where each class represents a mathematical group whose elements are the symmetry operations described earlier. γ = 90◦ . 120◦ and 180◦. mirror plane and glide plane). When one subjects it to a symmetry operation. Without proof. 7. and centered rectangular (a = b. b . Reflection at a plane. there is a symmetry element that stays fixed in space (the inversion center. The basic unit of a two-dimensional crystal is a parallelogram. One can readily work out that the only possible angles compatible with translational symmetry.5). the operation may be symbolized by r = −r . If the center is in the origin of the coordinate system. γ = 120◦). However. The parallelograms of Bravais lattices may be • • • • • oblique (a = b. Macroscopic symmetry is. the normals of the faces of the polyhedron do not change their directions. when the 14 Bravais lattices of the atomic world are combined with these symmetry elements.2. When the Bravais lattices are combined with the symmetry elements permitted for a plane (rotation axis. Any two vectors a . γ = 90◦ .1. rectangular (a = b. one arrives at 230 space groups. the rotation axis or the mirror plane. such as the following ones.Crystal structure 205 much smaller than the dimensions of a real (even an interstellar) crystal.3 Two-dimensional lattices The classification scheme is much simpler and easier to understand in twodimensions. figure 7. Rotation about an axis. However. quadratic (a = b. for a single molecule (not a crystal). are 60◦ . too. a crystal has the shape of a polyhedron. b with lengths a = b and angle γ between them describe a centered cell of a rectangular lattice but the cell with vectors a. one obtains 17 planar space groups. figure 7.


Structure and composition of dust


a a b1 a γ b3 b2


Figure 7.3. A two-dimensional lattice. There is an infinite number of ways to generate cells. Here it is done by the vector pairs (a, bi ) with i = 1, 2, 3, 4. All define primitive cells, except a, b2 which has an area twice as large.


Figure 7.4. A simple rectangular lattice is generated by the primitive vectors a and b.







Figure 7.5. In a centered rectangular lattice, a primitive cell is generated by a, b1 and has half the area of a conventional cell generated by a, b. Note that a , b with a = b are also primitive vectors.

as in figure 7.5 seems to better convey to us the symmetry. Likewise in three dimensions, the body-centered (bcc) and face-centered (fcc) cubic type can be generated from primitive rhombohedra (where all three lattice constants are equal, a = b = c) but then again, one loses the advantage of a rectangular coordinate system because the angles are no longer 90◦ .

Binding in crystals
7.1.3 The reciprocal lattice


Because the structure in a crystal is periodic, it is natural to expand the density ρ(x) in a Fourier series. If a, b, c form the primitive vectors of the lattice according to (7.1), then ρG eiG·x . (7.2) ρ(x) =

The ρG are the Fourier coefficients and the sum extends over all vectors G = ua∗ + vb∗ + wc∗ (7.3)

where u, v, w are integers and a∗ , b∗ , c∗ are the primitive vectors of the reciprocal lattice. The first, a∗ , is defined by a∗ = 2π b×c a · (b × c) (7.4)

so that a∗ is perpendicular to b and c and a∗ · a = 2π. The unit of a∗ is one over length. There are corresponding definitions and relations for b∗ and c∗ . Suppose a plane wave of wavevector k falls on a crystalline grain. Any small subvolume in the grain, let it be at locus x, scatters the incoming light. The amplitude of the scattered wave is proportional to the electron density ρ(x). Two outgoing beams with the same wavenumber k which are scattered by subvolumes that are a distance r apart have a phase difference ei k·r with k = k − k. The vectors k and k have the same length but point in different directions. To find the total scattered amplitude E s , one must integrate over the whole grain volume V , so modulo some constant factor Es =



d V.

With respect to an arbitrary direction k , the integral usually vanishes as a result of destructive interference. It does, however, not vanish if k is equal to a reciprocal lattice vector G of (7.3), i.e. k = k − k = G. (7.5)

Equation (7.5) is another way of formulating the Bragg law for reflection in crystals.

7.2 Binding in crystals
Solids are held together through electrostatic forces. Five basic types of bonds exist and they are discussed here. All five types are realized in cosmic dust.


Structure and composition of dust

7.2.1 Covalent bonding In a covalent or homopolar bond between atoms, a pair of electrons of antiparallel spin, one from each atom, is shared. The more the orbitals of the two valence electrons in the pair overlap, the stronger the atoms are tied together. The bonding is directional and follows the distribution of the electron density which has a significant high concentration between the atoms. One finds covalent binding in solids and in molecules. The wavefunctions in covalent bonding are not extended and only nearest neighbors interact. We sketch the simplest case which is presented by the H+ 2 molecule, although it contains only one electron, and not an electron pair. A stable configuration with two protons and one electron is possible only if the protons have the right distance and the electron cloud is properly placed between them. The motion of the protons can be neglected for the moment; it may be included afterwards as a correction and results in quantized vibrational and rotational ˆ states. The Hamiltonian H of the H+ -molecule then represents the kinetic energy 2 of the electron (mass m, momentum p) and the electrostatic potentials among the three particles (figure 7.3), p2 ˆ + e2 H= 2m 1 1 1 − − R R1 R2 . (7.6)

When one assumes the distance R between the protons to be fixed, the ˆ ˆ ˆ Hamiltonian of the electron, He (r), has the property that He (r) = He (−r) with r = (x, y, z). Any eigenfunction ψ is then even or odd if there is no degeneracy, and a linear combination of even and odd functions in the case of degeneracy. One therefore seeks ψ in the form of the symmetric or anti-symmetric functions ψs = ψ1 + ψ2 ψa = ψ1 − ψ2 .

Here ψ1 , ψ2 are the usual hydrogenic wavefunctions with respect to proton 1 and 2 with energies E 1 = E 2 . The electron is either at proton 1 or 2, one cannot say at which. When the distance R between the protons is very large, there is degeneracy and the system has zero binding energy (E b = 0). For smaller R, the degeneracy is removed. The electron is then in a double potential well and this leads to a splitting of energy levels. The analysis is completely analogous to the discussion of the double well in section 6.4.4. There is a binding (ψs ) and an unbinding state (ψa ) and their separation increases as R becomes smaller. The lowering of the potential energy as a function of R is [Lan74] E = 4χH Re−R−1 when R is expressed in Bohr radii, a0 , and χH denotes the ionization potential of hydrogen. The qualitative reason for the existence of a binding state (of lower energy) follows from the uncertainty principle: a hydrogen atom in its 1s ground state has a radius x = a0 and a momentum p = m e c/137, so x p = . The

Binding in crystals



e− R2 R1


Figure 7.6. A sketch of the configuration between the two protons and one electron in the H+ -molecule. The protons are labeled 1 (right) and 2 (left). 2

binding energy of the hydrogen atom is E b = χH = p 2 /2m e . If the electron can spread out and be at two protons, as in the molecule H+ , more space is available 2 to the electron and its momentum and the binding energy will be lower. The electronic wavefunctions and the electron energies are schemtically drawn in figure 7.7 for intermediate R. When one includes the repulsion among the protons and considers the full Hamiltonian (7.6), the binding energy of the H+ molecule has its minimum E b = 2.65 eV at a distance of about two Bohr 2 radii. This is the stable configuration of the molecule and E b is the dissociation energy. When R is further reduced towards zero, repulsion between the protons dominates and the system is again unstable. 7.2.2 Ionic bonding In ionic or heteropolar bonds, the adhesion is due to long-range Coulomb forces. In a crystal, the ions of opposite sign attract and those of the same sign repel each other but in the end, attraction wins. At short distances, there is an additional kind of repulsion but it acts only between adjacent atoms. It results from Pauli’s exclusion principle which restricts the overlap of electron clouds. Let us take a sodium chloride crystal (NaCl) as an example of ionic bonding. One partner, the alkali metal, is relatively easy to ionize (5.14 eV) and the other, the halogen Cl, has a high electron affinity (3.61 eV). Although the transfer of the electron from Na to Cl requires 5.14 − 3.61 = 1.53 eV, the Na+ and Cl− ions can supply this deficit. They can provide even more than that in the form of potential ˚ energy by coming close together. Indeed, at the equilibrium distance of 2.8 A, the binding energy per molecule equals 6.4 eV. As the ions of heteropolar bonds have acquired full electron shells, like inert gases, their electron clouds are fairly spherical and the charge distribution between the ions is low. Ionic bonding is, therefore, non-directional and usually


Structure and composition of dust


binding energy E1 anti−binding



Figure 7.7. Left: A sketch of the wavefunction ψ in the H+ molecule. It may be either 2 anti-symmetric (ψa ) and anti-binding, then the electron density is zero midway between the protons, or it may be symmetric (ψs ) and binding, then the electron density is enhanced at x = 0 reducing the repulsive potential between the protons. Right: As the protons come closer, the degeneracy is removed and the atomic levels split into an unbinding and binding state.

found only in solids. To estimate the binding energy of an ionic crystal, we write the potential ϕi at the locus of atom i resulting from all other atoms j in the form ϕi =
j =i

±q 2 + ri j

λe−ri j /ρ .
j =i


In the first term representing the interaction between ions of charge q, the plus sign refers to like, the minus sign to unlike charges and ri j is the distance between them. The second term heuristically describes the repulsive potential at very close distance. Because its range ρ is only about one-tenth of the equilibrium separation req between adjacent atoms, one can replace the second sum by nλe−r/ρ , where n is the number of nearest neighbors and r the nearestneighbor separation. The constants λ and ρ have to be determined from quantum mechanical calculations or from experiment. We put pi j = ri j /r and introduce the Madelung constant (±) A= . (7.8) pi j
j =i

If atom i is negative, there is a plus sign when atom j is positive and a minus sign when it is negative. For a crystal of N ion pairs, the total potential energy becomes q2 −r/ρ − A . tot = Nϕi = N nλe r

Binding in crystals
The equilibrium distance req follows from d
tot tot /dr


= 0, which yields


Nq 2 A req


ρ req


The Madelung constant A is evaluated in special routines. It must obviously be positive as tot is negative; for NaCl, A = 1.748. Pure ionic bonding is, however, an idealization; one finds it neither in molecules nor in crystals. Even when compounding alkali metals with halogenes, there is some homopolar component or overlap of orbitals. In the extreme case of CsF, homopolar binding still amounts to 5%. There are also no pure covalent bonds, even in H2 or diamond, because of the fluctuations of the electron clouds. In the real world, there are only blends between the two bonding types. 7.2.3 Metals Another type of strong binding exists in metals. One may imagine a metal as a lattice of positive ions bathed in a sea of free electrons; metallic binding is therefore non-directional. Sodium As as an example, we consider the alkali metal sodium (Na), the 11th element of the Periodic Table with an electron configuration 1s2 2s2 2p6 3s1 . The two inner shells are complete and the 3s valence electron is responsible for binding. In sodium metal, the 3s electrons form the conduction band. In an isolated sodium atom, the 3s electron is bound to the Na+ ion by an ionization energy χNa = 5.14 eV. The electron is localized near the atomic nucleus and its wavefunction ψ goes to zero at large distances from the nucleus. When, however, the 3s electron is in the potential of a lattice of Na+ -ions, as in a metal, its wavefunction ψ is very extended and subject to a different boundary condition: symmetry requires that the gradient of the wavefunction vanishes midway between atoms. The quantum mechanical calculations now yield a binding energy of the electron to the ionic lattice of 8.2 eV. Because sodium has a density of 1.01 g cm−3 , the concentration of 3s electrons is 2.65 × 1022 cm−3 and the Fermi energy after (6.54) E F = 3.23 eV. From (6.54) also follows the density of states, ρ(E) = V dN = dE 2π 2 2m e
2 3/2

E 1/2 .

The average energy E of a conduction electron, is therefore, E = N −1 ρ(E) d E = 3 E F 5 1.9 eV.

Consequently, in a metal each sodium atom is bound by 8.2 − χNa − 1.9 1.16 eV.


Structure and composition of dust Conductivity and heat capacity The free and mobile electrons of a metal are responsible for the high electric conductivity. In a constant electric field E, they are accelerated but collide after an average time τ with phonons (thermally oscillating atoms). The resulting current density J is equal to the charge density ne, where n is the number of free electrons per cm3 , multiplied by their drift velocity v. An electron experiences an acceleration eE/m e over an average time τ , so that v = τ eE/m e . Because Ohm’s law (1.112) asserts J = σ E, the conductivity σ is σ = ne2 τ. me

At very low temperatures, collisions of the electrons with lattice imperfections are more important than collisions with phonons. Because of free electrons, the heat capacity has a peculiar behavior at very low temperatures. Below the Debye temperature, which is typically several hundred K and thus not low, the specific heat C due to the lattice falls rapidly, like T 3 (see (8.39)). The heat capacity of the electron gas, however, declines less swiftly, it is proportional to T and, therefore, dominates at very low temperatures. Crudely speaking, when there are n free electrons per cm3 , heating of the metal from zero Kelvin to temperature T thermally excites those conduction electrons which lie T degrees below the Fermi temperature TF , altogether a fraction T /TF . Their total mean energy is then U ∼ nkT and, therefore, C= nkT ∂U ∼ ∝ T. ∂T TF

For an exact relation, one has to use the Fermi distribution (5.37). Energy bands and Bloch functions It is a general property of crystals that the electronic states are clustered in bands, with energy gaps between them. The basics of their formation was explained in section 6.4.3 on one-dimensional periodic potentials. To derive the energy states and eigenfunctions of an electron in the three-dimensional potential U (x), we first note that as U (x) comes from the lattice ions and is strictly periodic, it may, from (7.2), be written as UG eiG·x (7.9) U (x) =

where the G are the reciprocal lattice vectors of (7.3). The general solution for the wavefunction ψ(x) is sought in a Fourier expansion, ψ(x) =

Ck eik·x


Binding in crystals


where the wavenumber k fulfils the periodic boundary conditions and is connected to the momentum p by p = k. o The coefficients Ck are found by inserting (7.10) into the Schr¨ dinger equation: ˆ p2 ˆ ˆ + U (x) ψ(x) = Eψ(x) (7.11) H ψ(x) = 2m e and solving the resulting set of algebraic equations. In (7.11), the interaction among electrons is neglected. After some algebra, one is led to Bloch’s theorem, namely that the eigenfunction ψk (x) for a given wavenumber k and energy E k must have the form ψk (x) = u k (x) eik·x where u k (x) is a function periodic with the crystal lattice. So if T is a translation vector as given in (7.1), then u k (x) = u k (x + T). The wavefunction ψk (x) itself does not generally display the periodicity. It is straight forward to prove Bloch’s theorem in a restricted form, for a onedimensional ring of N atoms, where after N steps, each of the size of the grid constant, one is back to the starting position. It is instructive to consider the formation of bands from the point of view of the tightly bound inner electrons of an atom. When we imagine compounding a solid by bringing N free atoms close together, the inner electrons are afterwards still tightly bound to their atomic nuclei, only their orbits have become disturbed. Because of the disturbance, a particular energy state E i of an isolated atom will be split into a band of N substates due to the interaction with the other N − 1 atoms. The shorter the internuclear distance, the stronger the overlap of the wavefunctions and the broader the band; the energetically deeper the electrons, the narrower the band. There will be a 1s, 2s, 2p, . . . band. One can determine the energy states of a tightly bound electron by assuming that its wavefunction is a linear superposition of the atomic eigenfunctions. In this way, one finds the width of the band and the average energy in a band. The latter is lower than the corresponding energy in a free atom, which implies binding of the crystal. 7.2.4 van der Waals forces and hydrogen bridges The van der Waals interaction is much weaker than the bonding types discussed earlier and is typically only 0.1 eV per atom. It arises between neutral atoms or molecules and is due to their dipole moments. The dipoles have a potential U that rapidly falls with distance (U ∝ r −6 , see section 9.1).


Structure and composition of dust

If the molecules do not have a permanent dipole moment, like CO2 , the momentary quantum mechanical fluctuations of the electron cloud in the neighboring atoms induces one; its strength depends on the polarizability of the molecules. This induced dipole moment is usually more important even for polar molecules. Its attraction holds CO2 ice together and binds the sheets in graphite. Hydrogen can form bridges between the strongly electron-negative atoms N, O or F; the binding energy is again of order ∼0.1 eV per atom. For example, when a hydrogen atom is covalently bound to oxygen, it carries a positive charge because its electron has been mostly transferred to the oxygen atom. The remaining proton can thus attract another negatively charged O atom. The double helix of the DNA is bound this way but so also is water ice. In solid H2 O, every oxygen atom is symmetrically surrounded by four others with hydrogen bonds between them. Because of these bonds, the melting and evaporation point of water is unusually high compared, for instance, to H2 S, which otherwise should be similar as sulphur stands in the same Group VI of the Periodic Table right below oxygen. This thermodynamic peculiarity of water is crucial for the existence of life.

7.3 Reddening by interstellar grains
Our knowledge of interstellar dust comes from the following measurements: • • • • • Interstellar extinction. The interstellar extinction or reddening curve specifies how dust weakens the light from background stars as a function of wavelength. Dust emission. We receive this from all kinds of objects: protostars, old stars with shells, interstellar clouds and whole galaxies. Infrared resonances. These are features observed in absorption and emission which allow us to identify the chemical components of the dust material. Polarization. It refers to light from stars behind dust clouds but also to scattered radiation and dust emission. Scattered starlight. Examples are reflection nebulae, the diffuse galactic light or the zodiacal light of the solar system.

The items in italics are our five observational pillars. In this section, we are concerned with interstellar extinction and the clues it provides to dust properties, the other pillars are discussed elsewhere. We first describe the principles of stellar photometry because it is the basis for the reddening curve. 7.3.1 Stellar photometry The interstellar extinction curve is obtained from stellar photometry. The wavelength resolution in such measurements is usually poor, typically λ/ λ 10 but the broadness of the observational bands enhances the sensitivity. The standard photometric system is due to H L Johnson and W W Morgan and rooted

Reddening by interstellar grains


Table 7.2. The standard photometric system. The center wavelengths λc and the conversion factors wλ between magnitudes and Jansky after (7.14) are averages gleaned from the literature. The last column gives the conversion factors for a blackbody of 9500 K, see text. Band U B V R I J H K L M N Q Historical meaning Ultraviolet Blue Visual Red Infrared λc (µm) 0.365 0.44 0.55 0.70 0.90 1.25 1.65 2.2 3.7 4.8 10 20 wλ 1810 4260 3640 3100 2500 1635 1090 665 277 164 37 10 wλ (bb) 2486 2927 3084 2855 2364 1635 1116 715 294 184 46 12

in optical astronomy. It was later expanded into the infrared; there the choice of wavelengths was dictated by the transmission of the atmosphere of the Earth. The observational bands are designated by letters: U, B, V, R, I, . . . . Table 7.2 lists their approximate center wavelengths λc and the conversion factors, wλ , for translating magnitudes into Jansky and back (see (7.14)). The precise effective observation wavelength follows only after convolving the spectrum of the source with the transmission of the instrument and the atmosphere. In photometry, especially at shorter wavelengths, it is customary to express the brightness of an object in apparent magnitudes. These are logarithmic quantities and they were appropriate units in the days when the human eye was the only detector because according to the psycho-physical rule of W Weber and G T Fechner: the subjective impression of the eye changes proportionally to the logarithm of the physical flux. But to the pride of many astronomers, magnitudes are still in use today. As they are laden with five millenia of history, only he who has thoroughly studied the 5000 year period can fully appreciate their scientific depth. An exhaustive explanation of magnitudes is, therefore, beyond the scope of this book. The brightest star of all, Sirius, has a visual apparent magnitude m V = −1.58 mag. Capella, Rigel and Vega are around zeroth and Spica (α Virginis) is only of first magnitude. Generally, a step of one up in magnitude means a factor of 2.5 down in observed flux. Whereas the apparent magnitude m λ at wavelength


Structure and composition of dust

λ depends on the stellar distance D, the absolute magnitude Mλ , defined by m λ − Mλ = 5 log10 D pc −5 (7.12)

does not and thus measures the intrinsic brightness of a star. At a distance of 10 pc, the apparent and absolute magnitude are, by definition, equal. When there is intervening dust, one adds, to the right-hand side of equation (7.12) a term Aλ to account for the weakening of starlight through interstellar extinction: m λ − Mλ = 5 log10 Aλ is related to the optical depth τλ through Aλ = 1.086τλ. The strange conversion factor of 1.086 between the two quantities arises from their definitions: stellar magnitudes are scaled by a factor of 2.5, the optical depth by e 2.718. The term m − M in (7.12) is called the distance modulus. So ingenious astronomers have contrived a means to express length in stellar magnitudes. To convert a flux Fλ in Jy into an apparent magnitude m λ or vice versa, we use the relation wλ . (7.14) m λ = 2.5 log10 Fλ Fλ is expressed in Jy, wλ is given in table 7.2. The formula is simple, the difficulty lies in the calibration factors wλ . The values in the literature scatter by about 10% and, furthermore, they do not refer to identical center wavelengths because of the use of different filters. The magnitude difference between two bands is called the color. It is equivalent to the flux ratio at the corresponding wavelengths and thus determines the gradient in the spectral energy distribution. By definition, the colors of main sequence A0 stars are zero. The most famous A0V star is α Lyr (= Vega, D = 8.1 pc, m V = 0.03 mag, L = 54 L ). Such stars have an effective temperature of about 9500 K and they can be approximated in the infrared by a blackbody. Indeed, if we calibrate the emission of a blackbody at 9500 K in the J band and put all colors to zero, we find the conversion factors wλ (bb) listed in table 7.2. In the infrared, they are not very different from wλ (adjacent column); however, the discrepancies are large at U, B and V because of the many absorption lines in the stellar spectrum that depress the emission relative to a blackbody. 7.3.2 The interstellar extinction curve The standard color excess E B−V The interstellar extinction curve is obtained from photometry at various wavelengths λ on two stars of identical spectral type and luminosity class, one D pc − 5 + Aλ . (7.13)

7.2 How the extinction curve is defined Fixing the wavelength λ in the color excess E(λ. Ext(B) = 1 Ext(V ) = 0. E B−V AB − AV τB − τV (7.55 µm.16) Aλ is the extinction in magnitudes at wavelength λ. the center wavelengths of the B and V band.55 µm and normalizing E(λ. λ ) at 0. one receives from the stars the flux Fi (λ) = L(λ) 4π Di2 · e−τi (λ) i = 1. Obviously.Reddening by interstellar grains 217 of which is reddened (star No. . The normalization of the extinction curve by E B−V is important because it allows a comparison of the wavelength dependence of extinction towards stars which suffer different amounts of reddening. gives the color excess m(λ) − m(λ ) = −[τ1 (λ) − τ1 (λ )] E(λ.2. λ and λ . λ ) = which no longer contains the distance D. 1). whereas the other is not (star No. m i (λ) = ln L(λ) − τi (λ) + 2 ln Di + constant and because τ2 = 0. L(λ) and τi (λ) are the spectral luminosity and optical thickness. V ) and called the standard color excess.15) E B−V = (B − V ) − (B − V )0 and likewise for any other pair of wavelengths. the color excess is denoted by E B−V = E(B. 2. it is the result of absorption plus scattering but observationally one cannot separate the two. The intrinsic color is denoted (B − V )0 in contrast to the observed color B − V . Because the apparent photometric magnitude m is a logarithmic derivative of Fi (λ). which includes the effect of the selective weakening by interstellar dust. V = m V or B = m B . The band symbol is also used for the apparent magnitude. V ) = = .3. m at two wavelengths. 2). the difference in magnitude becomes m(λ) = −τ1 (λ) − 2 ln The difference in m(λ) between the two stars D1 D2 . for example. we can write (7. without dust emission along the line of sight. With this notation. Ext(λ) = Aλ − AV τλ − τV E(λ. At λ = 0.44 µm and λ = 0. In the case of pure extinction. one arrives at the extinction curve in its traditional form. V ) by E B−V .

218 Structure and composition of dust The quantity RV = AV = − Ext(λ = ∞) E B−V (7.3 Remarks on the reddening curve The interstellar extinction curve (figure 7. Aλ /AJ (the J band is at 1. If one plots instead of Ext(λ) the ratio τλ /τV then.55 µm.1 to 10 µm. moreover.17). one measures E B−V . The extinction curve refers to broad-band observations but nevertheless it is remarkably smooth.2.25 µm) or (Aλ − AJ )/(AV − AJ ). the smaller ones toward high latitude clouds. deviations occur only in the UV. all extinction curves look alike for λ ≥ 0. the larger values being found towards molecular. even in the infrared. they can differ substantially.2 µm however. The reduction in UV extinction for large RV suggests that the small grains have disappeared.6 µm−1 ˚ or 2175 A which is always well fit by a Drude profile. RV . are sometimes more suitable for displaying certain trends. Photometric data exist only for cloud edges. The reddening curve displays the following salient features: • • In the diffuse interstellar medium. which also appears in the distance modulus of (7. the optical depth is small and the extinction of background light weak.13). Ext(λ) of equation (7. assumes an RV value and then uses (7. Still other forms. one gets τλ Ext(λ) = +1 τV RV (7. There is also scatter 1 ± 0. The observed variations of RV range from 2. for example. Towards clouds. however. 7. Because K λ /K V = τλ /τV . The only resonance in the extinction curve is the broad bump at 4. containing the same information. −1 −1 ). It is not a directly observable quantity. At longer wavelengths. its width varies ( λ • • . The central position of the bump stays constant from one star to another to better than 1% implying ˚ that the underlying particles are in the Rayleigh limit (sizes < 100 A). is the most common quantity to characterize the opaqueness of a cloud. The curves are largely (but not totally) fixed over the entire wavelength range by the one parameter RV of equation (7. at shorter wavelengths.16) is.3. the dust begins to emit itself.17).1.1 to well over 5.8) is observationally determined from about 0.18) τλ /τV and Ext(λ) are mathematically equivalent. the shape of the curve Ext(λ) varies from one source to another. to first order. One can also express the reddening by dust through the mass extinction coefficient K λ normalized at the V band. simply. The visual extinction in magnitudes AV .17) is called the ratio of total over selective extinction or. To obtain AV from photometry. at optical and infrared wavelengths. fairly uniform over all directions in the sky and RV equals 3. their cores are too obscured.

9 µm. the extinction curve rises continuously. The length of the shift is proportional to the foreground AV and in figure 7. Beyond 2175 A. one could subtract scattering from extinction to obtain a pure absorption curve. As an example of a TCD.3 Two-color diagrams In two-colour-diagrams (TCDs). If one had good albedo measurements (which one does not have).3. say. There are altogether more than a hundred. The characteristic wiggle in the main sequence line in figure 7.4 and 0.8). So compared to a blackbody of the same color B–V. It is correlated with RV in the sense that the resonance is weak in clouds where RV is large. with RV = 5. This should happen several times during their lifetime as they pass from dense into diffuse medium and back. it greatly supresses the stellar flux. a stellar photosphere is much weaker at U .9.8 suggest that the grains are modified when they change their environment.1) shifts any star in the UBV diagram from the main sequence in the direction parallel to the arrow of figure 7. In the far UV (λ−1 > 6 µm−1 ). the curve rises again.Reddening by interstellar grains 219 • • • in the strength of the bump. This would provide a touchstone for any dust model. The reason for this behavior is that photoionization of the n = 2 level in hydrogen falls into the U-band. All luminous stars of type B0 or earlier cluster in a tight strip with B–V between −0.6). The latter would imply fluffy particles built up of smaller subunits (section 2. of variable ˚ strength.30 and −0. we show in figure 7. the identification of the DIBs is still unclear. stellar light is reddened because extinction ˚ decreases with wavelength. shape and width. As this is the dominant process for the photospheric opacity at this wavelength. it corresponds to AV = 5 mag. the slope and length of the arrow would be different .9. one plots one color against another. The position in the UBV diagram of blackbodies with temperatures from 3500 to 40 000 K is always well above the main sequence. The two extinction curves in figure 7. one finds superposed on the extinction curve diffuse interstellar bands (DIBs but they are not shown in figure 7.9 the UBV diagram of unreddened main sequence stars. TCDs are a simple and efficient tool with which to separate distinct or identify similar astrophysical objects. the broadest being about 2 A wide.32 mag. The modifications may be due to energetic photons. ˚ From the infrared to the hump at 2175 A. Over this range. Interstellar reddening with the standard extinction law (RV = 3. mainly between 0. A plausible explanation is coagulation of dust particles in clouds as a result of which the small grains disappear. In the case of another extinction law. the ratio τλ /τV first declines and here the light is made bluer(!).9 reflects the variation of this supression (called the Balmer decrement) with effective temperature. 7. At higher spectral resolution. condensation of gas atoms or collisions among grains. Although discovered 80 years ago.

9 and know. commonly defined in the infrared as α= d log(λFλ ) . d log λ (7. RV = 3. we can determine the amount of extinction by moving it up.8.1. Ext(λ) is zero at V and equals one at B. (for the same AV ). If we observe an object somewhere in the right-hand part of figure 7. 7. RV = 5 to the edges of molecular clouds. The observed interstellar extinction curve in the form τλ /τV and as Ext(λ) according to equation (7.220 Structure and composition of dust Figure 7. that it is a main sequence star. This dereddening is unique as long as the star is of type B5 or earlier so that the reddening path does not intersect the wiggle of the main sequence curve. until it reaches the main sequence.16). Whereas τλ /τV is normalized at V.4 Spectral indices Akin to a color is the spectral index.19) . refers to the diffuse medium. parallel to the arrow. for some reason.3. The ratio of total over selective extinction.

Hence. log λ2 − log λ1 (7.8) and A V = 5 mag of foreground extinction. therefore. α is calculated not as a derivative but as the slope of λFλ between two wavelengths λ1 and λ2 : α= log(λ2 Fλ2 ) − log(λ1 Fλ1 ) . The arrow shows the direction and distance by which a star is displaced from the main sequence under the standard reddening law with RV = 3.21) . The main sequence locations of a few spectral types are indicated. ln(λ2 /λ1 ) (7.9. quite accurate and may be used to find the position of a main sequence star that suffers a certain amount of extinction A V . the blackbody and reddening curve have been calculated. a spectral index depends on wavelength. see text.20) For a blackbody obscured by a foreground extinction τ (λ).1 (figure 7. in the limit of high temperatures. The figure is. the upper to 40 000 K. UBV diagram for unreddened main sequence stars (thick line) and for blackbodies (thin line). The lower tip of the blackbody line refers to a temperature of 3500 K. α = −3 − τ (λ2 ) − τ (λ1 ) . The data for the main sequence are taken from the literature.Reddening by interstellar grains 221 Figure 7. one has. In practice.

one measures towards a star the standard color excess E B−V and the total hydrogen column density NH to . Because α is sensitive to both T and A V .2 and 5 µm. Notice the different temperature ranges right and left. one also uses the index between 2.10 and 7. The curves show how the spectral index is modified as the foreground extinction increases from 0 to 20 mag and 40 mag. To obtain the absolute value of the cross section per gram of interstellar matter.2 to 20 µm after (7.5 The mass absorption coefficient The reddening curve yields only a normalized extinction coefficient.2 and 20 µm which serves as a classifier for protostellar objects. α of a blackbody approaches −3 at high temperatures. In the right box.20) of a source at temperature T that is observed through a foreground of visual extinction A V . one has to know K V .036 for T → ∞. Figures 7.4).20 or 12. In this case. K λ . K λ /K V . An important example is the spectral index between 2. a star.11 demonstrate how the indices change as a function of source temperature. The spectral index α in the wavelength interval from 2. Of course. to first approximation. The weak dotted lines are drawn to alleviate comparison between the two frames. for standrad dust (see figure 7. for instance. In the left box it is an optically thin dust cloud with the standard mixture of silicate and carbon grains (see section 12. it is used in the classification of protostars. the object is a blackbody or. 7. other wavelength intervals fulfil the same purpose.10.3. For this purpose.10) α → −3 + AV · 0.222 Structure and composition of dust Figure 7. for instance. Without extinction.

This value is easy to remember. (7. we know K λ in the range from about 0.22) As AV = 1. The amount of dust relative to the gas follows then from the condition (7. In combination with the interstellar reddening curve.9 × 10−22 cm2 per H-atom and (7.086 × NH K V . For technical reasons. As figure 7.8 × 1021 cm−2 produces a standard color excess .9 × 1021 AV cm−2 . one finds K V a mass extinction coefficient KV 4.23) also allows us to estimate the dust-to-gas mass ratio Rd if one makes reasonable assumptions about the composition of the grains.Reddening by interstellar grains 223 Figure 7. The average value over a large sample of stars is NH or with (7.10 but for the spectral index α between 2. their shape and sizes.11. Equation (7.2 and 5 µm.1.24) 200 cm2 per g of interstellar matter.17) and RV = 3.8 × 1021 E B−V cm−2 (7. NH 1.22) that a hydrogen column density of 5. The observations are performed in the far UV from a satellite and consist of absorption measurements towards early-type stars in the Lyα line of HI and the Werner and Lyman bands of H2 .23) 5.1 to 5 µm. one chooses stars that are only slightly reddened and located in the diffuse interstellar medium. which both atomic and molecular hydrogen contribute: NH = N(HI) + 2N(H2 ).

4 Carbonaceous grains and silicate grains 7. MgO and FeO. . 1%. The type of dust that is produced depends on the abundance ratio of carbon to oxygen: • If there is less carbon than oxygen. [C]/[O] < 1. cannot be made in the interstellar medium.4). Rd = Mdust Mgas 0. which has a density ρ ∼ 10−3 g cm−3 . • . the dust can only consist of the most abundant elements.2 × 10−5 cm−3 .5).23) and (7.4. as in M-type and in OH/IR stars. respectively. How dusty is the interstellar medium? Very dusty! If the atmosphere of the Earth. carbonaceous grains form. Fe. N stars). Observational evidence points towards the wind of red giants on the asymptotic branch of the Hertzsprung– Russell diagram as the place of their origin (section 9. R. as in C-type stars (these are carbon-rich objects like C. Si.8 µm and 18 µm. mainly silicates form.25) The lower limit may be appropriate for the diffuse interstellar medium. at least their seeds. The observational characteristic of silicates is the presence of an Si–O stretching and an O–Si–O bending mode at 9. the optical depth τ = ρ K V would become unity over a distance of only = 10 cm and we could not see our feet! 7. (7. O. According to equations (7. had the same relative dust content. Considering the uncertainties. so almost all of it resides in grains.224 Structure and composition of dust E B−V of 1 mag. this is in agreement with the dust-to-gas ratio derived from the Kramers–Kronig relation in (2.5 . the total dust volume in a column of 1 cm2 cross section with AV = 1 mag is Vdust ∼ 1. besides minor solid constituents like MgS. quite independently of its particular choice. Because at least half a percent of interstellar matter is in the solid phase. Silicon has a cosmic abundance [Si]/[H] = 3 × 10−5. the higher for dense clouds where grains are ice-coated. tiny refractory nuclei are created in the inner part of the circumstellar envelope which then grow through condensation as they traverse the cooler outer parts (section 9. In such an environment. . or Mg. N.1 Origin of the two major dust constituents Interstellar grains. they can only be modified there or destroyed. At first. Any dust model compatible with the interstellar reddening curve gives.5 g cm−3 .25). If [C]/[O] > 1. like C.95) but the method employed here is probably superior. no carbon is available for the solid phase because all C atoms are locked up in CO which is a diatomic molecule with an exceptionally strong bond. assuming a bulk density of 2. it is heavily depleted in the gas phase (depletion factor f depl (Si) ∼ 30).

3 µm but it is not detected in the interstellar medium. and a depletion factor f depl (C) ∼ 3. or somewhat less.2 eV) but it will be more than returned in the formation of a molecule. Smaller grains are probably prevented from penetrating deep into the solar system by radiation pressure and. this is the chemically relevant case where carbon has four covalent bonds. The promotion requires some energy (4. Al2 O3 ). as the Sun moves relative to the local interstellar cloud (it is of low density. one is in the gas phase. in units of . Generally. Some of them were detected by spacecraft at the distance of Jupiter and identified as belonging to the local interstellar cloud by the direction and value u (∼26 km s−1 ) of their velocity vector [Gr¨ 94]. correspond to µm-sized grains which we think are rather atypical for the interstellar medium. there are only two unpaired p-electrons. if they are charged. Two out of three atoms are built into solids. Because the isotopic abundance ratios are explainable in terms of nucleosynthesis in AGB stars or supernovae. the sixth element of the periodic system. is only seen in C-stars through its emission signature at 11. they have probably undergone considerable reprocessing in the solar nebula. The ground state of carbon is designated 3 P0 . It is a triplet (prefix 3) with total spin S = 1. Their masses. The last substance. so they cannot be brought to Earth and studied in the laboratory.3 cm−3 ). include metal oxides (MgO. these subparticles are hypothesized to be of interstellar origin. In the ground state.Carbonaceous grains and silicate grains 225 Carbon has a cosmic abundance [C]/[H] = 4 × 10−4. In the process of detection they were destroyed.4. by the interplanetary magnetic field. which could also be determined. In interplanetary dust particles and in primitive meteorites which have been collected on Earth. 7. The bonding of carbon to other atoms is of the covalent type and there are two kinds: . one 2s-electron may be promoted to a 2p orbital and then all four electrons in the second shell become unpaired.2 The bonding in carbon Carbon. However. graphite and PAHs. FeO. because of their distance. Other types of grains. one can identify inclusions which distinguish themselves by their isotopic pattern from the solar system. silicon carbide. Fe3 O4 . MgS and SiC. n H ∼ 0. dust particles of this cloud sweep through the solar system. However. has four electrons in its second (n = 2) uncompleted shell: two s-electrons denoted as 2s2 with angular momentum quantum number l = 0 and two unpaired p-electrons (2p2 ) with l = 1. If so. interstellar grains are beyond the reach of spacecraft. There are several forms of carbonaceous grains: mainly amorphous carbon. Carbon and silicates are the major constituents of interstellar dust. for whose existence there are either observational or theoretical indications. total orbital angular momentum L = 1 (letter P) and total (including spin) angular momentum J = 0 (suffix 0).

C6 H6 . But the two remaining 2p orbitals in the carbon atoms. which makes the CH2 groups stiff with respect to rotation (see figure 7. although the symmetry is no longer perfect. see later). So. where the angle from C over C to H equals 120◦. each carbon atom has three sp2 hybrid orbitals connecting through σ -bonds to the adjacent two C atoms and one H atom. HC–CH. Benzene. There are σ -bonds between C–H and C–C. In the identical hybridized sp3 orbitals. overlap to an additional π-bond.15). two adjacent C atoms are coupled by a σ . There is still one unhybridized 2p electron per C atom left. Acetylene is believed to be the basic molecule for the nucleation of carbonaceous grains in the wind of giant stars (see figure 7. There are two more π-bonds. Here are examples: • In methane. the carbon atoms are held together by a triple bond. the wavefunction of the two electrons has a lobe on each side of the internuclear axis. An sp2 hybrid is present in ethylene. The more bonds between two C atoms there are. which is a planar molecule. the C–H binding is through σ -bonds.226 • • Structure and composition of dust In a σ -bond.12). C≡C. likewise there are sp2 and sp hybrids. besides the sp2 hybrids. so altogether we have a C=C double bond (one σ and one π). The carbon atoms are sufficiently close together so that neighboring 2p orbitals can pair to form an additional π-bond. • • • • . is an especially simple and symmetric hydrocarbon (a so called PAH. one gets cos α = − 1 . 3 The situation is similar in ethane. the coupling of carbon is completely symmetric to all four H atoms and the difference between s and p valence electrons has disappeared. H3 C–CH3 . In the linear molecule acetylene. In a π-bond. in total.14). on average. the p-orbitals are perpendicular to the plane.5◦. The combination of one s and three p electrons is denoted as an sp3 hybrid. Each of the two C atoms has again four sp3 hybrid orbitals connecting to the other carbon atom and to the neighboring three hydrogen atoms. H2 C–CH2 . The angle α between the internuclear axes follows from elementary geometry. The s and p wavefunctions can combine linearly resulting in hybrid orbitals. so α = 109. the stronger they are bound and the shorter their internuclear distance (see figure 7. the C atoms are sp hybridized yielding C–C and C–H σ -bonds. The binding is therefore intermediate in strength between a single and a double bond. Because of the σ -bond between the two Cs. CH4 . In this planar ring (aromatic) molecule. the CH3 groups can rotate relative to each other (see figure 7. Because of the complete symmetry of the ring. the unhybridized electrons in the 2p-orbitals are not localized but equally shared (resonant) in the ring. They point from the C atom at the center of a tetrahedron towards the H atoms at its corners. the electron distribution has rotational symmetry about the internuclear axis.and half a π-bond. whose orbitals are perpendicular to each other and to the internuclear axis.13).

4.C2p) σ (Csp. 7. big circles. H1s). small ones.13–7.Carbonaceous grains and silicate grains 227 ethane σ (Csp3 . H atoms. Four types of carbon binding are exemplified in figures 7.13. In the description of the bond. σ (Csp2 .Csp) σ (Csp. As a C atom has only four nearest (and 12 the internuclear axes again 109.H1s) Figure 7. Each C atom is linked tetrahedrally to its four nearest neighbors through sp3 hybrid ˚ orbitals in σ -bonds.H 1s ) .H1s) Figure 7. the bonding is similar to methane.3. the angle between ◦ . π-bonds with broken lines. Here a single bond in ethane (C2 H6 ). σ -bonds are drawn with full lines.Csp 2) π (C2p.3 Carbon compounds 7.54 A. ethylene σ (Csp2 . a crystalline form of carbon. acetylene π (C2p.4.12. for instance.C2p) σ( Figure 7.Csp3 ) σ (Csp3 . C atoms.1 Diamond In diamond.5 Cs p 2 . Triple bond in acetylene (C2 H2 ). The distance between the atoms is 1.14. the type (σ or π) is followed by the atomic symbol and the participating orbitals. Double bond in ethylene (C2 H4 ).15.

.15. 3. This is to be compared with a close-packed structure.42A.16. Resonance structure in benzene (C6 H6 ) with three π-bonds shared among six C atoms. .17). the distance between ˚ neighboring sheets is 3. is possible and combinations thereof.228 Structure and composition of dust benzene σ (Csp .35 A. The side length of the hexagons equals 1. where in both cases each atom has 12 nearest neighbors and the volume filling factor amounts to 74%. Graphite is formed by sheets of carbon atoms. graphite sheets Figure 7.C2p) σ (Csp2 . . . next nearest) neighbors. In analogy to close packed spheres (figure 7. H1s) 2 π (C2p. each with an hexagonal ˚ honeycomb structure. the available space is filled to only 34% assuming that the atoms are hard spheres. also ABCABC . either face-centered cubic or hexagonal close-packed (see figure 7.17).Csp2 ) Figure 7.35A . besides the sheet sequence ABABAB .

. This gives an hexagonal close packed structure (hcp). all other locations are then fixed. (b) When a ball is over the upper cross.3 eV per atom) and due to its overall bonding isotropy. So diamonds are long lasting presents that will outlive the affluent donors as well as the pretty recipients of the gems. The bonding in diamond is strong (7.. is covered by an identical but horizontally shifted layer B (full). They are mined in Kimberlite pipes which are vents associated with volcanoes where the deep material is transported upwards so that it can be reached by man. On Earth. (a) When a ball is centered at the lower cross.17. The upward transport during the volcano eruption occurs sufficiently rapid without a phase transition. however. diamond is used for cutting and drilling.Carbonaceous grains and silicate grains 229 next higher layer B bottom layer A Figure 7. A ground layer A of equal balls (dashed). . the angles between the facets are not . As there are no free electrons. it is metastable in our environment but well separated by a large activation barrier from graphite. which is the standard gem with a circular girdle and 33 facets above and 25 facets below the girdle. one gets by repetition a sequence ABCABC . each touching its six nearest neighbors. one only needs to specify the position of one ball in the new layer. diamond is electrically insulating. The stiffness of the bonds also explains the exceptionally high thermal conductivity (several times greater than that of copper). diamonds are formed under high pressure and temperature at a depth of about 200 km. which is the energetically lower phase. . one obtains by repetition the sequence ABABAB . diamond is not in equilibrium at the conditions at the surface of the Earth. directly over a sphere in layer A. diamond is extremely hard with hardness number 10. Big raw diamonds are cleft parallel to octahedral crystal faces. In fact. into brilliants. . There are two ways to put a third layer on top of B. In the subsequent cutting. and a face-centered cubic lattice.

Their existence in interstellar space is hypothetical. supernovae. briefly PAH.4. C60 . hydrogen atoms are attached but only sporadically. The angles are chosen to maximize the fire (brilliance) of the stone.3 PAHs A polycyclic aromatic hydrocarbon. It is made of 12 pentagons (C5 ). The stacking is arranged in such a way that above the center of an hexagon in the lower sheet there is always a C atom in the upper one. is a planar compound made up of not too large a number of aromatic (C6 ) rings (chapter 12). 7. has the shape of a football. graphite cleaves well along these planes and owing to this property.23 g cm−3 and a very high sublimation temperature (Tev > 2000 K at interstellar pressures). depending on who is pressing. Diamond grains are also found in meteorites but they are extremely small ˚ (∼30 A). so that not a negligible fraction of C atoms is at the surface.4. the angle between ◦ . We reckon that in interstellar space about 10% of the carbon which is in solid form is graphitic. The origin of the meteoritic diamonds is unclear but probably interstellar (carbon stars. Graphite has a density of 2.3.35 A (figure 7.4.42 A. The most famous. A fascinating kind of PAHs are fullerenes. interstellar shocks?).4 Amorphous carbon Diamond and graphite are very regular (crystalline). Therefore. The distance from one ˚ sheet to the next is 3. At the edge. 7. The distance between adjacent C atoms is 1. 7. the coverage need not be complete. its name is derived from the Greek word for writing: when gently pressing a pen over a piece of paper. ˚ and 20 hexagons (C6 ). PAHs are unambiguously identified in the interstellar medium and account for a few percent by mass of the interstellar dust.3.1 A. as in benzene. like OH or more complex radicals.16).3. which produce the curvature.51 g cm−3 . The bonding is similar to that in benzene as again the π-bonds are not them 120 localized in the rings and the electrons of the unhybridized 2p orbitals are free to move in the plane.230 Structure and composition of dust related to the crystalline structure but to the refractive index n. The sheets are only weakly held together by van der Waals forces.2 Graphite Carbon atoms combine also to planar structures consisting of many C6 ˚ hexagonals. the material is said to be amorphous. This mobility endows graphite with an electric conductivity within the sheets. If there are many defects and if regularity extends only over a few atoms. tiny chunks peel off to form sentences of rubbish or wisdom. . They contain many H atoms and their density ρ is considerably lower than in perfect crystals for which ρ = 3. The diameter of the spheres is ∼7. Possibly one finds at the periphery also other species. Graphite consists of such C6 sheets put on top of each another. besides hydrogen. Fullerines do not contain hydrogen.

These amorphous particles should also contain hydrogen because it is present in the environment where they are created (wind of mass loss giants).3.5 The 2175 A bump in the interstellar extinction curve The bump at 4.Carbonaceous grains and silicate grains 231 pyroxenes Si O Figure 7. The splitting into a π. as in crystals but may more resemble worms because of widespread disorder. (Mg. two adjacent tetrahedra share one oxygen atom. The triangles represent the SiO4 units. as a potpourri of carbon compounds with all types of the bonds discussed earlier. When the electric vector of the radiation field is parallel to the plane of the carbon rings.1).6 µm−1 hump. Left: The building blocks of silicates are SiO4 tetrahedra.3.4.18. with many unclosed rings and an endless variety of carbon–hydrogen connections.and π-band. the π-band lies energetically above the σ -band. the π-electrons can be excited into the π ∗ -band.and π ∗ -band is analogous to the splitting into a binding and anti-binding state as illustrated in . abbreviated HAC.Fe)SiO3 . We may tentatively think of such hydrogenated amorphous carbon.and π-electrons in such grains form the σ . and this may cause the 4.2. pure carbon grains could only come from the hydrogen deficient atmospheres of WC stars or R CrB stars. Right: Other chain types are realized in silicates.8) is usually attributed to small graphite-like grains (section 4. too. Since the π-states are weakly bound. some of them stacked. respectively. Middle: In pyroxenes. here a double chain where more than one O atom per tetrahedron is shared (amphiboles). or as a disordered agglomerate of PAHs.0. In interstellar grains. which have a chain structure. an example is bronzite. ˚ 7. Most interstellar carbon grains are probably amorphous (aC). the σ . the chains are probably not very regular and linear.4.6 µm−1 in the interstellar extinction curve (figure 7. As we have seen in section 7.

The case r = 1 is depicted in figure 7. it takes more energy to promote an s electron in order that all four electrons become available for covalent bonding. therefore.23 < r ≤ 0. For crystals with 0. The ground state is also 3 P0 . r is different. there are no silicon ring molecules and double bonding (Si=Si) is rare (it involves delectrons). that oxygen fills basically all the volume and. The dust particles formed in oxygen-rich atmospheres are predominantly silicates.2 Coordination number It is intuitively clear that the relative size of adjacent atoms in a crystal is important for the crystal structure. the π ∗ -band is empty and responsible for electric conduction. In other crystals. one has N = 6 and an octahedron as coordination polyhedron. two s and two p electrons (3s2 3p2 configuration). for 0. the coordination number N equals 4 and the nearest neighbors form a tetrahedron. First. which is the number of its nearest equal neighbors.3. is four. controversial because of the overall messy character of the binding in carbon with its multitude of hard-toidentify bonds. .40 A. which assumes that the ions are hard spheres touching each other.4. We may idealize an SiO4 tetrahedron as an Si4+ ion symmetrically surrounded by four O2− ions. there is also substantial covalent bonding in the tetrahedron. like carbon.4 Silicates 7.232 Structure and composition of dust figure 7.4 ˚ than the radius of the O 0. only in a solid the levels are broadened because 2 of the interaction with many lattice atoms. Let us define r = ratio of the radii of neighboring atoms.73. They are the building blocks which form an ionic grid together with positively charged Mg and Fe cations. Second. Therefore.1 Silicon bonding and the origin of astronomical silicates Silicon stands right below carbon in Group IV of the Periodic Table.41/1. first.41.4. This implies. ˚ In silicates.62 A.41 < r ≤ 0. There are two basic reasons why silicon is chemically different from carbon.17. The assignment of the 4. in the third (n = 3) it has. however.3. second. silicon atoms are bigger. Note that whereas the π-band is filled. Silicates consist of negatively charged SiO4 groups in the form of ˚ tetrahedra with a side length of 2. however.41 A and much smaller 2− ion which is 1. Its inner two shells are filled.4. r = 0. so their p orbitals cannot come together closely enough to overlap and form a π-bond. 7. the radius of the central Si4+ ion is 0.4. that the coordination number N of a Si4+ ion. are summarized in table 7.7 for the H+ molecule.6 µm−1 hump to π → π ∗ transitions is.4. The coordination number and the coordination polyhedron for the limiting values of r . 7.

MgSiO3 ). (Mg.4).3 Silicate types The (negative) [SiO4 ] anions and the (positive) metal cations can be regularly arranged in various ways.19 is up-to-date or out-of-date . Because the O atoms are so big. two. We can now proceed to present a reasonable set of optical constants m(λ) for the two major interstellar solid substances. three or even all four oxygen atoms are shared with neigboring tetrahedra. at least. amorphous carbon and silicate. The repeating unit has the formula [SiO3 ]2− .73 1 Coordination number 3 4 6 8 12 Polyhedron around central ion flat triangle tetrahedron octahedron cube hexagonal or cubic close-packed 7. also of the orthorhombic crystal family (the form without Fe is called enstatite. The limiting ratio of the radius of the central ion to that of its nearest neighbors determines the coordination number and the kind of polyhedron around it. imagine that each Mg2+ or Fe2+ cation lies between six O atoms of two independent tetrahedra. One. in the literature they are.4. one indicates that magnesium is often replaced by iron.41 0. We mention bronzite. These six O atoms are at corners of an octahedron around the cation.17). • 7.Fe)2SiO4 which belongs to the orthorhombic crystal family. The question whether the choice displayed in figure 7. there are chains of SiO4 tetrahedra in which two O atoms are common to neighboring units.15 0.03 to 2000 µm.5. To visualize the three-dimensional structure of this mineral.Fe)SiO3.Fe). they form approximately an hexagonal close packed structure (figure 7. By writing (Mg. one also needs m(λ) for the two minor solid constituents: graphite and PAHs (section 12. We will call these data our standard set and it may be used as a benchmark for comparison between dust models.4.23 0. The latter are then isolated islands. Optical constants vary with time. The wavelength ranges from 0. Ratio r of ionic radii 0.5 A standard set of optical constants We have learnt in this section what dust is made of. The ratio r of the ionic radius of the metal cations to that of O2− is about 0. which is a sufficiently broad interval for most applications. For some applications. The prototype is olivine (Mg.4.3.Carbonaceous grains and silicate grains 233 Table 7. The basic silicate types are determined by the following criteria: • No oxygen atom is common to two tetrahedrons. In this mineral.

very low absorption in the near infrared.5.5. it declines proportionally to ν 1.2 µm. particles of different sizes must be invoked to explain its various sections. k) and dielectric permeability ε = (ε1 .1 The size distribution Because the extinction curve covers a large frequency range.55 . the wavelength dependence is more complicated. The absorption and extinction efficiencies resulting from figure 7. the 2175 A bump and . the optical region medium-sized and the UV.7 µm and then falls steadily at longer wavelengths. two resonances at 10 and 18 µm and a far IR decline proportional to ν 2 . The data fulfil the Kramers–Kronig relations of section 2. For silicate.19 for a typical grain size are shown in figure 7. There is also a flat part in the ultraviolet up to 0.5 Grain sizes and optical constants 7. degrades to secondary importance in view of our general ignorance about grain properties. The infrared part requires ˚ large grains. For amorphous carbon.5 in the UV and usually negligible beyond 3 µm. ε2 ) for silicate material after [Lao93] and amorphous carbon after [Zub96] (their type BE). For both substances. in the far infrared. the extinction is basically flat up to λ 0. Optical constant m = (n.234 Structure and composition of dust Figure 7. 7.20. the albedo is of order 0.19.

26) which is called the MRN size distribution [Mat77]. The result is widely accepted and follows from binning the radii into intervals [ai . Qualitatively speaking. composite grains made of these materials are also possible. the far UV small particles. The scattering efficiency can be found as the difference between extinction and absorption. Assuming that they are spheres and defining the size distribution n(a) such that n(a) da = number of grains per cm3 with radii in the interval [a.19. one assumes silicate and carbon material with optical constants similar to those in figure 7. one can ask how the grains have to be distributed in size in order that their wavelengthdependent extinction follows the observed interstellar reddening curve. a + da] one obtains a power law: n(a) ∝ a −q with q 3.Grain sizes and optical constants 235 Figure 7.20. The absorption (dots) and extinction (full line) efficiency for spherical dust grains of 0.1 µm radius with the standard set of optical constants from figure 7. the radii a of the relevant grains are given by 2πa/λ ∼ 1. smaller ones are inefficient as absorbers or scatterers.5 (7. Amorphous carbon = aC.19. silicate = Si. we expect that at wavelength λ. For the dust composition. . ai+1 ] with ai+1 > ai and optimizing the number of grains n i in each bin until agreement with the extinction curve is achieved. To quantify this supposition.

To solve (7. m (7. m) dµ = number of fragments in the interval µ to µ + dµ resulting from the collision of a body of mass m.29) n(a)a 2 da ∝ If q were greater than four. t) dt = probability that particle of mass m is collisionally destroyed during time dt.3 µm (7. t)P(m. t) = −N(m. For an MRN distribution.30). If fragmentation is the only process that determines the mass distribution. t) dm = number of particles in the interval m . m) dµ. P(m. because extinction becomes gray for large grains.5. one separates the variables: N(m. a− ˚ 10 A and a+ 0. t)P(µ. t) is governed by the formula ∂ N(m. the total mass or dust volume V is supplied by the big particles and the geometrical surface F by the small ones: V ∝ F∝ a+ a− a+ a− √ √ √ n(a)a 3 da ∝ ( a+ − a− ) ∼ a+ 1 1 √ −√ a+ a− 1 ∼ √ . a+ and a− . t) + ∂t M N(µ.27) but the outcome is not very sensitive to either boundary. roughly.5. and even less to a− because small grains λ) where only the total volume of the grains are in the Rayleigh limit (a− counts. .28) (7. the small grains would also contain most of the volume. N(m.30) M is the maximum mass of the particles and the total mass of the particles is conserved.5 by studying grains undergoing destructive collisions [Hel70.2 Collisional fragmentation One can theoretically obtain a size distribution n(a) ∝ a −3. m + dm. Consider a time-dependent mass distribution N(m. It is. Dor82]. t)ξ(µ. a− (7. therefore. which has an exponent q = 3. Not to a+ . one must additionally specify its upper and lower limit. .236 Structure and composition of dust To fix the size distribution. t) = η(m) · ζ (t). . no wonder that the smallest ˚ grains (a < 100 A) were detected through their emission and not from studies of the extinction curve. 7. ξ(µ. and not their individual sizes. t) such that N(m.

m/µ. is related to the ratio of initial mass over fragment mass. ξ(µ.Grain sizes and optical constants 237 The probability P(m. Both sides may be set to a constant. ξ(µ.31) η(µ) µ 3 + m 3 1 1 2 dµ. The lower boundary on the integral. The solution for ζ(t) is ζ(t) = ζ0 1 + tCζ0 where ζ0 is the value of ζ at t = 0 and C is a positive constant. m). we may write P(m. One expects that the number of debris.5 ≤ x ≤ 3). t) µ 3 + m 3 1 1 2 dµ = ζ(t) · γ (m) (7. m). we need to know the mass distribution of the debris. Equation (7. m . . To solve ˙ (7.32) The equation is completely separated as there is a pure time dependence on the left and pure mass dependence on the right. When one puts mx ξ(µ.5 ≤ x ≤ 2. If we assume.5 . is the minimum mass for destructive collisions. m) dµ. t) for destruction of a particle of mass m is obviously proportional to an integral over relative velocity vrel of the colliding particles multiplied by the collisional cross section. we recover.83 over a wide range of x values (0. Because for spheres m ∝ a 3 and dm ∝ a 2 da. in terms of sizes. µx+1 With a power law ansatz η(m) = m −y one finds with a little algebra that y = 1.31) with ζ /ζ 2 = −C for η(m).30) now becomes ˙ 1 ζ = −γ (m) + η(m) ζ2 M m η(µ) γ (µ) ξ(µ. as a first approximation. m) = θ x+1 µ empirical evidence suggests for the exponent that 0. The constant θ follows from the condition of mass conservation: θ 0 m mx µ dµ = m. very small particles are not destroyed. Q is some proportionality factor. (7. the MRN distribution n(a) ∝ a 3y−2 a −3. t) = Q with γ (m) ≡ Q M m M m N(µ. after each shattering event. that vrel is constant.

.238 Structure and composition of dust There are two dissatisfactory aspects to this scenario when applied to interstellar dust. for example. First. for instance. the scenario does not include further modifications in the interstellar medium. such as accretion or coagulation. Second. in the outflow of giants [Bie80]. which do exist as one observes. before they can be ground down by collisions or shocks to the saturation distribution n(a) ∝ a −3. varying values of RV or ice mantles.5 . the particles have to be born uncomfortably big.

as in LTE. the calorific properties of dust. But whenever it is heated. investigate. grain emission is also only a function of Td and. In section 8. for a dust cloud of uniform temperature.5 presents examples of the emission of very small grains. section 8.1) 239 .Chapter 8 Dust radiation We treat. we have to know the specific heat and the internal energy of the dust material. the physical background of dust emission. we show how to compute the temperature to which a grain is heated in a radiation field and add a few common examples as well as useful approximation formulae.3 illustrates the emission of grains for a typical dust type. which are those whose temperature does not stay constant.1. given by ν abs = K ν · Bν (Td ). Section 8. According to Kirchhoff. For evaluating the emission of very small grains. therefore. 8. Therefore. a grain in interstellar space is not at all in an LTE environment.1 Kirchhoff’s law 8.1 The emissivity of dust A grain bathed in a radiation field acquires an equilibrium temperature Td which is determined by the condition that it absorbs per second as much energy as it emits. This discovery had already been made in the abs 19th century. Finally. the excess energy is very rapidly distributed among the very many energy levels of the grain and the resulting distribution of energy states is given by the Boltzmann function and depends only on the dust temperature. it also presents a first example of radiative transfer. Td . Of course. later it was found that ν /K ν equals the Planck function. in section 8. We. (8. in local thermodynamic equilibrium (LTE) the abs ratio of the emission coefficient ν to the absorption coefficient K ν is a function of temperature and frequency only. for example by a photon.2. A single grain radiates according to Kirchhoff’s law and we strictly derive this law in section 8.4.1 under the assumption that a grain consists of an ensemble of coupled harmonic oscillators in thermal equilibrium. in this chapter.

v=0 To obtain 4π th (i. The quantity ν is called the emissivity. one: ∞ Pi. potential (V ) and interaction (H ) energy.4.v = 1.42). The dimensions change accordingly.v−1 is calculated according to the general formulae (6.1. The i th oscillator has levels vi = 0. each of mass m i and resonant frequency ωi .v of finding the i th oscillator in quantum state v is given by the Boltzmann equation (5. the eigenfunctions ψ are the . 2.v by the Einstein coefficient Ai v. the emissivity ν per unit volume is expressed in erg s−1 cm−3 Hz−1 ster−1 . 4π th (i.v Ai Av. equilibrium could never be established.v−1 hνi = 1 − e (8.2 Thermal emission of grains To understand how and why an interstellar grain emits according to equation (8. 1. the radiation integrated over all directions from oscillator i in the transition v → v − 1 at temperature T .v e−βhνi (v+ 2 ) = 1 − e−βhνi e−vβhνi = Z (T ) 1 with the partition function Z (T ) from (5. If the energy difference is lost radiatively.v−1hνi . In (8.v−1 (here the superscript i is not an exponent) and the photon energy hνi .8): Pi. Classically.1). v). v.2). Otherwise the oscillators would be decoupled and unable to exchange energy at all: starting from an arbitrary configuration.41) and (6.240 Dust radiation The emission over all directions equals 4π ν . The sum over all probabilities is. a photon of frequency νi escapes. In the thermal (canonical) description. of course. . . For example. qi and pi denote coordinates and momenta. the total energy E of the system is E = E kin + V + H 1 2 f i=1 pi2 + m i ωi2 qi2 mi (8.2) and consists of kinetic (E kin). the level population of the quantized harmonic oscillators is described by a temperature T . and can make transitions of the kind vi → vi − 1. The probability Pi. we assume that it consists of N atoms and approximate it by a system of f = 3N − 6 (see section 8.2) weakly-coupled one-dimensional harmonic oscillators. we have to multiply Pi. For a harmonic oscillator. It can refer to a single particle. 8.10) and β = 1/kT . The latter is small compared to E kin and V but it must not be neglected altogether.3) The Einstein coefficient Ai v. v) −βhνi −vβhνi i e = Pi. a unit volume or a unit mass. .

yields v K νi = c2 −βhνi e−(v−1)βhνi − e−vβhνi . th (i ) = th (i.v−1 = v A 1. the absorption coefficient in the transition v → v − 1 of the oscillators with frequency νi can be expressed through the Einstein B coefficients and the level populations (see (13. v i i K νi = Pi. towards the boundary their spacing decreases. according to (6. For the integrated radiation th of the particle over all frequencies.v Bv. c i i As the statistical weight of the levels is one.5) This is the total emission of the grain at frequency νi . we obtain th (i ) = hνi Ai 1.0 that for an ideal harmonic oscillator there is no upper limit to its maximum energy and νi is the same for all levels v.40).0 eβhνi − 1 .1. have to evaluate the integral Hv (y)y Hv−1(y)e−y d y 2 and its solution is given in (A.3).v−1 Bv−1. We. (8.v and a change from B to A coefficients.v − Pi. v).v−1 hνi . To compute the total emission per solid angle of oscillator i .4) All one has to know is Ai .61)). therefore. 8.1). Ai v. Note 1. the coefficient for the ground transition. we sum up over all levels v ≥ 1.v−1 1 − e 8πνi2 .v−1 = Bv−1.0 v≥1 v 1 − e−βhνi e−vβhνi = hνi Ai 1. The Einstein coefficient is thus proportional to the quantum number v and we may put i Ai v.16). one must add up all oscillators. (8.3 Absorption and emission in thermal equilibrium According to the theory of line radiation.Kirchhoff’s law 241 Hermitian polynomials Hn (y) of (A. v≥1 Because of the mathematical relations (A.0 (v ≥ 1). th = i th (i ). whereas a real oscillator has a maximum energy above which the system is unbound and only its lowest levels are approximately energetically equidistant. Bv.

This derivation of equipartition 2 presupposes. 8. A gas of N atoms possesses 3N degrees of freedom and each degree of freedom has an average energy E = 1 kT . E . as in a solid.5) and (8. 2 Furthermore. including particle i . api2 of particle i is found by averaging api2 over the phase space with the Boltzmann distribution as the weight: api2 = e−β(api +E ) api2 dq1 . v Note that the absorption coefficient K νi . thermal equilibrium implies equipartition of energy when for each particle i of the system the kinetic energy is quadratic in its momentum pi and the potential energy is quadratic in its space coordinate qi . . Quite generally. still contains the temperature via β = 1/kT . .1).0 v K νi = K νi = . As the atoms have only kinetic energy. 2 When the atoms are not free but bound to their neighbors by electric forces. we retrieve Kirchhoff’s law (8. Let us first consider a gas where the atoms are unbound. (8. The total energy E tot is now twice as large as for a gas and evenly split between V and E kin : E tot = E kin + V = 3NkT. . However. that energy between the members of the system can . Also note that K νi does include stimulated emission because it is incorporated into the formula v for K νi . When we combine the two formulae. the radiation of frequency νi interacts with all level pairs and in the total absorption coefficient K νi the temperature has disappeared. but we have to be aware that the frequency νi refers to a certain oscillator. can be written as E = api2 + bqi2 + E . v − 1). which is taken with respect to a particular pair of levels (v. altogether 2 E tot = E kin = 3 NkT . there is.1. the typical factor (1 − e−hνi / kT ) has canceled out. d p f 2 e −β(api2 +E ) dq1 .4 Equipartition of energy We conclude with remarks on the distribution of energy under equilibrium conditions.6) 8πν 2 v We might drop the subscript i in equations (8. Then the total energy E of the system.242 Dust radiation Summing over all levels v ≥ 1 gives the absorption coefficient K νi of the grain at frequency νi . and when the total energy of the rest of the system. . rather interestingly. there is also potential energy V . c2 Ai 1. however. The mean kinetic energy. 2β 2 Likewise. one finds bqi2 = 1 kT . the same average kinetic energy 1 kT . depends neither on pi nor qi . d p f = e−βapi api2 d pi 2 e −βapi2 d pi = 1 1 = kT. however.6).

2 Approximate absorption efficiency at infrared wavelengths When one wants to determine the dust temperature from equation (8. there are certain approximations to Q abs which allow us to ν evaluate Td analytically. our standard case. 2 then have different mean energies equal to hν/(ehν/ kT − 1) after (5. 8.The temperature of big grains 243 be exchanged continuously. a grain is hot when Q abs is high at the ν wavelengths where it absorbs and low where it emits.8). Simplifying the latter by a power law. at high temperatures.11). (8. we get Q abs = ν 8πa m 2 (λ) − 1 · Im λ m 2 (λ) + 2 = a Q0νβ . Different oscillators. for instance. 1.7) abs K ν is the mass (or volume) absorption coefficient of dust at frequency ν. by collisions with gas particles.2. 8.8). It is evident from (8. often with only a small loss in accuracy. they have to be added in equations (8.9) . Therefore.8) that it is not the absolute value of Q abs that determines ν the dust temperature in a given radiation field Jν .1 The energy equation Equation (8.7) and (8. of radius a and absorption efficiency Q abs (a). We neglect them for now.ν 2. (8. and Jν is the average of the radiation intensity over all directions. According to (3. Two efficiencies. to calculate the dust temperature Td from the balance between radiative heating and radiative cooling: abs K ν Jν dν = abs K ν Bν (Td ) dν. one first has to compute the absorption efficiency Q abs (a) for all frequencies from Mie ν theory and then solve the integral equation for Td . E = 1 kT . However. For spherical grains. ν (8. but constant factor. Qualitatively speaking. These procedures require a computer. When T is small and the energy levels are quantized. yield exactly the same Td . Q abs and Q abs . although in thermal equilibrium.1) permits us. one exploits the fact that dust emits effectively only in the far infrared where the Rayleigh limit is valid. the absorption efficiency Q abs of spheres is then proportional ν to the size parameter 2πa/λ multiplpied by a function that depends only on frequency.2. equivalently.2 The temperature of big grains 8.8) If other forms of heating or cooling are relevant.ν that differ at all frequencies by an arbitrary. it holds only for classical systems or. in a straightforward way.3). ν Q abs (a) · Jν dν = ν Q abs (a) · Bν (Td ) dν. For a quick and often sufficiently precise estimate.

See also figure 7. When the expression Im{. The ratio of absorption efficiency over grain radius. . Q abs /a. β = 1. We. Q abs /a. The optical constants are from figure 7.3 µm).1µm. propose for quick estimates at infrared wavelengths Q abs ν 10−23aν 2 [a in cm. The broken line shows the approximation from (8.10) is not bad at ν λ > 10µm for both interstellar silicate and amorphous carbon dust.9) varies inversely proportional to wavelength. therefore.10) Figure 8. for amorphous ν carbon (aC) and silicate (Si). . The curves display the ratio of absorption efficiency over grain radius. (8. This is the value we favour.5).} in (8. We see that the fit Q abs /a from (8.10). When the complex optical constant m(λ) deserves its name and is truly constant. ν in Hz].1 µm but are quite generally valid in the infrared.1 µm. β = 2. . which is the relevant range of emission.20 which shows Q abs ν for a = 0. for the ν particular value a = 0. A blackbody would have β = 0 and a Q 0 = 1. The curves have been calculated with a = 0. At wavelengths λ > 10 µm.244 Dust radiation Figure 8. Estimates of the exponent β range from 1 to 2. they are fairly independent of radius provided the grains are not very big (a ≤ 0.19.1 gives a feeling for the quality of the approximation.1. Values smaller than 1 are excluded for very long wavelengths because of the Kramers–Kronig relations (section 2.

9) and use a power law for the absorption efficiency. although our planet does not emit like a perfect blackbody and its clouds reflect some 30% of insolation. Approximate values are: ∞ 0 x 3+β d x ex − 1 6.89 if β = 1 122.08 if β = 2. Equation (8. we find that a blackbody r astronomical units from the Sun heats up to r −1/2 T = 279 K. AU For r = 1. β≥1 Q abs ∝ ν β ν Q abs is. (8.17).8) describing emission.13) 4πr 2 For example. But this is irrelevant for ν the evaluation of the integral on the right-hand side of (8. a blackbody is usually colder than any other object. as long as the grain temperature stays below a few hundred K so that the Planck function at these frequencies is small. a perfect absorber with Q abs = 1 everywhere.2. its temperature follows from L = 4σ T 4 . But this is not a law. overestimated at high frequencies. the ν energy equation (8. one can construct counter examples. The temperature is now independent of particle size.494 if β = 0 24.11) 8.1 Blackbodies Easiest of all is to find the temperature of a blackbody. If a blackbody is heated by a star at distance r and bolometric luminosity L. With the power law (8. (8. this value is close to the average temperature of the surface of the Earth.The temperature of big grains 245 When we follow the suggestion of (8. of course.80).9) for Q abs . . In a given radiation field.8) becomes ∞ 0 Q abs Bν (T ) dν = a Q 0 ν 2h c2 kT h 4+β 0 ∞ x 3+β d x .8) then reduces to ν Jν dν = σ 4 T π (8.2.12) σ being the radiation constant of (5.3 Temperature estimates 8. ex − 1 The right-hand integral is evaluated in (A.3.

47 × 10−6 aTd6 (cgs units).16) So the grain temperature falls off with distance from the star like Td = 4.246 Dust radiation 8. Jν is the diluted blackbody radiation of temperature T∗ . from 20 to 30 K requires a tenfold increase in flux. According to (8.14).11) for β = 2. the approximation ν (8. it remains to evaluate the integral on the left. Q abs ν so that 0 ∞ 1 Jν dν 1. and Jν in erg cm−2 s−1 Hz−1 ster−1 . Consequently. a moderate temperature difference implies a vast change in the energy budget.0 L a 1 6 r−3 1 (cgs units). a is expressed in cm.2. The average intensity over all directions is.2 Interstellar grains directly exposed to stars For real grains. Jν = 2 Bν (T∗ )π R∗ . 2 4πr 2 4 When we insert this expression of Jν into (8. Consider a grain at distance r from a hot star of luminosity L. L in erg s L 16π 2r 2 1. in view of equilibrium. (8.10) together with the integral (8.82). additionally. which is usually true in the interstellar medium.8) then transforms into ∞ 0 Q abs Jν dν ν 1. we may use. For example. at all wavelengths relevant for absorption. From 2 the position of the grain. for the absorption efficiency Q abs . the 4 integral over Jν is directly proportional to T∗ . also its total emission rate are proportional to Td6 . r in cm.15) and use L = 4πσ R∗ T∗ after −1 ) (5. For an early-type star or even the Sun.17) . There is a further simplification if the radiation field is hard. one may put.47 × 10−6aTd6 (cgs units).14). (8.15) If. ν in Hz. the stellar radius disappears and we find (a.3. the heating rate of a grain and. (8. the star subtends a solid angle π R∗ /r 2 and the intensity towards it is Bν (T∗ ). or a whole dust cloud. To determine Td in (8. to warm a particle. effective temperature T∗ and radius R∗ . therefore.14) In cgs units. (8. The equation is applicable as long as the grain diameter stays below ∼1 µm.47 × 10−6aTd6 (cgs units). Td in K. The energy equation (8.

rather in the space between the stars. one is from starlight peaking at ∼1 µm. 247 line of sight. several components.2.6 that really tiny grains may not attain an equilibrium temperature at all. Generally speaking. The figure contains none of our previous approximations. As an example. i. 4π ISRF Jν dν ∼ 0. We will see in section 8. This value is similar but not identical to that of a blackbody.4 Relation between grain size and grain temperature In the estimate (8. to either side of (8.5. much bigger than the particles found in interstellar space. Tlim . Tbb .8) becomes independent of ν size and the grain temperature levels off to some limit.19. in figure 8. giants and interstellar dust. By the latter we mean the environment in the solar neighborhood but outside clouds and not close to any star. the dust temperature changes with particle radius a like Td ∝ a − 6 .8) and is not constant. The radiation field expected there averaged over a solid angle is depicted in figure 8. 1 However. If the grains are big. of order 1.8).04 erg s−1 cm−2 .3 becomes meaningless. In view of the usual grain properties and geometry of the are often no less trustworthy than 8. For the size range supposed to prevail in the interstellar medium (a ≤ 0.2. Its main sources are stars of spectral type A and F.17).2. small grains are warmer than big ones. The two ISRF major ones are of comparable strength in terms of ν Jν . the radius a in the absorption efficiency Q abs (a) cancels out (see (3. The ISRF should be uniform in the galactic disk on a scale of 1 kpc. estimates of this kind sophisticated computations. the other from interstellar dust with a maximum at ∼200 µm. .The temperature of big grains If there is extinction along the luminosity L by a frequency average uncertainties regarding radiation field. Integrated over frequency and solid angle. this relation is valid only for grains of intermediate size. Then a plot of the kind of figure 8. Figure 8.3 demonstrates the size dependence of the dust temperature for these two environments. Q abs (a) in (8. one must replace the over L ν e−τν .e. the absorption efficiencies are calculated from Mie theory with optical constants from figure 7. the maximum temperature ratio is smaller.3)) and all tiny grains of the same chemical ν composition have the same temperature. configuration. because Q abs stays under the integrals ν in (8.3 µm). The temperature increases from tiny to huge particles by a factor of about three. If the particles are very small so that the Rayleigh limit is applicable to both emission and absorption. we compute grain temperatures near a luminous star and in the interstellar radiation field (ISRF). instead their internal energy fluctuates in response to the quantum character of the absorbed photons. One can discern.

2. further away colder than carbon grains.5 Temperature of dust grains near a star The variation in dust temperatures in a reflection nebula. The underlying dust model is emits at wavelengths > 3 µm a flux proportional to Jν described in section 12.375. the curves are very smooth. silicates are warmer. which is typically excited by a B1V star. is depicted in figure 8. Carbon has a very constant α of 0. • • • Writing the dependence of the temperature on distance as Td ∝ r −α . grains are very hot and emit also below 10 µm. 8.01 to 0. When we check the accuracy of the approximation formula (8. . The grains have radii of 0.01 and 0.1 µm lowers the temperature by ∼30%. it turns out to be satisfactory in both variables. This behavior can be explained with the help of figure 8. The approximate mean intensity Jν in the Milky Way at the locus of the Sun. At these wavelengths. Near the star. grain radius a and distance r . Small grains are always warmer than big ones. The curve is based on data from [Per87] but ISRF includes the emission of PAHs (spikes).248 Dust radiation ISRF of the interstellar radiation field (ISRF) Figure 8.1: at short distances. In the logarithmic plot. and there is a wiggle around r = 1015 cm for silicates.1 µm and consist either of silicate or amorphous carbon.2. the emissivity of silicate particles is small. so they have to be hotter than the carbon grains in order to get rid of their energy.2.17) with the help of figure 8. Interstellar dust exposed to a radiation field Jν ISRF . the slope α is not far from 1/3.4. almost linear in the case of carbon. An increase in grain radius from 0.4.

19. 8.18) Fν = Bν (Td ) · [1 − e−τν ] · .3. the intensity Iν towards a uniform dust layer of optical thickness τν and temperature Td is Iν (τ ) = Bν (Td ) · [1 − e−τν ] so that an observer receives from a solid angle the flux (8.19) (8. Bottom. For comparison. in the interstellar radiation field with mean intensity Jν 17 cm from a B1V star with L = 104 L after figure 8.1).2.2.2 K near the B1 star and Tbb = 3. a blackbody would acquire a temperature Tbb = 34.8 K in the ISRF. the optical constants are from figure 7.6 Dust temperatures from observations Whereas gas temperatures can often be obtained rather accurately from properly chosen line ratios. the experimental determination of dust temperatures is always dubious.The temperature of big grains 249 Figure 8.17). After the basic equation of radiative transport (see section 13. here the integrated mean intensity of the radiation field is about 7000 times stronger. In both boxes. The temperature of spheres of amorphous carbon (aC) or silicate (Si) as a ISRF function of grain radius: Top. The dotted line represents Td from the approximate formula (8. at a distance of 10 and T∗ = 2 × 104 .

The flux ratio at two frequencies. The observations at the two frequencies refer to the same astronomical object. Temperature of amorphous carbon (aC. which is usually true in the far infrared.01 µm for the upper full and dotted curve and 0.4) and the observations have to be carried out with different spatial resolution.1 µm for the lower ones. Particle radii are 0. (8. This apparently trivial condition is sometimes hard to fulfil when the source is extended relative to the telescope beam (see section 13. i. ν1 and ν2 .20) It is customary to extract from (8. If they did.1. do not lie both in the Rayleigh– Jeans limit of the Planck function. The optical depth towards the star is zero. The observational frequencies.20) the dust temperature under the following premises: • • • The emission is optically thin (τ 1).19. becomes B1 (Td ) [1 − e−τ1 ] F1 = F2 B2 (Td ) [1 − e−τ2 ] 1 2 .e.4. Optical constants are from figure 7. there is no absorption between the star and the grain. marked by the subscripts 1 and 2. ν denotes the frequency.250 Dust radiation Figure 8. . Td would cancel out because Bν (Td ) ∝ Td . full curve) grains as a function of distance from a B1 main sequence star with 104 L and effective temperature T∗ = 2 × 104 K. dotted curve) and silicate (Si. a certain region in a galactic cloud. for instance.

20) simplifies to F1 K 1 · B1 (Td ) = F2 K 2 · B2 (Td ) (8. In view of the numerous restrictions and because the exponent β in (8. over the particular frequency interval from ν1 to ν2 is known or. This quantity ν is proportional to the emissivity ν of (8. Should the assumed value for K 1 /K 2 be wrong.The emission spectrum of big grains • • 251 The dust temperature in the source is uniform. The blackbody intensity (Q abs = 1) ν is shown for comparison. A blackbody (β = 0).9) and the slope α are. related through α =β +2 independent of temperature. like a planet. K 1 /K 2 . has a spectral index α = 2.5 ˚ displays.3 The emission spectrum of big grains 8.1). In a double-logarithmic plot of Fν versus ν and when λ is large. we should always be sceptical about absolute values of Td . we may trust results of the kind Td (source A) > Td (source B) and their implications for the energy budget because they depend only weakly on the ratio K 1 /K 2 .3. for silicate grains of 600 A radius. If all requirements are fulfilled. the product Q abs Bν (T ). equivalently the exponent β in (8. it is defined by F1 ν1 α = . K ν itself is not needed. 8.1 Constant temperature and low optical depth As a first illustration of the spectral energy distribution emitted by dust. However. The ratio of the dust absorption coefficients.21) but with K 1 /K 2 = 1 because a star is. We recall here that the surface temperature of a star can also be obtained through photometry from an equation like (8. The flux ratio also determines the spectral index α of the energy distribution. F2 ν2 The exponent β in (8.5 represent three astronomical environments: . a blackbody.9). at very long wavelengths. to first order. without any physical correspondence. all curves are parallel for any T . figure 8. one gets a purely formal color temperature. The temperatures chosen in figure 8.9) is debatable.21) and yields Td in a straightforward manner. (8. The number 2 in this equation comes from the Rayleigh–Jeans part of the Planck function.

This is similar to Wien’s displacement law for a blackbody (see discussion after (5.10) (broken curve) and from Mie theory with optical constants after figure 7.5. the power law approximation for the absorption efficiency Q abs from (8. The Planck function Bν (Td ) as well as Q abs Bν (Td ) have the ν units of an intensity. especially in the far infrared. Emission of a blackbody (full curve) and of a silicate grain of 600 A = 0. only in the hot dust do we see the 10 µm resonance. Although the dust optical constants of figure 7.252 Dust radiation ˚ Figure 8.19 (dotted curve). ν The spectrum of a dust grain can be crudely characterized by the wavelength of maximum emission. the dust emission is very smooth. 60 K for warm dust in star-forming regions and 200 K for hot dust close to stars. The efficiency Q ν from approximation (8. For instance. Several points deserve a comment: • • • • • A dust particle emits at all wavelengths less than a blackbody.19 have some structure. can boost emission by orders of magnitude. • • • 20 K for the bulk of the dust in the Milky Way. A seemingly unspectacular temperature change by a factor of three. . say from 20 to 60 K.77)).6 µm abs of the silicate grain is calculated radius at different temperatures.10) gives acceptable results. Unless the dust is hot (²100 K).

55 µm.5 but the temperature is now higher. The variation of the optical depth with wavelength is depicted in the lower part of the figure. the intensity has been calculated as a function of wavelength and arbitrary optical depth (from τV = 0. the curve there is normalized at 0. the normalized optical depth of silicate grains with 600 A radius. the spectral distribution of the intensity is Iν = τν Bν (T ). Bottom.18).3. In figure 8. The intensity (in units erg s−1 cm−2 Hz−1 ster−1 ) towards a cloud of temperature T = 600 K filled with such grains. Top.1 to ∞) for a cloud consisting of ˚ identical silicate spheres of 600 A radius at a temperature of 600 K. the curves Q abs Bν (Td ) in figure 8.1 and 103 . To obtain the absolute value of Iν .6.2 Constant temperature and arbitrary optical depth If the dust in a cloud is at constant temperature and the emission optically thin.The emission spectrum of big grains 253 ˚ Figure 8. For small τν .5 are proportional to the observed intensity Iν ν as given by equation (8. one needs the optical depth τν .1 approximately reflects the optically thin case and may be compared with the plots in figure 8. As . The Planck function Bν (Td ) corresponds to τV = ∞. The wavelength scale starts at 0. The line with τV = 0. 8. The visual optical thickness τV varies between 0.6.55 µm where τλ /τV = 1.

At long wavelengths. the cloud becomes optically thick at 25 µm and emits like a blackbody for λ ≤ 25 µm when τV > 100. all forces exactly balance. If the oscillations are small. Because V and its first derivatives are zero. Iν asymptotically approaches the Planck function.1 Normal coordinates In a simple-minded model. one uses normal coordinates. ηi = x i − x i0 and put V (x0 ) = 0. 8. To find the eigenfrequencies of the oscillators.4 Calorific properties of solids To investigate the calorific properties of a a dust particle. one can approximate the potential by a Taylor expansion. If there are N mass points in the grain. the atoms in a grain are replaced by mass points connected through springs. ∂V ∂ xi = 0. we treat the conglomeration of atoms in the grain as an ensemble of harmonic oscillators. ij (8. designated x0 .254 Dust radiation τV increases. determine their eigenfrequencies and density of states and evaluate the energy of the grain under the assumption of thermal equilibrium. This is exemplified by the disappearance of the 10 µm feature. For instance. They are equidistant and have the same spectral slope Iν ∝ ν α with α 4. Any information about the emitters is then lost. .22) As the kinetic energy T is quadratic in ηi . x0 We introduce coordinates relative to equilibrium.4.23) . we describe their positions by one vector in Cartesian coordinates. x n ) n = 3N. 8. . the intensity curves for all τ run parallel. At the equilibrium position. which is always possible. so the derivatives of the potential V (x) vanish. . ˙ T = 1 2 i m i ηi2 ˙ (8. the first non-vanishing term is V (x) = 1 2 ∂2V ∂ xi ∂ x j ηi η j = 0 ij 1 2 Vi j ηi η j . x = (x 1 . .

When there are two kinds of atoms in the chain of mass m and M > m. yi2 ˙ V = 1 ωi2 yi2 . It is a standard procedure to obtain from the relative coordinates.1 Oscillators in a linear chain Instructive and well-known examples are provided by linear chains which present the simplest configuration in which atoms can be arranged. respectively. each with its personal frequency.1. It is straightforward to derive the dispersion relation 2κ 1 − cos(kd) (8. . In fact. It has n solutions and one thus finds n frequencies.25) This is an algebraic equation of nth order in ω2 . so called normal coordinates yi . n. T =1 2 2 i i So a grain of N atoms may be substituted by f = 3N − 6 independent oscillators. 8. and that the kinetic and potential energy are quadratic in yi and yi . with the property that each normal coordinate corresponds to a harmonic motion of just one ˙ frequency. . . in alternating sequence. a solution of the kind ηi = ai e−iωt yields a set of n linear equations for the amplitudes ai . as usual.26) ω2 = m where k = 2π/λ is the wavenumber and λ the wavelength. Let all atoms be of the same mass m. Let us consider a longitudinal wave. . ηi . The motion in any coordinate x i will be a superposition of n harmonic oscillations of different frequencies ωi and amplitudes ai . j For non-trivial values of ai . (8. through a linear transformation new. (8. (V j i − δ j i m i ω2 ) ai = 0. and connected through springs of force constant κ. . six solutions will be zero if there are more than two (nonlinear) atoms but the remaining frequencies refer only to internal atomic oscillations. |V j i − δ j i m i ω2 | = 0. with a constant distance d between them when they are at rest.Calorific properties of solids 255 one derives from the Lagrange function L = T − V of (6. the determinant must vanish.24) m i ηi + j Trying. The force on any mass point depends then only on the distance to its nearest neighbors on the left and right.1) the equations of motion ¨ Vjiη j = 0 i = 1.4.

from (5. . But in most applications. . (8. i = 1. the eigenfrequencies νi are closely packed and one may replace the sum by an integral: U (T ) = 0 νD hν ehν/ kT −1 ρ(ν) dν (8. At temperature T .27) M + m ± M 2 + m 2 + 2m M cos(2kd) ω± = mM where the minus sign refers to the acoustic and the plus sign to the optical branch. In the acoustical mode.30) . s. If n i oscillators have the frequency νi . When N ≥ 3. 1. .11).28) E i = hνi + hν / kT 2 e i −1 The zero-point energy 1 hνi is not necessarily small compared to the second 2 term in (8.2 Internal energy of a grain A point-like unbound atom has three degrees of freedom corresponding to the three independent spatial coordinates along which it can move. then s ni = f i=1 and the possible energy levels are E iv = hνi (v + 1 ) 2 v = 0. The total energy content U of the solid is obtained by summing over all oscillators: s hνi ni . . A solid body consisting of N atoms has 3N degrees of freedom.4. adjacent atoms swing in phase. If the body has N atoms.28) but drops out when calculating the specific heat and is. Let there be s different frequencies altogether. oppositely and out of phase. 2. We bin oscillators of identical frequencies. An atom in a solid can be considered as a three-dimensional harmonic oscillator. disregarded. . there are now two eigenfrequencies. . 1 hνi . although there are now forces between the atoms.29) U (T ) = ehνi / kT − 1 i=1 When N is large. . 8. we can put f 3N. The dispersion relation reads κ 2 (8. (8. the number of vibrational modes in a grain is f = 3N − 6 because one has to subtract 3 + 3 to account for the translatory and rotational motion of the grain as a whole. the mean energy of the oscillators of frequency νi is. For each wavenumber k. therefore. in the optical mode where frequencies are higher. there are f = 3N − 6 one-dimensional oscillators.256 Dust radiation the spectrum splits into what is called an acoustical and an optical branch.

30) to very small grains. (0. n = 0. y. cos γ ) At the walls. for instance. The conditions for a standing wave are lλ/2 = ax cos α and mλ/2 = a y cos β for integer l. where ρ(ν) denotes the number density of oscillator frequencies and νD the upper integration limit.7). az along the coordinate axes and one corner at the origin r = (x. a y . This implies ±1 = e−ikx ax = cos k x ax − i sin kx ax which leads to l λ = ax cos α 2 m λ = a y cos β 2 n λ = az cos γ 2 l. ρ(ν). Here a two-dimensional sketch for a rectangle with sides of length ax and a y . m. The extrapolation is convenient but far from accurate. the wave must have nodes because the atoms cannot move. y.Calorific properties of solids 257 y ay ky β k α kx ax x Figure 8. y. . where N is only of order 100. kz ) = k(cos α. z) = (ax . 1. like PAHs (section 12.2). A plane wave in the crystal of wavelength λ has the form ei(k·r−ωt ) = (r)e−iωt k = |k| = 2π .7. z) = 0. k y .3 Standing waves in a crystal The density of states. One has to be on the alert when applying (8. . . cos β. Modes have nodes at the walls of the crystal. The wave propagates in the direction of the wavevector k. is found by determining first the low-frequency part (sound waves). 8.4. Consider a rectangular crystal of N atoms with sides ax . . 2. . z) = 0 (see figure 8. This is easy and similar to finding modes in acoustics. λ k = (kx . m. Then ρ(ν) is extrapolated to high frequencies (optical waves) assuming the same functional dependence of ρ(ν).

m. If cl and ct are the sound velocities for the longitudinal and for both transverse waves.4. 3 3·8 λ 3c3 We divided by 8 because l. there are altogether three types of waves and we have to make separate counts for all three of them. Because the transverse wave has two kinds of polarization corresponding to motions with perpendicular velocity vectors. Thus all points (l. γ . Because cos2 α + cos2 β + cos2 γ = 1 it follows that l 2ax /λ 2 + m 2a y /λ 2 + n 2az /λ 2 = 1. n are positive numbers and fall into the first octant.4 The density of vibrational modes in a crystal 8. 8. and if we define the average c by ¯ 1 2 1 = 3+ 3 3 c ¯ cl ct we find for the frequency density in a crystal ρ(ν) = 4πax a y az 2 dZ = ν . c3 ¯ (8. n) of positive integers specifies one mode and can be represented by a point in a Cartesian coordinate system. m. β. and as a transverse wave in which the atomic motion is perpendicular to wave propagation.4. The direction into which the wave travels is given by the angles α. Their total number is approximately 4πax a y az 3 4π 8ax a y az Z≈ = ν .1 Longitudinal and transverse waves A disturbance can travel in a crystal in two ways: • • as a longitudinal (compressional) wave where the atoms move in the direction of wave propagation or opposite to it. m.4. n) corresponding to wavelengths greater than or equal to λ lie inside an ellipsoid of volume V = 32πa x a y az /3λ3 .31) .258 Dust radiation Each triple (l. c = λν denotes the sound velocity. dν c3 ¯ The number of eigenfrequencies in the interval ν to ν + dν is then dZ = 4πa x a y az 2 ν dν.

4.5 Specific heat The specific heat at constant volume of a system of f one-dimensional oscillators follows from (8. If n= is the density of atoms.35) It specifies how much energy is needed to raise the temperature of a body by 1 K. What the true frequency density might look like compared to the Debye approximation is qualitatively illustrated in figure 8. the crystal does not have to be rectangular but may have any shape.34) into (8.37) = 9k N V x 4ex d x.2 The Debye temperature 259 The cutoff νD that appears in (8. For a continuous distribution of modes.29).34) Actually. k (8.Calorific properties of solids 8.4. 8.30) to find the internal energy and specific heat. the number of all modes.8. we insert ρ(ν) from (8. kT (8.4. Cv = ∂U ∂T f =k V i=1 x i2 e xi ni [e xi − 1]2 xi = hνi .33) The frequency density may now be written as ρ(ν) = 9N 3 νD ν2. 4π (8. we get ¯ νD = c · 3 N a x a y az 9n . (e x − 1)2 There are two important limits: .30) for the internal energy of a crystal is the highest possible frequency. U (T ) = Cv (T ) = ∂U ∂T 9N 3 νD 0 νD hν 3 dν ehν/ kT − 1 T θ 3 0 θ/T (8.36) (8. Therefore the maximum of Z is Z max = 3N. There is a limit because Z obviously cannot exceed 3N. The Debye temperature θ is related to νD through θ= hνD .32) This is the Debye frequency. (8.

the upper bound θ/T approaches infinity and the total energy of the grain and its specific heat are (see (A. A true distribution might look like the dotted curve.41) Any improvement over the Debye theory has to take the force field into account to which the atoms of the body are subjected.38) 3 (8. C V = 3Nk. The latter is correct only for sound waves with wavelengths much larger than ˚ the grid size of the crystal (∼1 A). so U = 3NkT (8.17) for the mathematics): U= CV = 3π 4 NkT 5 12π 4 Nk 5 T θ T θ 3 (8.260 Dust radiation Figure 8. (8.34) (full curve). which is to be compared with the Debye approximation of (8.39) • The characteristic feature in C V is the proportionality to T 3 . At high temperatures. • At low temperatures. . It yields at low temperatures the famous T 3 dependence of C V . the formulae approach the classical situation where a three-dimensional oscillator has the mean energy E = 3kT .40) and the specific heat is constant and given by the rule of Dulong–Petit. There is a cutoff at νD because the number of degrees of freedom of all atoms is finite. The frequency density ρ(ν) of crystal vibrations.8.

νmin ] the PAH has just one frequency.4. One can estimate the minimum frequency νmin from the condition that in the interval [0. when the temperature goes to zero. With ρ(ν) from (8.42) which increases only linearly with frequency. only at temperatures above 20 K.43) x 3ex dx (e x − 1)2 The two-dimensional case is important for PAHs because they have a planar structure. With θz = 950 K referring to out-of-plane bending and θx y = 2500 K to inplane stretching vibrations [Kru53]. T (8. one can improve the model of the specific heat by using two Debye temperatures: C V (T ) = k N f (θz ) + 2 f (θx y ) f (y) = 2 y2 y 0 (8. For graphite sheets. the smaller νmin becomes. one finds a density of modes ρ(ν) = 6N ν 2 νD (8. C V eventually approaches the T 3 dependence.4 is repeated for a two-dimensional crystal of N atoms. not as ν 2 .4.42). Under normal interstellar conditions. 1= 0 νmin ρ(ν) dν. It also applies to graphite which consists of PAH sheets.45) The bigger the PAH.Calorific properties of solids 8. Therefore now C V ∝ T 2 and U (T ) = C V = 6k N 1 y2 6N 2 νD y 0 νD 0 ehν/ kT hν 2 dν −1 y= θ .44) x 3ex (e x − 1)2 d x.6 Two-dimensional lattices 261 When the analysis of section 8. Although the coupling between the sheets is relatively weak. only the lowest levels are populated. . Very low vibrational frequencies cannot be excited in small grains. however. there is good agreement with experimental data [Cha85]. Collisional excitation of the modes by gas atoms is possible provided kTgas ≥ hνmin . this gives kθ νmin = √ . h 3N (8.

. after [Cha85]) and PAHs without hydrogen atoms (dotted curve. At low temperatures.44)).262 Dust radiation Figure 8. 8.9 summarizes the specific heats C V (T ) which we use in the calculations of the emission by small grains.6 × 1022 atoms per gram. the curves flatten towards the Dulong–Petit rule (8.41). The optical behavior depends in a sophisticated way on the two dielectric functions ε1 (ω) and ε2 (ω) and on the particle shape. This happens because whenever an energetic photon is absorbed. As the grains get warmer. its temperature will fluctuate.9. C V = 3Nk. Figure 8. we need their optical and thermal properties. the grain temperature jumps up by some not negligible amount and subsequently declines as a result of cooling. after [Kru53] using (8. We will speak of very small grains (vsg for short) when we have in mind particles whose temperature is time variable because they are tiny. for PAHs C V changes like T 2 . The thermal behavior is determined more simply from the specific heat. graphite (broken curve. To compute their emsission. Graphite has 5.0 × 1022 and silicate about 2. similar to [Guh89]). graphite and silicate have a T 3 dependence. The specific heat per gram of dust for silicate (full curve.5 Temperature fluctuations of very small grains When a dust particle is very small.

we can express its average monochromatic emission per solid angle by ν = πa 2 Q abs ν Bν (T )P(T ) d T. of course. must be equal: Æ Æ Æ Npop = Ndepop . where the matrix element A f i denotes the transition probability that a single grain changes from state i to f .1). . They occur at a rate Pi A f i . Q abs Jν dν = ν Q abs Bν (Teq ) dν. where Teq follows from the steady-state balance between emission and absorption from (8. the temperature oscillates only a little around an equilibrium value Teq and in the limit of large grains. which is a function of temperature only. We bin U (T ) into N states U j of width U j with j = 1. An absorption or emission process implies a transition in U from an initial state i to a f inal state f . In equilibrium. T + d T and call P(T ) the probability density.8). 8.5. (8. for each level j the number of populating and depopulating events. . so in the case of a sphere of radius a. We are faced with the problem of finding P(T ) and we describe below its solution (an elaborate treatment can be found in [Guh89] and [Dra01]). .2 The transition matrix When a grain absorbs or emits a photon. .47) Although the emission of such a single grain is not constant over time. changes. Each state U j corresponds to a temperature T j or frequency ν j = U j / h with corresponding spreads T j and ν j .Temperature fluctuations of very small grains 8. . The probability particles in state U j is P j of finding an arbitrary grain in a large ensemble of equal to the number of all grains in level j divided by . It is. . ν Even for a very small particle we will assume that its radiation. Let us arbitrarily pick out one of them and denote by P(T ) d T the chance that its temperature lies in the interval from T . 2. (8. obeys Kirchhoff’s law (8.1 The probability density P(T ) 263 Consider a large ensemble of identical grains in some interstellar environment. the whole ensemble radiates at any frequency at a steady rate. at any time. N. . normalized: ∞ 0 P(T ) d T = 1.46) In normal interstellar grains of average size. Npop and Ndepop . P(T ) approaches the δ-function δ(Teq ).5. its internal energy U (T ).

47). the first sum after the first equal sign refers to processes that cool and the second to those which heat the grain. To find the probability density P(T ) required in equation (8. . With the purely mathematical definition Akj Ajj = − k= j we may write the condition Npop = Ndepop for all j as A j k Pk = 0. k= j Ndepop / Æ = Pj k< j Akj + P j k> j cooling Akj = P j heating In both formulae. . below those for heating.48) Only N − 1 of these N equations are linearly independent.46)).48) for P2 . k (8. . the number of transitions P j Akj from j → k is thus proportional to the width of the initial (via P j ) and final energy bins.51) hν Above the main diagonal stand the cooling elements. for each level j . for dust cooling from state j to a lower one k. abs 4πCν Bν (T j ) Akj = νk ν = ν j − νk k < j. . one may first put P1 = 1. (8. The energy balance requires for the cooling and heating rate of each . P j = 1. Likewise we have.50) abs Jν stands for the mean intensity of the radiation field and Cν is the absorption cross section of the grain. (8. i (8. As it should be.49) A matrix element Akj referring to dust heating ( j < k) is equal to the number of photons of frequency νk − ν j which a grain absorbs per Hz and second multiplied by the width of the final bin νk . solve (8. Akj = abs 4πCν Jν hν νk ν = νk − ν j j < k.264 Dust radiation Therefore. Npop / Æ= k> j Pk A j k + k< j Pk A j k = k= j Pk A j k Akj . PN and then rescale the P j by the obvious condition that all probabilities must add up to one (see (8.

This suggests that in cooling from state j one needs to consider only the transitions to the levels immediately below. (8. they should be carried out whenever the time interval between the absorption of photons with an energy comparable to the heat capacity of the grain is larger than the cooling time. Generally speaking. therefore. Therefore. have to be written as A j −1. it is good to know when such calculations are necessary. j = 0 abs 4πCν Jν dν · h(ν j +1 − ν j ) −1 do not violate energy conservation.48) the computationally rapid recursion formula (but see the simple trick described in [Guh89] to safeguard against numerical rounding errors).52) The total matrix A f i has thus acquired a new form where. j = ∞ 0 abs 4πCν Bν (T j ) dν · h(ν j − ν j −1 ) −1 . . their emission changes the grain temperature very little. (8. put all matrix elements A f i above the main diagonal to zero.8). except A j −1. But the latter. Unfortunately. . in practical applications it suffices to ignore cooling transitions with j → k < j − 1. 1 Akj Pk j = 1. . In fact. only the elements A f i with f = i − 1 are non-zero. One can. . j +1 k≤ j We mention that heating may not be reduced to transitions j → j + 1. One now immediately obtains from (8.Temperature fluctuations of very small grains level j . in order to fulfil the energy equation. Although heating elements of the form ∞ A j +1.53) P j +1 = − A j. this is not very . above the main diagonal. j .3 Practical considerations Calculating the emission of a grain with temperature fluctuations is not straightforward. As cooling proceeds via infrared photons which have low energy.5. 8. they would result in an unrealistically small spread around the equilibrium temperature Teq of (8. N − 1. Akj νkj = k< j 0 ∞ abs 4πCν Bν (T j ) dν cooling 265 Akj νkj = k> j 0 ∞ abs 4πCν Jν dν heating with νkj = |ν j − νk |. This would ignore the big energy jumps of the grain after UV photon absorption which are important for the probability function P(T ).

266 Dust radiation practical advice. The simplest grid has a constant mesh size.4 The stochastic time evolution of grain temperature The probability density P(T ) describes the steady-state temperature distribution of a large ensemble of grains. We are purposely vague about what is meant by hard but one may think of the UV range. Umin and Umax . Qualitatively. equal one. To minimize the computational effort. 8. An iterative procedure is. ν The right-hand side describes the difference between the power absorbed from the radiation field Jν and the cooling rate. However. it suffices to calculate the radiation of the particle from (8. This point is counterintuitive but is exemplified in figure 8. we will present examples that give some feeling for when temperature fluctuations are important and how they modify the spectrum. Radiative cooling is not balanced at every instant by heating and so the internal energy U changes for a spherical grain of radius a according to the first-order differential equation: dU = 4π πa 2 dt Q abs Jν (t) dν − ν Q abs Bν (T (t)) dν . are hard to determine beforehand and intuitive values like Umin = 0 and Umax = hνmax are often not adequate. however. The radiation field is weak to make the intervals between capture of energetic photons long. a grid of constant T seems to be better suited than one with constant U . An indicator for the computational accuracy. The temperature of an individual particle. one has to avoid a large transition matrix A f i and properly select the energy bins.1) with a constant temperature from (8. especially in problems of radiative transfer. The grains are tiny so that their heat capacity is small. either in temperature or energy. is the time-averaged ratio of emitted over absorbed energy which must. Generally. The boundaries of the grid. a grid of 100 energy bins is sufficient provided they are properly selected. one can find a better adapted grid by using as new boundaries those values of T where P(T ) has dropped by some large factor (for instance. 1012) from its maximum. as the former has at low temperatures a finer spacing in energy to handle the infrared photons. As P(T ) varies strongly with hardness and strength of the radiation field as well as with particle size. recommended: starting with a first choice that crudely brackets the maximum of P(T ). Replacing U by the specific heat . there are three requirements for temperature fluctuations to occur: • • • The radiation field is hard to ensure the presence of energetic photons. but not a precise one. is time variable. a good grid is sometimes not easy to find. of course. Whenever the function P(T ) becomes very sharp. Because U ∝ T α with 0 ≤ α ≤ 3. Evaluating the probability density P(T ) when temperature fluctuations are small is not only unnecessary but can also incur numerical difficulties.8).13. therefore.5.

however. T + d T . take into account that the electromagnetic field is quantized.e. To incorporate the quantum character of the photons. this ordinary differential equation yields the time evolution of the temperature. i. is continuous. To solve (8. . The number of photons with frequency ν j that are absorbed within each time step equals Nj = t · 4π πa 2 J (ν j )Q abs (ν j ) νj. At the beginning of the new time step i + 1. T (t).46). The emissivity ν can be computed as an average over time τ . Ui is written as Ui = a j hν j j where the a j are either one or zero. we assume that the cooling rate. although its time-averaged flux is constant. the N j are much smaller than one and represent also the probability for photon capture. the probability density P(T ) follows from counting how many times during the period τ the temperature attains values in the interval T .54) Given a starting value T0 at t = 0.35)). T (t). These numbers are then normalized according to (8. this gives Ti+1 . the particle energy is raised by an amount Ui that accounts for the stochastically absorbed photons during the new interval t. One must.54) easier. we may write 4π πa 2 dT = dt C(T ) Q abs Jν (t) dν − ν Q abs Bν (T (t)) dν . . For each time step. The chance that a j = 1 is N j . occurs sporadically through absorption of individual energetic photons (>1 eV) and leads to temperature jerks.54). we proceed as follows. at the end of the i th time step. hν j For short time steps t. They are computed from N j with a random number generator. the probability that a j = 0 equals 1 − N j . has been laboriously evaluated from (8. To make equation (8. Then we solve for the (i + 1)th time step lasting from ti until ti+1 = ti + t the differential equation 4π 2 a 2 dT =− dt C(T ) Q abs Bν (T (t)) dν ν with the boundary condition T (ti ) = Ti + Ti . the temperature is increased accordingly by Ti . When the evolution of the temperature. ν = πa 2 τ τ 0 Q abs Bν (T (t)) dt. ν .54) numerically from t = 0 until t = τ . one thus gets a temperature. however. Most of the grain heating. ν 267 (8. Let Ui and Ti be the energy and temperature of the grain at time ti . The total period τ is divided into small time steps t and the total frequency range into intervals ν j = νj+1/2 −ν j −1/2 .Temperature fluctuations of very small grains C(T ) (see (8. the flux of emitted low-energy infrared photons.

In the first case. has a hot atmosphere (Teff = 2 × 104 K) and guarantees a copious amount of energetic quanta. . which is close to but not identical to the value Tmax = 115.47)) is displayed as a solid line. we depict P(T ) by dots.10–8. The full and dotted curves for P(T ) (see previous explanation) practically coincide.1 K. πL Bν (Teff ). All examples deal with silicate grains because of their 10 µm resonance which occasionally shows up.4 or. • In the interstellar radiation field (ISRF) with a spectral shape as depicted in figure 8. L ν . Lν = and satisfies the condition L ν dν = L. 4 σ Teff In figures 8. more simply. the area under the curves equals one. The temperature excursions are small and amount to some 10% around a mean value. from the formalism developed in section 8. the emission coefficient ν refers to one grain and is plotted in the unit erg s−1 Hz−1 ster−1 . The true emission (from (8. where P(T ) attains its maximum.1 s.1 Small and moderate fluctuations The top panel of figure 8.5. Its monochromatic flux.9 K. For comparison.13. Altogether we followed T (t) over a time τ = 2 × 104 s partitioned into 105 steps of fixed length t = 0.2. otherwise by a solid line.10–8. is proportional to the Planck function. In the emission spectrum of the lower right panel the 10 µm feature is indicated. Here Teq = 116.6 The emission spectrum of very small grains We illustrate the stochastic temperature fluctuations and their effect on the spectrum for grains in two environments: • Near a B1V star.2.8).268 Dust radiation 8. the emission under the false supposition of constant temperature from (8. The probability density P(T ) in figures 8. It is luminous (L = 104 L ). 8.10 gives the chance of finding the grain within a temperature interval of 1 K width centered on T . it does not matter whether ones takes the temperature variations into account or assumes a (constant) equilibrium temperature Teq after (8. We show an arbitrary section of 400 s.10 displays the temperature variation T (t) of a single ˚ silicate grain of 40 A radius at a distance r = 1017 cm from the star.6. In this example.13 is also computed in two ways. The probability density P(T ) in the lower left panel of figure 8.5.1) is shown dotted. either from the stochastic time evolution of the temperature T (t) described in section 8.

although large.11. There are now discernible diffferences in the probability P(T ) depending on the way in which it is computed (see previous explanation). See text. the spectrum is a good approximation only at far infrared wavelengths. was finite. emissivity ν of the grain in erg s−1 Hz−1 ster−1 .5. time evolution of the ˚ temperature of a silicate grain of 40 A radius at a distance of 1017 cm from a B1V star.2). at 170 K. on average.4. say. although close to the full one. The dots (hardly discernible) plot the emission assuming a constant temperature.The emission spectrum of very small grains 269 Figure 8. Small stochastic temperature excursions: top. when the emission is evaluated under the false supposition of temperature equilibrium (dotted). In the bottom frame on the right. Because the total time interval over which we integrated. probability distribution P(T ) of the temperature shown on a linear scale (the dotted line is computed from section 8. the same grain is placed at a distance ten times greater from the star (r = 1018 cm). its effect is negligible here. the solid from section 8. temperatures far from the mean never occurred. colder. lower left. As one can estimate from the figure. however. In the mid infrared. One would have to wait very long to see the grain. the full curve includes temperature fluctuations. the dotted line displays a scatter. it is hit by an energetic photon only once every 100 s and the temperature excursions are no longer small. and is determined well only around the maximum of P(T ) and does not extend to probabilities below ∼10−5 .5. the errors . so the rate of impinging photons is reduced by a factor 100 and the particle is. lower right. Therefore.10. In figure 8.

one hesitates to assign an average temperature at all. Moderate fluctuations. The ordinate for P(T ) is now logarithmic. When looking at the temperature evolution T (t). There are now two disparate regimes: most of the time the grain is cold and cooling is slow but occasionally the grain is excited to a high temperature from which it rapidly cools. although mathematically this can be done.2 K after . The probability density P(T ) has turned asymmetric. the grain is again at a distance r = 1017 cm from the star but has a ˚ ˚ radius of only 10 A. The temperature fluctuations have increased.9 K and far from the equilibrium temperature Teq = 116.12.270 Dust radiation Figure 8. As in figure 8. which gives a better feeling for the scatter than the probability function P(T ). Compared to the two preceding examples where a = 40 A. are large (two powers of ten at 10 µm) because the grain is occasionally at temperatures far above the average.2 Strong fluctuations In figure 8. accordingly. 8.11. The absorption cross section is also 43 times smaller than before (the Rayleigh limit) and photon capture is. less frequent. 3 = 64 times lower and photon absorption the heat capacity of the particle is now 4 thus induces a much larger relative change in energy. the maximum is at Tmax = 51.10 but the same grain is ten times further away from the star.6.

a longer integration time would smooth it. we can follow in detail individual absorption events and the subsequent cooling. The energy reservoir U (T ) of a grain is proportional to T 4 when it is cold. As in figure 8. ˚ In figure 8. The dotted line in the bottom right-hand box bears no resemblance to the real emission (full curve).10 but the grain is now small (a = 10 A). beyond 55 K a sharp drop sets in. representing P(T ) as determined from T (t). Strong fluctuations. the same and equal to the absorbed flux.The emission spectrum of very small grains 271 ˚ Figure 8. The cooling rate varies after (8. In these examples.14) with the sixth power of the temperature.38). although the frequency integrated emission is. To evaluate the emission without taking into account the hot excursions no longer makes sense. (8. The probability density P(T ) tapers off gradually from its maximum at Tmax = 8. Note the strong 10 µm silicate feature. and above the Debye temperature U ∝ T (see (8. (8.8). The dotted line shows the probability P(T ) as calculated from the time evolution T (t).13. The dotted line.40) and . Absorption here is a rare event. even with regards to low-energy photons and occurs approximately every few hours. is jerky and inaccurate above 120 K because of the finite time over which we calculated the evolution. in both cases. Note the much larger time scale in this figure. even in the far infrared. the grain has a radius of 40 A and is heated by the weak light of the ISRF.3 K to ∼55 K.12.

8. τcool is of order 10 s.272 Dust radiation ˚ Figure 8. λ1 and ˚ λ2 . Of course. also figure 8. a grain of normal size (>100 A) becomes colder when the distance to the exciting star is increased. equivalently. the flux ratio is constant.13. i.9). It concerns their color temperature or. it also receives less photons farther from the star but its color temperature at shorter wavelengths is only determined by the hot phases corresponding to the spikes in figure 8. therefore. falls rapidly as T increases. The interesting point is: no matter whether these spikes are rare or common. In the spikes of the present examples. For a very small grain.6. obviously because it receives less energy. A silicate grain of 40 A radius in the ISRF.12. The cooling time is roughly τcool ∝ U (T ) . the flux ratio at two wavelengths.3 Temperature fluctuations and flux ratios Very small grains display in their emission another peculiar feature. When multi-photon events occur. only the emitted power in the wavelength band from λ1 to λ2 diminishes with distance. when an energetic photon is absorbed before . as long as they do not overlap. T6 and. In a reflection nebula.e. the situation is more tricky. Strong fluctuations.

The wavelengths in micrometres are indicated as subscripts in the emissivity. the ratio of the emission coefficients. the near infrared colors. is again distancedependent.10. such as that illustrated in figure 8. Figure 8.The emission spectrum of very small grains 273 Figure 8. like K–M corresponding . Flux ratios for graphite grains of various sizes as a function of distance to a B1V star. The grains are made of graphite and have radii between ˚ 5 and 40 A. Small particles undergo strong temperature fluctuations which lead to flux ratios that are almost independent of the distance from the star. the grain has had time to cool from a preceding capture.14 shows the flux ratio for several wavelengths as a function of distance from the star.14. (λ1 )/ (λ2 ). For all grain sizes.

274 Dust radiation to 2. the critical particle size is pushed up a bit and the flux ratios are constant only at larger distances.8 µm . But already at distances smaller than the typical dimension of a reflection nebula. they level off. .2 µm / 4. first fall as one recedes from the star. They continue to stay so far away from the star where the stellar UV radiation field resembles that of the diffuse interstellar medium. Near IR color temperatures do not change across a reflection nebula if the grain radius is ˚ ∼10 A or smaller. For mid and far IR colors.

which can be considered to be at rest.Chapter 9 Dust and its environment 9. Gas accretion has great astronomical consequences as the following examples demonstrate.1) A grain of geometrical cross section πa 2 .1 Grain surfaces 9.16)) vi = 8kTgas . accretes the gas species i at a rate Racc = n i πa 2 vi ηi (9. They may either rebounce from the grain surface or stick. Consider a gas of temperature Tgas containing a certain species (subscript i ) with particle mass m i . gas atoms and molecules continually collide with the dust particles. Accretion of H atoms is the only way in which molecular hydrogen can be made at all in relevant quantities. Such species may not then be observable. • • • • The transfer of energy from gas to dust and back which may strongly influence the temperature of the components.2) 275 . πm i (9. The depletion of molecular species in the gas phase as they freeze out on the grains.1. the number of accreted atoms per unit time is equal to the number of atoms leaving the grain. number density density n i and mean velocity (see (5.1 Gas accretion on grains In the interstellar medium. A change in the optical properties of the grains and thus their emission characteristics. The formation of new molecules. whereas otherwise they would be strong emitters and important coolants. In equilibrium.

we find that an atom that sticks on collision (ηi = 1) freezes out after a time τacc ∼ 2 × 109 yr. we first derive their interaction potential. their total cross section F for gas capture per cm3 is 3n H m H Rd F= . The growth rate da/dt of the grain radius should be independent of a. (9. increases. which is given by some fixed far away charges. . a− = 100 A. 108 yr). has at position x0 the electrostatic potential energy U = qφ(x0 ). If the gas density is high (n H > 104 cm−3 ). ˚ a+ = 3000 A. because such clouds always have densities above 102 cm−3 . depletion and mantle formation proceed more quickly (τacc ≤ 105 yr) than the dynamical processes which are √ characterized by the free-fall time scale (tff ∼ 2 × 107 / n H yr).4) τacc is shorter than the lifetime of a molecular cloud (τcloud ∼ 107 .3 km s−1 ). √ 4ρgr a− a+ Here Rd is the dust-to-gas mass ratio. either in atomic or molecular form and ρgr the density of the a− . The lifetime τacc of a gas atom grain material. although the mean grain size. 9. the energy becomes U= ρ(x) φ(x) d V.1. large grains hardly grow any bigger through accretion. which is the average over the total distribution.2.3) τacc = Fvi ηi 3m H Rd ηi n H vi ˚ Inserting reasonable numbers (Rd = 0. Therefore.2 Physical adsorption and chemisorption 9. Consequently. n H = n(HI) + n(H2 ) the total number density of hydrogen. Both atoms are electrically neutral but have dipole moments: let p1 refer to the surface and p2 to the gas atom.1 The van der Waals potential To describe the collision between an atom on the grain surface and an approaching gas atom. ρgr = 2.276 Dust and its environment where ηi denotes the sticking probability. . as the impinging atoms do not see the grain curvature. When the grains have a size distribution n(a) ∝ a −3.1. vi = 0.5 g cm−3 . localized around x0 . the matter will mostly accrete on the small dust particles as they have the larger total surface area. as in clumps. a charge distribution ρ(x).5 with lower and upper limit a− and a+ .007. . nH (9. A point charge q in an external electrostatic potential φ. and it was assumed that a+ before it is swallowed by dust may be defined by τacc Racc = 1 and thus becomes √ a− a+ 4ρgr 1 = . If instead of the point charge q there is.

Suppose the field E(x) is due to the dipole p1 on the grain surface. these minima are at the mid-points between four neighboring surface atoms (see figure 7. But when the two dipoles are very close.e. the potential attains minima at various locations privileged by symmetry.2) of a dipole p. In our case of colliding atoms. E(x) is then described by (3.1.5) Let us further assume that the dipole p2 of the gas atom is induced by p1 of the surface atom.2.1). On a regular surface. like 1/r 12. r3 r5 (9.2 The full potential including repulsion The absolute value of the van der Waals potential increases rapidly at short distances. we obtain U = qφ(x0 ) − p · E(x0 ) 277 where we have used the general definition (1.7) − U (r ) = 4D r r 2 with Dσ 6 = αp1 . a repulsive interaction sets in because the electrons tend to overlap and Pauli’s exclusion principle forbids them to occupy the same quantum state. r6 (9. reaches its √ minimum value −D at r = 6 2σ and remains negative as it approaches zero for r → ∞ (figure 9. the effective charge q vanishes. This potential bears a more empirical character and changes with distance even more abruptly.6) 9. changes sign at s = σ . . i. For a simple or body centered cubic lattice. atoms on the grain surface. φ(x) = φ(x0 ) + x · ∇φ(x0 ) = φ(x0 ) − x · E(x0 ) and substituting the expansion into the volume integral for U . The strength of the former depends then on its polarizability α (see (3.10) and (1. we get U (r ) = p1 · p2 3(p1 · r)(p2 · r) − . Denoting by r the vector from p1 to p2 . The parameter D defines the strength of the potential and σ its range.10). U (r ) becomes infinite for r → 0.8)): 2αp1 p2 = 3 r and the dipole moments will be parallel (p1 · p2 = p1 p2 ) so that U (r ) = − 2 4αp1 . much quicker than for monopoles. one has to include the contribution from all force centers. The net result of the combination between repulsion and attraction is σ 6 σ 12 (9.Grain surfaces Expanding φ(x) to first order around x0 .2). In a collision of an atom with the grain surface.

the surface must contain chemically active sites and the gas species should not consist of saturated molecules. The . Quite generally. for example. H2 or H2 O. we expect grains in cold and dense clouds to be covered by a sheet of physically adsorbed ice.1 eV to remove the atom or molecule from the grain surface. The minimum is at r = 6 2σ and U (r ) = −D. like H2 or H2 O but atoms. here with D = √ and 0. The interaction is therefore much stronger (from 0.1. This is not possible on a cold grain so that the process is probably unimportant there. particularly hydrogen. In principle. A much tighter coupling of a gas atom colliding with the grain surface is possible through chemical binding or chemisorption. Chemically active sites will have disappeared once the grain surface is covered by physical adsorption by one monolayer of.5.5 eV.7). It involves a profound change in the electron structure of the binding partners. As a precondition for chemisorption. The potential U (R) between dipoles after (9. 9.278 Dust and its environment Figure 9. it needs typically only 0.25 σ = 0.3 Physical and chemical binding Physical adsorption is rather weak.2.5 to 5 eV) and the range of the chemical potential shorter than in the case of physical adsorption. the mean binding energy of atoms and molecules in an ice mantle ˚ several A or more thick is less than 0. The exact value depends on the composition and structure of the surface and on the type of gas species. Therefore. corresponding to 1000 K if the binding energy E b is expressed as a temperature T = E b /k. radicals like OH could also chemically bind to the grain surface but prior to the reaction an activation barrier must be surmounted. The curve is relevant for physical adsorption.1.

F = m gr . CO2 .9) if the surface atom of mass m s has its equilibrium position at x = 0. F(r ) = −U (r ). then encounters the repulsive part. the force F with which the gas atom is rejected becomes F(r ) = F0 − 0 F0 r 2b 0 ≤ r ≤ 2b r > 2b with F0 = D + Ek .1).4 A. 9. One can approximately evaluate the process by studying the interaction of the gas atom (subscript g) with just one surface atom (subscript s). b To fix the initial conditions. NH3 and others. Its equation of motion. it must transfer to the dust particle in the collision an energy E s greater than its kinetic energy E k far from the surface. otherwise it will rebound. The gas atom passes in the collision first through the attractive potential where its kinetic energy increases. 0 ≤ η ≤ 1. For the interactive potential U (r ) between the two atoms we use (9. where r = x g − x s > 0 is their distance and D the binding energy. Let us neglect the attraction in the potential and approximate its repulsive part by a parabola [Wat75].7). One finds the energy E s which is transferred to the surface atom during the impact by numerically integrating the equations of motion. obviously.1. is attractive and negative when r is large and repulsive and positive when r is small. the constant b is then called the slope parameter and is of order 0.Grain surfaces 279 sheet may be envisaged to be made either of pure water ice or to be a composite of different ices. The repulsive potential U is often also written as U (r ) = −D + (E k + ˚ D)e−r/b . U (r ) = −D + −D D + Ek (r − 2b)2 4b2 0 ≤ r ≤ 2b r > 2b.3 The sticking probability We estimate the probability η that an impinging gas atom stays on the grain surface. which is almost like a wall (see figure 9. ˙ ¨ . An analytical solution under simplifying assumptions is illustrative.8) (9. mixed in a certain proportion. The force associated with the potential. and rebounces at the distance rmin where U (rmin ) = E k . let the gas atom at time t = 0 pass the point r = 2b with a velocity r = − 2(D + E k )/m g . For the gas atom to stick. such as H2 O. We imagine the surface atom to be attached to a spring of force constant κ so that it oscillates at frequency ωs = κ/m s . Such a simple mechanical system is governed by the equations of motion ¨ m s xs = − κ xs − F m g xg = F ¨ (9. In the quadratic equation.

the integrations can be evaluated analytically because for ω0 = ωs . 2 With t1 and f (t) = −F(t) from (9. 1 ˙ f (t) ξ − i ωs ξ = ms which has the solution ξ(t) = eiωs t 0 t (9. ξ0 = ξ(t = 0) = 0. π = t1 ω0 (9. it undergoes a forced oscillation described by ¨ m s x s + κ x s = f (t). Putting where ωs = κ/m s is the vibration frequency of the surface atom.280 Dust and its environment then has. −1 ω0 = τcoll . the solution r = 2b · (1 − sin ω0 t) where the frequency ω0 = F0 /2bm g (9.11). The infall of the gas atom is reversed at the turning point r = 0 at time ˙ t = π/2ω0 where also r = 0.12) √ ξ = x s + i ωs x s ˙ 1 f (t)e−iωs t dt + ξ0 . Its inverse defines the collision time. Altogether the force varies with time like F(t) = F0 sin ω0 t 0 0<t< else . ms In our case. . one can replace the second-order differential equation (9.12) by one of first order. Because the energy of the surface atom is 2 2 E s = 1 m s x s + ωs x s = 1 m s |ξ |2 ˙2 2 2 the energy which it receives in the collision becomes Es = 1 2m s t1 0 f (t)e−iωs t dt . for times t > 0. During the impact. e−iωs t sin ω0 t dt = e−iωs t 2 2 ωs − ω0 ω0 cos ω0 t + i ωs sin ω0 t .11) Now consider the collision from the point of view of the surface atom which is subjected to a force f (t) of equal strength but opposite sign.10) is of order 1014 s−1 .

Elaborate estimates suggest that under most astronomically important circumstances sticking of the atoms is likely. To use a sticking coefficient between 0. generally smaller than one. It cannot travel freely along the surface because to move. • • When ω0 ωs . Other atoms will then also absorb energy and the assumption of just two colliding bodies is inadequate. m g /m s This simple theory can be refined in various ways: one can take into account the reaction of the crystal. A physically adsorbed gas atom or molecule finds itself in a potential minimum.13) for whether the energy loss of the gas atom. use a more suitable interaction potential because a gas atom generally hits a spot somewhere between several surface atoms. Two numerical examples for the case ωs = ω0 are presented in figure 9. which then appears. typically U0 ∼ 0. The value of U0 is a few times smaller than the binding energy E b of the gas atom.3E b .13) Es = 2ω0  ms  ω0 ωs . the Boltzmann distribution gives the population ratio .4 Thermal hopping.13) that the sticking efficiency is largely determined by the binding energy D for physical adsorption and by the mass ratio of gas-to-surface atom. the collision time is long compared to the oscillation When ω0 −1 period ωs of the crystal.  2 2 (ω − ω0 )2 In the evaluation of the integral for ω0 ωs . is bigger than its initial kinetic energy E k . • As under most interstellar conditions E k D. it has to overcome the surrounding potential barrier of height U0 . it may quantum mechanically tunnel through the barrier (see [Tie87] for details). The atom can jump over the barrier classically by means of thermal excitation. alternatively. consider oblique incidence and the possibility that the gas atom repeatedly recoils from the grain surface. evaporation and reactions with activation barrier The more or less evenly spaced atoms on the surface of a grain act as force centers and the surface potential has a semi-regular hilly structure. we obtain  ω0 ωs 4  2  π /4 ω0 = ωs mg 4 (D + E k ) (9.1. or not. it was assumed that the term cos(πωs /ω0 ). For the limiting cases. the collision time τcoll is very short and e−iωs t is practically constant. • The adsorbed atom vibrates and thereby swaps its kinetic and potential energy with a characteristic frequency ν0 . Nevertheless.Grain surfaces 281 This can be verified immediately by taking the derivative. For two states separated by an energy difference U0 . ωs . vanishes on average. this ratio is 1. one can get a feeling from (9.2. we learn from formula (9.1 and 1 presents an educated guess. implying sticking. E s . 9.

the time ratio of how long the system resides. Two examples of the time evolution of a one-dimensional collision along the x-axis between a gas atom (subscript g) and a surface atom (subscript s). enlarged 20 times.9) by a Runge–Kutta method.282 Dust and its environment −1 τcoll = ω0 . the gas atom. temperature of grain and gas: Ts = 20 K. the elongation of the surface atom increases. m s /m g = 50 and E s /E k = 1. ωs . The forces between the two particles follow from the U12–6 potential of equation (9.7). Figure 9. The collisional frequency is defined by ω0 = 2(D + E k )/σ 2 m g .7). on average.13) according to which the gas atom rebounds for E s /E k < 1. which is needed for the atom to overcome . For the lower level. the residence time is also the time thop .10) which contains the range parameter b.2. slightly different from (9.2 A.2. is 1. The time coordinate on the abscissa is in units of the vibration period of the surface atom. Bottom: Rebounce with m g = m H . It is evident from the figure that when the gas atom loses energy in the collision. Curves are obtained by integrating the two second-order differential equations (9.0. The vibrational frequency of the surface atom. Tg = 50 K.58×1014 s−1 and equal to the collisional frequency ω0 .05 eV. in these two states. binding energy D = 0. of upper over lower level or. Top: Sticking with m s /m g = 20 and E s /E k = 3. whereas here we use the range parameter σ after (9. the dash-dots. Full curves denote the surface atom. The inverse of the latter can be considered as the effective collision time. or otherwise ˚ sticks. Further parameters are: range of the repulsive potential σ = 0. equivalently.8) and (9. Both numbers for E s /E k agree qualitatively with the approximate formula (9.

If the atom takes N steps of constant length a equal to the distance between two surface atoms. that a reaction in which a molecule on the grain surface is chemically transformed but which requires an activation energy E a .Grain surfaces 283 the energy barrier and move to a neighboring site. If the mean residence time −1 in the upper level is ν0 .4 and 9. However. the tunneling time −1 ttun ∼ ν0 exp 2a 2mU0 is short for small U0 . According to (6. one gets the evaporation time scale −1 (9. thermal hopping is very −1 quick and thop ∼ ν0 . it has no preferential direction when jumping from one site to the next and its path is a random walk (see figures 9.14) U0 by the binding energy E b . A more accurate expression in the case of a symmetric harmonic potential is ν0 = 2n s E b /π 2 m • where n s denotes the surface density of sites. the levels are degenerate (the U here is denoted V in section 6. otherwise thop increases exponentially. The time ttun is very sensitive to all numbers . the time scale for thermal hopping is −1 thop ∼ ν0 eU0 / kT (9.5). the root mean square deviation from the starting position is σ = N 1/2 a.4). For finite U0 . If U0 < kT .16) 9.61)). When an adsorbed atom moves on the surface of a grain by thermal hopping.14). The gas atom on the grain surface then has a high mobility and is not fixed to any particular site. an atom has the possibility to quantum mechanically tunnel through the potential barrier as discussed in section 6. • (9. Replacing in (9.14) where T is the grain temperature. For ν0 . there is energy splitting with an amount E 2 − E 1 that increases as U0 is lowered (see (6.1. When the potential barrier U0 is infinite. as before.63). m and E b are the mass and binding energy of the adsorbed atom. thermal hopping does not work when the grain is cold and kT falls below U0 . Therefore. occurs on average after a time −1 tchem ∼ ν0 e Ea / kT . one may use the frequency of lattice vibration of the grain. typically a few times 1013 s−1 . One can argue.15) tevap ∼ ν0 e Eb / kT . the adsorbed atom needs a time N 2 thop to move a distance Na from the starting point.5 Tunneling between surface sites Because of its exponential dependence on grain temperature in (9.4.

67 × 10−24 g) ˚ physically adsorbed on an ice layer with a spacing 2a = 2 A between force 12 s−1 . tunneling to a site n grid spacings away is not a random walk. U0 /k 10−11 s and E 2 − E 1 10 K. an order of magnitude shorter than the crude estimate given here. Putting the potential centers and vibrating at a frequency ν0 = 3 × 10 100 K.1. n · ttun . the ttun estimates are coarse. Inserting numbers for a ˚ grain of 1000 A radius. The case of a one-dimensional double potential well can be generalized to a perfectly regularly structured grain surface. ttun 4 E . Let us evaluate the separation of the energy levels. The full analysis gives a very similar residence time for an atom in a particular well. its linear dimension is of order N 1/2 and the mobile particle covers in a random walk such a distance in a time τscan N ttun. The time to reach a site n spacings away increases from n ·ttun without scattering to (n 2 / )ttun when defects are taken into account because one needs (n/ )2 steps of length . E 2 − E 1 . we find tscan º 10−6 / s.63) for a hydrogen atom (m = 1. the energy difference between the two split states |E 1 − E 2 | is replaced by the width E of the lowest energy band. and the tunneling time ttun from (6. It consists of as many sublevels as there are interacting wells.18) Provided that n. tunneling can only be important for the lightest atoms.61) and (6. sites surrounded by a high potential which the atom cannot penetrate and where it is deflected in its motion. Obviously. we obtain barrier U0 to 0. i. i. (9.e. but this discrepancy does not worry us considering the uncertainties and simplifications.e. (9. Because of the exponential terms. thus leading to a random walk.17) When it is applied to a hydrogen atom on ice.e.6 Scanning time For an ideal surface. In the earlier expression for ttun .3 times the binding energy E b . in practice for hydrogen or deuterium. to the lattice spacing 2a. Let be the mean free path before the atom is scattered. so scanning is quick.284 Dust and its environment in the exponent. A real grain has defects on its but the well is reached after a time tn surface. . The defect sites invariably arise because the grain is bombarded by soft X-rays and cosmic rays. 9. one calls τscan the scanning time. i. If the surface contains N sites altogether. The mobility of other species is restricted to thermal hopping. Suppose the grid constant of the atomic lattice is of unit length. the calculations yield E/k 30 K and thus ttun 10−12 s. the mass of the gas atom m and the barrier height U0 . where the potential is periodic in two dimensions [Hol70].

20) 2 V is the actual impact velocity and σe = πaeff the effective cross section. This critical value is determined by 2 m e v0 Z e2 = . (9.2. . The enhancement in the cross section follows immediately from the two equations describing conservation of energy and angular momentum for an electron with grazing impact. The number of electrons striking and then staying on the grain: Ye n e vσe (v) = Ye n e πa 2 4π me 2πkT 3/2 ∞ v0 v3 1 + 2Z e2 am e v 2 e−m e v 2 /2kT dv (9.13) at the temperature T of the plasma. the processes that make the grain positive and negative exactly balance. The term in the brackets determines the change over the pure geometrical cross section σgeo = πa 2 . Its cross section for capturing electrons of mass m e and velocity v is σe (v) = πa 2 1 + 2Z e2 am e v 2 . Consider a spherical grain of radius a and charge Z .21) n e is their density. Ye their sticking probability and the bracket .1 Charge equilibrium in the absence of a UV radiation field First we neglect the radiation field. notably • • • the impact of an electron the impact of a positive ion and the ejection of an electron by a UV photon.19) (9.2 Grain charge A grain in interstellar space is not likely to be electrically neutral. 1 Z e2 1 m e v2 = m e V 2 − 2 2 a aeff v = aV. 2 a . The bracket has a value greater than one when the dust is positively charged (Z > 0) and less or equal unity otherwise.Grain charge 285 9. 9. denotes an average over the Maxwellian velocity distribution (5. Mechanisms are at work that tend to alter its charge. In equilibrium. . In the case of negative grain charge (Z < 0). some electrons are repelled and only those whose velocity exceeds a critical value v0 reach the surface at all.

We notice that in a hot plasma. if Z ≥ 0. Heavier ions make the grain more negative but only a little.30) and (A. In equilibrium.28) for the ions and (A.3 that for a fixed temperature. (A. Z adjusts itself in such a way that.2 The photoelectric effect 9. Y bear the subscript i. The impact rate of electrons on a grain of charge Z is n e v σeff . For a plasma with equal density and sticking probability for electrons and singly charged ions.1 The charge balance In the presence of a hard radiation field of mean intensity Jν .21). equals the work Z e2 /a necessary to liberate one unit charge. by order of magnitude. one has to include a term in the charge equilibrium equation (9. σ. one finds akT (9. n. There is a corresponding equation for ions. we can write for the brackets that appear in (9.26).5 2 . the mean kinetic energy of a gas atom.22) that accounts for the fact that UV . We see in figure 9. Without photoemission.286 Dust and its environment Otherwise.32) for the electrons. As one would expect.2. the degree of ionization and the density of the plasma do not appear in this formula. There are corresponding formulae for ion capture where the quantities m. Solving this equation for x = Z e2 /akT and assuming that the ions are protons. where + Z e2 (9. large grains bear a greater charge than small ones but the potential U= Ze a does not change.2. vσ (v) e = vσ (v) i . we get exp Z e2 akT = me mi 1 2 1− Z e2 akT . n e = n i and Ye = Yi . e Interestingly. kT .23) σeff = σgeo (1 e2 /akT /akT ) if Z > 0 Z e if Z < 0 is the effective cross section and v = (8kT /πm e )1/2 the mean electron velocity. we must put v0 = 0. the grain must be negatively charged (Z < 0) because the electrons are so much faster than ions. 9.2. When we evaluate the integrals vσ (v) using the relations (A.24) Z −2. the charge can become very large. (9. the average refers to a Maxwell distribution for electrons and on the right for ions.22) On the left.

yν . hν (9. represents a threshold frequency for photon absorption leading to electron emission. the energy hνt is of order 10 eV and includes the work to liberate an electron from the solid (∼4 eV) and to overcome the potential U of the positively charged grain. νt .Grain charge 287 Figure 9.3. The equilibrium charge of a grain (in units of the charge of an electron) in a hydrogen plasma of temperature T for two particle radii. provided one knows the parameters νt and yν . The second parameter. photons can chip off electrons from the bulk material of the grain. one can then neglect the impact of positive ions. This certainly happens near a star of early spectral type but also in diffuse clouds permeated by the average interstellar radiation field. Putting the sticking probability Ye of electrons equal to one. . Photoelectric processes are absent. is the yield for photoemission. The first.25) We divided under the integral by hν because we wanted the number of absorbed photons. the grain will be positively charged.25) for the charge Z . we find for the charge balance with respect to impinging electrons and absorbed photons that ∞ νt ne v 1+ Z e2 akT = 4π Jν Q abs ν yν dν. If photoemission is strong. One can immediately solve (9.

E ν . after subtraction of the electrostatic grain potential U . the higher the chance for deexcitation.3 The photoelectric effect and gas heating Photoemission can also be important for heating the interstellar gas. The material constants are uncertain but various evidence points towards yν ∼ 0. exceeds the average thermal energy 3 2 kT of a gas particle. The yield yν for electron emission induced by photons of frequency ν is now defined via the equation yν div S d V = C V V e−x/ le div S d V (9.2. The mean free path of electrons in the bulk material. In the end. when such an electron has reached the surface.1 [Wat72].2. The likelihood P that the absorption leads to emission of a photoelectron is assumed to have the form P = C exp(−x/le ). The heating rate 2 due to one dust grain is therefore H = 4π πa 2 ∞ νt Jν Q abs ν yν E ν − U − 3 kT dν.27) . one gets d Wa = − div S d V where S is the Poynting vector. le .288 Dust and its environment 9. The deeper the subvolume d V below the grain surface. Because of the factor ˚ e−x/ le .2 The photon yield The yield yν for photoemission may be estimated from the following physical picture based on classical electrodynamics [Pep70]. it is evident that small grains (radii a ∼ 50 A) are much more efficient in ˚ photoemission than big ones (a ∼ 1000 A). is collisionally imparted to the gas. When one applies the Gauss theorem (A. The energy absorbed by a subvolume d V of a grain is given by (2. The factor C incorporates the following two probabilities: for excitation of an electron to a ‘free’ state and. 2 hν (9. one can compute the internal field of the grain (the relevant formula is (2. Let the subvolume be at a depth x below the surface. 9. for penetrating to the outside and not being reflected.45)) and thus the Poynting vector.2. As the mean kinetic energy of a photoejected electron. the electron is thermalized and its own average energy will then also be 3 kT .2. is of order ˚ 30 A. The exponential term exp(−x/le ) gives the probability that the electron reaches the surface at all and is not de-excited in any of the scattering processes on the way.52). the excess energy.13).26) where the integrals extend over the whole grain volume V . From Mie theory. possibly shorter for metals and longer in dielectrics.

in three-dimensional space.2 The drag on a grain subjected to a constant outer force Let us consider the one-dimensional motion of a heavy test particle (grain) through a fluid or gas in more detail. unaffected and purely Maxwellian. whereas in the rest frame of the gas it is at zero velocity. The y. The gas has a number density N at temperature T and its atoms have a mass m M. . starting from position R0 = 0. If the vectors Li are of constant length L but in an 2 arbitrary direction. there it is the dominant heating mechanism. .5 the verification of formula (9. Of course. In a steady state.29) On the right stands an accelerational term M x and a dissipational term. The number of atoms in the velocity range [vx . x = 0 and F = µV .28) by induction because the average of R N · L N−1 obviously vanishes.3. ¨ ˙ (9. Let the grain advance in the positive x-direction. grows like R2 = N L 2 . Figure 9. N(vx ) dvx = N m 2πkT 1/2 e−m(vx +V ) 2 /2kT dvx . The maximum of N(vx ) is now at vx = −V . 9. one has arrived at the position R N = R N−1 + L N . of course. 9.27) alone but only from the balance with the cooling processes.14) but has an offset in the exponent. a sequence of steps defined by the vectors Li .3. Suppose one applies a constant outer force F on the test particle of mass M. After N steps. ¨ We determine the force F in (9.and z-axes are.4 shows a two-dimensional random walk and figure 9. vx + dvx ] that hit the grain head-on (vx < 0) equals −πa 2 vx N(vx )dvx and each atom imparts in an .3 Grain motion 9.Grain motion 289 The photoeffect has to be invoked to explain the fairly high temperatures of 50 . the mean square of the distance. The ¨ coefficient µ in the latter depends on the properties of both the fluid and the test particle. 100 K observed in HI regions.29) needed to move a spherical grain of geometrical cross section πa 2 at a constant velocity V through a gas. The velocity distribution N(vx ) of the gas atoms along this direction with respect to the grain is no longer given by equation (5.1 Random walk Suppose one makes. RN . N (9. It then experiences a drag from the fluid or gas molecules that is proportional to its velocity x = V and the equation of motion of ˙ the test particle reads: F = M x + µx. the temperature does not follow from the rate H in (9.28) in a numerical experiment.28) It is straightforward to prove (9.

290 Dust and its environment Figure 9. (9. one gets the retarding force on the grain. −46). one has to integrate over all velocities vx < 0. Substituting in (9. 0) and ending after 2700 steps of equal length L = 1 at the big dot near (5. which leads to the expression πa 2 Nm • m 2πkT 1/2 0 −∞ 2 vx e−m(vx +V ) 2 /2kT dvx .30) w = vx + V and taking the momentum difference between front and back. F = πa 2 NmV 8kT mπ 1/2 m 2πkT 1/2 0 ∞ we−mw 2 /2kT dw .30) Suppose the grain moves much slower than sound. inelastic collision. (9. Random walk in two dimensions starting at the coordinate center (0.31) . the momentum mvx . To obtain the total momentum transfer.4. The term that does not cancel out under this operation is F = πa 2 4NmV and therefore.

(9.8) but there the letter F means flux). πa 2 .Grain motion 291 Figure 9.4 is the jittery line. The drag is proportional to the geometrical cross section of √ grain. In elastic collisions. the drift velocity. • When the grain moves highly supersonically.5. the forces would be larger by about a factor of two.28). If L ∗ is the stellar luminosity and r the distance of the grain to the star. we may approximately set the cross section for radiation pressure. the and the drift velocity V . According to (9.7) and (2.32) F is now proportional to the square of the velocity. The actual distance in the numerical experiment of figure 9. equal to the geometrical cross section σgeo = πa 2 . V = 2 Nm 8kT 4πcr . With the mean gas velocity v = 8kT /mπ. for subsonic motion. C rp . in a random walk one travels after N steps a mean distance √ N L (smooth line). An astronomically important case of an outer force acting on dust particles is provided by radiation pressure (see formulae (2. corresponding to an efficiency Q rp = 1. there is only a force acting on the front side and the momentum transfer is simply F = πa 2 NmV 2 . the friction coefficient becomes µ = πa 2 Nm v . In the case of direct illumination by a star. is then mπ 1/2 L∗ .

35) Grains smaller than acr are expelled. too. so 2 M Vx2 = M Vy = M Vz2 . the gas density is always low and the gas drag unimportant. one finds that in stellar environments supersonic drift speeds are easily achieved. If M and V denote the mass and velocity of the dust particle. The kinetic energy in a 20 K gas only V = 28 cm s of the grain. equally distributed among the three degrees of freedom.33) (9. E kin. N and T .1 µm) are acr 3. Evaluating this formula for typical numbers of L ∗ . the grains perform a Brownian motion. .5 .33)).5 blown away by radiation pressure. The removal is always rapid as one can show by integrating radius falls with M∗ the outward acceleration v = 3L ∗ /16πcρgr ar 2 (see (9. For instance.5 g cm−3 has −1 : it is practically at rest. From the balance between the two.3. L ∗ σgeo G M∗ M = 2 r 4πcr 2 one obtains a critical grain radius independent of the distance: acr = or acr µm 0. (9.24 L∗ L M∗ M 3L ∗ 4πc Gρgr M∗ −1 (9.34) . so all particles with sizes of interstellar grains (∼0. However.292 Dust and its environment N. 2 The Brownian velocity V follows from V = 3kTgas . is. It has a translatory and a rotational part but for the moment we neglect rotation. on average.3 Brownian motion of a grain Under equipartition between dust particles and gas. Around main sequence stars. ˙ 9. M (9. In such circumstances. C rp is much smaller than the geometrical cross section and V accordingly smaller. For low mass stars L ∗ ∝ M∗ and the critical 2. in many configurations grains are not directly exposed to starlight but irradiated by infrared photons to which the starlight has been converted through foreground matter. 3 2 kTgas = E kin = 1 M V 2 .36) ˚ Even a small grain of a = 100 A with bulk density ρgr = 2. A grain is attracted towards the star of mass M∗ by gravitation and repelled by radiation pressure. It is a fundamental result of thermodynamics that in equilibrium the mean kinetic energy of a gas atom is equal to the mean kinetic energy E kin of a grain. a 10 M star has 2.4 mm. m and T refer to the gas.

at a certain instant. In three dimensions. where p2 is the average change in the square of the momentum. must be constant in time because the particle has no memory about its past. per unit time due to collisions with gas atoms. as in a random walk. it will have completely lost its memory about p0 . p2 = 6µkT.3. 2 dt dt 2 dt On taking averages. (9. when all directions are equal. tdis = µ= M . the rate of change of the mean square distance. therefore. when the mass of the colliding atoms equals the mass of the grain.29) is also correct with respect to the Brownian motion of interstellar grains when the retardation is not due to an external force but to internal friction.40) . in a gas of number density N. in three dimensions. (9.29) by x. we get p2 = 2µkT . So in equipartition. p2 . tdis (9. this equation shortens. the friction coefficient can be expressed through the damping time tdis .Grain motion 293 The one-dimensional equation (9.e.37) To find the mean distance that a grain travels when there is no external force. After a disorder or damping time tdis. On ˙ averaging. In a quasi-stationary state at constant velocity V . to 1 µ d x 2 /dt = M x 2 = kT . Second. that F x = 0 because x and F are unrelated and the force F due to collisions with gas atoms is completely stochastic in its direction. the momentum p0 . i. the associated heat loss F V must be compensated by a permanent acceleration p2 /2M. (9. the square of the distance r grows linearly with time. M d dx2 µ dx2 − Mx2 + ˙ . we multiply equation (9. The latter always involves dissipation of energy.38) r 2 ≡ x 2 + y2 + z2 = µ F x = M x x + µx x = ¨ ˙ 9. the disorder or damping time is M . d x 2 /dt. 6kT t. Therefore. 2 Therefore. Thus we require F V = p2 /2M and because M V 2 = kT in equilibrium at temperature T .4 The disorder time Suppose a grain has. One impinging gas atom of mass m M and velocity v changes the momentum of √ the grain statistically by mv. when M V 2 = mv 2 the momentum of the grain is profoundly altered after M/m collisions. first.39) Nmvπa 2 Because F = µV = M d V /dt M V /tdis. and Z atoms by mv Z . we note.

grain ˚ parameters.40).and 2 y-component of the velocity. the chance that during one time step the grain is hit by an atom within a certain velocity interval. mass M = 1. radius 100 A.5 × 109 s.39) equals 1. the time evolution of the mean square deviation r 2 = x 2 + y 2 . The actual occurrence of such an event is prompted by a random number generator.14) into 100 velocity intervals. The parameters are such that a gas atom impacts the grain about once every 1000 s. The computations pertaining to figure 9. The figure shows the x. Figure 9. One finds good agreement between numerical experiment and (9. Numerical experiment of a two-dimensional Brownian motion: gas parameters.05 × 10−17 g. hydrogen atoms at T = 100 K with number density 104 cm−3 . either from the front or back. To follow the stochastic evolution. It is roughly the interval after which there is a change in the direction of the velocity vector of the grain by 90◦ or in its kinetic energy by more than 50%. of course. and likewise for N(v y ).38) only after integrating over many disorder times tdis because one disorder time corresponds to one step in the random walk of figure 9. initially 1 M Vx = kT and V y = 0. Because of the large 2 time interval displayed in the plot. we bin the velocity distribution of the gas atoms N(vx ) of (5. Therefore one can check whether r 2 (t) fulfils the theoretical prediction of equation (9. is small.6 displays a numerical experiment of a two-dimensional Brownian motion assuming inelastic collisions. The disorder time according to (9.4.6 also yield.294 Dust and its environment Figure 9.38) with the dissipation constant µ from (9.6. . Choosing a time step of 103 s. the actual jitter is graphically not fully resolved.

29) we find. and ρ and ν are the density and kinematic viscosity coefficient of the gas (see the empirical relation (15.41). a force F = 6πηav (9.89)). that µ = 6πηa. the gas density ρ = Nm and the velocity squared. we consider gases of sufficiently high density such that the mean free path of the atoms after (15.41). • A laminar flow of velocity v around a sphere of radius a exerts. This somewhat strange result can be understood from (15.4) or planetary atmospheres. F is now proportional to the cross section πa 2 . on the surface of the body. A sphere rotating at velocity v at the circumference is retarded by a torque τ = Fa with F also approximately given by (9.Grain motion 9. the retarding force is F = ηA v . D (9. When the flow is turbulent.74) and equation (15.5 Laminar and turbulent friction 295 In this section. For a rectangle of area A = bl and with normal perpendicular to the velocity v. in which the gas velocity increases from zero. according to Stokes law.41) where η = ρν.3. if one writes for a sphere: F = νρ ∂v 2πa 2 ∂a • and replaces the derivative ∂v/∂a by v/a. to v at the boundary of the layer. To obtain the coefficient 6π rigorously is tiresome.93) and is of order one.41) is linearly proportional to the velocity v but also to the radius a. by comparison with (9.32).90) is smaller than the dimension of the body moving through the gas. The coefficient cW depends on the Reynolds number (15. But even in a turbulent flow there is a laminar surface layer of thickness D.42) The work required to renew the kinetic energy of the surface layer continually is 1 v2 = v 2 bv Dρ. For the coefficient µ in (9.74). So the discussion does not apply to grains in the interstellar medium but is relevant to dust particles in accretion disks (section 15. the decelerating force becomes F = cW πa 2 ρv 2 as in (9. The retarding force F in (9. Fv = ηbl D 2 .

With the appropriate numbers for air (kinematic viscosity ν 0. we study a falling rain drop. When one takes into account that the velocity in the surface layer increases linearly from zero to v. For small Reynolds numbers. η = ρν −4 g cm−1 s−1 ). vρ Inserting D into (9. when the drop has a radius a ∼ 0. a (9.45) is needed when evaluating the evaporation of small grains. 6πηaV = πa 2 ρV 2 . The Reynolds number Re = aV /ν of (15.42) gives a force that is in between the purely laminar and turbulent cases.2 × 10 and laminar deceleration become equal. for large ones.1 µm in radius. to the surface requires a few hundredths eV. −3 g cm−3 .151T erg cm−2 . it is more than ten times the atmospheric pressure at sea level for a rain drop 0. . to have (for the same mass) a surface larger than necessary. the turbulent 1.15 cm2 s−1 . Not to be a sphere. It is slightly temperature dependent.6 A falling rain drop As an example. The rain drop is pulled by gravity with the force F = 4πa 3 gρw/3 (gravitational acceleration g = 981 cm s−2 . (9. To create an area d A of surface takes the work dW = ζ d A (9.43) where ζ 75 erg cm−2 is the surface tension of water. V ∝ a 1/2.e. the terminal velocity of the falling drop changes like V ∝ a 2 .3.93) is then slightly below 10. one finds by integration for the thickness of this so called Prandtl layer D= 6ηl . a more precise formula is ζ(T ) = 116.44) One roughly estimates that to lift a molecule from the interior of the drop. water density ρw = 1 g cm−3 ). where it is surrounded by other water molecules on all sides. This has the advantage over an interstellar grain that its properties are known and that it is shaped like a perfect sphere.8 × 10 density ρ 1. almost. would be energetically quite disadvantageous. By equating the p d V work (here d V = infinitesimal volume) to the work required to create new surface.45) The pressure can become high.296 Dust and its environment It is assumed that the side l of the rectangle runs parallel to the flow and b perpendicular to it.1 mm and sinks at a velocity V ∼ 1 m s−1 . Equation (9. 9. it also follows that the pressure p inside a drop of radius a is given by p= 2ζ . i.82 − 0.

4πr 2 In thermal balance. For instance. the distance of the grain to the star shrinks according to L ∗ σgeo dr =− · .5 g cm−3 for the grain material. (9. radius a and geometrical cross section σgeo = πa 2 circling a star at distance r with frequency ω and velocity v = ωr .47) τPR = L ∗ σgeo or. during the lifetime of the Sun (∼ 5 × 109 yr) only bodies we get v = G M∗ /r and . the stellar photons that are absorbed travel in a radial direction and carry no angular momentum. If we associate with the absorbed energy E a mass m phot = E/c2 . assuming a density of ρgr = 2. mπc2 2 r (9. dt c 4πc2r 2 m Because for a circular orbit √ v2 G M∗ = 2 r r dr d = . the absorption coefficient of the grain is C abs σgeo and the particle absorbs per unit time the energy E= L ∗ σgeo . whereas the emitted photons do because they partake in the circular motion of the grain around the star. the same amount is re-emitted.7 The Poynting–Robertson effect 297 Consider a grain of mass m.48) yr µm L AU The Poynting–Robertson effect is an efficient way to remove dust in the solar system. we find the time τPR that it takes a initial radius r to the stellar radius R∗ grain to fall into the star. (9.46) dt 2πr mc2 When we integrate the equation dt = −(2πmc2/L ∗ σgeo )r dr from some r . dt 2r dt Due to the loss of angular momentum. in more practical units. The angular momentum of the grain is = mr 2 ω = mr v. Let M∗ and L ∗ be the mass and luminosity of the star. Seen from a nonrotating rest frame. Because there is mostly optical radiation. the angular momentum of the grain decreases per unit time through emission by L ∗ σgeo E d = −m photr v = − 2 r v = − .Grain motion 9. a τPR L ∗ −1 r 2 = 1700 .3.

298 Dust and its environment with a diameter greater than 6 m (!) can have survived within the orbit of the Earth. they are born. In such a reference frame. hot gas or shocks. including dust. a short period compared to the Hubble time.1 Mass balance in the Milky Way Grains do not live forever but have a finite lifetime. The continuous mass loss of gas in star formation must be offset by an equal mass input. Coagulation: in dark clouds. Birth: . The total gas input in the Milky Way from red giants is about 1 M yr−1 and about 1% of this is expected to be in the form of dust. gas atoms condense onto grain surfaces to form mantles. modified and destroyed in processes such as mainly in old (evolved) stars with extended shells but possibly also in novae and supernovae (SN). IRG and ISN .4. The Poynting–Robertson effect may also be described by an observer in a frame corotating with the grain. It comes mainly from planetary nebulae which supply about 80% of the gas. As part of the dynamic. respectively. Therefore the numbers given here are not precise at all but they do show the order of magnitude. is processed in stars. all interstellar matter. mainly in star formation. As gas and dust are intimately linked. The star formation rate implies that within one billion years. red giants (∼20%) and supernovae (∼10%?). Let us denote the input rates of dust due to planetary nebulae. and express them in solar masses per year. OH/IR and carbon stars) as high temperature condensates. grains collide.01 M yr−1 . 9. This phenomenon arises because one has to add the velocity of the photon and the grain according to the rules of special relativity. the stellar photons approach the grain not exactly along the radius vector from the star but hit it slightly head-on in view of the aberration of light. stick and form larger particles. Particles existing today smaller than 6 m must have been replenished from comets or asteroids. The rates at which this happens are quite uncertain. Most grains probably form in the wind of red giants (like M stars.1. we begin with a few remarks about the mass balance of all interstellar matter in the Milky Way.4 Grain destruction 9. Accretion: in dark clouds. In a steady state. continually changing interstellar medium. So the input rate of dust from red giants is IRG ∼ 0. four important numbers are listed in table 9. red giants and supernovae by IPN . Destruction: partial or complete. The photons thus decrease the angular momentum of the grain and force it to spiral into the star. The Poynting–Robertson effect also works when the photons are not absorbed but isotropically scattered. dust formation is balanced by dust destruction.

suppose that 3 M of heavy elements are ejected per SN explosion.1. .1).2 Destruction processes Grains are destroyed in various processes (recommended reading [Sea87] and [McK89]) such as the following ones. But by far the greatest sink is star formation and the destruction rate tSF gives the maximum lifetime of a grain.01 M yr−1 .03 M yr−1 . tSF = Mgas ∼ 109 yr.03 yr−1 Planetary nebulae. possibly η ∼ 0. atoms are chipped off only with an efficiency η smaller than one. The solid is heated up to the condensation temperature. This happens near luminous stars. also in the diffuse medium (section 14.Grain destruction 299 Table 9. to a certain extent in HII regions (section 14. a considerable fraction is locked up in CO. are less effective than giants in forming grains so that probably IPN < IRG . The loss rate of dust associated with astration equals ISF = Mdust /tSF ∼ 0. τSFR (9. For each input rate.49) Mdust .1. • Evaporation. To estimate the contribution of supernovae to the dust balance. one gets a dust input rate from supernovae comparable to that of red giants of ISN ∼ 0. they are cycled at least 10 times between cloud and intercloud medium. Global parameters of the Milky Way relevant to the mass balance of dust.4. although they inject more gas into the interstellar medium. Sputtering.4). H2 O or otherwise and is unable to form grains. one can define a formation time scale t by t= 9. either neutrals or ions. Total gas mass Total dust mass Star formation rate Supernova rate Mgas ∼ 5 × 109 M Mdust ∼ 3 × 107 M τSFR ∼ 5 M yr−1 τSN ∼ 0. Although this is all potential dust material.3) when the grains are extremely small or have volatile mantles. I • If grains should live a billion years. If one assumes that each explosion creates 0. The threshhold impact energy for liberating an atom from the solid phase is a few times larger than the binding energy of an atom (∼5 eV). But even for impact energies above the threshhold.3 M of dust and if there are three supernova events per century in the Milky Way (see table 9. Atoms are ejected from a grain by colliding gas particles.

2.1) and adopts a gas density n = 1 cm−3 . the component of B parallel to the shock front. The collisional velocities between the grain and the gas atoms are now non-thermal and non-thermal sputtering dominates. . If one determines the rate of impinging protons using (9. The final velocity vgr is limited by drag forces from the gas. The observational evidence comes from clouds of high velocity (v > 100 km s−1 ) where more than half of Si and Fe is in the gas phase. whereas the binding energy is only 5 . the shock becomes radiative when v ≤ 200(n 2 /E 51 )1/14 km s−1 . grains will be vaporized if the efficiency for knocking off atoms is greater than 1%. the momentum vMSNR is constant. • To estimate the importance of shock destruction. one finds that at least small grains are likely to be eroded. the grain velocity vgr increases because its magnetic 2 moment µ = mvgr /2B is conserved and the grain is accelerated (betatron acceleration). The decelaration is proportional to the inverse of the grain radius. whereas in normal clouds only a fraction of order 1% of these atoms is in the gas phase. 2 at later times. A supernova rate . As B .4. 9. the remnant of mass MSNR and velocity v is adiabatic and its kinetic energy E = 1 MSNR v 2 conserved. The mechanism of grain destruction depends on the type of shock: • In a fast shock. In a low-velocity shock. small grains may survive. During the first (Sedov) stage. so big grains become faster than small ones and are more easily destroyed. A shock with v ∼ 300 km s−1 corresponds to a temperature T ∼ 3 × 106 K. cooling is slow. Sputtering is thermal because the velocities of the gas atoms are thermal. . E 51 is the energy expressed in units of 1051 erg. 10 eV for refractory material. A charged grain of mass m will gyrate in the magnetic field B. The shock is radiative and the density jump is much greater than four. usually in grain–grain collisions.300 • • • Dust and its environment Grain–grain collisions after acceleration in a magnetic field. is compressed. According to [McK89]. In a gas of density n = 1 cm−3 and with the standard explosion energy E 51 = 1. the cooling time is smaller than the expansion time of the supernova remnant. Shattering which is the destruction of a grain into smaller units. A velocity of 100 km s−1 corresponds to ∼1000 eV kinetic energy per atom.1 Destruction in shocks Besides star formation. The shock is adiabatic or non-radiative and has a moderate density jump of four. Photodesorption which is the ejection of grain atoms by photons. High-velocity clouds are interpreted to be fragments of an expanding and shocked supernova shell. we turn to the theory of supernova blasts. Grains may also collide with one another. one finds for the mass of the supernova shell MSNR ∼ 103 M at the onset of the radiative phase. grains are mostly destroyed in shocks associated with supernova remnants (SNR).

Grain formation 301 τSN = 0.1% of all interstellar matter. however. 9.5 Grain formation We derive approximate equations that describe the nucleation of monomers into large clusters. in agreement with observations. this requires an additional dust source for which accretion in the interstellar medium.5 × 109 yr). like liquid and gas.4. is usually invoked. the material constants are poorly known. 104 K. For the total dust in the Milky Way.50) . The examples refer to water because its properties are experimentally well established but the physics apply. Therefore the mantle material with a low binding energy is confined to dark clouds. are in equilibrium at temperature T .2 Sputtering in the coronal gas Sputtering of refractory grains also occurs in the hot coronal gas (T = 106 K) where all atoms are very energetic (∼100 eV).3 Sputtering of grain mantles Similarly simple computations show that volatile grain mantle material which has low binding energies per atom (∼0.5. with some modifications. n > 10 cm−3 ). the associated timescale (∼4 × 108 yr) is substantially shorter than the joint dust injection time scale from supernovae and red giants (∼1. negligible because it contains only ∼0.03 yr−1 then implies that all interstellar material of the Milky Way is processed in a time tSNR = Mgas ∼ 2 × 108 yr. 9. destruction in the coronal gas is. . however. There. 9. especially in dark clouds.2. Assuming there a gas density ˚ n = 10−3 cm−3 .1 eV) is effectively sputtered in the diffuse interstellar medium (T ∼ 103 . In a steady state. Q dP = dT (Vg − Vf )T (9. also to interstellar grains. MSNR τSN As the high-velocity clouds indicate an efficiency for dust destruction of 50%. 9.2. Low-velocity shocks are also efficacious. grains of 1000 A size are eroded in ∼109 yr.1 Evaporation temperature of dust When two phases. the vapor pressure P changes with T according to the Clausius–Clapeyron equation.4. .

302 Dust and its environment 1 r p ∞ Vg. and Q is the heat necessary to evaporate 1 mol. the total effective work done is. Then the vapor is condensed in another heat bath at slightly lower temperature T − d T and pressure P − d P. which is the phase transition between the solid and gas phases.7).r Figure 9. where Vg and Vf are the volume per mol in the gas and f luid phase. In this way. T ). the Boltzmann constant k and L are related through k= R . During the phase transition from fluid to gas. with obvious modifications. ∞ 2 pr Vg. the substance expands from Vf to Vg and the vapor pressure P does the work P (Vg − Vf ) by lifting a piston. and the gas constant R. also valid with respect to sublimation.7. The shaded area marks the fluid. L is Loschmidt’s number. and thus applies to the evaporation of interstellar grains and their mantles. A reversible cycle employing two vessels to find the vapor pressure of a drop of radius r . Finally. see text.02 × 1023 molecules. We remember that 1 mol consists of L = 6. in view of the first law of . one brings the system back to its initial state (P. respectively. L (9. one returns the latent heat Q and the work (P − d P)(Vg − Vf ). The fundamental formula (9.51) The Clausius–Clapeyron equation is. this requires heat Q. one totally vaporizes at constant temperature T one mol of a liquid in a vessel closed by an adjustable weight (of the kind as depicted on the lefthand of figure 9. • As the internal energy of the system is the same at the beginning and the end of the cycle.50) is derived in a thought experiment from a Carnot cycle: • First.

equal to the evaporation pressure. P0 and T0 have to be determined experimentally. respectively. NH3 and CH4 by ∼30 K. water ice evaporates at T ∼ 120 K as one can read from figure 9.50) by Because Vf Q d ln P = dT RT 2 which has the solution P = P0 e−T0 /T T0 ≡ Q/R. no heat engine can do better. If it is ice.54) is very general and does not say anything about the details of a specific phase transition.4 cal to liquify it at 0 ◦ C. it takes 79. One gram of H2 O. In cooling the vapor by T = 100 K from 100 to 0 ◦ C.54) with properly chosen constants.1 cal to vaporize the water.184 × 107 erg the total sublimation energy per H2 O molecule at 0 ◦ C is. Because 1 cal = 4. Ammonia and methane go into the gas phase already at about 90 and 70 K.8.54) (9. silicates (1800 K) and graphites (2500 K).64 × 10−13 erg corresponding to a temperature of 6260 K. At hydrogen densities of order ∼106 cm−3 . we can approximate (9.8 assuming an H2 O abundance of 10−7 . Typical evaporation temperatures of other astronomically important species are: hydrogen (3 K). This agrees nicely with the fit parameter T0 = 6170 K in figure 9. They can be neatly fit over the whole range by equation (9. Vg . therefore. Laboratory data for water ice from 0 ◦ C down to −98 ◦C are displayed in figure 9. kT = 8.49) with dU = 0).22 × 1022 molecules. (9.Grain formation thermodynamics (see (5. The evaporation temperature Tevap in interstellar space of a species X frozen out on grain mantles follows by setting its partial pressure in the gas phase. (9. has N = 3. Therefore. according to the second law of thermodynamics. then 100 cal to heat it up the water to 100 ◦C and another 539. the result is rather insensitive to the gas density. P0 e−T0 /Tevap = n X kTevap .53) The affinity to the Boltzmann distribution is not only formal but physical because atoms in the gas are in an energetically higher state than atoms in a solid. CO. assuming six degrees of freedom per molecule. one gains 3Nk T = 1.52) and.8. Equation (9. n X kT . CO2 (20 K). d P · (Vg − Vf ) = Q − Q . The efficiency of converting heat Q into work is η= d P · (Vg − Vf ) dT = Q T 303 (9. in whatever form.37 × 109 erg. .55) Because of the exponential dependence of the vapor pressure on temperature. A rise in density from 106 to 1012 cm−3 increases the evaporation pressure for H2 O.

easier in grains of small radius r and their vapor pressure pr is higher than that over a flat surface. The upper horizontal line is for n X = 105 cm−3 . Here. The fit is extrapolated to lower temperatures.8.7). Evaporation is.0 × 1013 dyn cm−2 or 3. the full curve is a fit according to (9. p∞ .2 Vapor pressure of small grains Condensation and evaporation of small interstellar grains has the peculiarity that the transition surface between the solid and the gas is not flat but curved. To compute pr . This has fundamental consequences.5. The lower horizontal line shows the partial gas pressure.0 × 1010 mm of Hg column.54) with T0 = 6170 K and P0 = 4. (1) In vessel 1 is a small liquid sphere of radius r at temperature T in equilibrium with its vapor of pressure pr (figure 9. one performs an isothermal reversible cycle of four steps. Vapor pressure of ice. therefore. Experimental points (dots) from [Wea77]. 9. Such an H2 O pressure prevails in interstellar space when the total gas density n H = 106 cm−3 and the water abundance (in the gas phase) is 10−7 . The volumes per mol of gas and .304 Dust and its environment Figure 9.43). the formulae for Wi are elementary. P = n X kT in mm Hg column for a number density n X = 0. we write down the work Wi that is being done in each step.1 cm−3 . As molecules leave the grain. the surface area shrinks by an amount d A thereby creating the energy d W = ζ d A according to (9.

3 Critical saturation To estimate the critical saturation parameter scr > 1 at which clusters are created. (3) The 1 mol of vapor is pressed into vessel 2 where the liquid has a plane surface and is under its vapor pressure p∞ .r and Vf . 2ζ /r .57) is greater than one. is much bigger than pr − p∞ .56). Let v0 denote the volume of one molecule in the liquid phase. Condensation of the first seed grains is thus possible only when the partial pressure p exceeds p∞ or when the saturation parameter s (capital S for the entropy) defined by s= p p∞ (9. 9. At pressure pr given by (9. we take from vessel 2 with a syringe 1 mol of liquid. a smaller one will evaporate. W2 = RT ln( pr / p∞ ). (2) Isothermal expansion from pressure pr to p∞ of the 1 mol of gas that has been vaporized in step (1). we calculate the work W needed to form in a reversible cycle a drop of radius . W3 = −RT. When 1 mol has become gaseous. so r is the critical cluster radius. a bigger drop will grow indefinitely.Grain formation 305 fluid are Vg. respectively. W4 = Vf p∞ . W1 = pr Vg. As the cycle was isothermal (d T = 0).r − pr + 2ζ r Vf . The vapor is then said to be supersaturated. a sphere of diameter 2r is in equilibrium with the vapor. Because RT = pV and because the pressure in the drop. (4) To complete the cycle. so all the time r = constant. = p∞ kT r (9. we obtain ln pr 2ζ v0 .5. We evaporate the droplet but in such a way that while evaporating we inject with a syringe into the sphere against the internal pressure p = 2ζ /r of (9.45) and the outer pressure pr the same amount of liquid as is being lost to the gas. the sum Wi must be zero.56) This tells us that the vapor pressure pr = p∞ ·e2ζ v0 / kT r equals p∞ for big spheres but increases exponentially when the radius becomes very small.

3 (9. the term would be proportional to n 2 . to subtract from the system the heat Q = W . only connect to their nearest neighbors. The drop consists of n atoms and is formed in a vessel with vapor at pressure p and temperature T .59) . so W = Wi . according to the first law of thermodynamics. (4) Form there from the liquid a drop of radius r (and bring it back to the first vessel). Although the entropy of a macroscopic system can only increase. The work in each of the four steps is denoted by Wi . which goes as n 2/3 . Therefore. It is proportional to the number of molecules n because the molecules possess shortrange forces and.43) equal to Boltzmann constant k multiplied by the logarithm of the number of states. For longrange forces. S decreases because a liquid represents a state of higher order than a gas. (3) Press the n atoms into another vessel. Formation of a seed grain comes about by such an entropy fluctuation. As the entropy is after (5. introduces a correction because atoms on the surface are attracted only from one side. S = k ln . also at temperature T but containing a liquid of flat surface under vapor pressure p∞ .57).51)) S = Q/T = 4πζ r 2 /3T. When we insert the saturation parameter s = p/ p∞ of (9. one has. therefore. As the energy of the system stays constant over the cycle. W2 = −nkT ln( p/ p∞ ). the probability P = 1/ for the formation of a seed is given by e−S/ k .58) The term −nkT ln( p/ p∞ ) is the potential energy of the drop. (1) Remove n gas atoms from the vessel. W3 = nkT.306 Dust and its environment r . (9. on a microscopic level all processes are reversible and S fluctuates. The other term 4πζ r 2 . W4 = 4πζ r 2 . (2) Isothermally expand these n atoms from p to p∞ . For the sum we get W = −nkT ln( p/ p∞ ) + 4πζ r 2 = 1 ζ · 4πr 2 . during formation of the droplet the entropy of the system falls by (see (5. simple algebra yields P= 1 = e−S/ k = exp − 4πζ r 2 3kT = exp − 2 16πζ 3v0 3k 3 T 3 ln2 s . W1 = −nkT.

˙ (9. the exponent S/k ∼ 116/ ln2 s. When v0 is the volume of one molecule. The number density of the gas molecules (they are monomers) is. mean velocity and collisional cross section of the vapor atoms.56). therefore. A small cluster or droplet of n molecules is an n-mer with concentration cn . So J = N 2 vσ P is the rate at which seeds form and the condition N 2 vσ P ∼ 1 yields scr . One finds typical values of the critical saturation parameter for water at room temperature around 10 which implies in view of (9. there are.60) This is a very general equation. For carbon compounds (ζ ∼ 1000 erg cm−2 . If atoms impinge on a drop at a rate βn An and evaporate from it at a rate αn An . αn and βn .61) . all the physics is contained in the coefficients for evaporation and accretion. one speaks of homogeneous nucleation. A cluster has a radius rn and a surface area An = 4πrn . The probability P is most sensitive to ζ /T and. v0 = 9 × 10−24 cm3 ) at T = 1000 K.57). If N. in contrast to heterogeneous nucleation when different molecular species or ions are present. n and rn are related through nv0 = 4π 3 r . It is exactly this property which allows us to estimate the critical value scr as follows. ˚ o 9. v and σ denote the number density.5.1) and pn /kT from (9.Grain formation 307 For water (ζ = 75 erg cm−2 . where v is the mean velocity of (9.0 × 10−23 cm3 ) at temperature T = 275 K. If the gas has only one kind of atoms or molecules of mass m. N 2 vσ atomic collisions each leading with a probability P to the formation of a seed. the factor N 2 vσ has very little influence. The outcome of the simple calculation depends entirely on the exponent in (9. S/k ∼ 500/ ln2 s.4 Equations for time-dependent homogeneous nucleation Let us study the creation of seed grains in a kinetic picture. per second. if ζ /T is fixed. 2πmkT (9.56) a critical seed radius rcr of a few angstr¨ ms. 3 n Consider a gas of constant temperature T and pressure p = c1 kT with saturation parameter s = p/ p∞ > 1 after (9. the concentration cn changes with time like cn = −cn βn An + αn An + cn−1 βn−1 An−1 + cn+1 αn+1 An+1 . v0 = 3. then pn αn = √ 2πmkT p β= √ . 2 denoted c1 . to s. changing from practically impossible to highly likely in a narrow interval s.59) which has to be of order one. If we assume that atoms are added to the 2 2 grain at a rate πrn v c1 and leave at a rate πrn v pn /kT .

.62) We can now transform equation (9. G n is. Jn = J s = constant > 0.3. 2 (9. Figure 9.63) 0 The equilibrium concentrations cn are given by the Boltzmann formula 0 0 cn = c1 e − G n / kT (9. for a system at constant temperature and pressure. (9. 0 Jn = −cn β An cn+1 0 cn+1 − cn . When they have climbed the barrier.308 Dust and its environment In this case. nothing changes with time and ∂/∂t = 0. it only has to be a positive constant.60) into the deceptively simple looking form cn = Jn−1 − Jn .64) where G n is the work to be expended for creating an n-mer out of monomers (see (9. In equilibrium. All values at n ∗ will. We will flag equilibrium concentrations by the superscript 0. pn is the vapor pressure of an n-mer according to (9. The conditions for steady state are less stringent than for equilibrium because Jn need not vanish.65) G n = 4πζ rn − nkT ln( pn / p∞ ).5 Equilibrium distribution and steady-state nucleation When the system is in phase equilibrium. detailed balance holds. Clusters of size smaller than n ∗ are inside a potential well and fight an up-hill battle when growing. One defines a particle flux or current from n-mers to (n + 1)-mers by Jn = cn β An − cn+1 An+1 αn+1 .60) disregards collisions between clusters.9.5.58)). ˙ Equation (9. Jn = 0 and 0 0 cn β An = cn+1 αn+1 An+1 . they will grow unrestrictedly. So n ∗ is the critical size.56). 0 cn (9. only molecules impinge on a drop but that is a valid assumption as one can easily verify. The function G n is depicted in figure 9. for non-equilibrium conditions.10 plots the dependence of n ∗ on the saturation parameter s. it has its maximum at n ∗ where ∂ G n /∂n = 0. 9. The last relation also enables us to write. their size is n ∗ and for n > n ∗ . β is constant for all n and the sticking coefficient is one. They have to overcome the barrier of height G ∗ . the difference in free enthalpy. in the following.5 on the equilibrium conditions of the state functions. In view of the discussion in section 5. be marked by an asterisk. In a steady state (subscript s).

(9. it just has to be somewhat greater than n ∗ . s = p/ p∞ = 4 and the surface tension ζ = 75 erg cm−2 . The curve reaches its maximum at n ∗ . cs Js cs = n − n+1 0 0 0 cn β A n cn cn+1 − ∂ ∂n cn 0 cn . In these plots. The broken curve depicts the equilibrium 0 concentration cn of n-mers on an arbitrary linear scale. T = 300 K.9. one has to invoke a Maxwellian demon who removes large clusters with n ≥ L > n ∗ by gasifying them.Grain formation 309 2 Figure 9.65)). . The demon supplies the boundary condition for the cluster distribution. from (9. s cn = 0 for n ≥ L. the minimum is at n = n ∗ . The full curve shows G n /kT where G n = 4πζ rn − nkT ln p/ p∞ is the change in free enthalpy when n water molecules condense into a drop of radius rn (see (9.63). Under this condition.66) To realize a steady state in a thought experiment. we get. One does not have to be very particular about L. As a second boundary condition we impose s 0 c1 /c1 = 1 which means that only a very small fraction of the total mass is in clusters.

The latter may also be appproximated analytically from the relations: 1 Js β L 1 dn 0 A n cn Js 0 βc1 1 L e G n / kT An dn Js 0 β A ∗ c1 1 L e G n / kT dn which follow from (9.67).310 Dust and its environment Figure 9. L−1 Js n=1 cs cs 1 L = 1 − 0 =1 0 0 cn β A n c1 cL (9.10. Consequently. on the saturation parameter s = p/ p∞ for water vapor at 273 K. has its maximum and ∂ G n /∂n = 0. The dependence of the critical cluster size n ∗ .9. G n . the surface 2 area An = 4πrn can be taken out from under the integral and replaced by A∗ . the difference in the Gibbs potential between gas and droplet.68) . For clusters of size n ∗ . A Taylor expansion of G n around the maximum n ∗ .66).67) we can directly compute the steady-state nucleation rate J s . defined in figure 9. Because G n has a fairly sharp peak at n ∗ . Gn with G∗ = G∗ + 1 ∂ 2 Gn (n − n ∗ )2 2 ∂n 2 2 v0 4πζ 2 16πζ 3 r∗ = 3 3 kT ln s (9. by performing a sum over (9.

61). Equation (9. so v0 = 4πa0 /3. Figure 9. of course. the steady-state nucleation rate J s .12 that for small n-mers. Figure 9. over many powers of ten. 9. entirely determined by the exponent.71) .5. The number of clusters that grow per unit time and volume from size n to n+1 is then independent of n and one readily finds J s = v0 2ζ πm 1/2 0 (c1 )2 exp − 4µ3 27 ln2 s .69) The exponent in square brackets equals − G ∗ /kT and µ is defined by µ= 2 4πζ a0 kT (9.12 shows the deviation of the equilibrium from the steady-state concentrations. upon replacing the integration limits (1.Grain formation and ζ v2 ∂ 2 Gn = − 04 ∂n 2 2πr∗ 311 yields. the ratio cn /cn tends to zero. the ratio cn /cn s /c 0 )/∂n evaluated at n declines with an approximately constant slope b = ∂(cn n ∗ s 0 over a region of width n = −1/b.5. with the saturation parameter s.70) 3 where a0 denotes the radius of a monomer.6 Solutions to time-dependent homogeneous nucleation 9. The expression in (9. Let us write the time derivative of the concentration of n-mers as old cn − cn cn = ˙ τ old where cn is the old value of the concentration a small time step τ ago.25). For large n. are as simple as in (9.67).60) can easily be solved when the coefficients for evaporation and accretion.1 The evolution towards the steady state The time-dependent system of equations for nucleation presented in (9. ∞) to obtain the standard integral (A. cn . L) by (−∞.60) can immediately be brought into the form An cn−1 + Bn cn + Cn cn+1 + Dn = 0 (9. we want to find the new one.6. (9. αn and βn . Around the critical cluster size n ∗ .69) is quite a satisfactory approximation to the sum in (9. equilibrium and steady-state s 0 concentrations are the same. the latter being calculated from L−1 s 0 cn = cn J s i=n 1 ci0 β Ai . We learn from figure 9.11 illustrates the immense change in J s . The magnitude of J s is.

Bn . where An .92 × 1017 dyn cm−2 . Cn . The boundary conditions are suggested by figure 9. Putting cn = γn cn−1 + δn and inserting it into (9.72) We determine the concentrations cn in the size range g≤n≤L with 1 g n ∗ (as far as possible) and L > n ∗ . L . surface tension ζ = 75 erg cm−2 .12: At the upper end.54) and equals 1. Below s 5. An is not the surface area of an n-mer). Dn are known coefficients (so for the moment.69) and displayed on a linear (top) and logarithmic scale (bottom). Variation of the steady-state nucleation rate J s (s−1 cm−3 ) with saturation parameter s = p/ p∞ for water vapor at T = 273 K. above catastrophic and s 5 represents the critical value of the saturation parameter.11. Bn + γn+1 Cn (9. nucleation is negligible. J s is computed from (9. The pressure p∞ is derived from (9.71) yields γn = − An Bn + γn+1 Cn δn = − δn+1 Cn + Dn .312 Dust and its environment Figure 9. we take c L = 0 because c L /c0 = 0.

g + 1.65) for the smallest clusters where it cannot be correct.13. At n = n ∗ . .72). (9. . .5×1017 cm−3 and a critical size n ∗ = 71. The steady state is reached after a relaxation time τrel that can be shown by an analysis of equation (9. converge towards their steady state values J s given in (9.12.66) to be of order τrel ∼ 2 r∗ 2 4βζ v0 . .73) s In the particular example of figure 9.63). We see that the fluxes Jn . We then find cn for n = g + 1. the ratio is about one half.13 presents a numerical experiment for water vapor at constant temperature T = 263 K and saturation parameter s = 4. The asymptotic values of the concentration ratios c /c 0 are 1 cm n n qualitatively the same as in the steady-state distribution of figure 9. At the lower end. Figure 9. . L − 1 from (9. By choosing g considerably greater than one. Therefore. the abundance of clusters is very close to 0 equilibrium and we put cg = cg . we start with a configuration where only very small clusters exist and put cn = 0 for n > g. and this allows to compute γn = δn = 0 for n = L − 1. we avoid using the enthalpy G n of (9. At time t = 0. . . .9.12 (although the latter is computed for somewhat different values of T and s). These two values 0 imply a monomer concentration c1 = 3.69).Grain formation 313 Figure 9. here for water vapor at T = 273 K and saturation parameter s = p/ p∞ = 4. s−1 . The ratio of the cluster abundance in a steady state to the cluster abundance in equilibrium as a function of cluster size n. γ L = δ L = 0. as defined in (9. τrel ∼ 2 × 10−7 s and Jn −3 .

At time t = 0. the smallest grains 0 are in equilibrium.6.73) is short compared to the time it takes to grow big stable drops. The smallest clusters for which we compute the time evolution consist of g = 12 molecules. 9. cn /cn = 1 for n ≤ g = 12. and bigger clusters are absent (cn = 0 for n > g). Water vapor at 263 K supersaturated by a factor s = 4. the critical cluster size n ∗ = 71 and the concentration of the monomers c1 = 3. normalized to the equilibrium value.314 Dust and its environment Figure 9. evolve in time.5.5 × 1017 cm−3 .13. the largest of L = 120.9.2 Time-dependent nucleation using steady-state fluxes As the relaxation time τrel in (9. Under these conditions.62). the steady-state flux J s represents a good approximation to all . The bottom plot shows how the concentrations of various cluster sizes. The top panel depicts the corresponding particle fluxes Jn of (9.

nucleation rises in a spike as rapidly as shown in figure 9. is unrealistically long.Grain formation 315 fluxes Jn after the time τrel when transient effects have died out.11. Te To illustrate the formation of grains.54). t J s (t )N(t ) dt .76) s(t) = T (t) T0 /T (t )−T0/Te e . Until the moment of crtitical supersaturation. (9. the gas temperature is denoted by Te and the time t is set to zero. therefore. (9. these clusters grow at a rate 2 ˙ (9. At this instant. As the gas cools further.75) with Runge–Kutta asssuming an adiabatic expansion with a cooling rate T = Te − wt with Te = 292 K and w = 100 K s−1 .247 s. the saturation parameter s equals one. reaches the vapor pressure P∞ over a flat surface.9) and the time it takes to form one critical cluster.75) c1 (t) = −J s (t) · N(t) ˙ which leads to a monomer depletion since time zero of c1 (t) − c1 (0) = − 0 κ−1 If the cooling is due to an adiabatic expansion. Only when the saturation parameter approaches the critical value scr at time tcr do clusters become bigger than the critical size n ∗ and catastrophic nucleation sets in. s increases. where κ = C p /Cv is the ratio of specific heats. depriving the gas of its atoms . The time ˙ dependence of the saturation parameter s follows via P∞ as given by (9. The rate at which monomers are depleted through steady-state nucleation is. The Jn may then be replaced by J s and this greatly simplifies the further analysis in which we study nucleation in a cooling gas. Then at tcr = 0.75) the term γ T T γ −1 /K γ where γ = 1/(κ − 1). P = c1 kT . Results are plotted in figure 9. When the partial gas pressure of the condensible component. we integrate equations (9. Such a situation prevails in the outward flowing wind of mass loss giants where most interstellar grains are formed.74) R = πa0 v c1 (t) √ where a0 and v = 8kT /πm are the radius and mean velocity of a monomer. grains do not form and the decrease in the gas density and temperature and the rise of the saturation parameter are due only to the adiabatic expansion at the prescribed rate.14. the steady-state nucleation rate J s is negligible because all clusters are below the critical size n ∗ (see figure 9. At first. K = T /c1 is a constant.74) and (9. For t ≥ tcr . One then has to add to the density ˙ decrease c1 (t) of (9. Let N(t) be the number of monomers in clusters that were formed at time tcr and and let R(t) be their radius. so 4π R 3 /3 = Nv0 . roughly given by the inverse of J s .

At t = tcr = 0. the number density of the gas molecules c1 .74) whence they are stable. s. The maximum values are 7. 9. p0 . The ordinate is for all four variables linear with the zero point indicated. We plot as a function of time the saturation parameter s.6 × 1017 cm−3 .3 for s. m. the saturation parameter s is near its critical value of 5. µ = 9.5.0 and the temperature has fallen to 268 K. c1 ∼ . T . T0 ) but one can drastically reduce the parameter space to only two variables: µ from (9. 103 s−1 cm−3 for J s and 280 µm for R. the steady-state nucleation rate J s and the radius R of the largest drops. Nucleation in an adiabatically cooling water vapor.7 Similarity relations The number of variables that enters the nucleation calculations is quite large (v0 .7 and ∼ 6 × 106 . The radius at which their growth levels off (at t = 0. at an accelerated pace. The saturation continues to increase after tcr because of cooling but eventually falls for lack of gas atoms. From that moment on.70) and defined below [Yam77]. the models presented here for the condensation of water near room temperature can be carried over to grain formation in late-type stars (T ∼ 1000 K. c1 . At t = 0.14. At the maximum of the steady-state flux J s . clusters grow beyond n ∗ according to (9.65 s) represents the typical final grain size (one can also determine a size distribution). Therefore.247 s.316 Dust and its environment Figure 9. The gas expands adiabatically cooling by 100 K every second. the gas is at Te = 292 K and just saturated (s(0) = 1) with a monomer concentration c1 (0) = 6. ζ. v .

(9. one finds for the time dependence of the saturation parameter s(t) = where 1 τsat T0 d ln T . it may be approximated by a linear function in time.76) can be transformed into the new equations: dρ =y dx 1− y = 4 (9.77) 4µ3 27(x + ln y)2 ) dx .7. is fixed by the ratio of surface tension over temperature and thus incorporates grain properties. As the gas temperature changes only a little during the brief period of catastrophic nucleation when T ∼ Tcr .74) and (9. Tcr dt c1 (t) t /τsat e c1 (0) The cooling rate d ln T /dt is evaluated at Tcr . ˙ such as gas density c1 and cooling rate T but it also reflects the physics of the dust via the constant T0 in the expression for the evaporation temperature of the grain material (see (9. When one inserts this T (t) into the evaporation pressure p∞ = p0 e T0 /T of (9. By defining y = c1 (t)/c1 (0) x = t/τsat 2 = τsat c1 (0) πa0 v 3R ρ= a0 the nucleation formulae (9.54). . The second.54)).78) 81 µ π 1/2 0 x y 2 ρ(x ) exp − They contain only the two independent parameters µ and .5. is determined by the conditions in the environment. oversimplifications and . The final grain size is largely influenced by (see definition of ρ). ζ ∼ 103 dyn cm−2 ) as long as µ and are comparable. 9. The first. µ. T (t) = Tcr − at.Grain formation 317 109 cm−3 .1 Cautioning remarks The quantitative results of homogeneous nucleation theory outlined here are fairly speculative because of the intrinsically exponential behavior of the process of grain formation and a number of dubious assumptions.

Observations consist of infrared spectra that do not carry detailed information about the dust. it is there and not graphitic ˚ because of the absence of the 2200 A feature. which was neglected in our derivation of (9. The surface tension of the smallest clusters is unknown. Second. immense fudge factors (up to 1017 for the concentration of the critical clusters) are cited in the literature. Because of this uncertainty. However. So when it comes to the question of how much one learns by applying homogeneous nucleation theory to grain formation in the outflow of red giants. an honest answer is sobering. is still controversial. Abr74]) the following ones. Replacing ζ by ζr = ζ /[1 + δ/r ]. The latent heat of the condensing molecules should not be neglected. nucleation in the wind of giants is not homogeneous but proceeds in a vast network of chemical reactions involving many different species for which we do not know the rate coefficients. All they tell us is that silicates do form in M-type stars. does not solve the problem. there is no information in the spectra about grain size or the dependence of growth on the physical parameters of the medium. astronomical real life is even more complicated. • Besides such intrinsic uncertainties in the theory of homegenous nucleation. • • • The smallest clusters are not spheres. where n is the number of molecules in the cluster [Kuh52].57). A drop also has macroscopic motion but it is not clear how its translation and rotation replaces the internal degrees of freedom. such as (recommended reading [Fed66. The thermodynamics of small clusters. and where we expect carbon dust. where δ is a curvature ˚ correction of atomic length (∼1A) and r the grain radius. We mention that because of the macroscopic motion. First. Another reason for this dissatisfactory state of affairs is the limited possibility to compare observations with theory. stellar photons are present in the wind that create new reaction channels and the radiation field is not in equilibrium with the gas.318 Dust and its environment unsolved problems. the proper expression for the free enthalpy. the correct formula for the vapor pressure of a drop is not (9. in particular. . as can be inferred from the presence of the 10 µm band.57) but ln( pr / p∞ ) = 2ζ v0 /kT r − 4/n. the concept being even ill defined.

We call the cylinder infinite when on halving its length 319 . we can treat dust grains as infinite circular cylinders. a. Many reddened stars appear to be polarized. The problem of computing the electromagnetic field in the interaction of light with an infinite cylinder for any ratio of cylinder radius over wavelength was first solved by Rayleigh [Ray18] for normal incidence and by Wait [Wai55] for oblique incidence. which we have already discussed and illustrated. 10.1. When starlight passes through a dust cloud. otherwise the net polarization would be zero. It is also possible to handle coated or multi-layered cylinders. for example elongated particles shaped like cigars.1 Normal incidence and picket fence alignment Consider a very long circular cylinder of length L much longer than its radius. • To explain the latter two phenomena. up to 10% or so. They must also be aligned. although the emission from the stellar photosphere is certainly not. the weakening of the flux depends on the orientation of the electric vector. The far infrared radiation of some clouds shows a small degree of polarization. such an idealization provides quantitative estimates for the real particles. the numerical code is still fairly simple. one has to evoke the presence of nonspherical grains.1 Efficiency of infinite cylinders As an approximation to elongated particles. The achieved degree of polarization can be very high. Through emission. We skip the mathematical derivation of the cross sections. 10. Through extinction. Although interstellar grains are certainly neither exactly cylindric nor infinitely long (at best they are elongated).Chapter 10 Polarization Electromagnetic radiation can become polarized by interstellar dust in three ways: • • Through scattering. or the wavelength λ.

C .5. is often disregarded in the context of polarization because. unpolarized light becomes slightly linearly polarized after interaction with the cylinder. in every aspect that of a sphere. the display of Q ext is rearranged.1 with figure 4.03. which is about the optical constant of silicate material between 0. except that we now have two Qs differing by an amount Q ext = Q ext − Q ext .e. the cross section C is also halved. We adopt m = 1. which is the center of the visual band V.03 but we now keep the wavelength λ fixed and vary the radius. we will also consider in this chapter examples of carbon particles. because there the observed interstellar polarization typically attains its maximum.7 + i 0. taken with respect to 1 cm3 of dust material (see . Q ext is only a function of the size parameter x = 2πa/λ. When light impinges under normal incidence. In the bottom frame of figure 10. we see that the extinction efficiency of a cylinder mimics.7 + i 0.1. However. there are two extinction cross sections depending on the polarization of the incident light: • • C ext : the cylinder axis is parallel to the electric and perpendicular to the magnetic vector of the incident electromagnetic field. we mostly talk about extinction efficiencies and sometimes omit the superscript ‘ext’.2. ⊥ The efficiencies Q ext and Q ext reach their first maximum at size parameter ⊥ x 2. By comparing the top of figure 10. it is hard to align.1) Q= 2a L In this section. It refers to the same optical constant m = 1.320 Polarization L. the other major chemical component of interstellar dust. C/L is then constant. The efficiency Q is defined as the cross section over projected area. the volume coefficients K and K ⊥ . (10. Note that Q ext may be negative.1 µm and that much thinner or thicker cylinders are ineffective. whereas the difference Q ext has a flat peak near x 1.3 and 2 µm. i. therefore. ext C⊥ : the cylinder axis is perpendicular to the electric and parallel to the magnetic vector of the incident electromagnetic field. In this case.55 µm. In figure 10. Because for a given dielectric permeability. the cylinder axis is perpendicular to the direction of wave propagation. They are not equal and. The figure informs us that in polarizing visible light. which shows the efficiencies Q and Q ⊥ . one speaks of picket fence alignment. Carbon.1. we further learn that there is a direct proportionality between the cylinder radius and the wavelength λmax of maximum polarization. What counts is the difference ext C ext = C ext − C⊥ . We choose λ = 0. When there are many such cylinders with their axes all parallel. over the whole visible region and part of the near IR.1 we present Q ext and Q ext for infinite ⊥ cylinders with circular cross section. without any magnetic inclusions. silicate grains of cylindrical shape are most efficient when their radii are about 0. To derive from figure 10.

7 + i0.1. The bottom frame shows the dependence of Q ext on radius a for a wavelength λ = 0.03. only rearranged. as a function of the size parameter x = 2πa/λ for infinite cylinders of radius a and optical constant m 1.1). Q ext = Q − Q ⊥ . This gives K ∝ a −1 . section 2. if at some wavelength Q ext is equal for thin and thick rods. Therefore. the thin ones are more efficient polarizers (with respect to the same total dust mass).Efficiency of infinite cylinders 321 Figure 10. .55 µm and the same m. The data for Q ext are exactly the same as in the top box. one has to multiply Q by the geometrical cross section 2a L under vertical incidence and divide by the cylinder volume πa 2 L. The top frame shows the extinction efficiencies Q and Q ⊥ and their difference.

We can figure out in a quiet moment that θ is given by cos θ = cos cos φ. Q and Q ⊥ converge and for extremely oblique incidence they tend to zero. Figure 10.1. when the cylinder has a very large radius.2) . the wavevector lies in the (x. presented by the page of this book. y)-plane of a Cartesian coordinate system. for reasons of symmetry. Normal incidence corresponds to θ = 90◦ . and its momentary location is characterized by the angle φ (see figure 10. customary to call the quantity Q = C/2a L the efficiency.322 Polarization 10. Anyway.6 that peaks at an orientation angle of 0. there will be.9). its axis can form any angle θ with the wavevector k. There is a sharp resonance for the curve with the size parameter x = 0. whereas at θ = 0. While the cylinder spins. Formulae to compute Q for normal incidence are found.3).55◦. then cross sections are different for electric vectors along the x. the direction of wave propagation. so one has to form an average over a rotation cycle.and z-axes. It is also. on average. the angle θ between k. so Q(θ ) ∝ sin θ for a λ.2 Oblique incidence If the cylinder is oriented arbitrarily. At an angle = 0◦ . in the case of oblique incidence. light falls directly from above on the page and then. with Q = C/2a L. for instance. in [Boh83]. Furthermore. (10. for oblique incidence in [Lin66]. the efficiencies Q and Q ⊥ as well as their difference Q are rather constant. The incidence is generally oblique but we have just learnt how to handle this. y)-plane. 10. the efficiency reflects the geometrical contraction.2 shows the efficiences Q and Q ⊥ as a function of orientation angle θ for size parameters x = 2πa/λ in the range where polarization has its maximum. and the cylinder axis changes. although this is not in accord with the definition for Q in (2.3 Rotating cylinders Grains in interstellar space are knocked about by gas atoms and rotate. Suppose the cylinder circles in the (x. The sinusoidal dependence of Q on θ for large size parameters is already indicated in the curve with x = 2. where C is divided by the projected geometrical cross section 2a L sin θ . Q(θ ) → 0 for θ → 0. k is parallel to the cylinder axis.1. Let the cylinder be illuminated by an unpolarized flux whose wavevector k lies in the (y. Therefore we also have to evaluate the cross section of a spinning cylinder. no polarization. At = 90◦. During rotation the incident angle θ varies. Around normal incidence down to orientation angles of 40◦ . z)-plane and makes an angle 90 − with the z-axis. When the cylinder is tilted further.

the lower shows Q ⊥ when E is perpendicular to the plane defined by the wavevector and the cylinder axis. The bottom left panel is a magnification for x = 0. the wavevector and the cylinder axis are in one plane. for a size parameter x = 1. .Efficiency of infinite cylinders 323 Figure 10. An orientation angle θ = 90◦ corresponds to normal incidence. There are three pairs of curves for three size parameters. The top frame shows the extinction efficiencies for polarized light of obliquely illuminated infinite cylinders.6. The bottom right panel gives the relevant quantity for polarization. the polarization reverses.2. Where Q and Q ⊥ (dots) cross over. the upper curve gives Q when the electric vector E. Q − Q ⊥ .2.

agrees qualitatively with the more sophisticated models that include rotation.3. makes an angle 90 − with the z-axis which is perpendicular to the (x. An infinite cylinder rotating in the (x.4. let Q (θ ) be the efficiency of the cylinder for oblique incidence under the angle θ . when the electric vector. The direction of the incident light.e. as before.3). which refers to rigid cylinders. An example is shown in figure 10. In the integral. Polarization will become weak when the rotation axis begins to point towards the observer. by some 30% compared to rotation with = 0 and by more when is larger. We see that the picket fence. Naturally. equivalent definitions hold for Q ⊥ (θ ) and for Q y . y)-plane. sin Q the time average of Q y − Q x .3) Denoting by Q = Q y − Qx = Q (θ ) − Q ⊥ (θ ) cos 2ϕ dφ. The angle ϕ is just the projection of φ onto a plane perpendicular to k. Let Q x be the efficiency when the electric field swings in the plane containing k and the x-axis. . Q is proportional to the degree of linear polarization that the rotating cylinder produces.324 Polarization y φ x Figure 10.2) and (10. y)-plane. Now. the cylinder axis and k lie all in one plane. Then Q y = Q cos2 ϕ + Q ⊥ sin2 ϕ Q x = Q sin2 ϕ + Q ⊥ cos2 ϕ so that Q y + Q x = Q (θ ) + Q ⊥ (θ ) Q y − Q x = [Q (θ ) − Q ⊥ (θ )] cos 2ϕ. This again applies to silicate at optical wavelengths. tan φ . i. given by the wavevector k. we get tan ϕ = 2 π π/2 0 (10. θ and ϕ are both functions of φ as specified in (10. Q ext is greatest for the picket fence.

denotes the angle between k and the plane of rotation. their efficiencies Q and Q ⊥ and the efficiency of a sphere of the same radius a are. Q⊥ Q (sphere).1. Q ext = Q ext − Q ext . with frequency-dependent dielectric permeabilities from figure 7. here the cylinders do not rotate but are fixed in space and their axes are parallel to the electric vector and perpendicular to the direction of wave propagation given by the wavenumber k.1 µm. at optical wavelengths.03 as a function of size parameter for various inclinations of the rotation axis. Within a factor of two. the situation is radically different.Efficiency of infinite cylinders 325 Figure 10.19. The particles consist of our standard dust materials.5.4 Absorption efficiency as a function of wavelength When cylinders have a diameter 2a 0. amorphous carbon and silicate. The curve labeled picket fence can also be identified in the top frame of figure 10.7 + i0. which is a typical size of interstellar grains. When the cylinder is spinning. of ⊥ a rotating infinite cylinder with refractive index m = 1.1. This is demonstrated in figure 10. The difference in the extinction efficiency. quite similar. Q In the mid and far infrared at wavelengths λ a. Whereas for both . 10.4. where we have calculated the Qs over a broad wavelength region for cylinders as well as spheres.

There are tremendous differences in the dust temperature. where the electric vector of the light is along the cylinder axis. The cylinder radius is a = 10−5 cm. one has to take an average over the solid angle and over the polarization direction. The efficiency of a sphere (dotted) of the same radius is shown for comparison. we have computed. In figure 10. the right. its emission and the degree to which the emission is polarized. substances.1). the emission ν of cylinders near a B1V star and compare it with that of spheres. the efficiency Q . of infinite cylinders (full lines) under normal incidence of light.5.7.4) where θ equals 90◦ when the light falls at a right angle on the cylinder. When the cylinders in a cloud are randomly oriented or randomly rotating in space.6. The left panel is for silicates. Q abs is ν plotted in figure 10.8) when one wants to determine the temperature of the cylinders in a radiation field Jν . they are much stronger emitters in the far infrared. The absorption efficiencies. Q and Q ⊥ . a factor of three (!) for amorphous carbon and a factor of 1. This has consequences for the temperature of dust. amorphous carbon.5 for silicate. Q ⊥ and Q(sphere) are comparable at all λ. from equation (8. is in the far infrared much larger than Q(sphere).326 Polarization Figure 10. The mean efficiency is given by Qν = 1 2 π/2 0 Q ν (θ ) + Q ν⊥ (θ ) sin θ dθ (10. especially if they consist . It must be plugged into equation (8. extinction is usually irrelevant at these wavelengths. Despite the fact that cylinders are much colder than spheres.

the extinction of a linearly polarized wave depends on the orientation of the electric field vector.6. The average Q ν of (10.1 Effective optical depth and degree of polarization p(λ) In a cloud with anisotropic grains that are aligned to some degree. 10.5. an extreme example but particles with realistic elongations also show considerable enhancement in their far infrared emission (one may consult figure 3. of amorphous carbon which has a large absorption efficiency because of its metallicity. Let τmax be the optical depth in the direction of maximum attenuation.Linear polarization through extinction 327 Figure 10. otherwise as figure 10. Infinite cylinders are.5) . When the unpolarized light from a background star traverses the cloud. τmin .2.3).4) for randomly oriented rotating cylinders and for spheres (dots). of course. the optical depth has its minimum. the initial stellar intensity I∗ is weakened to the observed intensity Iobs = 1 I∗ (e−τmax + e−τmin ) = I∗ e−τeff 2 where τeff defines the effective optical thickness. τeff = − ln 1 −τmax 2 (e + e−τmin ) > 0. perpendicular to it.2 Linear polarization through extinction 10. (10.

so they are not infinite. Teff = 20 000 K) and are either spheres (dots) or randomly rotating cylinders (full curve) but always of radius a = 10−5 cm. the polarization goes to unity.7. just very long. strongly reddened stars. 2 The (positive) degree of polarization is (see (2. The grains are at a distance of 1018 cm from a B1V star (L = 104 L . τeff = 1 (τmax + τmin ). The upper limit is set by the difficulty of observing faint. The total volume of the cylinders equals also 1 cm3 .4 K.328 Polarization Figure 10. (10. e−τmin + e−τmax (10.cyl) = 30.6) The difference in optical depth.sphere) = 56. .sphere) = 44. τmax − τmin .3 K. See also previous figure.9 K. T (Si.cyl) = 19.7) p = 1 (τmax − τmin ) 2 However. is usually small (although τmax and τmin themselves need not be so) and the polarization may then be approximated by for τmax − τmin → 0.3 K.71) and (2. The observed percentage of polarization rarely exceeds 10%. In the case of weak extinction (τmax 1). when τmax − τmin is large. The dust temperatures are: T (aC. the higher the degree of polarization. Emission in erg s−1 ster−1 Hz−1 of 1 cm3 of amorphous carbon (right) and silicate dust (left). T (aC. The more foreground extinction there is.77)) p= e−τmin − e−τmax . p=1 for τmax − τmin → ∞. T (Si.

2 The Serkowski curve A central question is how the polarization. (10.8) is squared. In section 10. as an example.8) is just a mathematical construct to fit empirical data. from Mie theory. Dots represent measurements.8. the cross sections of infinite cylinders.8 presents. It is also easy to see that the parameter k determines the width of the curve. The Serkowski curve approximates polarization measurements of most stars quite well. p(λ). It has three parameters: λmax . the full curve is the best fit Serkowski curve. nothing more and certainly not a law. we calculated. multi-wavelength observations of the star HD 161056 in the spectral interval from 0. The data are nicely fitted by the expression p(λ) = exp −k ln2 pmax λ λmax .6 µm. Reproduced from [Som94] with permission of the American Astronomical Society. Equation (10. pmax and k. it does not matter whether we write λ/λmax or λmax /λ in the exponent.2.Linear polarization through extinction 329 Figure 10. We now apply these results in an attempt to give a theoretical . Nature also takes the liberty of behaving otherwise. 10.8) This functional form of p(λ) is called the Serkowski curve after one of the pioneers in the field.1. Linear polarization of the star HD 161056 [Som94].13 to 3. As the logarithm in (10. The meaning of λmax is obvious—it is the wavelength where the degree of polarization reaches its maximum pmax . Figure 10. varies with wavelength as stellar light is extinguished by aligned grains in foreground clouds.

in the line of sight. Figure 10. For this purpose. Ng . we would be at a loss to explain the observed variations in k and λmax . Naturally. Their origin is trivial as this quantity must depend on the degree of grain alignment and is at small optical depth proportional to the number of particles. one gets a continuous range of values for the three parameters ( pmax .7 + i 0. we expect strong fluctuations in pmax . and then fits the measurements for each star by a Serkowski curve. It also testifies that such artificial geometrical bodies. This quick result is gratifying and guides us in understanding why interstellar polarization has the functional form of equation (10. On the left. on the right. In either configuration. the plots already look quite similar to a Serkowski curve.7) that Q ext (λ) is generally proportional to the polarization p(λ) and thus has the same wavelength dependence.330 Polarization explanation to the Serkowski curve. are more subtle. We also conclude from figure 10. representative of silicate at optical and near IR wavelengths. λmax ). Therefore.03. the size distribution is not nearly as important as it is for modeling the extinction curve. they rotate with the rotation axis inclined by an angle = 45◦ . When one observes the wavelength dependence of polarization towards a large and diverse sample of stars. to achieve good agreement it takes only a few cylinder radii with proper size steps. say a = 0. the cylinders form a picket fence. depends only on x = 2πa/λ. Their maxima are spaced by x 0.1 µm.4 which displays the difference Q ext = Q − Q ⊥ in the extinction efficiency of rotating cylinders as a function of size parameter x = 2πa/λ.9 that rotation.4 is m = 1. Indeed.7 but as the difference in the extinction efficiency. Of course. the agreement in figure 10. we go back to figure 10. Rotation mainly reduces the height of the curve (degree of polarization). p(λ) will be smooth and close to the Serkowski curve everywhere. which is not a bad assumption for many purposes. as opposed to the picket fence. for a size distribution of cylinders. we conclude that . We infer from equation (10.9 between model and Serkowski curve is not perfect. Q ext .8). There are broad and deep wiggles. The refractive index of the cylinder material in figure 10. the Serkowski curve gives a remarkably good fit.4 but rearrange them by fixing the cylinder radius to some typical size. k and λmax . whereas the parameters λmax and k are not much affected. in their polarizing capabilities over a wide wavelength interval.9 shows two alignment scenarios. does not change the shape of the wavelength dependence of polarization greatly. As variations are common. locally it is far from it. The variations in the other two parameters. from a tenth to a few micrometres. akin to real interstellar grains. We do not expect them to be related to Ng and the influence of alignment is uncertain and probably slight. If interstellar grains were the same everywhere. so that only λ appears in the abscissa. and also for various inclination angles of the rotation axis. For modeling polarization. k. like infinite cylinders. are. When we use exactly the data of figure 10. we know that a spread in cylinder radii by only 50% will effectively smear out the wiggles.

which is proportional to the polarization p(λ).03 (full curve). The full curve is a rearrangement of the curve in figure 10.1 µm. they fit more or less. The left panel shows rigid cylinders under normal incidence. intrinsic differences exist among the polarizing grains in various environments.1 µm. A dependence of k on cylinder radius a is not evident. 10. Also shown are Serkowski curves after (10. Q ext = Q − Q ⊥ . We will argue in the discussion of figure 10. the maximum wavelength λmax is proportional to grain size.3 Polarization p(λ) of infinite cylinders Figures 10.Linear polarization through extinction 331 Figure 10.7 + i0. The computations are performed for two materials. nothing essentially changes. In both frames. If we allow for rotation and an inclined rotation axes.6 µm and k = 1. The wavelength dependence of linear polarization produced by aligned infinite cylinders with optical constant m = 1. . there exists an interesting statistical correlation between k and λmax [Whi92]. changes with wavelength. Nevertheless. for infinite cylinders.35 to 0. k 1.10 and 10.2.9 µm but values cluster at 0.10 that. For field stars. λmax has a large scatter from approximately 0. the difference in the extinction efficiency.8) (dotted lines) with parameters λmax = 0. A good explanation for this formula is still missing. for cylinders.11 display how.4 for = 45◦ and is obtained by putting a = 0. the right. silicate and amorphous carbon.2. the cylinder radius a = 0. The plots refer to picket fence alignment.9.66λmax .55 µm. rotating cylinders with rotation axis inclined 45◦ to the line of sight.

or a bit less.9. Let us first look at the polarization curve of silicate cylinders with radius a = 0. if the polarizing dust particles in interstellar space are silicate cylinders. almost identical (dots in figures 10. The optical constant m(λ) is from figure 7.19.2. in both cases. The cylinder radii change from 0. they must have diameters of ∼0. only there the optical constant m(λ) was fixed and the wavelength display linear. 10.6 µm.1 µm.332 Polarization Figure 10. The dotted line represents a least–square fit by a Serkowski curve with λmax = 0. This is the full curve in figure 10.3. Consequently. the parameters k and λmax are. In least–square fits to p(λ) by a Serkowski curve. and cover a wide wavelength range.10 and determines the location of the first peak (the one at λ < 2 µm).10.10. Computations were done for three cylinder radii.10 two other peaks . m(λ) is variable and the wavelength scale logarithmic. 12 and 23 µm). The one around 0. there appear in figure 10. It has three bumps (at about 0.9 and 10. whereas in figure 10.6 µm is very similar to the bump in the left frame of figure 10. as indicated. one finds that the relation between λmax and the cylinder radius a is roughly linear.6 µm and k = 1.6.2 µm.1 Silicate cylinders Both silicate and amorphous carbon grains have their peculiarities. Besides the bump at 0.55 µm. which should comprise the astrophysically relevant sizes.025 to 0. Q ext (λ) for infinite silicate cylinders in picket fence alignment.10). When one smooths the oscillations in the curves of figure 10. a.10.2.2 µm. to yield the Serkowski curve with the standard λmax = 0.

λmax .2. but now for cylinders of amorphous carbon (aC).23i is fixed and has the value of aC at 0. For cylinder radius a = 0.025 µm. where most polarization curves peak. the overall shape of p(λ) does not agree with .7 µm silicate absorption feature but its position is substantially offset by about 2 µm to the red.19). k = 0. for the other cylinder radii. the fit parameters are λmax = 0. these bumps would vanish. its strength increases with a. The bumps in the mid infrared have disappeared. The curve p(λ) now has only one peak that occurs. the optical constant m(λ) = 1.10.2 Carbon cylinders Carbon grains are depicted in figure 10.5 µm.025 µm. 10. The first. they are λmax = 2.55 µm. But even when carbon rods of a certain radius produce the right wavelength of maximum polarization. For cylinder radius a = 0. in the mid infrared which are not predicted by the Serkowski curve. As figure 10. for a = 0.8. Their origin is of an entirely different nature and due to resonances in the bulk material (see figure 7.2. sharper peak is associated with the 9. Dots show Serkowski curves. at much larger λmax compared to silicates. for the same cylinder diameter.55 µm.Linear polarization through extinction 333 Figure 10. m(λ) is from figure 7. If m(λ) were constant. The location of the polarization bump is independent of the cylinder radius. however.11.3.81 µm.85 + 1.19 and variable. k = 1.1 µm. nevertheless the polarization at these long wavelengths is several times stronger than for silicates. ˚ Extremely thin carbon rods (a ≤ 200 A) are required to produce a hump at 0.11.

The cylinder cross section C is about 25% greater than C⊥ (see figure 10. It is also straightforward to compute coated ellipsoids.95) have C /C⊥ = 2. So the ratio C /C⊥ is not far from unity and the cross sections are not very sensitive to the direction of the electric vector. Cigars. Coated ellipsoids will thus be superior in fitting observations. although we will not do so. In the infrared region.4 Polarization p(λ) of ellipsoids in the Rayleigh limit Although infinite cylinders can reproduce the basic features of optical polarization. However. For example. They allow an investigation of a much greater variety of shapes. their mantle thickness and the optical constants of core and mantle. 10. They further widen the parameter space: one can then vary the axial ratio of the ellipsoids. Maximum polarization occurs when the cylinder radius is comparable to the wavelength. it is unsatisfactory that their shape is so artificial. the computation of cross sections is easy because the interstellar particles are then small compared to wavelength and we may apply the electrostatic approximation of section 3.36 + i 0. therefore. which does not seem to be an unreasonable size.034. These numbers are independent of cylinder radius as long as a λ. 10.41+i 0. include cylinders in the limit of infinite axial ratio a/c. critically ask whether a refinement of the dust model in the direction of complicating shape and composition deepens or muddles our understanding of dust.61.3. We can improve the infinite cylinder model by treating ellipsoids as a more realistic alternative.2. the ratio C /C⊥ can be large.12 depicts how the optical depth τ (λ) and the polarization p(λ) produced by small silicate cigars change with wavelength.2.3. polarization is well described by the Serkowski curve which can be quantitatively reproduced by applying Mie theory to infinite cylinders. Figure 10. we can fit only the peak of the Serkowski curve. can be excluded not only because λmax is far off in the near infrared but also because p(λ) overshoots on the red side and is much too broad (k = 0.025 µm. Likewise grains with a = 0. silicate cylinders at 10 µm (optical constant m = 1.3 The ratio C /C⊥ at optical and infrared wavelengths In the optical region.1 µm. where m = 3.3 mm.1). a fit to observations becomes astronomically relevant not by its mere agreement with the data but by its ability to predict and to constrain our conception about dust. however.5). For instance. One must. In the infrared. amorphous carbon is not likely to be the prime polarizing agent.9 and for amorphous carbon cylinders the ratio approaches 104 (see figure 10.334 Polarization observations. this value goes up to 39. a subclass with major axes a > b = c. where the particles are small compared to wavelength. On these grounds. in the plot of p(λ) for a = 0.8). At λ = 1. The particles are spinning with their rotation axis perpendicular to the line of sight ( = 0 in .

The curve p(λ) gives the degree of polarization one would observe in a column with τ9. τ (λ) is proportional to the time-averaged cross section C = 1 (C⊥ + C ).1). The optical depth is one at 9.12.5 µm (curve maximum). no other grains are asumed to be present. τ is one at 9. τ (λ) is the effective optical depth after (10.6). a cloud with an optical depth of three at 20 µm. 2 and p(λ) to the difference C = C⊥ − C (see formulae (3. Optical thickness τ and degree of polarization p in the infrared as a function of wavelength for small rotating silicate cigars. At low optical depth. The polarization p(λ) for other optical depths may be estimated from interpolation remembering that p(λ) goes to zero for τ (λ) → 0 and approaches one for τ (λ) 1.49) and (3. about the shape of an egg). From figure 10. The size of the particles is irrelevant as long as they are in the Rayleigh limit. the nomenclature of section 10.5.13. In this particular example. the right box. the curves are less interesting and display a monotonic decline with wavelength. When the spheroids are made of amorphous carbon (not shown). The dielectric permeability εSi (λ) is taken from figure 7. corresponding to τV ∼ 100 mag.5 µm.Linear polarization through extinction 335 Figure 10.50)). would produce a polarization of 50% under perfect grain alignment. The results for silicate pancakes are practically identical.19. The rotation axis is perpendicular to the line of sight. It allows us to read off p(λ) if the optical depth τ (λ) at wavelength λ equals 1 or 3. The optical .5 µm. weak elongation.13 we conclude that if interstellar dust is made entirely of weakly elongated particles (a/c = 2/3. which is approximately true outside the peak at 9. a/c = 1. A different presentation of the polarization produced by perfectly aligned spinning spheroids is given in figure 10. The left box shows moderate elongation with axial ratio a/c = 3.5) and p(λ) is defined in (10.5 µm = 1 under perfect spinning alignment.

one nevertheless expects a considerable degree of infrared polarization in those cases where there is a suitable infrared background source. Although alignment in real clouds will not be perfect.13.19. polarization p(V ) is then close to one but could not be observed because of the large extinction. However. Full curves are for cigars. there exist significant offsets in the resonance peaks. so p(λ) always refers either to τ (λ) = 1 (lower pair of curves in each box) or τ (λ) = 3 (upper pair of curves). Optical constants are from figure 7.12 and 10. Furthermore. The percentage of polarization due to perfectly aligned spinning spheroids of silicate and amorphous carbon. dotted ones for pancakes. here the degree of polarization p(λ) is plotted for fixed effective optical depth τ (λ). Note that in contrast to figure 10.3) but has a great influence on polarization.336 Polarization Figure 10.13 and similar calculations for small amorphous carbon spheroids (not shown) we learn that polarization through extinction and extinction itself are qualitatively similar in their wavelength dependence. From figures 10.12. one finds that a small or moderate axial ratio (a/c ≤ 3) has little affect on extinction (see figure 3. .12. if necessary with the help of a ruler. as can be checked by scrutinizing figure 10.

only on the size parameters a/λ. Q ext (λ) for prolate spheroids (full curves) and a cylinder (dashed curve) under normal incidence in picket fence alignment. the maxima of the curves shift to shorter wavelengths and the degree of polarization ( Q) falls.2. The cigars have principal axes a > b = c with uniform c = 0. The numerical code was kindly provided by N V Voshchinnikov [Vos93]. and demonstrate how their axial ratio influences the polarization. the body is a sphere and there is no polarization.15. expressed through the difference in extinction efficiency Q.55µm (visual band) and vary the axial ratio a/c ≥ 1. in the visual band. as it should be. and c/λ. To be specific. the cigars begin to resemble cylinders.14.Linear polarization through extinction 337 Figure 10. we assume constant values c = 0. When all axes are equal. the curves also become narrower. If we keep the small axes b = c fixed and increase a. the polarization.1 µm but various axial ratios a/c ranging from 1. When a/c decreases.1 µm and λ = 0. By and large.14 and 10. For an axial ratio of a/c = 8. We restrict the discussion to cigars.41 to 8. The results depend. the response in amorphous carbon particles to a change . The optical constant is fixed and appropriate for silicate in the visual band. interstellar grains of spheroidal shape are not small relative to wavelength and computing their cross sections entails some numerical effort [Asa75]. 10. Results for silicates under normal incidence are displayed in figures 10. and on the refractive index m. of course. which have principal axes a > b = c.5 Polarization p(λ) of spheroids at optical wavelengths Unfortunately. is similar to that of an infinite cylinder.

those for a/c = 1. This enables direct comparison of all polarizing efficiencies.2. A Serkowski curve (k = 1. and it is hard to tell one from the other. we must convert Q into a mass coefficient. 10. For instance. λmax = 0. spheroids of variable axial ratio show a greater diversity in their polarizing properties than infinite cylinders but it also becomes harder to infer from the shape of the Serkowski curve size and chemical composition of the grains. the latter the degree of polarization p.6 µm) is added to study the shape of the polarization curves better. The Qs for cigars with axial ratio 4 and 8 have been omitted. the width of the curve p(λ) is influenced by the optical constant and the particle shape.41 and 2 are enhanced so that the maxima of all curves are about equal.6 Polarization and reddening It is instructive to combine photometric with polarimetric observations. To determine K . Naturally.15.5). The former yield the effective optical depth τeff of (10. we have multiplied Q cyl in figure 10.15. If we want to know how much polarization we get from a unit volume of spheroidal dust particles.14.338 Polarization Figure 10. in the ratio a/c is similar. assuming that they are spheroids. Both are produced by the extinction of dust particles and both . So K is proportional to Q/c. we multiply Q by the geometrical cross section πac and divide by the volume of one grain. In order that the cylinders also refer to a unit volume. 4πac2/3. K .14 by 8/3π. This is a derivate of figure 10.

and gives the polarization per unit reddening.9) also as p(V )/τV ≤ 0.7). furthermore. AV . β ≡ τmax −1 τmin º 0. through AV = RV E B−V = 1. (10. We can determine a theoretical upper limit to β by assuming that all grains are cylinders and optimally aligned. where pmax /E B−V must attain this maximum value of 6%. is related to E B−V . (10. which gives. pmax E B−V º 9%. There is an important empirical upper limit. . Without pretending to be precise.10 to 10.062.10) This inequality implies that the optical depth at visual wavelengths in two perpendicular directions never differs in interstellar space by more than ∼6%.1 and 10.086τV where τV is the corresponding optical depth. or τeff and p. This number is only four times greater than what actually is observed under favorable conditions and thus surprisingly low when one considers that some of the real grains do not polarize at all and those that do are not infinite cylinders but only weakly anisotropic.3 Polarized emission Figures 10.12 present the theoretical wavelength dependence of polarization due to extinction by silicate cylinders and ellipsoids.Polarized emission 339 depend on wavelength. no detectable effect is expected because the extinction optical depth becomes too small. the quantity β However. pmax p(V ) and the visual extinction in magnitudes. from (7. in those cases. 10.15) combines photometry and polarimetry. These plots include the Serkowski curve but also cover the mid infrared region.9) Because the highest polarization is generally reached in the visual. the orientation is never perfect and possibly changing along the line of sight.031. When extended further into the far infrared. It is a measure of the polarizing efficiency at visual wavelengths along the line of sight.17).4 that the maximum value of β = Q/Q is about 25%. using equation (10. we can express the condition (10. The ratio of pmax over standard color excess E B−V as defined in (7. we can read from figures 10. 9% is observed.

Ca + Cc . Even for arbitrarily small τ .12) p depends on wavelength but not on τ . p(λ). τmax + τmin (10. becomes p= e−τmin − e−τmax . Let the grains have major axes a ≥ b ≥ c with corresponding cross sections Ca . This is quite contrary to polarization by extinction which grows with τ and vanishes for τ → 0. We. Their cross sections for light when the electric vector lies along or perpendicular to the cylinder axis.4). temperature Td and intensity Iν = Bν (Td ) · [1 − e−τν ] the degree of polarized emission. As the optical depth increases.3. can differ by orders of magnitude.16). dust not only absorbs but also radiates and polarized emission of aligned grains can be observed in the far infrared and millimeter spectral region. Even for a dielectric.12) with cylinders. so p(λ) does not change much. We start the discussion of polarized dust emission according to equation (10. It also turns out that when the rotation vector is inclined towards the observer. expect a very high degree of polarization in the case of complete alignment (see figure 10.13) = p= τmax + τmin  Ca − Cc   pancakes (Ca = Cb > Cc ) . The (positive) degree of polarization can then be written as p τmax − τmin .5. as is evident from figure 10.1 The wavelength dependence of polarized emission for cylinders However.2 where τmax and τmin denote the optical depth at the position angle of maximum and minimum attenuation. In most cases and certainly at submillimeter wavelengths. the degree of polarization drops less swiftly than in the case of extinction (see figure 10. 10. C and C⊥ . the emission in a direction perpendicular to their rotation axes is polarized to a degree:  C −C c  a cigars (Ca > Cb = Cc )  τmax − τmin  Ca + 3Cc (10. Cc . therefore.340 Polarization 10.3. the ratio C /C⊥ may be above 10. one can also easily compute spheroids in the small particle limit as outlined in section 3.11) We abide by the nomenclature of section 10. the polarization is finite. Towards a uniform cloud of optical depth τν . 2 − (e−τmin + e−τmin ) (10. Rotation of the cylinders roughly halves the value of C /C⊥ with respect to picket fence alignment but C /C⊥ is still large. Cb .2 Infrared emission of spheroids At low frequencies. the polarization can only decrease. the optical depth is small. If they rotate about their axis of greatest moment of inertia and are fully aligned.3.

40) produce very substantial polarization (∼20%). the polarization curves are flat at long wavelengths.19.3 Polarized emission versus polarized extinction Sometimes towards a dust cloud one detects polarized submillimeter radiation and simultaneously polarized light at much shorter wavelengths from embedded or background stars. the associated angles have values between 0◦ and 180◦ . There is perfect spinning alignment for cylinders. They are either infinite cylinders (the two top curves in each frame). 10. Some results are shown in figure 10. cigars and pancakes and the rotation axis is perpendicular to the line of sight. the two are given by (10. which is equal to the degree of polarization.Polarized emission 341 Figure 10. left box) or amorphous carbon (aC.3. and by the direction where the electric field vector has its maximum. cigars (all other full curves) or pancakes (all other dotted curves) of varying axial ratios a/c. It is easy to see that the polarization vectors associated with these two processes.11) and are .16 and we point out that • • even spheroids of small elongation (a/c = 1. right box) with optical constants from figure 7.7) and (10. If psub is the submillimeter polarization due to emission and pV the visual polarization due to extinction. are perpendicular to each other.16. The percentage at which thermal emission by dust is polarized in the far infrared. Only the dotted top curves in each frame show cylinders arranged in a picket fence. The particles consist of silicate (Si. A polarization vector is defined by its length. emission and extinction.

The light is oriented in such a way that it can be decomposed into two equally strong waves. depend on the direction of the electric field vector. This produces a number of interesting phenomena which can all be understood if the molecules are treated as harmonic oscillators. It splits an incident unpolarized beam into two polarized waves with different refraction angles (this is actually a special kind of birefringence known as anomalous refraction). which determine m or ε after (1. n. Calcite.77). CaCO3 . • When the real part of the refractive index. the propagation of a linearly polarized wave is sensitive to the orientation of polarization because the building blocks of matter are anisotropic as. one speaks of birefringence. is the best known example. for example. linearly polarized along the principal directions. E and E⊥ are in phase.342 Polarization βsub pV = τV psub βV related through where at both wavelengths we use the quantity β = τmax /τmin − 1 defined in equation (10. 10. in all crystals of the non-cubic lattice or substances consisting of long and aligned fibrous molecules. one obtains βsub /βV . differs in two directions. Imagine a slab filled with a medium that has unequal indices n and n ⊥ in two perpendicular directions and thus unequal phase velocities. • We will be dealing with birefringence only. As pV . The gem tourmaline is renowned for this property. it leaves with some degree of circular polarization. inside they will fall out of phase. the initially linearly polarized wave presented by the sum E + E⊥ has been converted into a circularly polarized wave. psub and τV can be determined observationally. The optical constant. a slab with the right thickness to do just that is called a λ-quarter plate. in principle. When the phase difference amounts to π/2. k. is anisotropic. the phenomenon bears the name dichroism. Let us shine linearly polarized light through the slab. m = n + i k. information on the particle shape. too. The particular structure of the molecules in the medium has the effect that the motions of the charges. this ratio contains. At the front of the slab. Quite generally. Because the influence of grain elongation on β is markedly different at visual and far infrared wavelengths. which specifies the maximum change in the optical depth with direction. when light that is linearly polarized (but not along a principal direction) enters a birefringent medium. depends then on orientation. When the imaginary part of the refractive index.10). E and E⊥ . .4 Circular polarization In some media.

which is responsible for the phase velocity. C pha = λ2 Im{S(0)}.14) When the wave has crossed a distance l. n−1. . we exploit the complex scattering function S(θ. determines. The electric vector is rotated to the right when the left-circular wave is faster and to the left otherwise. Let us define the volume coefficient for phase lag. according to (2. so that K pha = NC pha .22). can formally be interpreted as the optical constant of the interstellar medium. one tenth. Here the phase velocity of the medium is different for circularly polarized waves with opposite senses of rotation. the grains not only reduce the intensity of radiation but may also change the phase of the wave. optical activity. m = n + i k.2. pha The optical thickness over this distance l equals NlC ext . Its value in the forward direction. A wave of wavenumber k = 2π/λ that passes through a plate of unit thickness and refractive index m = n + i k is phase-shifted with respect to a wave traversing a vacuum by φ = k · (n − 1). Q pha /Q ext and φ will be smaller. The imaginary part. very roughly.4. an optical depth τ produces a phase lag φ=τ C pha . Therefore. by putting K pha = φ. K pha .1 The phase shift induced by grains When a dust cloud is traversed by a plane wave. the index of refraction. Consequently. the phase lag in a cloud of visual optical depth τ due to perfectly aligned cylinders is of order φ ≈ τ/10. 10. Im{S(0)}. which we mention for completeness.22). from (2. φ) discussed in section 2.Circular polarization 343 There is another related phenomenon.1)) is for silicate cylinders. we get. If there are also non-polarizing particles in the line of sight. each with a cross section for phase lag C pha . To understand this. If there are N particles per cm3 . C ext Near the wavelength λmax of maximum linear polarization in the Serkowski curve. S(0). the ratio Q pha /Q ext = C pha /C ext (see (10. the components E and E⊥ of the electric vector E are out of phase by an angle φ = l(K pha − K ⊥ ) = l N C pha . A linearly polarized wave entering an optically active medium exits it still linearly polarized but with its orientation shifted by some angle. 2π (10.

pha (10.1. This is provided by a cloud with grains aligned in one direction.344 Polarization Figure 10. Circular polarization contains also information how the alignment changes along the line of sight but it is buried and to find it one has to dig deep. The curve Q ext (full curve) is identical to the one in figure 10. Linear and circular polarization and phase lag of infinite cylinders in picket fence alignment as a function of size parameter x = 2πa/λ. the more the better. The optical constant m is fixed. Therefore circular polarization is a second-order effect and weak. the second for the phase lag. The percentage of linear polarization is usually small.4. . One needs another cloud where the particles are aligned in some other direction to produce the phase shift. pcirc (λ) ∝ Q ext (λ) − Q ext (λ) · Q ⊥ pha (λ) − Q ⊥ (λ) . the Qs are the efficiencies.2 The wavelength dependence of circular polarization Interstellar circular polarization proceeds in two stages: • • One needs linearly polarized light to start with.15) The first factor stands for the linear polarization. pcirc (λ). is given by the product of two first-order effects. Its wavelength dependence.17. As usual. 10.

seem superfluous. which is given via the index k.19. Figures 10. nothing new in circular polarization or in its dependence on wavelength.19 illustrate the behavior of circular polarization for infinite cylinders in picket fence alignment. and .Circular polarization 345 Figure 10. we do not know the linear polarization at all wavelengths. Optical constants from figure 7.23) as well as the differences n −n ⊥ and k −k ⊥ are connected through the Kramers– Kronig relations. pha pha Q pha = Q − Q ⊥ . either as a function of size parameter or of wavelength: Q ext = Q ext − Q ext . When one computes the index n with the help of the Kramers–Kronig relation.17–10. However. pcirc (λ). Because of the proportionalities (n − n ⊥ ) ∝ (C pha − C⊥ ) pha ext (k − k ⊥ ) ∝ (C ext − C⊥ ) there is. The indices n and k of the interstellar medium as given in (2. The figures display the following quantities. in principle. This whole section might.22) and (2. such a point of view is too academic because. Linear and circular polarization and phase lag of infinite silicate cylinders with 0. in reality. one arrives at circular polarization. the difference in the extinction efficiency which is ⊥ proportional to linear polarization.18. therefore. the difference in the phase lag efficiency.1 µm radius in picket fence alignment. It suffices to know the linear polarization.

the product describing after (10. The 10 µm silicate resonance is very weakly discernible in circular polarization. • • . We summarize the noteworthy points about the figures: • • The efficiency for circular polarization.15) circular polarization. Q circ (λ). Q circ = Q ext Q pha . In either case. in a size distribution the spikes would disappear. in silicate cylinders of 0.346 Polarization Figure 10. However. the sense of rotation changes and either Q ext or Q pha vanish. for example. Linear and circular polarization and phase lag of infinite carbon cylinders with 0. can have several local extrema. although the optical constants are completely different. the sign reversal of pcirc is close to but not exactly at the maximum of linear polarization.55 µm wavelength and in carbon cylinders of 0.19. accompanied by a change of sign.4 µm.05 µm radius in picket fence alignment.19. Optical constants from figure 7. This occurs. The curves for silicate are rather jittery.05 µm radius at 1.1 µm radius at 0. Where Q circ (λ) reverses sign.

Chapter 11 Grain alignment To produce polarization by extinction. If we plot. We summarize in the first two sections the tools necessary for understanding the mechanical and magnetic processes.1. Iz . Although the impact of alignment studies on the rest of astronomy is humble. I y . there will obviously be no preferential direction and the time-averaged polarization vanishes. The question of how alignment works has not been settled satisfactorily. some of them. One usually assumes the Davis–Greenstein mechanism which involves cumbersome details and is rather speculative. at least in praxi. for example. I . depends on the axis about which it rotates. If the rotation vector points towards the observer. at least. 11. not the grains themselves. One can always find a rectangular coordinate system. a child can ride a bicycle with ease. they spin and it is their rotation axes that have to be aligned somehow. As they are not sitting still but receive stochastically angular momentum in collisions with gas atoms.1 Euler’s equations for a rotating body Rotating bodies seem to misbehave in the sense that their motion evades intuition: we push a spinning top one way and it goes the other. so somehow we can cope with rotation. The moment of inertia. 347 . The moments along these principal axes are labeled Ix . a long cylinder spinning about an axis perpendicular to its symmetry axis. like an interstellar grain. it is a delightful topic because of its physical richness. if the rotation axis is at a right angle to the line of sight. A freely rotating body. in which I is greatest about the z-axis and smallest about the x-axis. Take. However. interstellar grains must be elongated or flattened and aligned. Afterwards we describe them emphasizing the ideas behind them and not the final numbers which are rather vague. one gets maximum polarization. However. rotates about its center of mass.1 Grain rotation 11. from the coordinate center. fixed to the body.

no matter how irregular the body.5) . F = p. When we resolve the angular velocity with respect to the instantaneous values along the principal axes. I y ω y . The product τ · ω also gives the power needed to sustain rotation at constant angular velocity ω against a torque τ due to friction. ωz ). be a sphere.4) ˙ This equation is analogous to E = F · v for translational motion.348 Grain alignment the moment of inertia in any possible direction as an arrow whose length is given by the value of I . x y 2 (11. dt (11. the tips of the arrows form the inertia ellipsoid. where v is the velocity. ω = (ωx . L. dt (11. the angular momentum L stays constant in space and the tip of the vector L marks there a certain fixed point. For example.2) The equation of motion of a rotating body describes how its angular momentum L changes when a torque τ is applied: τ= dL . ω y . Without a torque. the time variation in kinetic energy E of a freely spinning top is given by dE = τ · ω. As viewed from the rotating body. Likewise. therefore. Q is not constant but moving and the vector L as seen from the rotating body varies like dL dt = L × ω. the inertia ellipsoid of a cube or a regular tetrahedron must. the total angular momentum L. With respect to a coordinate system connected to the rotating body. This is a very regular figure. changes like dL dt =L×ω+τ body (11. Iz ωz ) 2 2 2 L 2 = Ix ω2 + I y ω2 + Iz2 ωz x y 2 E rot = 1 (Ix ω2 + I y ω2 + Iz ωz ).3). L changes in space according to (11. body If there is a torque.1) (11. if we replace τ by the force F and L by the momentum p. its square and the kinetic energy of the body are: L = (Ix ωx .3) ˙ It corresponds to the equation for translatory motion. for reasons of symmetry. let us call it Q.

• Formula (11. For our purposes. where • . The axis of the Earth is inclined against the normal vector of the ecliptic by ψ = 23. ω and the figure axis describe circular cones around L. i.7) (11.5◦.6) to eliminate ω y ˙ gives ωx ∝ ωx and likewise for the y-component ω y ∝ ω y . Its rotation is not really free because the Sun and the moon exert a torque τ on the equatorial bulge of the Earth which causes the polar axis to slowly precess. where F = 0 implies boring uniform linear motion.3) can still be solved with pencil and paper but the rotation axis and the principal axes perform complicated wobbling motions around L.Grain rotation or in components ˙ τx = Ix ωx − ω y ωz (I y − Iz ) ˙ τ y = I y ω y − ωz ωx (Iz − Ix ) τz = Iz ωz − ωx ω y (Ix − I y ). altogether ¨ ¨ ωx = A sin t where A is a constant and = ωz ω y = A cos t I x − Iz . quite unlike the translation analog. From (11. Iz are unequal. in qualitative agreement with the Chandler period. In the general torque-free case. bodies where the moments of inertia about two principal axes are equal. The precession velocity ωpr follows immediately from (11. ωpr = τ/L. Obviously. ˙ Formulae (11. This implies time-variable stresses and. dissipation. are not trivial. Ix (11. 2π/ωz = 1 day and Iz /(Iz − Ix ) = 305. ˙ The solution of (11. The moment of inertia Iz around the polar axis is greater than Ix around an axis that lies in the equatorial plane because the Earth has a bulge at the equator. The body is then symmetric about the z-axis.8) one finds 1 yr.6) are known as Euler’s equations.2 Symmetric tops 349 (11. 11.8) applies to the wandering of the pole of the Earth. I y .3) as we want to derive the change of the rotation axis with respect to the stars. consequently. which is called the figure axis.6) The kinematics of free rotation.1.5) or (11.6) is easy for τ = 0 and Ix = I y .e. equation (11. we use equation (11. it suffices to consider symmetric tops. where all momenta Ix . ωz is then zero. To estimate the effect.8) During the rotation. For our planet. It precesses around the figure axis with angular velocity .3). The length of the angular velocity vector ω and its component ωz are constant but ω changes its direction in space as well as in the grain. say Ix = I y . which means without a torque. Taking the time derivative of the first equation of (11.

L = Iz ωz is the angular momentum of the Earth.350 Grain alignment figure axis angular momentum axis fixed in space L momentary rotation axis ω zI z ωz ω θ ω1 ω 1I x Figure 11.5 cos ψ. and into a perpendicular component ω1 with moment of inertia I1 = Ix . the figure axis and the angular momentum vector L all lie in one plane. T = 2π/ωpr ∼ 26 000 yr. the angular velocity vector ω can be decomposed into ωz .1. which is constant. Rotation of a symmetric top with figure axis z and moments of inertia Ix = I y . Nowadays the polar axis is directed towards the Northern Star and gives . The torque is a tidal effect inversely proportional to the third power of the distance r . the rest to the Sun. the latter two form an angle θ. In the plane defined by the figure axis and the momentary rotation axis. The angular velocity ω. the factor sin 2ψ changes into 1. about two-thirds of the effect are due to the moon. Inserting numbers yields for the period T of the Platonic year. When one corrects for the fact that the torque oscillates as the Sun and the moon apparently orbit the Earth. One easily finds ωpr = G Iz − I x τ = 2 sin 2ψ L ωz I z M + r3 ¯ M r3   where G is the gravitational constant and M/r the ratio of mass over distance for the Sun or moon.

e We will mainly be concerned with pure spin angular momentum where J = S.9) m=g 2m e c with the Land´ factor g = 1. not its length J .13) (in which the vector J is the current density) to an electron in circular motion of angular momentum L.11) Because τ stands perpendicular to the angular momentum J. So the magnetic moment m associated with the spin of an electron (S = 1 ) equals Bohr’s magneton: 2 µB = e = 9. in the case of pure spin momentum (S). The field B exerts on the dipole m a torque τ = m × B.116% α/2π. The average kinetic energy of a gas atom is then equal to the average rotational energy E rot of a grain: 3 2 kTgas = E rot = 1 I ω2 2 . in a straightforward way.1. be carried over to rotation. to first order in the fine structure constant α. one gets e L (11. (11. In future millenia.12) ωpr = J 11.10) Quantum electrodynamics gives an upward correction. 2m e c (11.1. it amounts to 0. the latter is usually not parallel to the net magnetic moment m because of the different Land´ factors for S and L. This result is also true in quantum mechanics with e respect to the orbital momentum L of the electron. When there is equipartition between dust particles and gas. If one applies the classical formula (1.3 on translatory Brownian motion may. Therefore J precesses around B with angular frequency mB . the grains perform translatory and rotational Brownian motion. the polar star will no longer point north and one will have to be more careful when walking away far from home at night.27 × 10−21 [cgs]. g = 2. 11. (11. the torque changes only its direction. When S and L couple to form the total angular momentum J = S + L.3 Atomic magnet in a magnetic field An atomic magnet in a field B is another example for a rotation which is not force-free. However.Grain rotation 351 guidance to nocturnal wanderers.4 Rotational Brownian motion The results of section 9.

The rotational velocity of a dust particle in Brownian motion is ωBrown = 3kTgas . volume 10−15 cm−3 and bulk density 2.5 g cm−3 .352 Grain alignment Table 11. If we adopt grains of moderate elongation. almost all angular momentum will reside in rotation perpendicular to the symmetry axis z.13) ω 2 I x = ω 2 I y = ωz I z . For example. we get. The disorder time tdis of equation (9. Moment of inertia for spheres. in the case of a thin rod (Ix = I y Iz ).39) stays practically the same with respect to angular momentum.1. we see from the condition of equipartition (11. Note that Idisk = 1 Idisk ⊥ . For real interstellar grains. for their Brownian motion.1 for a few relevant configurations.14) Moments of inertia are given in table 11. Sphere of radius r Thin cylinder of length l with rotation axis through the mass center and perpendicular to the symmetry axis Cylinder of radius r rotating about its symmetry axis Disk of radius r with rotation axis through the mass center and perpendicular to the disk Flat disk with rotation axis through the mass center and parallel to the disk Isphere = 2 Mr 2 5 1 Icyl ⊥ = 12 Ml 2 Icyl = 1 Mr 2 2 Idisk ⊥ = 1 Mr 2 2 Idisk = 1 Mr 2 4 and E rot is. this may be a gross underestimate if they rotate suprathermally (see later). . The gas atoms that collide with the spinning grain retard its rotation by exerting a torque τ . and that the ratio Icyl ⊥ /Icyl = l 2 /6r 2 can take up 2 any large value. from (11. Indeed. one defines a friction coefficient ζ through τ = ζω where τ is the torque needed to sustain a constant angular velocity ω and −1 ζ = I tdis . on average.13) that Ix < I y < Iz implies L x < L y < L z . In a likewise manner. I (11. x y The angular momentum L tends to point along the axis of greatest moment of inertia.14) in a gas of 20 K ωBrown ∼ 3 × 105 s−1 . cylinders and disks of mass M. equally distributed over the three degrees of rotational freedom: 2 (11.

1. = tdis 2I (11. To see why. In thermal balance. for example. If one can estimate the mean intensity of the radiation field. Because of the particular form of the grain. hν Inside a molecular cloud. Each photon carries off a quantum in an arbitrary direction. We mention another randomizing process.Grain rotation 353 Again. Tdust < Tgas and have a uniform sticking coefficient η over their surface. the discrepancy between NUV and NIR is less severe. 5 2 NIR √ The expression follows from NIR tdis = I ω with I ω2 = 3kT . besides grain collision.4)) must be compensated by a spinup L 2 /2I . A fraction η of the gas atoms. If the grain emits per second NIR infrared photons. and its rotational energy may be orders of magnitude greater than 3 kTgas. in a stationary state. as assumed earlier. from the dust temperature.2) that are cooler than the gas. where only a few energetic photons fly around. stays after the collision for a while on the grain before it leaves with a lower temperature (momentum) than it had before the impact. the associated disorder time is 8πkTρgr a 5 . Usually the dust is heated by the UV field. is typically 50 times smaller than NIR . 11. we get I ω2 = 3kT and L 2 = 6ζ kT . NUV .15) In thermal equilibrium at temperature T . the angular momentum of the √ grain has changed in a random walk by an amount N . that is always at work: infrared radiation. imagine S-shaped grains of mass Mgr (figure 11. gas atoms approaching from the right and hitting the top of the S may collide twice and their chance for an inelastic . the number of UV photons absorbed per second.5 Suprathermal rotation An interstellar grain is probably not in energy equilibrium with the gas. This process may dominate over collisional randomization for small grains.14) from Brownian 2 motion [Pur79]. NUV follows from tdis = NUV = 4π πa 2 UV Jν Q ν dν. After N emission processes. It then rotates much faster than estimated in (11. the heat loss rate ωτ (see (11. where L 2 is the average change in the square of the angular momentum per unit time due to collisions: ωτ = L2 I ω2 . where 0 < η < 1.

If 2L L ∼ we use the formula (11. on average. A sticking probability η that varies the way as shown on the righthand of figure 11. Consequently.354 Grain alignment Figure 11. If there are certain favorable (active) sites on the grain surface. collision is higher.15) for the frictional loss and put L 2 2I ω·nva πa 2 ·m a va a. Therefore. minor inevitable irregularities. the ejection of the . less momentum to the grain than those striking the bottom half of the S. Because of the immense factor 109 . without a temperature gradient. are sufficient to boost E rot by powers of ten. the grain’s rotational energy E rot is not 3 kTgas but 2 E rot i.2 will also do. like unevenly distributed cavities over the surface or small variations in η. Suprathermal spinup does not really require the dust particles to be Sshaped. they impart. where an H atom gets trapped until another comes by so that they can leave together as H2 . The fat stretches on the bar indicate places of enhanced sticking probability. The particle is said to rotate suprathermally.2. colder than the gas and with a constant sticking probability η. The same holds for a less fancy-shaped particle if η is not uniform but displays in its distribution rotational symmetry (right). is accelerated clockwise (left). we find indeed ωa ∼ va . The dust particle rotates faster and faster and a limit is set only by friction when the lateral velocity of points on the grain surface becomes comparable to the velocity of gas atoms. The process leads to a spinup. As a result. • One is the formation of molecular hydrogen. There are two other possibilities for spinup that do not need a temperature difference between dust and gas. A grain in the form of the letter S. the S will rotate clockwise.e. Should the grain be hotter than the gas. Mgr 3 kTgas 2 ma 109 times greater. it will rotate anti-clockwise and. Then each atom transfers approximately the angular momentum m a va a to a grain of radius a. the mechanism does not work at all.

According to the classical explanation.e.1 Diamagnetism When a diamagnetic particle moves in a static inhomogeneous magnetic field in the direction of growing field strength. for example olivine. it experiences a repulsive force.Magnetic dissipation 355 • H2 molecule imparts angular momentum to the grain. random excitation.2. some clockwise and some anticlockwise.2 Paramagnetism Paramagnetism is relevant astrophysically and shows up when atoms have a permanent intrinsic magnetic moment as a consequence of unfilled inner electron shells. from the Lenz law.2 Magnetic dissipation 11. farther away from its starting position. The atoms with the strongest paramagnetism are the lanthanides. therefore. Here again one gets suprathermal rotation if the yield for this process changes over the surface. in the opposite way to a diamagnet and its µ is a little bit greater than one. It can easily be built into silicates. Even when the accommodation coefficient is perfectly uniform. speeds up the rotation in the same way as a randomly kicked ball moves with each kick.16). The magnetic permeability µ. Mg2−x Fex SiO4 . This element has a high cosmic abundance and must be mostly solid because of its heavy depletion in the gas phase. The explanation applies to all atoms and. The kinetic energy of the ejected electron is typically ∼1 eV and thus less than the energy hν of the UV photon but the momentum of the electron is a hundred times greater than hν/c. 11. with x not far from one (see section 7. The amount of silicate grains is presumably limited by the cosmic abundance of Mg. diamagnetic. in some way. The field leads to an additional electronic current and to a magnetic moment which points. in the opposite direction to the outer magnetic field. We will not come across diamagnetism any more. Another mechanism is the ejection of electrons by UV photons. all matter is. the magnetic flux through the electronic orbits of the atoms in the particle increases during the movement and thus induces an electric field. is for a diamagnetic particle a little smaller than one. The electrons of an unfilled outer shell usually pair in binding and give no effect. 11. The sites will not be exactly balanced over the surface and they endow the grain with a net rotation.2. other effects may override it. A paramagnetic substance is pushed in the direction of increasing field strength. If all iron in the interstellar medium is in . however. defined by the relation B = µH of (1. on average.4. which is some 20% higher than that of iron. The paramagnetism of interstellar dust is probably due to iron. i.4).

50 × 1022 cm−3 ˚ 2.133)) U = −m · B = −m B cos θ. has a potential energy (see (1.4 µB eff.2. The ion Fe3+ has a ground term 5 6S 5/2 (L = 0. For the value of Npm it is assumed that all iron is condensed into silicate particles. Parameters of iron pertaining to the magnetic properties of interstellar grains.2. The third (n = 3. M-shell) could accommodate 2n 2 = 18 electrons but contains only 14.87 g cm−3 Nbulk = 8. Fe atoms constitute some 10% of all atoms in such grains and their number density there should then be Npm 6. totalling 26. The electron configuration of iron.16) Suppose a grain has N such atoms per unit volume. There are two more electrons in the 4s subshell. Abundance: cosmic abundance depletion in gas phase number density in paramagnetic silicates Bulk iron: crystal structure γ -form α-form Curie temperature density number density lattice constant (bcc) nearest neighbor distance saturation magnetization effective magnetic moment static susceptibility Electron configuration: Fe atoms Fe2+ ions Fe3+ ions [Fe]/[H] = 3. J = 2 ). moment: m 5.2 × 10−5 >90% Npm 6.9 µB ground state 5 D4 ground state 5 D4 ground state 6 S5/2 silicate particles. mag.22 µB per atom χ0 12 n = 4 shell: 2s electrons n = 3 shell: 2s. the multiplicity is 5 and the total angular momentum J = 4 ).48 A Msat = 1750 G m 2.2 × 1021 cm−3 .87 A ˚ 2. Its magnetization M in . multiplicity 6. is summarized in table 11.2 × 1021 cm−3 1183K≤ T ≤1674K T ≤ 1183K cubic face-centered (fcc) body-centered (bcc) Tc = 1043 K ρ = 7. from [Kit96]. The magnetic moments of the electrons in the third shell do not balance resulting in a net moment and thus paramagnetism. (11.356 Grain alignment Table 11. as of Fe2+ . mag. oriented under an angle θ to a field B. The ground state of atomic iron. 6p and 6d eff. is 5 D4 (D stands for orbital angular momentum L = 2 . moment: m 5. An atom of magnetic dipole moment m. together with some of its other parameters. The two inner shells of iron are complete.

L(x) x/3. only a cluster of atoms can show this property. • If S = 1 and the field is arbitrary. The magnetization M in a ferromagnet is very strong even when the applied magnetic field Ha is weak or absent.18) χ= B T 3k So far we have neglected the quantization of space. Equation (11.136) and its derivation was given in section 1. It has 2S + 1 equally spaced energy levels. the magnetic moment equals twice Bohr’s 2 magneton. m = 2µB . Below 1183 K. In weak fields. We mention two cases for a system with spin angular momentum S only. has a Curie temperature TC = 1043 K.137). well above any common dust temperature. A single iron atom is not ferromagnetic. This is the result of a collective.Magnetic dissipation the direction of B is then M = Nm L(x) with x = mB . in ferromagnetism there is an additional interaction among the electrons themselves formally described by an exchange field He . and M = Nm tanh x • with x = mB . C Nm 2 M = with C = . It only shows up below the Curie temperature TC . kT (11.18) becomes C= N S(S + 1)m 2 . This is a fictitious field insofar as it does not appear . the constant C in (11.6. Co and Ni but also in alloys of these elements and some of their oxides.17) corresponds to (1.2. above face-centered (γ -form) but this form is.3 Ferromagnetism Ferromagnetism appears in certain metals. the relevant element for us. 3k (11. When taken into account. Whereas in paramagnetism only the applied field acts on the electrons and tries to align them.17) The Langevin function L(x) is defined in equation (1. mB kT and the susceptibility χ = M/B depends on temperature according to Curie’s law. (11. never ferromagnetic. kT 357 (11. for small arguments. only those directions for which the projection of the angular momentum along the direction of B differs by multiples of are allowed.19) If the field is weak and S arbitrary.20) There are some minor changes if there is also orbital momentum L. The magnetic susceptibility is then slightly modified. of course. the grid is body-centered (α-form). Crystalline iron possesses a cubic grid. as in interstellar space. like Fe. 11. nonmagnetic effect of the electrons. Iron.

21) . iron is paramagnetic and the dependence of the magnetization M on the total field. One gets an idea of how the exchange field comes about from Heisenberg’s theory for just two electrons: The relevant quantity is the exchange integral which occurs because the electrons are indistinguishable and which determines whether the spins are parallel or antiparallel. where ferromagnetism begins to fail. or ten thousand times weaker. The conception of an exchange field He is useful. It exerts on the electrons an apparent force which makes their spins parallel. (11. He + Ha. at the Curie point (∼1000 K). the kinetic energy of the atoms must be comparable to the potential energy of a Bohr magneton in the exchange field. M= C (Ha + λM). The exchange field is much stronger than the applied field. At its root is Pauli’s exclusion principle. Ha Any attempt to explain the huge magnetization M by the interaction of the magnetic dipoles associated with the electron spins fails by orders of magnitude. 11. Let us assume that it is proportional to the magnetization.4 The magnetization of iron above and below the Curie point Above the Curie point. the dipole field of ˚ a Bohr magneton at the distance r ≈ 2. M = χ Ha it follows that χ= C T − Tc with Tc = λC.358 Grain alignment in Maxwell’s equations. However. He so the total effective field is Ha + λM λM.2. Indeed. µB H e kTc . follows Curie’s law. Therefore we estimate that He to be of order 107 G. T As the magnetic susceptibility χ combines the applied field Ha with the magnetization. He = λM.5 A between neighboring iron atoms is 2µB /r 3 ≈ 103 G. Ferromagnetism can only be understood quantum mechanically.

21). For bulk iron. With rising temperature. Within each domain.20). known as the Weiss law. like magnetite (Fe3 O4 ) or maghemite (γ –F2 O3 ). This phenomenon is known as spontaneous magnetization. the effective magnetic moment per atom is m 2.50 × 1022 cm−3 . (11. We now have a relation between the Curie temperature Tc and the magnification factor λ. is then lowest.5 Paramagnetic dissipation: spin–spin and spin–lattice relaxation After these preliminary remarks about magnetism. the magnetization below the Curie point is not uniform over the crystal but divided into small domains.2. The crystal is partitioned into domains because its total energy. But in an adjacent domain the magnetization has a different orientation so that the effective M over the whole crystal may be much lower than Msat .1 µm size. In a ferromagnet consisting of several domains. one proposes that neighboring electron spins are not parallel. so the saturation magnetization Msat (Fe) 1750 G. there is almost perfect alignment and the magnetization is close to its saturation value. When the temperature approaches Tc from above. Solving it. we use (11. which expresses M as a function of m B/kT but replace B by Ha + λM. as in a ferromagnet but anti-parallel without fully compensating each other.22) This scenario of spatially constant magnetization applies to single domains and thus only to very small particles. first gradually. except that the saturation magnetization is much lower. are ferrimagnetic. although the susceptibility χ is still large. To explain it. In a mono-crystal of iron of only 0. χ has a complicated behavior depending on the applied field Ha and also on the magnetic history.19). Equation (11. To evaluate the magnetization below the Curie point. the total energy would be higher. The one most frequently evoked for the alignment of interstellar grains is paramagnetic relaxation [Cas38. We then have an equation for M. then rapidly and goes to zero at T = Tc . we discuss and quantify processes of magnetic damping. 11.22 µB . Dav51. Akin to ferromagnetism is ferrimagnetism. is only approximate. which includes the magnetic field outside the crystal. Some iron oxides that might be found in interstellar grains.Magnetic dissipation 359 Here C is the same as in (11. if the crystal were one big single domain. Spi79]. one gets values of M which are different from zero even without an applied field (Ha = 0). Jon67. the susceptibility rises steeply. With Ha = 0. the number density N = 8. There are two thermodynamic systems in a grain: . the spontaneous magnetization M as a function of T has its maximum at zero temperature given by the saturation value Msat = Nm when all moments are perfectly aligned. M falls.

the spin system gets disturbed. τss is more or less independent of it. • τsl .8. it will lose energy. Equilibrium with the lattice system is established much more slowly after the spin–lattice (sl) relaxation time τsl ∼ 10−6 s. serves as a heat reservoir and is always relaxed. In equilibrium.6 The magnetic susceptibility for spin–lattice relaxation We next calculate the susceptibility for spin–lattice relaxation using the thermodynamic relations for magnetic materials of section 5.23) . the dipoles are better aligned in the new stronger field H and their potential energy U is. heat flows from the lattice to the spin system at a rate −α(TS − TL ) (α > 0). while keeping its direction. The heat flow is proportional to the temperature difference d T = TS − TL . It is at temperature TL . The above numbers for τss and τsl are very crude but certainly τss While τsl varies inversely with temperature. 11.2. When the field is suddenly increased from H to H > H . For TS = TL . Equilibrium is characterized by a Boltzmann distribution e−U/ kTS at temperature TS . has much more energy than the spin system. The magnetic moments may be due to electron spin as well as orbital angular momentum. The coupling between spin and lattice is mediated via phonons. (2) The lattice system. Suppose a bias field H0 has produced a constant magnetization M0 and superimposed onto H0 there is a small periodic perturbation so that the total field and the total magnetization are: H (t) = H0 + he−iωt M(t) = M0 + hχe−iωt . lower than before the disturbance. they correspond to changes in the quantum numbers of the lattice vibrations. TS = TL . In a constant magnetic field of strength H . This process is adiabatic. the temperatures of the two systems are equal. there is some net orientation of the dipoles. the level populations of the dipoles stay the same so that m H /kTS = m H /kTS . there is no heat exchange with the lattice. Once equilibrium has been established (TS = TL ). therefore.360 Grain alignment (1) The spin system. Associated with the two systems are two damping constants: • The spin system comes to equilibrium very quickly at a new temperature TS after the spin–spin (ss) relaxation time τss ∼ 10−10 s.16) U = −m H cos θ is the potential energy of a dipole of moment m aligned with the field at an angle θ . The excess energy has been transferred to the lattice. The ensemble of magnetic moments arises from the electrons of the paramagnetic atoms. It consists of the vibrating atoms or ions in the grid of the solid body. It stops when the two temperatures are equal.3. where after (11. (11. If the spin system was initially hotter.

Therefore the susceptibility for spin–lattice relaxation is: χ1 (ω) = χad + χT − χad 2 1 + ω2 τsl χ2 (ω) = (χT − χad ) ωτsl . χ approaches the polarizability α. It is positive. as it should be. After a sudden disturbance. The reason being that the . χad and χT denote the adiabatic and isothermal value of the susceptibility.55) which we derived before. 2 1 + ω2 τsl (11.23) follows from equation (5. because χT = χ(0) > χad = χ(∞). where d T = TS − TL = ∂T ∂H dH + M ∂T ∂M dM H one obtains the complex susceptibility χ(ω) = −α + i ωCM dM = dH α − i ωCH ∂T ∂H · M ∂M ∂T . The product of the two brackets in the expression for χ(ω) yields −χT according to (5. The formula is correct whether a bias field H0 is present or not.Magnetic dissipation 361 The perturbations d H = he−iωt and d M = hχe−iωt imply a heat input δ Q to the spin system given by (5.69). 2 Here V is the grain volume.146) confirms that we are dealing with some kind of Debye polarization. χ2 is proportional to ω.66). However. It also agrees with the energy dissipation rate (1. dδ Q = −i ωCH dt ∂T ∂M d M − i ωCM H ∂T ∂H dH M is equated to the heat flow −α d T . α The power W which has to be expended for the magnetization of a body in a variable field of the kind given by (11. W = 1 ωV χ2 h 2 . H CH and CM are the specific heat at constant field and constant magnetization. When the time derivative of the heat input.63). because in the case of a small susceptibility.68) and (5. A look at the permeability of (1. the very process of dissipation by spin–lattice relaxation does not work for purely paramagnetic interstellar grains. At low frequencies.24) The imaginary part χ2 stands for the dissipational loss. the return of the spin system to equilibrium with the lattice proceeds exponentially on a characteristic time τsl which is related to the constant α and the specific heat CH through τsl = CH .

the magnetic dipoles have to adjust their precession to a field that changes its direction but not its strength. in the range from ω = 105 s−1 for Brownian motion up to 109 s−1 in the case of suprathermal spinup. also of interest to know χ(ω) at all frequencies and with respect to all dissipation processes so that one can check whether dust absorption or emission due to para.147) because the magnetization of the grain is due to orientational alignment of the dipoles in a time-variable field. of course. and there is no interaction with the lattice.e. and a hyperfine splitting interaction (accounting for the influence of the magnetic moments of the nuclei of the paramagnetic atoms). a quantum-mechanical exchange interaction (which tends to make spins parallel as in ferromagnetism). Consequently. The former is inversely proportional to temperature and follows from Curie’s law (11.5). the form ωτ χ2 (ω) = χ0 . It is. In section 11. our troubles are not in vain.6.362 Grain alignment bias field H0 is absent. where h is now the interstellar magnetic field. has. therefore.7 The magnetic susceptibility in spin–spin relaxation The rotation of paramagnetic dust particles may be damped via spin–spin relaxation. The dissipation comes about because the alignment of the dipoles always lags behind the new field direction (see figure 11. The spinning particles only feel the time-variable part.25) 1 + ω2 τ 2 To extract numbers. We wish to estimate the values of χ(ω) for magnetic grain material at the frequencies at which interstellar grains may rotate. The underlying microscopic processes are very complex and their treatment is beyond the scope of this text.3 we will revive spin–lattice relaxation by proposing an internal bias field in the grain material due to ferromagnetic inclusions. Nevertheless. So their potential energy U in the field and the level populations remain the same. 3kT (11. the spin temperature TS does not oscillate. We merely mention that there is: • • • • a magnetic dipole–dipole interaction (the paramagnetic atoms are the source of a magnetic field in which other magnetic dipoles precess). which is responsible for dissipation.2. 11.5)).or ferro-magnetism can ever become important (see second term in (3. d H = he−iωt . The magnetic susceptibility χ(ω) for spin–spin relaxation must have the functional dependence of the Debye permeability after equation (1. (11. i.26) . one must know χ0 and τ . d T = 0. as discussed in section 1. As the grains rotate. Stark splitting of the magnetic sublevels (arising from the electric field of the crystal). The imaginary part χ2 (ω).18): χ0 = Npm m 2 .

One thus finds χ0 τ = 0. Their average contributions to B loc cancel but the root mean square deviation does not vanish. hydrogen nuclei. Npm . therefore. As the ground state of Fe3+ is classified as 6 S5/2 . one sets it equal to ωpr because this is the characteristic time for a change in the magnetization perpendicular to the applied field. χ0 is equal to χad in equation (11.27) χ2 (ω) = χ0 4 × 10−12 . In contrast to the perfect cubic lattice discussed in section 3. It is remarkable that in the product χ0 τ the density of paramagnetic atoms in the grain. To first order. In particular. of course. Note that the electrons do not precess in the weak interstellar magnetic field but in the much stronger local field B loc created by all paramagnetic atoms (see section 3. This field is.24) (see also (5. only the nearest matter. cancels out and. which are building blocks of the otherwise non-magnetic H2 O molecules. As fields from dipoles fall off quickly and proportionally to r −3 .69)).49 /kT and ωτ ω (11.4). besides electronic paramagnetism which we have had in mind so far. neither does the Bohr magneton show up in the product χ0 τ . it produces a field B loc 700 G.4. have unsaturated nuclear moments. and the effective magnetic dipole moment of an iron atom are taken from table 11. The full expression should be used for magnetic emission at wavelengths λ = 1 cm or shorter. −1/3 we assume the field to come from a dipole at a distance r = Npm .9) and (11. So the mechanism is independent of the strength of the magnetic moments and. They are a grain constituent when the grain has an ice mantle.2). Another form of the susceptibility for spin–spin relaxation uses the fact that the frequency distribution function defined by f (ω) = χ2 (ω) ω . which is the average separation between Fe atoms (see table 11. Npm . includes.Magnetic dissipation 363 When the volume density of parmagnetic atoms. the static susceptibility becomes χ0 = 0.4 × 10−11 s. Because spin–spin relaxation is so fast. therefore. nuclear paramagnetism.12) so that τ −1 ωpr 2µB B loc 8. When its moment is 2m Npm directed towards the precessing electron. Interestingly. The electron spins precess around the applied field at frequency ωpr . T 1 + ω2 τ 2 The approximation on the right holds at low (rotational) frequencies. violently fluctuating because the spin directions are tossed about in collisions with phonons. does not appear in the expression for χ2 (ω).2. To get −1 a handle on the spin–spin relaxation time τ . we assume that the outermost five electrons in the 3d shell have only spin momentum and put g = 2 in (11. the paramagnetic atoms are dispersed irregularly over the grain.045/T .

The polarization resulting from scattering by dust is usually strong. We encountered an example of ϕ(t) in equation (1. and keeps only the second moment 2 . 11. One should also note in these figures. When one compares the magnetic field lines with the optical polarization vectors in other spiral galaxies. especially in the lower right of the figure. in the figures of [Mat70] for stars with a large distance modulus. In extinction measurements of the Milky Way. has the same direction (projected on the sky) as the electric field vector of polarized starlight. a galaxy which has beautiful spiral arms and is viewed face-on. for example. When.364 Grain alignment is mathematically fully determined by all its moments: ωn = f (ω)ωn dω · f (ω)dω −1 .3 and 11. derived from the polarization of radio synchrotron emission. it seemed natural to propose that magnetic fields are somehow responsible for the grain alignment. revealed that the polarization vector from interstellar extinction runs predominantly parallel to the galactic plane. For plausible reasons. match the magnetic field vectors only poorly. But for a factor of order unity.28) with τ = ω2 − 2 . χ2 (ω) resembles Debye relaxation at frequencies ω τ −1 . As the magnetic field. The magnetic field follows the arms nicely but the optical polarization vectors. This is exemplified. the large local excursions.143) of section 1. in line with experiments. which describes how the magnetic moment M decays when a constant magnetic field H is suddenly switched off. The reason that their orientation is not as uniform as in the Milky Way is probably due to the fact that one also sees scattered light in external galaxies. which comprise several thousand stars both in the southern and northern hemisphere. . f (ω) is related through Fourier transformations to the relaxation function ϕ(t). scattered light is absent and cannot contaminate the polarization. where the decay was exponential. especially when it occurs at right angles.4 for M 51. This is demonstrated in figures 11. however. however. and is completely unrelated to magnetic fields. ϕ(t) = χ0 e− 2 ω t . one approximates the relaxation 1 2 2 function by a Gaussian. one obtains ω χ2 (ω) = χ0 ωτ 1 π 2 1 2 e−ω 2 τ 2 /2 (11. the coincidence is less impressive.3 Magnetic alignment The polarization surveys of the Milky Way.6.

18).3. its magnetization M changes. For simplicity. we consider. ω and L.11)) τ =m×B directed perpendicularly to m and B. which is assumed to be small. Radio and optical polarization in M51. The radio polarization is fairly high (∼20%) and the corresponding vectors are shown by the large dashes.Magnetic alignment 365 Figure 11.3.85 µm (adapted from [Sca87]).11)). they follow the spiral arm pattern. a cylinder with its angular velocity and momentum vectors. The magnetic B-vectors at 2.8 cm [Nei90] are overlaid on the electric polarizarion vectors at the center wavelengths of 0. from (1. But χ is complex and the magnetization M stays a small angle χ2 /χ1 . To determine τ . The field exerts. on the dipole. pointing the same way but perpendicular to the cylinder axis. the susceptibility χ taken at the rotation frequency ω. in the constant interstellar magnetic field B a magnetization M and a dipole moment m = V M.5. because M = χB. The electric polarization is small (∼1%) and represented by short (unfortunately. as in figure 11. also black) dashes. 11. The torque tries to line up the dipole with the field. a torque (see (11. If the grain spins.1 A rotating dipole in a magnetic field A particle of volume V acquires. we need the magnetization M and. the magnetization would adjust instantaneously to the direction of B and the torque would vanish because of the term B × B = 0 (see (11. If χ were real.

This leads to the following equations of motion for the rotating magnetic dipole: dL χ2 = τ = V M × B = V (ω × B) × B dt ω d E rot χ2 = τ · ω = −V (ω × B)2 . operated by Aura Inc.12)).366 Grain alignment Figure 11. Use of this image is courtesy of the Palomar Observatory and the Digitized Sky Survey created by the Space Telescope Science Institute. . The companion on top is NGC 5195. behind the field B (see (1. This produces the magnetization M2 = χ2 ω×B ω which does not disappear in the vector product with B.4. dt ω (11. B1 B. as viewed by an observer in the rotating grain. the torque τ and the change in angular momentum dL are perpendicular to B and lie in the same plane as B and ω (see figure 11. Therefore. B2 . it is wasted into heat. the magnetization responds as if there were two magnetic field components: a major one.29).30) The rotational energy decreases. perpendicular to the first and to ω and of length B2 = (χ2 /χ1 )B sin θ .29) (11. producing a magnetization M1 = χ1 B and a small one. and is reproduced here with permission from AURA/ST Scl. for NASA. An optical picture of the spiral galaxy M51 from the Palomar Sky Survey shown in the same orientation and on approximately the same scale.5). Because of the double vector product in (11.

Magnetic alignment 367 B ω. there will be no more losses. This leads to a torque τ on the grain. When the alignment is complete (ω B). The inverse trel is.29): trel = Iω . This is the Davis–Greenstein mechanism. B does not exceed several times 10−6 G in low density clouds (n H < 103 cm−3 ). A paramagnetic cylinder rotating in a magnetic field B acquires a dipole moment that is not fully aligned with B. I /V . 11. Note that a higher rotation speed. V χ2 B 2 (11. ω and L lie parallel to each other and perpendicular to the axis of the spinning cylinder.31) −1 A typical number is 106 yr. by order of magnitude. But there are limits to these numbers. ω. although it implies enhanced dissipation.2 Timescales for alignment and disorder We estimate a relaxation time trel for magnetic alignment from the condition L/trel = d L/dt from (11. from diverse independent evidence. does not decrease the relaxation time trel because χ2 ∝ ω at low frequencies. L are in the same plane. As a result. B. the rate at which the angle θ between the field and the angular momentum vector changes. We see that magnetic alignment is favoured by a low ratio of moment of inertia over volume. A minimal value of I /V is given by a sphere and. All vectors τ . the standard scenario for magnetic alignment.5. When we compare trel with the time tdis after which disorder is established through collisions with gas atoms and assume Brownian motion so that tdis is . and a strong magnetic field.3. L θ τ Figure 11. the cylinder lines up with its long axis perpendicular to the field B.

tdis . when new layers are deposited or active sites are destroyed or created. The subparticles are possibly synthesized in supernova explosions and somehow built into interstellar grains. the angular momentum of the fast rotating particle must stay stable in space over a time greater than trel .1.3 × 10−21 B2 1 2 ω . Even if they contain only a few tens of iron atoms.39). 11. trel /tdis would be substantially above unity. for example. B2 Because the Serkowski curve mandates a typical grain size a of order 10−5 cm.368 Grain alignment given by (9.3. disorientation will set in when the surface properties change. we find trel 3 = tdis 10 8km a π 1 2 2 Tgas an ω B 2 χ2 1 Tgas an 7. The rotation speed ω does not appear in this formula.39).27) yields 1 trel tdis 2 × 10 −9 2 Tdust Tgas an . they . Therefore one has to search for a more effective magnetic alignment mechanisms.31) is not altered by the spinup as long as χ2 ∝ ω. one can readily check that in the environment where we actually observe optical polarization. will be much increased and no longer be given by (9. as discussed in section 11. trel /tdis . This implies that the grains change their angular momentum chaotically before paramagnetic relaxation has had time to do its job and they would not be aligned. For alignment to be effective. As the relaxation time trel of (11. they have a very high kinetic energy and the time to create disorder. This may occur. down and thus point a way out of the dilemma [Jon67]. 11.2. trel /tdis must be smaller than one. Of course. Because inhomogeneities on the grain surface drive the suprathermal spinup. Suprathermal rotation.3 Super-paramagnetism Interstellar dust may contain not only isolated iron atoms. otherwise orientation is poor. which would only produce paramagnetism of possibly insufficient strength but also tiny pure iron subparticles in which the collective effect of the electron spins dominate.3. χ2 (11. Time scales for these processes are debatable. Pure paramagnetic spin–spin relaxation with χ2 from (11. effective grain alignment according to the Davis–Greenstein mechanism with damping from paramagnetic spin–spin relaxation seems possible. supra–paramagnetism and ferromagnetic inclusions can bring the ratio of relaxation time over collisional disorder time.1 Magnetic alignment in suprathermal rotation Should the grains be rotating suprathermally with angular velocity ω 105 s−1 .32) m a and n are the mass and density of the gas atoms.

corresponding to weak damping. However.4 Ferromagnetic relaxation In a more general way. it must overcome a potential barrier Ubar . A grain with such ferromagnetic inclusions may then display super-paramagnetism. in the expression for the static susceptibility (11.e. each consisting of n atoms of magnetic moment m a . not restricted to paramagnetic relaxation. the higher the barrier. one assumes that the magnetization M is forced by a periodic outer field H to oscillate harmonically. The phenomenon is halfway between ferro. 11. respectively) separated by Ubar . The more atoms there are in the cluster. Ubar (= nkθ ) becomes. The magnetic moment of each cluster is at its saturation value m cl = nm a but the whole particle is paramagnetic because its magnetization follows equation (11.11 K and γ = 10−9 s determined experimentally. 2 2 ¨ ˙ (11.34) For small frequencies (ω ω0 ).and paramagnetism in the sense that the clusters are ferromagnetic single domains but their magnetization changes spontaneously by thermal perturbations so that the grain.25). M = χ0 H . Therefore the relaxation time has the form τ γ eUbar / kT with θ = 0.77)) χ2 (ω) = χ0 ωτ 1 − (ω/ω0 )2 2 + ω2 τ 2 .33) implies with τ ≡ γ /ω0 (see (1. (11. When a cluster changes its magnetization. the static susceptibility of the grain as a whole becomes χ0 = n Npm m 2 a 3kT . If τ is small. i. Suppose the grain has Ncl iron clusters.3. It is very sensitive to the number n of iron atoms per cluster. a pronounced . because not all directions of magnetization are equal. these are the cube edges [100] and the body diagonals [111]. as a whole. As n increases. There is an easy one and a hard one (for iron. As obviously Npm = Ncl n.17) with the Langevin function L(x).25) can become very large for small frequencies. the expression again approaches Debye relaxation (11. 2 In a static field.33) M + γ M + ω0 M = ω0 χ0 H.Magnetic alignment 369 already behave like ferromagnetic single domains. It may be overcome by thermal excitation and the rate at which this happens is proportional to e−Ubar / kT . they show spontaneous magnetization.26) the moment m a of a single Fe atom has to be replaced by the cluster moment m cl and the volume density of magnetic atoms Npm by the number of clusters Ncl in the grain. Therefore χ0 increases by a factor n over the static susceptibility with isolated iron atoms. χ2 (ω) after (11. behaves like a paramagnet. Equation (11.

3 × 1011 s−1 . the changing magnetic field causes the transition layer between adjacent domains (Bloch walls) to rearrange and the wandering of the walls dissipates energy. the precession τ −1 (4π Msat /3)(e/m e c) 1. the low-frequency limit of χ2 (ω)/ω becomes ∼10−7 which is four orders of magnitude greater than for spin–spin relaxation. and χ2 (ω) = χ0 ωτ . of course.1% would increase the susceptibility tenfold. As now frequency ωpr −10 . γ = 2ω0 after (1. as in (11.63). Then there are only two parameters. in order for the resulting field to be considered at a constant bias.35) For a grain with ferromagnetic inclusions. be as strong as the fluctuating field from the isolated iron atoms in the paramagnetic grain material. the presence of single-domain ferromagnetic inclusions has the interesting consequence that it provides a bias field H0 for the adjacent paramagnetic material [Dul78]. [1 + (ωτ/2)2 ]2 (11. τ and ω0 with physical processes and to determine their values from a combination of experimental data and theory. they are stable as ferromagnetic single domains. The effect on the damping of the grain as a whole depends. • However.54)). a field (see table 11. If the system is critically damped. A spherical single domain of radius a has. at a distance r from its center.33). as we did for paramagnets.2 and an estimate for τ of order 10−8 s from the laboratory. The field produced by the inclusions must. on the volume fraction f Fe of the iron inclusions but even f Fe = 0. The difficulty is to associate χ0 . The moving walls can phenomenologically be described by a harmonic oscillator. The latter is equal to 4π Msat /3 (see the electric analog in (3. This mechanism. This enables paramagnetic dissipation by spin– lattice relaxation to operate in the otherwise paramagnetic grain material.22µB 3r 3 7000 a r 3 . for which we estimated above 700 G. χ0 and τ . ˚ If the iron clusters have sizes below ∼200 A. Putting the saturation magnetization Msat = 1750 G and χ0 = 12 (table 11. Equation (11. is much more efficacious in enhancing the ratio χ2 (ω)/ω than spin–spin relaxation.2). One can squeeze out some semi-theoretical number for χ2 supposing.370 Grain alignment resonance appears at ω0 . two scenarios exist: • If the iron clusters are big enough to be divided into several domains. .34) says nothing about the dissipation mechanism. especially at low frequencies as evidenced by figure 11. that dissipation comes from precession of the magnetic moments in the local field.2) 4πa 3 Nbulk 2. With the static susceptibility χ0 = 12 from table 11. particles with single-domain iron inclusions seem not to χ2 (ω)/ω ∼ 10 be terribly effective in damping.6. which we had so far discarded. of course.

Ix < I y < Iz . χ = χ1 + iχ2 . 11. be most effective polarizers. The dots show the total dissipational loss rate W ∝ χ2 ω when sl ss both processes are activated (here χ2 = χ2 + χ2 ). it follows readily from (11. the axis of maximum inertia (the z-axis) tends to become parallel to L. for example.31). the broken line depicts for this case χ2 /ω which determines the alignment efficiency after (11.25χT . For the relaxation times it is assumed that τsl = 100τsl .1) and (11.5 Alignment of angular momentum with the axis of greatest inertia There is likely to be damping within the particle independent of the outer magnetic field. therefore.6. we further adopt ss χad = χ2 (0) = 0. each ferromagnetic inclusion provides a bias field for spin–lattice relaxation in the paramagnetic material over a volume which is about ten times bigger than its own.Magnetic alignment 371 Figure 11.7) and using the angle θ between the figure axis and L . The rotational energy E rot of the grain then decreases and is turned into heat. Consequently. Therefore. With the help of (11. Assuming. When such grains are then magnetically aligned.3. their orientation will be almost perfect and they will. The ordinate is arbitrary. A sketch of the magnetic susceptibility.2) that the angular velocity ωz grows at the expense of ωx and ω y . while the angular momentum vector L stays constant in the absence of an external torque. as a function of frequency in paramagnetic dissipation for spin–spin (ss) and for spin–lattice (sl) relaxation.

8)).3. the rotational energy E rot of (11. we first recall the famous experiments by Einstein and de Haas and its counterpart by Barnett. • When an unmagnetized iron cylinder with inertia I is suddenly magnetized by an outer field. will probably not show such alignment. As the total angular momentum of .39) for its value for Brownian motion). N electrons flip their spin direction from up to down.3.1 Mechanical damping via variable stresses Interstellar dust particles are not perfectly rigid bodies and the centrifugal forces in a spinning grain change the distances among the atoms. 11. however. so they will always rotate about their axis of greatest moment of inertia.1). Interstellar grains in Brownian motion. on the rotational velocity. (11.6. I x Iz If one knows the dissipation rate W = d E rot/dt.2 Magnetic damping via the Barnett effect More efficient is a gyromagnetic mechanism which is based on the coupling between the spin system and the macroscopic rotation of the body.36) θ dθ dθ For the alignment to work.372 Grain alignment (see figure 11. crucially. 11. 11. To see how it works. It turns out that for suprathermally spinning inelastic grains. A perfect and cold crystal is very elastic with hardly any loss but a real grain has imperfections and mechanical deformations inevitably drain some of the rotational energy into heat. it is necessary that t be shorter than the disorder time tdis (see (9. the atomic magnetic moments and their associated angular momenta line up. the mechanical damping constant of the grain material and.2) can be written as 2E rot = L2 + L 2 A cos2 θ Ix A= I x − Iz .6 Mechanical and magnetic damping We discuss two further dissipation mechanisms: one mechanical. t is much shorter than tdis .3. The previously introduced time t depends on the stiffness of matter. the rotation vector ω moves with respect to the grain coordinates and so the stresses arising from the centrifugal forces are time variable with frequency of order ω (see (11. one can work out the time t which it takes until the z-axis and the angular momentum L are parallel: t ∼ t d E rot dt ∼ = W −1 = −W −1 L 2 A sin θ cos θ ∼ −W −1 L 2 A. Suppose that in the process of magnetization. the other gyromagnetic. When rotation is not about a principal axis.6.

27) and B from (11. the centrifugal force may be neglected and one may put v = x x . To understand the linear relation (11. e (11. Therefore. When ˙ ˙ ω is small. When an unmagnetized cylinder is rotated at angular velocity ω. each atom of mass m endows the grain upon collision with a small angular momentum (11.38) The prime refers to the uniformly rotating system in which there additionally appear a Coriolis and a centrifugal term (second and third on the right).39) dLgas = r × mv. 11.Non-magnetic alignment 373 • the cylinder is zero before and after the magnetization. we mention. ¨ that in a rotating frame K the acceleration x of a point mass is different from the ¨ acceleration x as seen in an inertial frame K : ¨ ¨ ˙ x = x + 2ω × x + ω × (ω × x). Rotating the atom is thus equivalent to exposing it to a magnetic field of strength B = (2m e c/e)ω. . in the cylinder. this is Larmor’s theorem.37) between ω and B.36) is inversely proportional to ω2 and much shorter than the disorder time tdis . first. it feels a magnetic field B.1 Gas streaming When gas streams with a velocity v past a dust particle. we find that the alignment time t in (11. each of its electrons is subjected to the additional force: 2m e v × ω + (e/c)v × B. the cylinder acquires an angular momentum L = N and rotates afterwards at a speed ω = N /I (Einstein–de Haas effect). in a stellar wind. 11. a magnetization M = χB (Barnett effect).4. which is parallel to ω and has the absolute value B=ω 2m e c .37). The sum of the two is zero when ω + (e/2m e c)B = 0.37) The field B then produces. The first term is due to rotation and the other represents the Lorentz force of the magnetic field.4 Non-magnetic alignment We discuss two other ways in which grains may be aligned in the interstellar medium. They are interesting in themselves but except for special configurations.30) with χ2 from (11. when we take an atom and apply a magnetic field and furthermore slowly rotate the atom. for example. When we apply this result to an interstellar grain of volume V and compute the dissipation rate W = χ2 ωB 2 after (11. (11. not likely to be responsible for polarization.

7 finds that the starlight is polarized. An observer (right) views a star (left) through a cloud of cylindrical grains. there is no favorite φ. Because of the vector product. gas streams at supersonic velocity v relative to the grains. all changes in angular momentum.8 falls almost along the x-axis. φ) so that the direction of streaming is given by θ = 0 and the direction towards the observer by θ = π/2 and φ = 0 (figure 11. If the grain is very elongated.7). The observer in figure 11. we have. L gas will be limited by friction with the gas. When the number N of collisions is large. An observer who looks at the star from below in the direction of the velocity vector v in figure 11. Let us choose spherical polar coordinates (θ. The net polarization is due mostly to grains with φ 0. lie in this plane but are randomly distributed there.374 Grain alignment θ φ v Figure 11. When there are many particles. If the grain is a rod of length l. 2 The situation is depicted in figure 11.7. dLgas .8. φ). If the grain is hit by N atoms. Directions are given in the spherical polar coordinates (θ.7 would detect no polarization. Therefore the angular momentum is perpendicular to the x-axis which implies that the grain rotates in a plane with φ = constant. while those with φ π/2 do not contribute. all such planes are equally likely. dLgas lies in the plane θ = π/2 which is perpendicular to v. approximately. Perpendicular to the line of sight. Cylindrical grains will rotate in planes φ = constant and this results in a net orientation of their rotation axes and in polarization of the starlight [Gol52]. So the final angular momentum Lgas is also in the plane θ = π/2 and its length is the result of a random walk: √ L gas = N · d L gas . the vector r in figure 11. . d L gas = 1 lmv.

The latter dominates when > d L abs .2 Anisotropic illumination When the grains absorb radiation exclusively from one side. indicated by the curved arrow. the situation is similar to gas streaming. on top of their momentum hν/c. an intrinsic angular momentum that is imparted to the grain in each absorption event. and not perpendicular to it. This condition is fulfilled when the grain is smaller than the wavelength. the other angular momentum . The average (root mean square) angular momentum of the grain after the capture of N photons √ becomes N . Photons have. its momentum is hν/c. . Only now the impinging particles are photons and not gas atoms. The thermal motion of the gas has a disordering effect as it introduces random velocities between gas and dust particles. But the photon also transfers angular photon momentum and this leads to rotation out of the plane of the drawing. 2 c There is another effect related to photons that creates independent additional alignment [Har70]. one transfers momentum hν/c. as in the vicinity of a star. 11. There are two alignment processes through photons.8. It is directed along the line of photon propagation. The same is true for a photon coming from the same direction as the atom. as d L abs before. So each transfers the angular momentum d L abs = 1 hν l· .4.Non-magnetic alignment 375 x r v Figure 11. and their momentum is hν/c instead of mv. But such disorientation may obviously be neglected when the streaming motion is supersonic. either in a positive or negative direction. An impacting atom of momentum mv causes the cylinder to rotate in the plane of the drawing.

Inside a molecular cloud.376 Grain alignment Each emitted photon carries off a quantum in an arbitrary direction. the grain will acquire from this process an angular momentum L emis . a grain in the diffuse medium. . As the energy budget of heating and cooling is always balanced very well. where only a few energetic photons fly around. So L emis will √ be some 100 = 10 times larger than L abs . which is heated by few UV and cooled by many IR photons. very roughly. emits. the discrepancy between L emis and L abs is less severe. In the end. a hundred times more photons than it absorbs. But the associated vector Lemis has no preferential orientation.

is poor. by which they can be uniquely identified. mainly gas temperature and density. the lines also contain kinematic information via the Doppler shift in their radial velocity.1. so they are intermediate between molecules and grains and constitute a peculiar and interesting species of their own. Although their overall spectral signature was subsequently observed in very diverse objects (HII regions. The smallest grains. the emission is restricted to infrared wavelengths.1 What are PAHs? These are planar molecules. Because of the strong disturbance of any oscillator in a solid. PAHs. Because of their sharpness. The first observation of PAHs.1 Thermodynamics of PAHs 12. each grain emits very many lines but they overlap to give an amorphous continuum. star burst nuclei and other external galaxies). These features are the topic of the present chapter. Being a solid. usually with hydrogen atoms bound to the edges. their proper identification took place only in the mid 1980s. Resonances are also observed from silicate material and ice mantles. reflection nebulae. The spectroscopic information from dust. evolved stars. from the UV to the microwave region. PAHs is a short form for 377 . The smallest consist of barely more than a dozen atoms. made of a number of benzene rings. the transitions are broad and do not allow us to extract velocity information. planetary nebulae. without proper identification. however. which show very characteristic bands are a fortunate exception. cirrus clouds. Precise chemical allocation is also not possible but individual functional groups partaking in the transition can be identified and exciting conditions may occasionally be derived.Chapter 12 PAHs and spectral features of dust Gas atoms or molecules emit and absorb narrow lines. Moreover. Observing several transitions simultaneously allows us to infer the conditions of excitation. 12. dates back to the early 1970s when in the near and mid infrared a number of hitherto unknown features were found [Gil73].

which is at the left and right edge of the ellipse.1.378 PAHs and spectral features of dust unit cell p E + δE E q Figure 12. When . 12.1. for a given q. The accessible states are in a narrow range from E to E + δ E. There are then two coordinates. In interstellar space. given by the shaded annulus. position q and momentum p.2). according to equation (8.48). PAHs are excited by the absorption of a single ultraviolet photon upon which they emit their characteristic infrared lines. For f > 1. one finds the highest density of cells. Polycyclic Aromatic Hydrocarbons and they are called this way even when the hydrogen atoms are missing. the situation is multi-dimensional but otherwise analogous. Figure 12. i. in view of (5. i.2) as the sum of kinetic and potential energy of all atoms which are approximated by harmonic oscillators.v = ρ(E. and all states lie in an annulus around an ellipse. The oscillating particle is most likely to be found near the greatest elongation.1 illustrates the locus of the states in phase space when the system has only one degree of freedom ( f = 1). v) is the density of states at energy E subject to the condition that oscillator i is in quantum state v.2 Microcanonic emission of PAHs Let us investigate the possible states of a PAH of total energy E. v)/ρ(E). It is given in equation (8. The population of the cells in phase space is statistical and the probability of finding the i th oscillator of a PAH in quantum state v is. The ellipses correspond to total (kinetic plus potential) energy E and E + δ E. There the velocity is small and. The unit phase cell has the size δqδp = . Pi. respectively. The phase space accessible to a one-dimensional oscillator with energy between E and E + δ E is. ρ(E) is the density of states at energy E and ρ(E.

one obtains [Whi63] (E + a E 0 ) f −1 (12. i. and further refines the expression by adding to E the zero-point energy E 0 of all excited oscillators. which has a simple and fairly symmetric structure (see insert in figure 12. i.v = a ρ(E. ρ(E. (12. (12.1. There is one degree of freedom less and E. . The vibrational modes can be computed if one knows the force field (force constants) or the potential in the molecule. v) ξ = ( f − 1)ξ 1 − v + ρ(E) 2 f −2 f −2 ( f − 2)! j =i −1 hν j . The fudge factor a is of order unity. i.2 is based seven vibrational types are considered.v hνi Ai v. CH stretching. v) is also calculated after (12.079 ln(E/E 0 ) E0 = 1 2 j hν j . There are altogether f = 3 · (14 + 10) − 6 = 66 degrees of freedom.2) E + a E0 The i th oscillator emits in the transition v → v − 1 the power εmic (E. easy to handle theoretically.0 v≥1 v Pi. C14 H10 .87 + 0.3 The vibrational modes of anthracene As an illustration. εmic (E. i. therefore.Thermodynamics of PAHs 379 one extends expression (5. Besides the four obvious ones. all bonds at its edges are saturated by hydrogen atoms. The PAH is built up of three benzene rings. ξ= hνi . as well as the zero-point energy E 0 . have to be diminished accordingly.42) for ρ(E) to the case when the oscillators are not identical but each has its own frequency νi .v . v) = (E − vhνi ) + a(E 0 − 1 hνi ) 2 Hence the relative population of level v is Pi.3) 12.2) and is. v) = Pi. In the model on which figure 12.v−1 and the total microcanonic emission from oscillator i is. • • CC stretching. This gives ρ(E.1) but now one has to exclude oscillator i . we consider the PAH anthracene. i ) = hνi Ai 1.1) ρ(E) f ( f − 1)! j =1 hν j with a 0. therefore.

The distribution of the number of modes over energy is displayed in figure 12. we gave the Debye temperatures. by definition. For reasons of symmetry. This range is completely isolated and due to CH stretching. Despite the simplicity of the model. It is clear that atoms in similar force fields must swing in a similar manner.2. each having its eigenfrequency. for a graphite plane. all CH stretches of the ten hydrogen atoms in anthracene are crowded between 3030 and 3040 cm−1 . the frequency is roughly ω = √ κ/m (see (12. The polycyclic aromatic hydrocarbon anthracene has 66 degrees of freedom. which.2. The frequencies of the 66 oscillators by which the molecule is replaced are found by introducing normal coordinates (section 8. When κ is the force constant of CH binding and m the mass of an H atom. For instance. Hydrogen atoms are also involved in the modes around 1000 cm−1 which are due to a CH bend. They correspond to 66 oscillators (modes). The wavenumbers are tabulated in [Whi78] and grouped here into intervals of 100 cm−1 width. These are CCC and CCH bending as well as CCCC torsion. is at • • .4. is the Debye frequency. The rest of the histogram is due to the skeleton of carbon atoms and surprisingly smooth. θz = 950 K and θx y = 2500 K. the achieved accuracy is good. three mixed modes in which several atoms take part in the oscillation are also included. the high frequency cutoff. In section 8.380 10 PAHs and spectral features of dust number of modes anthracene C 14 H10 0 1000 2000 3000 wavenumber [1/cm] Figure 12. Some vibrational modes lie very close together because of the symmetry of the molecule. at most four of these CH frequencies can be different but these four are really distinct because they correspond to sites where the surroundings of the H atoms have different geometries.25).8)).4) and solving equation (8. the theoretical frequencies agree well with laboratory data. Excluding the CH bend at 3000 cm−1 . • • CC out-of-plane bending and CH out-of-plane bending. It contains several noteworthy features: • The modes strongly cluster around 3035 cm−1 .

2. The best example is presented by the C–H stretches at 3. if one looks closer. Because of anharmonicity. is permitted.28 µm (v = 1 → 0.4.Thermodynamics of PAHs • 381 1800 cm−1 and corresponds nicely to θx y (see (8. which when fulfilled makes life so much easier.3 demonstrates that if the PAH has absorbed a typical stellar photon of hν 4 eV. We also learn from the microcanonic results of level populations that those oscillators in the PAH which correspond to the major bands (table 12. and thus also for line emission. from figure 12. 66 E= i=1 hνi hνi / kT e −1 . At long wavelengths. The lowest energy is 73 cm−1 (137 µm) and again in qualitative agreement with the estimate for the minimum frequency νmin from (8.1.45) if one uses θz = 950 K. For higher photon energies. If the shift due to anharmonicity is small. we consider a fictitious molecule with 330 degrees of freedom but with the same relative mode distribution as in figure 12. they have to be taken into account. the thermal assumption is usually bad. see table 12. the highv lines blend with the ground transition producing an overall broadening of the band. Figure 12.3 µm.33)). We see. we can calculate from this equation the temperature of a thermalized anthracene molecule of energy E. There are now for each energy interval five times more modes. The shift is sufficiently large so that the v = 2 → 1 line can occasionally be seen well detached from the v = 1 → 0 transition.42 µm (v = 2 → 1). Let the PAH absorb a photon of energy E. If the photons are more energetic (E > 10 eV). this is always the statistically correct way. v ≤ 1.4 Microcanonic versus thermal level population We have learnt how to compute the microcanonic population of levels. for this PAH. However. E is equal to the sum of the mean energies of all 66 oscillators (see (5. the situation .1) are mostly found in the lowest quantum numbers. whereas it grossly underestimates the population numbers at short wavelengths. To estimate how big PAHs behave. Because all νi are known. we may indeed use the Boltzmann formulae (dotted lines) for the level population. to first order. When the levels are thermalized. not important. that for such a big PAH. We now apply our skills to anthracene and check whether. The v = 2 → 1 line is always weaker because of low level population but note that the Einstein A coefficient increases proportionally with v. It is acceptable only for lowfrequency vibrations. This implies that effects produced by the anharmonicity of the oscillator potential are. 12. the spacing between the upper levels decreases and the lines from higher v-states are shifted towards the red. the assumption of thermalization. things only improve. the modes extend into the far infrared. when it is excited by a photon of 4 eV.1) and at 3.11)). as for the C–H out-of-plane bend at 11.

if the size of the PAH is further increased.2). the number of states whose energy does not exceed E is. roughly . As the PAH gets bigger.41). 12. Of course. To an ordinary grain made of N = 105 . it must eventually be thermalized like a normal grain.1.1 eV (13 µm wavelength). Dots show the microcanonic fractional population as approximated by (12. always Nv = 1.5). For example. Note that while such a highly excited PAH cools. see also text. The dotted lines. Level population Nv in anthracene for three oscillators (their wavelengths are indicated) after absorption of a photon with E = 4 eV from the ground state. . one may certainly assign a temperature but it is doubtful whether one can do so in the case of a PAH.45) and νcut after (12. which only have a meaning at integral quantum numbers v. 1010 atoms. steadily declines. give the thermal results at a temperature T = 1325 K. the assumption of thermal emission gets poorer. for a PAH with f = 100 degrees of freedom that is excited to an energy of E = 10 eV and emits IR photons of 0.3. improves towards thermalization. the threshold for collisional and photoexcitation. .382 PAHs and spectral features of dust Figure 12.5 Does an ensemble of PAHs have a temperature? The concept of temperature is only valid if the system has many degrees of freedom. from (5. which may bear more affinity to a big molecule than to a solid body. νmin after (8.

somewhat artificially. major (top) from minor bands (bottom). If the photons were monoenergetic. as before.1 15.1 23.8).0 14.2 1.1. when exciting events are rare.1 Damping constant γ (1012 s−1 ) 20 14 22 6 4 7 3. The cross sections integrated over the band. the levels are not thermalized.3 11. If one cannot assign a temperature to a single PAH. prior to absorption of a photon.0 2. So one has to check in a microcanonic treatment. Furthermore. The space divides. Approximate numbers for PAH resonances. The kinetic temperature of the gas does not then describe the population of the internal levels of the atoms. excitation through energetic photons is much too rare an event and gas collisions are negligible altogether.7 8. whether the PAH is approximately thermalized or not.1 12 2.5 1.8 13.2 7.0 2.2 21.7 16. so would the energies of the PAHs freshly excited from the .6 11.3 15. as it should be for a good temperature but one of order unity.9 12. the PAH is usually in its ground state.7 1. See also figure 12.2 1.0 vibrational type C–H stretch C–C stretch C–C stretch C–H bend in-plane C–H bend out-of-plane C–H bend out-of-plane C–H bend out-of-plane C–H bend out-of-plane C–C vibration ? C–C skeleton C–C skeleton C–C skeleton C–C skeleton C–C skeleton C–C skeleton C–C skeleton Ns (E/ f ω) f . However. are per H atom for C–H vibrational modes. for example. The excitation then carries the signature either of the Planck function or of Maxwell’s velocity distribution and it is completely described by the Boltzmann formula (5. maybe one can to an ensemble of PAHs in the way one assigns a temperature to the level population of an ensemble of gas atoms? At least this is possible when the atoms are in a blackbody radiation field or when their density is high.6 5.4 18.5. at low gas density. σint = σ (λ) dλ.2 7.9 5 3 2 3 3 3 3 Integrated cross section σint (cm2 cm) 12 14 51 27 41 27 47 12 1.3 6.5 4 12 5. and per C atom for C–C modes. This is not a fantastically large number.Thermodynamics of PAHs 383 Table 12. Center wavelength λ0 (µm) 3. For PAHs.

384 PAHs and spectral features of dust Figure 12. therefore.e. i. say.2 PAH emission 12. Otherwise as figure 12. Collisions with gas atoms are rare and the gas temperature is too low anyway to thermalize the PAH. ground state be. immediately after the UV photon has been . Before the IR photons are emitted. Although the actual exciting photons are not monoenergetic but come from the interstellar radiation field and have a considerable spread. But by and large.1 Photoexcitation of PAHs In the diffuse interstellar medium. The energy of the absorbed photon equals 4 and 12. The dotted lines are calculated under the assumption of thermalization at T = 545 K (left box) and T = 1000 K (right box). mainly by emission in the fundamental modes (see table 12. Level population Nv of a PAH with f = 330 degrees of freedom. a singlet S0 . Then various and complicated things happen. The vibrational modes are. distributed over all available vibrational modes. Afterwards the molecule is highly excited and in a new electronic singlet state Si .2. their frequencies do not mimic thermal conditions. a neutral PAH molecule is almost all the time in its electronic ground state. also in their lowest level. The emission lasts typically for 1 s. 12.4. very fast.6 eV. the outcome is that the ultraviolet energy is thermalized. It has the same relative mode distribution as anthracene (see figure 12.3. Their distribution in energy would not show the spread typical of thermal systems but resemble a δ-function.1).2). once a day. some on time scales of 10−12 s. and then degraded into infrared photons. Emission of infrared lines from the PAH molecule is initiated by absorption of a UV photon of energy E UV .

˚ then L/s = 2 A and λcut = 4m e ( L )Lc s ∼ 330L h ˚ [A].4) ˚ Here a is measured in A and NC is the number of carbon atoms in the molecule.5) . A PAH may also be negative or bear multiple charges. so if there are s free electrons. One can understand the existence of the threshold by considering a free electron in a linear metallic molecule of length L. the gas temperature. 12. of course. some of the basic ideas are outlined in section 9. En = h2n2 8m e L 2 n = 1. the electron density and. . the excitation scheme is similar.PAH emission 385 absorbed. the energy hνcut = d En dn n= n=s h2s h2 . The electron is trapped in the molecule and its energy levels are given by (6. A transition with n = 1 has.2. the PAH may change from the electronic state Si without radiative loss to another singlet state. For PAHs of circular shape and size a . the crude relation λcut = 1630 + 370 NC = 1630 + 450 a ˚ [A]. . The phase space available to electrons with energy E ≤ E n is hn/2. all levels up to n = s are populated. Light from wavelengths greater than λcut is simply not absorbed and thus cannot produce the infrared resonances. on the UV cross sections of the PAH. one finds empiricallly. therefore. For a singly photo-ionized PAH. The degree and kind of ionization depends on the radiation field. in the laboratory. Small PAHs (NC < 50) are only excited by UV radiation.52). In this case.2 Cutoff wavelength for electronic excitation There is a minimum photon energy for electronic excitation of a PAH corresponding to a cutoff wavelength λcut . (12. larger ones also by optical photons. 2. . because the triplet system has a metastable ground level T1 . Or it may undergo intersystem crossing to a triplet state. a denser population is forbidden by Pauli’s exclusion principle. The ionization details may not be decisive for evaluating the IR emission of PAHs but probably have an influence on the cutoff wavelength and thus on the conditions for excitation (see later). only the electronic ground state is then a doublet because of the unpaired electron.2. . (12. = 2 4m e L 4m e ( L )L s ˚ If the atoms have a size of ∼2 A and if there is one optical electron per atom. some of the energy E UV will eventually go into phosphorescence from the triplet T1 to S0 and thus be lost for the IR features.

an exact value is not required.2. with a critical internal energy E cr and because E cr > E 0 . Big PAHs are easier to ionize than small ones. ν0 and E 0 . The critical temperature Tcr can be found from E cr via equation (5. the oscillator i in a state v ≥ vdis is given by equation (12. 12. The infrared emission rate of PAHs. E 0 = 2. E 0 = 2. the PAH is stable. an atom detaches itself from the PAH above a certain vibrational quantum number vdis .24 eV and declines to ∼6 eV for molecules with 100 carbon atoms. ν0 = 1015 s−1 [Joc94]. one. in a PAH with f degrees of freedom and total energy E. In slight modification of (12. kcr is associated. can be determined experimentally.386 PAHs and spectral features of dust This qualitative and simplified model of a linear metallic grain already yields the right order of magnitude for the cutoff. E 0 may be loosely interpreted as the bond dissociation energy. E cr So there is a rough linear relationship between the degrees of freedom of the PAH and the critical energy. if however acetylene evaporates. . For example. can thus be considered as the critical dissociation rate. kcr .3 Photo-destruction and ionization Because the PAHs are so small. In a strong and hard radiation field. therefore.6). with respect to C2 H2 loss. one has. f E cr = i=1 hνi hνi / kTcr e −1 . Ionization after absorption of an energetic UV photon is another reaction channel. writes kdis(E) = ν0 1 − E0 E f −1 . Because of the anharmonicity of the potential of a real oscillator. (12. the excited PAH is hot and has a certain chance to evaporate. and it has the overall effect of stabilizing the molecule against destruction.2). with respect to hydrogen loss.11). the ionization potential of benzene is 9. When kdis(E) > kcr .9 eV. When relaxation through emission of IR photons is faster than dissociation. the aromatic cycles are unscathed. The chance P(E) of finding. After UV absorption. from (12. the carbon rings are broken. The two factors. As kcr enters logarithmically. and P(E) is proportional to the dissociation rate kdis(E) of the molecule.2). evaporation will occur.8 eV and ν0 = 1016 s−1 .6) ν0 has the dimension of a frequency. which is of order 102 s−1 . besides dissociation and infrared emission. ln kcr ln ν0 − ( f − 1) E0 . approximately. PAHs will not exist at all. absorption of a UV photon can lead to dissociation. If hydrogen atoms detach.

L = 105 L ). more variable in intensity from source to source and their origin is less clear. the number of hydrogen and carbon atoms. and N the number of C or H atoms. i.e. harsh radiation field.2. under what conditions do PAHs survive in a given. . which are all located at the periphery of the PAH. we estimate that PAHs with only 30 carbon atoms can survive.1 compiles their vibrational type.e. Table 12. center wavelength and integrated line strength but it also lists some of the weaker or minor resonances which are. As the physical processes connected with PAH destruction are most complicated.4 µm which was mentioned in section 12. therefore. respectively. f evap ∼ 10−8 and Tcr ∼ 2500 K do not seem to be unreasonable numbers and one may use such an approach to decide whether PAHs evaporate or not. one may be tempted to use a purely phenomenolgical approach and simply assume that PAHs evaporate if the fraction of PAHs above some critical temperature Tcr exceeds a certain value f evap . There are two classes of bands: one involves hydrogen atoms. The strength of the emission is. near an O star (distance r = 1016 cm. a PAH without hydrogen atoms will have weaker C–C bands than one with hydrogen atoms (and the same NC ). 12. As the total amount of absorbed and emitted energy is determined solely by the carbon skeleton. NC should be above 200.106)). It is not listed in table 12. γ ν2 2 π 2 [ν 2 − ν0 ]2 + [γ ν/2]2 (12.1 because it is an overtone.4 Cross sections and line profiles of PAHs A PAH has several hundred vibrational modes but one typically detects only half a dozen major resonances around the frequencies where the strong modes cluster. respectively. We approximate the line shapes by Lorentzian profiles (see (1.46) has to be computed anyway for the evaluation of PAH emission. The most prominent representative of the minor features is the band at 3.4.2).PAH emission 387 For the astronomically relevant question. one has to balance their destruction rate with the rate at which they are built up by carbon accretion. the other only carbon atoms. In the interstellar radiation field (figure 8. i. σint = σ (λ) dλ the cross section of the particular band integrated over wavelength. The damping constant γ determines the band width and is probably the result of an overlap of many transitions with frequencies clustering around ν0 . when ∞ Tcr P(T ) d T ≥ f evap . however. The probability function P(T ) of (8.1. proportional either to NH or NC .7) where ν0 = c/λ0 is the center frequency.

for the heavier CN. the stretching band lies at ∼10 µm. hydrogen atoms are attached to the edge of the carbon skeleton. They allow us to identify certain characteristic groups in molecules of which the grains are made. The features are grain-size independent and serve as a tool to unravel the chemical composition of interstellar dust. the feature becomes narrower because collisional damping with neighboring atoms is reduced.3 Big grains and ices A few infrared resonances can also be observed in grains of normal size. For example. the resonance is at ∼10 µm. the fundamental frequency ω between two atoms of mass m 1 and m 2 connected by a spring of force constant κ is given by ω = 2πν = where the reduced mass is µ= κ/µ (12. the stretching mode of the light CH molecule occurs at ∼3 µm. three (trios) or four (quartets) adjacent H atoms.388 PAHs and spectral features of dust 2 1 4 2 3 Figure 12.8) m1m2 . The number of neighboring H atoms is indicated for each (peripheral) carbon ring. Likewise.5. Its width and center frequency depend weakly on temperature: when the solid cools. . in a triple C≡C bond at ∼5 µm. the center wavelength increases from approximately 11. 12.3 to 13. In a PAH. m1 + m2 (12. Due to the coupling between the vibrating H atoms. In a simplified picture. The exact frequency of the C–H out-of-plane bend depends on whether the hydrogen atoms are isolated (mono positions) or whether there are two (duos). the force constant κ and thus the frequency ν in (12. At the same time.8) increase because the lattice shrinks. in a single C–C bond.1).6 µm (see table 12. and fairly sharp in a crystal where high structural order leads to long-range forces.9) The frequency is high when the molecules are strongly coupled and the masses of the involved atoms small. for example OH in water molecules when water has frozen out. The resonance is broad in an amorphous material. Let us consider the vibrations in a characteristic group that consists of just two molecules.

the atoms are squeezed in by other molecules and cannot rotate freely. absent in molecular clouds is the one at 3.3.7 . The vibrational spectrum of diatomic molecules in an ice mantle has several peculiarities. from the depression of the 10 µm feature in figure 12. In a solid. τV /τ9.7 µm due to an Si–O stretching vibration in silicates. CH3 OH.4 µm. each consisting of many rotational lines (see section 14. τV τ9.6 we estimate that the object IRS9 in NGC 7538 suffers about 35 mag of visual extinction corresponding to τ9. It is usually attributed to an aliphatic (the carbon atoms are arranged in a chain.4 µm 389 The most prominent and widely studied resonance is the one at 9. The stars have shells of carbon dust that produce a smooth 10 µm emission background with a temperature of ∼1000 K and without the 9. one can determine the visual extinction AV from the 10 µm absorption. To stay on the grain and not to become photodesorbed. CO.3. . For free molecules. Towards star-forming regions. Only hindered rotations may be possible in . Knowing the ratio τV /τ9. the absorption coefficient ratio κ18 /κ10 0.7 µm feature. all oxygen is bound in CO and silicate grains are absent.7 2. NH3 . Another feature that is widely observed in absorption in the diffuse medium but.Big grains and ices 12. The mantles consist of a mixture of molecules like H2 O.1 and figure 14. it is observed in absorption.6.1 µm feature due to OH stretching seems to require shielding by more than 8 mag of visual extinction. To derive the ratio of the visual to the 9.7 18 (12. quite interestingly. CO2 . The method still works when stellar photometry is no longer possible. For instance.7 µm optical depth.1). towards oxygen-rich mass loss giants it is seen in absorption and occasionally in emission. 12. so these branches are supressed.10) with a scatter of about 10%.2 Icy grain mantles Ice mantles form in cold and dense molecular clouds as a result of gas accretion onto refractory cores (section 9. there is an O–Si–O bending mode which is weaker and much more difficult to observe from the ground because of atmospheric attenuation. At around 18 µm. For example. in varying abundance ratios. In these stars. for example.2. one measures the extinction through a cloud that lies in front of a carbon star. An impressive example of absorption by ice mantles is presented in figure 12. CH4 .and a P-branch. the 3. they have to be well hidden from the UV radiation field.1). It is the strongest indicator of silicon in space.1 The silicate features and the band at 3. and not in a ring) C–H stretching mode. there is generally an R.5. in CH3 but its exact identification has not yet been settled. On average. . . provided there are no hot grains along the line of sight that emit themselves at 9. in a gas.7 µm.7 .

The center frequency is also influenced by the composition of the matrix into which the molecule is embedded. which the molecule switches from one position to another.390 PAHs and spectral features of dust Figure 12. by rotating. to account for the interstellar polarization. without modifying them at certain wavelengths to improve fits. if employed in radiative transfer calculations. The infrared spectrum towards the infrared object IRS9 in the star-forming region NGC 7538. They are all capable of explaining the basic data on extinction. lies in its relative simplicity. see chapter 10.4 An overall dust model We are now in a position to present a tentative but complete model of interstellar dust with respect to its interaction with light. The resonances at 10 and 18 µm are due to refractory silicate cores. yield successful matches to observed spectra. A dozen of rival dust models can be found in the literature. which is energetically different.6. only 12 types of . its advantage. 12.) Altogether. The other features come from molecules that have frozen out onto refractory grains and are now in ice mantles. if any. scattering and infrared resonances and. (Therefore. Our model certainly does not produce superior fits. It uses published sets of optical constants. It will be applied in chapter 16 to compute the spectral energy distribution of dusty astronomical objects. Reproduced from [Whi96] with permission of EDP Sciences. we have to turn some of the spheres into infinite cylinders. The grains have the simplest possible structure and shape: they are uncoated spheres. polarization.

All major dust models agree. where one particle type differs from another either by its size or chemical composition.1. not insure against common error. K λ slumps but increases again when the wavelength falls below the cutoff for electronic excitation. fraction of built in refractory organic compounds).An overall dust model 391 Figure 12. such as enthalpy and vibrational modes. The PAH parameters are described in the text and in table 12. coating. each PAH = 3%. size and optical constants of the grains. None of the models is certain because hard checks on their validity are missing. All models can be refined by varying the shape. K λ refers to 1 g of interstellar matter. The mass absorption coefficient. This further enhances the agreement with observational data but does not necessarily bring new insight. λcut . Below 3 µm. of a mixture of two kinds of PAHs. Such concensus on important issues does. dust particles are needed. however. Differences between models show up mostly in the details of the chemical composition (mixing of components. under the assumption of a likely grid structure for the atoms ([Dra01] and [Li01]). A more promising attempt is to increase the internal consistency of the models.7. K λ . on the solid state abundances. . including PAHs. in principle. very small grains with temperature fluctuations and big grains consisting mainly of silicate and carbon. in the wavelength range where the infrared resonances are with an abundance YC observed. Steps in this direction have been undertaken with respect to the smallest particles (PAHs and vsg) by computing grain properties. the approximate grain sizes and the minimum number of dust components: one needs PAH–ike particles. Interstellar grains are not within our reach and we cannot verify our assumptions in the laboratory. degree of fluffiness.

broken curve) as a function of wavelength for the dust model of the diffuse medium.19. also called MRN grains (section 7. a− = 150 A.392 PAHs and spectral features of dust Figure 12. Absorption and scattering cross sections are calculated from Mie theory employing optical constants from figure 7. a+ = 1200 A).8.4. we adopt 5–10%.5 g cm−3 . The total extinction coefficient (MRN+vsg+PAH. ˚ form a population of very small grains (vsg) with radii less than 100 A.1 The three dust components Our dust model for the diffuse interstellar medium consists of three components with the following properties: • Big grains. It refers to 1 g of interstellar matter adopting a dust-to-gas ratio Rd = 140. The ˚ ˚ ˚ bumps at 2200 and 700 A are due to very small graphite particles (a− = 10 A. One ˚ ˚ a+ = 2400 A) can read the amount of scattering as the difference between extinction and absorption. Most of the solid matter in the interstellar medium resides in big grains. ˚ and big amorphous carbon grains (aC.5 × 10−4) and silicate particles ([Si]/[H] = 3 × 10−5).5). Their chemical structure has been discussed in section 7. a− = 300 A.5). both with bulk density ∼2. 12. They consist of amorphous carbon (abundance in solid [C]/[H] = 2. Very small grains (vsg). They are spheres and their radii ˚ range from ∼200 to 2400 A with an a −3. Big grains attain a constant temperature in any radiation field.5 size distribution (section 7. They • . Also shown are the extinction ˚ (full curve) and absorption (dotted curve) coefficients for big silicates (Si. A few percent of the solid matter.4. a+ = 40 A).

which is somewhat steeper than for MRN grains (but this is an arguable minor point). as demonstrated extensively in sections 8.8. sizes and UV properties of vsg and PAHs but it would then lose some of its simplicity. for them. and not amorphous. so optical constants of graphite are appropriate.11) .5 and 8. besides speeding up the computations. we put. fH/C . for NC > 30. we consider only two: small ones with NC = 40 carbon atoms. makes it easier to elucidate the influence of the PAH size on the spectrum. We adopt. and big ones with NC = 400. which is the ratio of hydrogen-to-carbon atoms. Although there must be a plethora of different species. • may be regarded as the small-size extension of the big particles. Finally. Such a restriction. The interstellar extinction curve: observations (dots) and model (curves). The fit could be substantially improved by modifying the abundance. The temperature of the PAHs also fluctuates violently and has to be calculated according to the recipes of section 8. PAHs.8/ NC . The main reason why one has to include vsg is their timevariable temperature.6. In estimating PAH properties. we widely follow [Sch93]. we add PAHs to the dust mixture.9.An overall dust model 393 Figure 12. f H/C = 2. The extinction coefficient of our dust model is displayed in figure 12. although the concept of a continuous solid medium may no longer be fully valid. Optical cross sections are still calculated from Mie theory. (12.5. and assume that the carbon component is graphitic. For the hydrogenation parameter. a size spectrum n(a) ∝ a −4 .

03 µm. f H/C is thus proportional to the number of carbon atoms at the periphery of the PAH.14) for NC = 400 (40). where the H atoms are found. The formula gives f H/C = 0. so C PAH does not depend on frequency and is proportional to the number of carbon atoms.4)). C PAH . The MRN curve is more or less identical to the broken curve of figure 12. we disregard differences between neutral ond ionized PAH species.394 PAHs and spectral features of dust Figure 12. although this is a severe simplification. there the dust consists of pure (unmixed) silicate and amorphous carbon grains with a− = 0. the upper limit a+ is indicated.44 (0. For the absorption cross section of a single PAH. except in the far UV due to slight differences in the vsg abundance.10. The line labeled MRN shows the diffuse medium for reference. there is probably a large scatter in the hydrogenation parameter due to variations in the radiation field. ice and vacuum. we put C PAH = NC × 10−17 cm2 . which is usually in the UV (see (12. In reality.3 µm. YC a certain kind of PAHs. Mass extinction coefficient per gram of interstellar matter (IM) as a function of wavelength for composite grains which are fluffy aggregates of silicate. The cross sections in the resonances are . over the total number NC .8.02 µm and a+ ∼ 0. Ionized PAHs can be excited by less energetic photons and have greater infrared absorption coefficients than neutral ones [Li01]. PAH Each of the two PAH types has an abundance YC of 3–6% relative to PAH gives the mass fraction of all solid carbon that is in the big grains. irrespective of their geometrical arrangement. carbon. At wavelengths shorter than λcut . The lower limit a− of the size distribution is in all curves 0.

5. The optical constants of silicate and amorphous carbon are from figure 7. the lines with a+ = 30 µm or 3 mm might be appropriate. we compute the mass extinction coefficient of dust that consists of such icy and porous grains. The absorption coefficient of PAHs per gram of interstellar matter is displayed in figure 12. Average optical constants are calculated from the Bruggeman mixing rule (section 2. amorphous carbon (aC). vsg. Consequently. their size increases and the dust mass doubles (Rd = 1 : 70). All grains are fluffy and composed of four components: silicate (Si).6. Si plus aC. Nevertheless. The vacuum fraction is always constant. the reproduction of the interstellar extinction curve is not bad. 12. Silicate has twice the volume of carbon. 12. the ice mass is equal to the mass of the refractory components. However.2 Extinction coefficient in the diffuse medium Figures 12. The effect of all three processes (fluffiness. f Si / f aC = 2.4 and 2. When the grains are very big. these cited abundances for all three dust components (big grains. for instance. frosting and grain growth) on the particle cross sections is investigated in section 2. There grains are modified with respect to the diffuse interstellar medium by coagulation and by ice mantles. the vsg are irrelevant.3 Extinction coefficient in protostellar cores We extend our dust model to cold protostellar clouds. Assuming that helium accounts for 28% of the gas mass.10. f vac = 0.6. one expects a continuum (see the smooth frequency distribution in the modes of even a small PAH in figure 12.6). as in stellar disks. possibly even triples (see discussion on icy and fluffy grains in section 2.1 µm.5. Without UV radiation. In all grains.9 further quantify our dust model. PAHs) imply a total dust-to-gas mass ratio in the diffuse medium of Rd = 1 : 140. the upper limit a+ is varied. The PAHs have disappeared and are dirty spots in the ice mantles. . In the far infrared. Several ice resonances are visible.19. The particles are spheres and have a size distribution n(a) ∝ a −3. The deficit in the far UV is irrelevant for all models of infrared sources where the observer does not receive far UV radiation. the one at 3.5 ˚ with lower limit a− = 0.6).03 µm = 300 A. ice and vacuum.1.8 and 12.7.5 times bigger than the volume of Si plus aC.3 µm is an educated guess for dust in protostellar clouds before it is warmed up by the star and the ices evaporate. this continuum is usually swamped by the emission from big grains and not observable. for ice from the tables of [Ber69] and [Leg83].4.An overall dust model 395 taken from table 12. four for the big Si grains and two for the very small grains. The curve marked a+ = 0. at λ > 25 µm. In figure 12. whereas the ice volume is 2.4.2) and Cλ should fall off like λ−m with m between one and two. We use only four particles sizes for the big aC grains.

most photons created within it do not reach us on a straight path but are scattered.1) Iν (x. To extract from the observational data the full information on the physical conditions in the star-forming region. 13. for a homogeneous medium. when one observes a star-forming region.1): Consider a surface element dσ at location x with normal unit vector n. both of which can. within the frequency interval ν . from the solid angle d in the direction e. estimates on temperature. . (13.Chapter 13 Radiative transport This chapter deals with the transfer of radiation in a dusty medium.1. absorbed and re-emitted before they are collected in the telescope. . It receives. density and masses. ν + dν. one has to understand how the photons propagate within the source. discuss the general transfer equation and sketch how to derive.1 Radiative intensity and flux The intensity Iν (x. e) has the dimension erg s−1 cm−2 Hz−1 ster−1 and is a function of 396 . Finally. It is defined in the following way (see figure 13. under minor assumptions. For example. We present a set of equations for spherical dust clouds that treats radiative cooling and heating in an energetically consistent way and outline how to solve it numerically. e) cos θ dσ d dν. the power d W = Iν (x. We summarize the elements of radiative transfer. e) is the fundamental parameter describing the radiation field. we also discuss the radiative transport in molecular lines because line observations are always complementary to dust observations when investigating stars in their making. which is inclined by an angle θ against n.1 Basic transfer relations 13. be handled in one dimension. We append solution strategies of radiative transfer for two other astronomically relevant geometries: dusty accretion disks and a dust-filled star clusters.

• • • place x. say.Basic transfer relations 397 dΩ n e θ dσ Figure 13. Let Iν (θ. The direction e towards it can be expressed by two angles. Jν = 1 4π Iν d = 1 4π π 2π dθ 0 0 4π dφ Iν (θ. The total flux which we receive on the Earth is given by obs Fν = S Iν (θ. In practice. If the intensity is constant over the source. direction e and frequency ν.4) . (13. the integration may be performed by mapping the source with a telescope of high spatial resolution.1. φ) sin θ. Imagine an extended astronomical source in the sky of solid angle S .2) where d = sin θ dθ dφ is the element of solid angle. (13. Defining the radiative intensity Iν . φ) can be taken out of the integral and then obs Fν = Iν S.3) Note that Iν is independent of the distance D of the source because the solid angle obs are both proportional to D −2 . Iν (θ. we receive less light but the intensity in the direction of the Sun stays the same. If we recede in a space ship S and the flux Fν from the Sun. φ) be the intensity distribution over the source. φ) d (13. θ and φ. • The mean intensity Jν is defined as the average of Iν over the total solid angle 4π.

1) stands for ‘true’ emission only and is thus different.6) as d Iν = −Iν + Sν . the intensity Iν decreases ext ext by an an amount K ν Iν and increases by ν . An object of low optical depth (τ 1) is transparent and radiation reaches us from any internal point. An object of large optical depth (τ 1) is opaque and we can only look at its surface and do not receive photons from its interior. The source function is thus the limiting value of the intensity for high optical thickness. φ) cos θ sin θ. τν . (13.1.1. When we introduce the source function Sν = we can rewrite (13. its value is given by the integral τν = so along the line of sight S2 S1 ext K ν ds.1 The source function When a light ray propagates in the direction e along an axis s.9) ν ext Kν (13. its intensity Iν changes.5) 13.7) ext dτν = K ν ds.2 The transfer equation and its formal solution 13. When light travels from point S1 to point S2 .1): Fν = π 2π dθ 0 0 dφ Iν (θ. respectively. (13. Note that the coefficient ν defined in (8. It is weakened by extinction and reinforced by emission and by light that is scattered from other directions into the direction e. We see from (13. . and their dimensions are cm A fundamental quantity in light propagation is the optical depth or optical thickness. and ν comprises emission and scattering. This is written in the following way: d Iν ext = −K ν Iν + ν . It depends on frequency.6) ds ext K ν is the coefficient for extinction. Both K ν and ν refer to unit volume −1 and erg s−1 cm−3 Hz−1 ster−1 .398 • Radiative transport The net flux Fν through a unit surface follows from (13.2.6) that per unit length. dτν (13.8) This equation tells us that the intensity increases with optical depth as long as it is smaller than the source function (Iν < Sν ) and decreases when it is stronger. (13.

6) and (13. An example is starlight weakened on the way to us by an intervening dust cloud. it may be taken out from under the integral yielding (13. s 0 K ν (s ) ds (13. and for constant emission coefficient ν .14) is only a formal solution to the transfer equation (13. This situation applies quite well to submillimeter radiation from an interstellar dust cloud. 13. The obvious solutions to (13.10) where Jν is the mean intensity of the radiation field after (13. The hard part is to find the source function Sν (τ ) in formula (13. . ext Pure emission without extinction (K ν = 0). (13. which is the sum of absorption and scattering. If Sν (τ ) is constant. we have not.11) (13. and increases because the grains radiate thermally and scatter light. the source function becomes Sν = abs sca K ν Bν (T ) + K ν Jν ext Kν (13. the observed intensity grows linearly with the path length s. in reality.2 Analytical solutions to the transfer equation We solve equation (13. Emission plus absorption. In this most general case. Equation (13.6) for three scenarios: • Pure extinction without emission ( ν = 0). the intensity of a light ray decreases because of extinction.14) The index for the frequency has been dropped for convenience of notation.Basic transfer relations 399 In a dusty medium. If scattering is isotropic and the grains are at temperature T .15) Iν (τ ) = Iν (0)e−τ + Sν (τ ) · [1 − e−τ ].2. Although it appears as if we had mastered the radiative transfer problem.12) • Iν (0) is the intensity of the background star if it were not attenuated.4).9). we multiply (13. (13.14). The observed intensity diminishes exponentially with optical depth.1.9) are: Iν (s) = Iν (0) exp − Iν (τ ) = Iν (0) · e−τν . now Iν (s) = Iν (0) + s 0 ν (s ) ds .9) by eτ and integrate by parts which leads to I (τ ) = I (0)e−τ + 0 τ S(τ )e−(τ −τ ) dτ .13) • Without a background source (Iν (0) = 0). which is the standard scenario.

Sν = Bν (T ). .15). out of a solid angle . Although Tb is not a direct physical quantity.2.3 The brightness temperature Observations at wavelengths greater than 100 µm fall in the domain of radioastronomy.19) Bν (T ) is the Planck function. However. Bν (T ) They imply that without a background source and if the cloud is translucent (τ 1). of course. the intensity is directly proportional to the optical depth: doubling the column density. 13. and thus its mass. The limiting solutions to (13.3 A dust cloud of uniform temperature In a dust cloud. the only exception is a blackbody for which Tb is independent of ν.17) yields an estimate of the source temperature and its optical depth.18) Iν (τ ) = for τν 1. one. At these wavelengths. according to (13. observes. Towards a uniform cloud of optical depth τν . the flux Fν (τ ) = Fν (0)e−τν + Bν (T ) · [1 − e−τν ] . (13. equation (13.16) are: Bν (T )τν + Iν (0) for τν 1 (13.1. Because Tb expresses an intensity. instead of weird numbers in awkward units. emission occurs only in the infrared.1. on wavelength. all information about the grains is wiped out and the object appears as a blackbody of temperature T . There it is common to replace the intensity Iν by the brightness temperature Tb which is defined through Iν = Bν (Tb ). the intensity Iν (τ ) = Iν (0)e−τν + Bν (T ) · [1 − e−τν ]. therefore. it depends.16) If the background is a point source of flux Fν (0). the source function is then equal to the Planck function at that temperature. if the source is opaque (τ 1). its use offers practical advantages: • One now has for the intensity handy numbers in a simple unit (Kelvin).400 Radiative transport 13. grain scattering is practically absent. For dust at temperature T . doubles the strength of the received signal. one receives. (13. In this case. the intensity Iν carries information about the amount of dust but also about the spectral behavior of the grains via the frequency dependence of the optical depth. (13.17) When the background is absent or its spectrum known and multi-frequency observations are available.

in the Rayleigh–Jeans approximation. Likewise one defines the solid angle of the main beam. The total solid angle of the antenna is defined by A = 4π Pn (ϑφ) d (13. therefore. indicates its physical (thermodynamic equilibrium) temperature. in directions close to the normal of the antenna surface. especially in the case of line radiation. 0). Fortunately. For the moment.5. where the integral extends only over the main beam. T is the grain temperature Td . Tb generally. 13. In a one-dimensional analog. Now we return to the common occupation of astronomers: they do not send messages into space but receive radiation from there.1. but not always. the situation can be compared to the diffraction pattern on a screen behind an illuminated slit: the intensity has a central maximum and the range around it down to the first zero points on both sides corresponds to the main beam of a telescope. one may. as in a radar antenna. (13. φ) denote the flux emitted by the antenna in the direction of the sky coordinates (ϑ. T is the excitation temperature Tex (see (13. In the direction (ϑ0 . Tb = Iν c2 /2kν 2 . the spatial distribution of the electromagnetic .2)). most of the power is emitted into the main beam. we replace the detector at the focus by an emitting device. which is parallel to the normal of the antenna surface. Note that. mb . in the case of line radiation. φ0 ). φ) = P(ϑ. and. The antenna pattern is the result of diffraction phenomena. always use sky coordinates such that (ϑ0 . When we put in equation (13. The normalized function Pn (ϑ. some radiation is spilled into the side lobes.66)). φ) attains its maximum value Pmax . To understand it. mb < A . φ) Pmax is called the antenna pattern and determines the telescope beam. P(ϑ. At low frequencies (hν/kTb 1). φ).Basic transfer relations • • 401 When the source is optically thick. the brightness temperature of a source of optical thickness τ becomes Tb (τ ) = T0 e−τ + T (τ ) · [1 − e−τ ]. φ0 ) = (0. is the main-beam-brightness temperature Tmb .5. we have to briefly discuss the standard measuring apparatus in radioastronomy. according to the reciprocity theorem of section 3.20) In the case of dust. of course.4 The main-beam-brightness temperature of a telescope The quantity in which radioastronomers express the strength of a signal.15) Sν (τ ) = Bν (T ). the brightness temperature Tb is directly proportional to the observed intensity. however.21) with d = sin ϑ dϑ dφ (see (2. Let P(ϑ. It consists of a parabolic antenna with a detector at its focus.

equally fathomless in its subtleties. therefore. If. either because the telescope is flying on a satellite or. φ) has been eliminated. astronomers do not worry about this and pretend to have all ηs under control. φ)P(ϑ. In practice. Tmb . Bobs = Bν (Tmb ) or to its Rayleigh–Jeans approximation. any ambitious calibration attempt. denoted TA . φ) d mb P(ϑ. requires the introduction of an efficiency ηkitch . When the cloud is heated from inside by a star. The main-beam-brightness temperature Tmb and the brightness temperature Tb are related. Tmb = S mb Tb and then the antenna temperature increases when one uses a smaller beam. however the beam is much larger than the source ( A S ). φ) be the intensity distribution on the sky and let us assume that the influence of the atmosphere of the Earth on B(ϑ. Bobs = Tmb 2kν 2 /c2 . φ) d . 10%. It is justified to say that the antenna temperature constitutes the hightech counterpart to stellar magnitudes (section 7. one should be aware that the absolute calibration accuracy at 1 mm wavelength is. 5% and in the submillimeter range. To account for it.3). although the received physical power does not change with mb . When a source of constant brightness temperature Tb subtends a solid angle S much bigger than the solid angle A of the antenna. at best. if it stands on the ground. It is often necessary to state explicitly which conversion is used. as they should be when names are so similar. The observational data consist either of maps or of single point measurements towards different positions on the source obtained with different spatial resolution. even from the kitchen of the observatory. on top of a dozen other ηs. The major reason why TA evades a precise determination lies in the fact that the telescope picks up a little bit of radiation from everywhere. Nevertheless. although historically the definition of TA was somewhat different. It is converted into the main beam brightness temperature. The observed brightness Bobs is then the following mean over the main beam: Bobs = mb B(ϑ. Let B(ϑ.402 Radiative transport field does not change when one switches from emission to reception. by setting it equal to the Planck function. 13. . one measures Tmb = Tb . at best. The main-beam-brightness temperature Tmb is also called the antenna temperature.2 Spherical clouds The dust emission of a cloud is observed at various wavelengths with the goal of deriving its internal structure. by way of calibration.

(13. the dust temperature T (r ). the basic quantity.5). for the mean intensity and the flux: Jν = 1 2 1 −1 1 −1 Iν (µ) dµ Iν (µ)µ dµ. where θ is the inclination angle to the radial vector. However.17)) are no longer acceptable. also depends on direction θ . It is a simple exercise to show that for spherical symmetry 1 Note that volume coefficients for absorption. the problem is one-dimensional.22) (13. In practice. Easy-to-model geometrical source configurations are infinite slabs. so altogether I = Iν (r. θ ). the dust density ρ(r ). however. the frequency-dependent intensity I . The structure of the medium then depends only on one spatial coordinate. The second moment reads1 Kν = 1 2 1 −1 Iν (µ)µ2 dµ. Jν and Fν are the zeroth and first moment of the intensity. we obtain. equals 1 and the quantity (4π/c)Kν gives 3 the radiation pressure. Putting µ = cos θ . It may then be better to write a code oneself. We now present the equations and sketch the numerical strategy for such an endeavour in spherical symmetry. .1 Moment equations for spheres In a spherical cloud. The radiative transfer may be solved in many ways and fast and efficient public computer codes exist. Fν = 0.2. even at the expense of less professionality. the mean radiative intensity Jν (r ) and the flux Fν (r ) are only functions of distance r to the cloud center.24) In the case of an isotropic radiation field. one may encounter difficulties using a public code because it usually has to be modified to meet the demands of a specific astronomical problem. 13. infinite cylinders and spheres.Spherical clouds 403 it is obviously inhomogeneous and estimates of the dust column density and temperature based on the assumption of a uniform cloud (see (13. scattering and extinction are denoted by the slightly different symbol K ν .4) and (13. Then the ratio Kν fν = Jν which is called the Eddingon factor. A more refined model is needed. Iν = Jν and the net flux is zero. from (13.23) Fν = 2π (4π/c)Jν is the radiative energy density. (13.

6) becomes µ ∂ Iν 1 − µ2 ∂ I ν ext + = −K ν Iν + ∂r r ∂µ ν. (13.25) because there is no dependence on the angle µ. one needs physical guidance.25) by µ0 = 1 and integrating over all directions gives 1 d 2 ext (r Fν ) = −4π Jν K ν + 4π ν . The introduction of the next higher moment (multiplication of (13. the radiation field is streaming radially outwards and Jν = K ν = Fν /4π. For high optical depth. Mean values for the extinction coefficent K ν in the case of low and high optical depth are found in the following way: • In an optically thick configuration. for low optical depth.28) must be.28) equals 4σ T 3 /π. equivalent to KR Fν dν = K ν Fν dν.25) with µ1 and integrating over 4π yielding ext 3Kν − Jν Fν K ν dKν + =− dr r 4π (13. Instead. . we cannot solve it for Jν (r ) because it contains the new variable Fν (r ). the definition of K R in equation (13. one uses the Rosseland mean K R (T ) defined through 1 ∂ Bν (T ) 1 K ν ∂T dν = . However.2 Frequency averages One can radically simplify the numerical problem by taking frequency averages.80). For instance. to be able to use formulae that do not depend explicitly on the direction angle µ.28) ∂ Bν (T ) K R (T ) dν ∂T where σ is the radiation constant of (5. If Fν is the energy flux at frequency ν per unit area and time.27) but the problem of having one more unknown than equations remains because now Kν comes into play. The Rosseland mean gives the greatest weight to the lowest values of K ν . 3 13. As Bν (T ) dν = the denominator in (13. σ T 4 /π. and indeed is.404 Radiative transport equation (13. as in a room darkened by thick curtains: it is not their thickness but the open slits between them that determine how much light enters from outside. r 2 dr (13. One could try to find an additional equation by multiplying (13.25) Multiplying (13. although not evident at first glance. the radiation field is isotropic and f ν = 1 .2. either.25) by µ2 ) will not remedy the situation.26) This equation looks much simpler than (13. where the mean intensity Jν is close to the Planck function Bν (T ). (13.

one gets a quadratic increase with temperature. 13. the equation for the radiative energy transport in the stellar interior may be approximated by using the average opacity K R and assuming Jν = Bν (T ) and f ν = 1 . The coordinates p and z. K P (T ) = K ν Bν (T ) dν .2.08 × 1022 K 0 T 2 . A dust-to-gas ratio Rd = 0. Light is propagating at constant impact parameter p parallel to the z-axis either to − + the left (Iν ) or to the right (Iν ).14 × 1021 K 0 T 2 K R = 1. which is fine for interstellar dust (section 8. In a tenuous cloud. (13.27).01 and a matter density ρgr = 2. 3K R (T ) dr (13. one may compute the coefficients analytically.2). Bν (T ) dν (13. the Planck mean K P .10).29) corresponds to Fick’s first law of diffusion. is the radiation pressure. When we put for the absorption efficiency Q ν = 10−23 aν 2 after (8. equation (13. one takes. and let K R refer to 1 g of interstellar matter.31) The two values are astonishingly close.2. the Planck mean K P may be used to assess the effective optical depth of clumps in a molecular cloud. which are used instead of r and µ. from formula 3 (13. Our choice of coordinates is illustrated in figure 13. where a is the grain radius in cm.32) The Rosseland mean opacity may be acceptable in protostellar cocoons or accretion disks.29) • We point out that because Pr = aT 4 /3. K 0 = 7.84). are connected to the radius by r 2 = p2 + z 2 (13. (13.5 × 10−24 R/ρgr . applied to a photon gas. K P = 8.30) Using the approximation K ν = K 0 ν m .3 Differential equations for the intensity We give here a set of equations for the intensity that retains the full information on frequency and direction.Spherical clouds 405 For example.5 g cm−3 leads to K R ∼ 3 × 10−4 T 2 cm2 per gram of interstellar matter. as a frequency average. they all do.6). determined by the temperature gradient in the star: F =− 16σ T 3 d T .33) . It is now not a question of which frequency penetrates deepest. with a from (5. it may be better under such conditions to use the enhanced estimate K P ∼ 10−3 T 2 cm2 per gram of IM. Then the total flux F = Fν dν is. As grains in early-stage protostellar condensations are probably fluffy and possess ice mantles (section 2. For m = 2. Instead one has to take the average over the emitted energy.

34) (13.35) If the dust particles scatter isotropically. K ν . z ∗ ) − Iν ( p. The coordinate system for the intensity in a spherical cloud of outer radius R. z ∗ ) = F∗ν π with z ∗ = 2 R∗ − p 2 .e.9)) + d Iν ext + = −K ν · [Iν − Sν ] dz − d Iν ext − = K ν · [Iν − Sν ]. z) With a star of stellar surface flux F∗ν and when p ≤ R∗ (see (5.2.406 Radiative transport p R I− I+ r ze z Figure 13. + − at z = 0.36) • The formulation of the second boundary condition depends on whether there is a central star or not. (13. and the intensities obey the differential equations (see (13. The ray I + enters the cloud at ( p. In view of the symmetry of the spherical cloud.38) .34) and (13.2. for z > 0. scattering and extinction referring to 1 cm3 of interstellar space. The cloud is immersed into a radiation field of isotropic intensity I0 . for reasons of symmetry.74)). (13.35) have to be solved only in the right-hand quadrant of figure 13. the source function is given by abs sca ext (13. z e ) = I0ν with z e = R 2 − p2 .10). z e ). Light rays at constant impact parameter p are directed either to the left (I − ) or to the right (I + ). + − Iν ( p. (13. where z = z e . i.37) Iν ( p. z) = Iν ( p. (13. the incoming intensity − Iν is equal to the intensity of the external radiation field I0ν : − Iν ( p. one has. Without it or when there is a star but the impact parameter p is larger than the stellar radius R∗ . dz (13. −z e ) and leaves at ( p. K ν and K ν are the volume coefficients for absorption. The equations must be complemented by two boundary conditions: • The first states that at the cloud’s edge.

39). z) 1 + − 2 Iν ( p.40). the intensity Iν from (13.42) From the first. Jν (r ) and Sν (r ) in the following way: First.5 Practical hints • As starting values for the first cycle (i = 1).Spherical clouds 13. z) .10) the source function Sν (r ). Updating the source function Sν (r ) (see (13. Solving (13. z) = − 1 + 2 Iν ( p.39) (13. For the temperature. z) − Iν ( p.4 Integral equations for the intensity 407 Instead of solving the differential equations (13. the optical depth τ increases from left to right. advisable to use the corresponding integral equations (see (13.34) and (13. From Iν and Iν we form the quantities u ν ( p. and thus after (13. Using the inner boundary condition. one still has to determine the dust temperature. r ) r 2 − p2 d p. the mean intensity of the radiation field Jν (r ).35).8)) finishes iteration cycle i + 1. one gets an updated mean intensity Jν (r ) = 1 r r 0 pu ν ( p. we then calculate. z) + Iν ( p. In (13. so τ (z = 0) = 0 and τ (z > 0) > 0. The boundary conditions stay the same. + + − for all p.39).39) and (13. the optical depth t equals zero at the cloud surface (t (z e ) = 0) and grows from right to left. In (13. z) = vν ( p. (13. we determine for each impact parameter p the − intensity Iν from (13.40) I − (t) = e−t I0 + 0 t S(x)e x d x . .2.40).14)) I + (τ ) = e−τ I + (0) + 0 τ S(x)e x d x (13. u ν . The new Jν (r ) is then inserted into equation (8. one may put T (r ) = constant if there is no central star and the outer radiation field is not too hard and strong.41) (13. an initial mean intensity Jν (r ) = 0 should be acceptable. One arrives at a full solution to the radiative transfer problem iteratively. We find improved values of T (r ). so t (z < z e ) > 0. it is. (13.8) or into Jν (r )K ν (r ) dν = Bν (T (r ))K ν (r ) dν which yields new temperatures when solved for T . 13. for reasons of numerical stability.2.22).43) It is not difficult to see that this agrees with the definition of Jν from from (13. Suppose that after iteration cycle i we approximately know the grain temperature T (r ).40) is not eneough.

heating comes from an outer isotropic radiation field of intensity I0ν . as a rule of thumb. (13. At each radius r . it is proportional to h 4 and that is worse when the grid is coarse.44) This formula can be derived from (13.23) and (13. as one recedes from the star. T (r ) ∼ 20(L ∗ /r 2 )1/6 should work when r and L ∗ are expressed in cgs units (see (8. one can compute. there therefore exists a cutoff frequency with the property that only photons with ν < νcut are energetically relevant.2). whereas those with ν > νcut . The accuracy is then proportional to the square of the step size h. the condition L(r ) = L ∗ should be numerically fulfilled everywhere to an accuracy of ∼1%. the dust may be quite hot. the radiation field softens and the mean free path of the photons increases. The integrated flux through a shell of radius r is L(r ) = 4πr 2 Fν dν. In Simpson’s rule (quadratic interpolation). the net flux Fν (r ) = 4π r2 r 0 • • p vν ( p. especially in the outer envelope. there are only far infrared photons and the grid may be coarse. To select a sufficiently fine radial grid which is also economical in terms of computing time. Nevertheless. however. shorter.408 Radiative transport With a star. . For example. for each radius r . The iterations converge when the changes of the temperature T (r ) become smaller with each new cycle. Use linear interpolation (trapezium rule) for evaluating integrals. Numerically. the number of points and their spacings must be adapted during the iterations. The integrated flux L(r ) is then zero because what enters at the cloud surface must also leave. As a check on the accuracy of the calculations. in a cloud of 50 mag of visual extinction with a central hot star. r ) d p. At the cloud edge.42). |L(r )| should now be small compared to 4π R 2 π I0ν dν • which is the total flux from the external radiation field through the cloud surface. the dust column density between neighboring grid points must be small near the star because UV radiation is absorbed very effectively. are unimportant. νcut may be defined by and computed from ∞ νcut Jν (r ) dν = α ∞ 0 Jν (r ) dν (α ∼ 0. Without an embedded source. • If the cloud has a central star of luminosity L ∗ . Generally speaking.17)). whose mean free path is.

T j +1 . . if at all. A disk is passive when it is only heated by the star and internal dissipational losses are negligible.3.8). radius R∗ and temperature T∗ whose mid-point is at (r = 0.3.3 Passive disks 13. in figure 13.1%) bigger than r1 . there are N +1 impact parameters pi = ri with i = 1. . z) given in cylindrical coordinates (r. . on the grain type.8) in the form σ Q abs Jν dν = Q(T ) T 4 ν π where σ is the radiation constant of (5.Passive disks 409 • • • Choosing the radial step size in such a way that the optical depth at the cutoff frequency is of order 0. z = 0). one needs only two impact parameters ( p = 0 and p = R∗ ) for the star. of course. we consider. only in the near infrared. r N = R.1 Radiative transfer in a plane parallel layer All stars are surrounded during their early evolution by an accreting disk. . . ν The sequence T j . For N radial grid points r1 = R∗ . Because to first order the intensity over the stellar disk does not change. a small section at radius r from the star. To determine the disk emission. Very close to the star. The second radial grid point r2 should only be a tiny bit (0. r2 . N and p0 = 0. . . . For heavily obscured sources (AV > 20 mag). quickly converges towards Td . Writing (8. the exact value of the evaporation radius has little influence on the spectrum. one may evaluate the grain temperature Td in the local radiation field and add a flag at all radii to each grain type indicating whether Td is above or below the evaporation threshold Tevap . .80) and Q(T ) is defined by Q(T ) = Q abs Bν (T ) dν ν Bν (T ) dν one gets an improved estimate T j +1 for the new iteration cycle j + 1 from T j4+1 = T j4 Q abs Jν dν/Q(T j ). r3 . . z). T j +2 . . The flag then controls whether the grain contributes in the next cycle to the extinction and emission or not. Let it be illuminated by a central star of luminosity L ∗ . dust evaporates. Imagine a flat and thin disk with density structure ρ(r. The dust temperature Td is computed from (8. To account for dust evaporation. 13.1 automatically ensures that the grid is fine only where necessary. it depends. . The exact radius revap is part of the solution of the radiative transfer. Let T j be the value of the grain temperature in iteration cycle j .

its width is ∗ /2π The first boundary condition of the transfer equations states that at the top of the disk the incoming intensity I − is zero except towards the ring where it is equal to Bν (T∗ ). 2 r In order not to break the slab symmetry. The star is far away to the left at a distance r and its light falls on the disk under some small grazing angle αgr .49) for a flat disk and in (13. The basic transfer equation (13. Then one calculates from these intensities a new vertical temperature run T (z).40) for I + and I − . The second boundary condition. The situation is symmetric with respect to the mid-plane which is indicated by the dotted line at z = 0. Viewed from the disk surface (figure 13.3). . The position of the observer is also shown. follows from symmetry requirements. at large distances. The formalism of the radiative transfer developed for spheres can be carried over to the slab geometry. The starlight falls on the disk under a small effective grazing angle αgr . the procedure is iterated until T (z) converges. This ring encircles the αgr .39) and (13. whole sky at an elevation αgr . Again one starts by solving the pair of equations (13. the stellar hemisphere is replaced by a luminous ring of the same intensity and solid angle. which is evaluated in equation (13.3. which are now the intensities along straight lines under various angles θ with respect to the z-axis. A small portion of a thin disk.9) is solved under all inclination angles θ for incoming (I − ) and outgoing (I + ) radiation. z light fr α gr 0 om sta r I + I − θ Figure 13.410 Radiative transport . I + = I − at z = 0. a solid angle ∗ = 2 1 R∗ π 2.53) for an inflated disk. the upper hemisphere of the star subtends.

1 − eτν (µ) µ dµ the temperature Tmid can be numerically evaluated from the condition f ∗ = 4π Bν (Tmid )E ν dν. a typical value would be AV ∼ 5 mag.47)). heights + at the top of the disk. + (13. The integrated incident flux in the z-direction is then f∗ = f ν dν.45) Iν (z = 0. including the scattered part. one may encounter numerical difficulties. z and angles θ .16)). can leave. µ) = Bν (Tmid ) · 1 − eτν (µ) .3. One can immediately write down the intensity that enters the top layer from below under the direction µ = cos θ . Furthermore. for all directions θ . to become absorbed but small enough so that all re-emitted radiation. (see (13. z. and possibly apply a correction for foreground extinction. θ ) for all frequencies ν.46) . z) and the intensities Iν (r0 . 2π Bν (Tmid ) Defining the last integral as Eν ≡ 1 0 1 0 1 − eτν (µ) µ dµ. θ ). Here τν (µ) is the optical depth of the mid layer.1 Disks of high optical thickness When the opacity in the vertical direction is large. The mid layer is assumed to be isothermal at temperature Tmid . Without scattering. 13. One thus also obtains Iν To obtain the total flux that an observer receives when he views the disk under a certain angle θobs. therefore. The optical depth of the top layer in the z-direction is chosen to be sufficiently large for all starlight.4).Passive disks 411 The radiative transfer at radial distance r0 yields the temperature structure + − T (r0 . all the flux f∗ is absorbed in the top layer and then reemitted. z. (13. We then recommend the following strategy: Divide the disk in its vertical structure into a completely opaque mid layer sandwiched between two thin top layers (figure 13. the other half will leave the top layer into space. Iν (r0 . one has to solve the radiative transfer at all radii r and then sum up the contributions. one has to include the direct emission from the star. which has infrared wavelengths. so that the radiative transfer inside it is very simple (see (13.23)). Half of it will penetrate into the mid layer. The monochromatic flux received by the top layer from below is. as happens in young stellar disks where AV is 104 mag or greater. Let f ν be the monochromatic stellar flux that falls in z-direction on the disk (see (13.1.

a surface element of unit area receives out of a solid angle d pointing towards the star the flux Bν (T∗ ) d sin θ sin φ (figure 13.412 Radiative transport z α gr 0 UV lig ht from IR photon I+ star θ top layer AV = 5mag IR photon I− very opaque isothermal mid layer top layer AV = 5mag Figure 13. Note that the zero point of the z-axis is different here from that in figure 13. About half of the re-emitted IR photons heat the mid layer which is assumed to be isothermal at temperature Tmid .3. The UV light from the star on the left is completely absorbed in the top layer.5). The assumption of isothermality of the optically thick mid layer is valid because a temperature gradient would imply in the diffusion approximation of (13. A small portion of a geometrically thin but optically very thick disk. If the disk is flat and thin or if it has a constant opening angle. Bν (T∗ ) denotes the uniform stellar intensity. However.4.45) for I + . One calculates the radiative transfer for the top layer (z ≥ 0) using the boundary condition (13. the surface element is illuminated by the flux f ν = Bν (T∗ ) θmax 0 sin2 θ dθ 0 π sin φ dφ = Bν (T∗ ) · θmax − 1 2 sin 2θmax (13.3.47) = Bν (T∗ ) arcsin(R∗ /r ) − (R∗ /r ) 1 − (R∗ /r )2 . By the whole upper half of the star. 13.1. . there can be none because there are no internal sources and the disk is symmetric about its mid plane.29) a net energy flux.2 The grazing angle in a flat disk We compute the monochromatic flux f ν that falls in the z-direction on the disk and heats it.

18) and (A. f ν = 2 x 3 Bν (T∗ ) and 3 αgr = 4 R∗ .5.Passive disks 413 R* S θmax θ r φ disk M P star Figure 13.48) 1). The factor sin θ sin φ takes into account the inclination of the disk towards the incoming light. where θ is the angle between the lines P M and P S. αgr . the surface element S subtends the solid angle d = sin θ dθ dφ.47) over the entire plane outside the star. To evaluate the . Fν = ∞ R∗ 2πr f ν dr ∞ R∗  = 2π Bν (T∗ )  2 R∗ R∗ 2 r arcsin − R∗ 1 − 2  dr = 1 π 2 R∗ Bν (T∗ ). At large distances (x ≡ R∗ /r the arcsin x − x 1 − x 2 2x 3 /3 and. In the figure. a unit area is illuminated by an infinitesimal surface element S of the star by the flux Bν (T∗ ) d sin θ sin φ. At position P on the disk.49) To obtain the total flux Fν falling on an infinite disk. therefore. 2 r r The integrals are solved using formulae (A. we integrate (13. We express f ν through an effective grazing angle. 3π r (13. Seen from P.19). where ∗ is√ solid angle of the star. S is drawn near φ = 90◦ and the dotted line near S is a circle with θ = constant. defined by fν = ∗ · Bν (T∗ ) · αgr (13.

Tdisk(r ) = T∗ 2 3π 1/4 R∗ r 3/4 . we define the height H by the condition that the visual optical depth AV from there towards the star is one. To derive the grazing angle αgr for this general case. the effective grazing angle αgr of the incident radiation increases over the expression (13. π (13.49) by an amount β − β = arctan H dH − arctan dr r d 4 R∗ +r 3π r dr dH H d − =r dr r dr H r H r so that now we have to use in (13. (13. one finds that each side of an infinite disk intercepts exactly one-eighth of the stellar luminosity L ∗ = 2 4 4π R∗ σ T∗ . it will have a surface temperature 4 Tdisk(r ) = 4 T∗ arcsin(R∗ /r ) − (R∗ /r ) 1 − (R∗ /r )2 . so it is desirable to have a good estimate for H and its derivative d H /dr .51) 13. has the form αgr 4 4 T . The surface temperature of a blackbody disk.52) The level z = H represents the photosphere of the dusty disk and it is the approximate place where stellar photons are absorbed. the disk height H increases with the distance r from the star (subsection 15.48) αgr = . Assuming that the gas and dust are well mixed. the radiative heating rate has a r −3/4 -dependence.50) For r R∗ . A flared disk intercepts more stellar light than one with constant H /r . the disk is said to be inflated or flared. If the disk is a blackbody. (13. when the gas is in hydrostatic equilibrium in the z-direction. Summing Fν up over all frequencies.53) is usually greater than the first.414 Radiative transport expressions at the upper bound r = ∞.3. (13. the dust disk is then not flat nor is the ratio H /r constant. we set x = 1/r . expand around x = 0 and let x go to zero.4.53) The second term in (13. (13.3). whether it is inflated or not.2 The grazing angle in an inflated disk In a Kepler disk. If ρ(z) denotes the dust density distribution in vertical direction and K V the visual absorption coeffcient. 1= KV αgr ∞ H ρ(z) dz. The disk thickness is 2H . Indeed.54) Tdisk(r ) = ∗ π ∗ .

Its origin may be either rapid star formation or accretion onto a massive black hole.Galactic nuclei 415 z P β β . the configuration is intrinsically threedimensional because the dust temperature varies not only on a large scale with galactic radius r . when the luminosity comes from a star cluster. the hard primary radiation. When the luminosity of the black hole dominates. Let us assume that the hot spots are spheres. so they are distributed over the galactic nucleus like raisins in a . The small shaded sphere represents the star. H r Figure 13. which is the distance to the center of the stellar cluster but also locally with the separation to the nearest star. absorbs only the mean interstellar radiation field Jν (r ) and will thus be cooler. the temperature of a grain located in the immediate vicinity of a star is determined by the distance to and the properties of that star. whereas a grain at the same galactic distance but not ISRF very close to any star. 13. The spherical symmetry is broken on a small scale because the surroundings of a star constitute a hot spot (abbreviated HS) in the interstellar medium. In both cases. like the center of the Milky Way. The bold line marks the top of the flared disk atmosphere. one may. It turns out that their total volume fraction is small. This circumstance must be taken into account in some approximate manner. Their presence significantly changes the spectral appearance of a galactic nucleus.1 Hot spots in a spherical stellar cluster Another class of objects whose infrared appearance one is eager to model and which possess a fair degree of spherical symmetry are galactic nuclei.6. to first order. is converted into infrared photons by dust. For a surface element P on a flared disk. r and H are the radial distance and height above the mid-plane. approximate it by a spherical central source with a power-law emission spectrum and use the formalism outlined in section 13. each with a star of monochromatic luminosity L ν at the center. Many galactic nuclei have large infrared luminosities as a result of internal activity. However. from the star or the black hole.4 Galactic nuclei 13. Loosely speaking. the effective grazing angle αgr of stellar light is enhanced with respect to the flat disk by β − β .2 for single stars.4.

respectively. however. The corresponding values are denoted n OB (r ). a dust grain is mainly heated by the interstellar radiation field (ISRF). a space density n ∗ (r ) and a monochromatic and bolometric luminosity L ∗ and L ∗ .4. of course. The radius R HS of a hot spot is determined by the condition that outside. also on the monochromatic stellar luminosity L ν . heating by the star dominates. altogether typically 109 . ν • If the stellar atmospheres are blackbodies of temperature T ∗ .2 Low and high luminosity stars It is useful to divide the stars in a galactic nucleus into two categories. (13. consider a nucleus of integrated luminosity L nuc containing N identical stars of luminosity L ∗ . Stars of the first class have small or moderate luminosity. If τν is the optical depth from the star to the boundary of the hot spot. ν They represent the population of old stars. To see why. where ν L∗ = L ∗ dν. independent of N.55) This equation can readily be generalized to a mixture of dust grains. The radius R HS of a hot spot is a function of galactic radius r . the intensity of the interstellar radiation field in the galactic nucleus is. one only has to introduce in the numerator of the source function (13. Therefore. are very bright and not so numerous. There are rarely more than 106 in a galactic nucleus and their space density n OB (r ) . ν • The low luminosity stars are very numerous. R HS follows from ISRF Q ν Jν dν = 1 (4π R HS )2 Q ν L ν e−τν dν.56) the volume emission coefficient ∗ ν (r ) = n ∗ (r ) · L ∗ . They are young and luminous and were formed in a starburst. only the product ν ν ∗ ∗ ∗ ν (r ) = n (r ) · L ν . When we fix L nuc = N L ∗ but increase N and thus lower L ∗ . Stars of the second class are of spectral type O and B. L OB and L OB .416 Radiative transport cake. A large population of low-luminosity stars may be smeared out smoothly over the galactic nucleus and the structure of their hot spots need not be evaluated. In view of the definition √ for R HS in (13. whereas inside. the dust distribution within the hot spot.55). To account for the radiation of these stars. the total volume N(R HS )3 of all hot spots decreases as N −1/2 . The OB stars. but it depends. one finds that R HS ∝ L ∗ ∝ N −1/2 . each containing identical objects. and on the grain type via the absorption efficiency Q ν . Note that L ∗ need not be specified. one may put L ∗ ∝ Bν (T ∗ ). to first order. 13. and the contribution of their hot spots to the overall spectrum may be neglected.

40)) I (τ ) = I (0)e−τ + 0 τ S(x)e−(τ −x) d x (13. . the source function in a spherical galactic nucleus with isotropic dust scattering becomes (see (13. The hot spot is illuminated at its edge by an interstellar radiation field (ISRF).56) It is straightforward to generalize to a dust mixture with different temperatures Td and to include small grains with temperature fluctuations. γ (r ) = 4π OB n (r ) · R HS (r ) 3 3 is small. The inner boundary condition is now always + = I − and. The emission of their hot spots is not negligible and has to be evaluated explicitly.36). To obtain L HS (r ). I + and I − . which fixes the outer boundary condition (13. This is done along lines of constant impact parameter for the intensities in positive and negative direction. They ISRF yield the mean intensity Jν (r ). The cloud radius R HS is determined from (13.57) is not strictly correct. ν Altogether. a good approximation as long as the volume fraction γ of the hot spots. at the inner boundary. equation (13. γ (r ) shrinks when n OB (r ) becomes large. One then solves with this source function the integral equations (see (13. Before solving the radiative transfer on a large scale in the galactic nucleus. γ is of order 10−3.57) for the spherical nucleus as a whole. For typical space densities of OB stars in starburst nuclei. equal to the luminosity L OB = ν L OB dν of a single OB star but L OB and L HS are different because much ν ν ν of the hard stellar UV flux is converted by dust into infrared radiation.10)) Sν = HS ν (r ) + ∗ ν (r ) + abs sca ISRF K ν Bν (Td ) + K ν Jν . I − = 0 (see (13.39) and (13. however. The ν frequency integral L HS dν is. at the outer edge. typically one star per pc3 or less in the starburst region. I As the galactic nucleus in this model consists of two phases.37)). we calculate for each galactic radial grid point the ν radiative transfer of a spherical cloud centrally heated by an OB star.36) and (13. It gives. ext Kν (13. a dusty medium interspersed with hot spots. of course.38) with the flux from the surface of the OB star.Galactic nuclei 417 is moderate.55). we use (13. The volume emission coefficient due to the hot spots is HS ν (r ) = n OB (r ) · L HS . one has to determine the luminosity L HS (r ) emerging from a hot spot.

(13. The frequency-dependent absorption coefficient results from the difference between upward and downward induced transitions and can be written as hν (ν). The basic equation of radiative transfer (13.5. n i .5 Line radiation 13. solid angle d and frequencies in the interval [ν. the other of emission.3. The emission and absorption coefficient can be expressed through the Einstein coefficients A j i and B j i for spontaneous and induced emission. line radiation arises from transitions between discrete energy states.59) In our applications. The molecules absorb radiation not only at exactly the center frequency ν j i but over some interval. The center frequency ν j i of the line is at hν j i = E j − E i . Consider the levels j and i with statistical weights g j . and population numbers n j . energies E j > E i . dν + dν] induces. How effectively they absorb away from the line center is described by the absorption profile (ν) (see section 6.1 Absorption coefficient and absorption profile In a gas of atoms or molecules. per unit length. the levels themselves are broadened by collisional perturbations. In other environments. This is a normalized function. the energy levels are rather sharp and the line broadening is due to the Doppler shift arising from the motion of the molecules (relative to their local standard of rest).6). Following the discussion in subsection 6.60) K ν = n i Bi j − n j B j i c .3.418 Radiative transport 13. H2 H π (13. gi . Iν Bi j Ni (ν) d c dν upward transitions.58) which for a Gaussian of half-width 2H has the form √ (ν − νi j )2 ln 2 (ν) = √ · exp − ln 2 . d Iν = −K ν Iν + ds ν is still valid but K ν refers now only to absorption. (ν) dν = 1 line (13.1).1. like stellar atmospheres. a beam of light of intensity Iν . Scattering of the kind that an atom absorbs a photon from level i to j and then returns to the same i is treated as two independent processes: one of absorption.

we see from equation (13. the emission coefficient ν ν (13. Usually one may put (ν) = (ν).65) is equivalent to the Boltzmann equation (5. (13.9). Bν (Tx ) = Sν . 13.63). . The ratio of the level populations defines an excitation temperature Tx . ni gi (13.64) and the profile function has canceled out.65) When we insert Tx into the Planck function.60) and (13. in view of (13.66) Different transitions of the same molecule have usually different excitation temperatures. Sν = ν Kν = 2hν 3 c2 g j ni −1 gi n j −1 (13.2 The excitation temperature of a line When the radiative transfer equation is written as d Iν = −Iν + Sν dτν it describes the variation of the intensity Iν with optical depth τν .5.62) may be expressed as (ν) hν 4π (13.64) that the source function is just the Planck function at temperature Tx .Line radiation 419 The term −n j B j i gives the correction for stimulated downward transitions and (without proof) is linked to the same . and then (13. equal to the kinetic temperature of the gas T . nj g j (E j −Ei )/ kTx = e . The integrated line absorption coefficient is hν K = K ν dν = n i Bi j − n j B j i (13. The source function Sν becomes. whereas the induced photons fly in the same direction as the photons in the incident beam. Likewise.63) = n j A ji with the profile function (ν). Only in thermodynamic equilibrium does one have one value of Tx . When for a particular transition Tx = T .61) c and the optical depth element dτν = K ν ds = K (ν) ds. Note that spontaneous emission is isotropic.

Generally. second.5.20). One uses the so called bi -factors to indicate the deviation of the population of a level i from its LTE value denoted n ∗ . radiative transitions due to spontaneous and induced emission and to induced absorption (section 6. if we replace in equation (13. d Tb = −Tb + Tx dτ then has.420 Radiative transport the corresponding levels are said to be thermalized.61) becomes K = n i K 0 1 − e−hν/ kTx .19) from the intensity. (13.3 Radiative transfer in lines For line radiation. the absorption coefficient K in (13. in a uniform cloud.61) depends on the statistical level population n j .69) ni Thus.71) Iν = Bν (Tb ) the transfer equation (13. Employing the excitation temperature and setting g j c2 K0 = A ji (13. (13. all subject to certain selection rules.68) The brackets contain the correction for stimulated emission. . i ni bi ≡ ∗ .67) gi 8πν 2 the integrated absorption coefficient of (13. (13. there are collisional transitions. the solution Tb = Tx (1 − e−τ ) + T0 e−τ . (13. b j −hν/ kT e = e−hν/ kTx .70) bi Because the brightness temperature Tb is defined by (13. 13. (13.73) (13. first.3). These are.74) (13.9) can be written as d Bν (Tb ) = −Bν (Tb ) + Bν (Tx ). one cannot assume LTE conditions but has to determine the abundance n j of each level j from the balance between processes that populate and depopulate it. dτν The Rayleigh–Jeans approximation.68) the excitation temperature Tx by the kinetic temperature T .72) The term T0 e−τ representing a background source of strength T0 has been added to stress the analogy to (13.

The total number of CO molecules is fixed by the additional condition N n j = n(CO).1).1 Level population of carbon monoxide 421 Let us consider the excitation of rotational levels in a simple diatomic molecule. The set of equations for the equilibrium population n j then reads N n k Akj + Bkj u kj + n i Bi j u i j + n(H2 ) m=0 n m Cm j N = n j A j i + B j i u j i + B j k u j k + n(H2 ) m=0 C jm (13. Because gi C j i = Ci j e−(Ei −E j )/ kT gj we are led to Ci j n i − n j i gi (E j −Ei )/ kT = 0. the main collisional partners being helium atoms and ortho. 1.3. terms connecting to i = j − 1 disappear because level i = −1 does not exist. j =0 (13. . . however. The inverse rate Ckj follows from g j C j k = gk Ckj e−(Ek −E j )/kT . Collisions.75) for j = 0.2. N − 1. 2.Line radiation 13.38).75): • When the density is high. The radiative energy densities u i j in the lines are taken with respect to transitions between levels i and j and are defined in (6. Radiative transitions coupling to rotational level j are then restricted by the selection rules to levels j ± 1. For j = 0. The left-hand side populates. the right-hand side depopulates level j . CS or CO (see section 14.2 LTE population We point out limiting cases of (13.3. for example. can go from j to any k = j and back.76) 13. The collision rate C j k from j → k is a function of temperature only. Let there be N + 1 levels altogether and let us put i = j −1 k = j +1 C j j = 0. e gj . The number of such transitions per s and cm3 is given by n j C j k .5.and para-hydrogen. .5. . radiative transitions are negligible and collisions dominate.

purely thermal and optically thin molecular lines from a cold interstellar cloud of 10 K would have a width of typically 0. 13. we now have to follow the intensity not only in the right quadrant of figure 13. u i j . In the line profile (13. without knowledge of the . T ) B j i = 0 • which also implies thermalization of the levels at temperature T .3. observed lines are usually a few times wider. in spherical symmetry. They must. dominated by other. In the case of a blackbody radiation field. Broadening is. This changes the boundary conditions in an obvious fashion. non-thermal motions. the local line profile (ν) appears in the expression for the absorption and emission coefficient. may be computed locally. therefore. one. therefore. K ν and ν . One solves the radiative transfer equations (13.2. u ν = B(ν. nevertheless.60) and (13.3. therefore.75) and (13. However. opposite signs in the near and far quadrant.39) and (13. and repeats this sequence until convergence is reached. however.5.422 Radiative transport These equations are fulfilled if the brackets vanish which is true when all levels are in LTE. When collisions are negligible. their precise origin being unclear. puts H2 = 2kT 2 + ξturb ln 2 m where m is the molecular mass and ξturb the most likely turbulent velocity. 13. turbulent character. For example.75) and (13. For the radiative transfer we again use the integral form (13.59).4 The Sobolev approximation When the velocity field v in a cloud has a large gradient.76).13)). usually of Gaussian shape which indicates that the motions have a random.40) for all lines and afterwards the abundance equations (13.3 The microturbulent approximation The equations (13.40).5. nj gj 8πh ν j i + n j − ni gi c3 B(ν j i . According to (13.76) for the abundances n j contain the radiative energy densities u i j .63). In the microturbulent approximation it is assumed that the velocity distribution of the non-thermal motions follows a Gauss curve and that the mean free photon path is much bigger than the scale length of turbulence. The observed profiles are. the radiative energy densities in the lines.39) and (13. from z = 0 to z = z e as for dust but from −z e to z e because the line-of-sight velocity has. T ). The function (ν) is rarely of purely thermal origin.1 km s−1 (see (5. be solved together with the radiative transfer.

radius r and direction µ. ds dr r . For a radial inflow.61) and ν0 the frequency at the line center. (13. the mean βe is given by an average over e−τ . two interacting points along the line of sight (figure 16. which is all reasonable. To find the effective β. For a spherical cloud of radius r . where θ is the inclination angle to the radial vector and µ = cos θ .8).75). with s = 0 at point x. Ryb70]. one has to determine τ as a function of frequency ν. consider a point x in a cloud. Because the velocity gradient is large. As there is also emission along the line for s > 0. Let τ be the optical depth from x towards the edge of the cloud along some axis s. photons cannot escape and J equals the source function of (13. If β = 0. βe = 1. If β = 1. The escape probability βe depends on the optical depth τ but not on the radiation field. The mean intensity J (or the radiative energy density u i j ) in the case of large velocity gradients is derived from a photon escape probability β by assuming that J is proportional to the source function S. βe = e−τ = 1 τ τ 0 e−x d x = 1 − e−τ . one must further average βe over all directions e. If τ is small.77) β= 2 −1 τ To evaluate this integral. we get 1 1 1 − e−τ dµ. Here e denotes the unit vector in the direction of the s-axis. however. If K is the integrated line absorption coefficient of (13. J = (1 − β)S. τ If τ is large. This significantly simplifies the procedure. To determine β. Let βe be the corresponding escape probability given by βe = e−τ . Atoms do then not interact with one another radiatively over large distances because their emission is Doppler-shifted by comfortably more than the linewidth. The standard example is the outward flow in a stellar wind. photons leave unimpeded and there are no induced transitions in (13. βe = 1/τ . The velocity gradient along the s-axis is given in spherical symmetry by dv dv v = µ2 + (1 − µ2 ) . all relevant contributions to the optical depth τ come from the immediate vicinity of point x.64).Line radiation 423 radiation field in other parts of the cloud [Sob60. the optical depth becomes τν = K ∞ 0 ν − ν0 + ν0 s dv c ds ds = K c ds ν0 dv ∞ ν−ν0 (x) d x = K c ds ν0 dv and we see that it is independent of the frequency ν. there are.

If it is not.424 Radiative transport This expression must be positive. be found from (see (13. The intensity which one detects along a line sight at a certain frequency ν is only due to the emission from a small region of size . Integration over all lines of sight and all frequencies yields the line profile which one would observe towards such a cloud. .15) and (13. where |dv/ds| is about equal to the local linewidth. and may. therefore.60)): Iν = Sν 1 − e−K ν . one takes |dv/ds| which corresponds to flipping the direction vector of the s-axis.

The flat disk is enveloped by a halo of some 200 globular clusters. 14. one-third is absorbed by dust grains and transferred to infrared radiation (²10 µm). spectral appearance. we give in the remaining chapters an overview of the Milky Way. Together these components constitute the interstellar medium.1 Global parameters The Milky Way is a rotating stellar system of some 1011 stars that has been created about 15 × 109 yr ago.Chapter 14 Diffuse matter in the Milky Way 14. About two-thirds of the starlight leave the Milky Way unimpeded. The center of the Milky Way harbours a huge stellar cluster. the nuclear bulge. Usually. In each of them. the most abundant element: 425 . magnetic fields and photons. like luminosity. The space between the stars is not empty but filled with gas. chemical composition or radiation field. on a time scale of ∼1010 yr. dust grains. It exists in several phases which differ in all basic parameters like temperature.1 Overview of the Milky Way To put interstellar dust into a broader astronomical context. with a diameter of order 2 kpc which contains ∼1% of the total stellar mass of the Milky Way. one classifies the interstellar medium according to the state of hydrogen. gas mass or supernova rate change only gradually. The total power radiated by the Milky Way is estimated to 4 × 1010 L and has its origin in the luminosity of the stars.1. After an initial and violent phase it has settled into a quasistationary state where global parameters. relativistic particles (moving close to the speed of light). the dust is modified by the specific environment. density. It is a flat disk about 30 kpc across and several hundred parsec thick with a global star formation rate of ∼5 M per year. talk about the interstellar medium in general and about star formation in particular because it is there that the study of interstellar dust finds its most important application.

The dust then changes its environment leaving. astronomically fundamental and philosophical. the fact that there is a lineage between us and dust is probably the ultimate reason for our interest in it. because it strongly influences the physical and chemical conditions of the interstellar medium (thermal balance. Together. hydrogen is ionized (H+ ) and forms a hot plasma.4). it is also essential for the dynamics of the gas. When did astronomers realize the relevance of interstellar dust? As a legendary beginning one may count the year 1784 when Friedrich Wilhelm Herschel (the father of John Frederick William) looked at the dust cloud ρ Ophiuchus and exclaimed: Hier ist ein Loch im Himmel (There is a hole in the sky). one finds in them all kinds of molecules. Besides H2 . The passage of spirals wave initiates star formation.426 • • • Diffuse matter in the Milky Way In molecular clouds. why is it so important? There are many reasons: practical. radiation field). the three phases account for 99% of the mass of the interstellar medium or about 5% of the stellar mass of the Milky Way. As a result of its rotation. hydrogen is mostly atomic (H). Assuming a dust-to-gas mass ratio of 1:150. and because man is made of interstellar dust. molecular abundances. The scientific investigation of dust started with the Sky photography of E Barnard and M Wolf in the 1920s which revealed the existence of dark .1. Although not an argument of natural sciences but of humanities. and of order 1% in HII regions. without aiming at completeness. in the Milky Way about 30%. The total gas mass is Mgas ∼ 5 × 109 M .2 The relevance of dust If there is so relatively little dust. in this phase. its mass in the Milky Way amounts to Mdust ∼ 3 × 107 M . In the diffuse or HI gas. In HII regions. the Milky Way has formed a spiral pattern which can be admired in face-on external galaxies (figure 11. because it converts the radiation from stars into the infrared. half of interstellar the mass is in molecular clouds. the warm and tenuous medium and entering a cold and dense phase. 14. because it determines the spectral appearance of stars during their formation. Some are listed here. in some galaxies up to 99%. hydrogen is mostly molecular (H2 ). for example. The dust is just a small admixture. because it is the basis for the formation of planets. half in the diffuse gas. Roughly speaking. This affects the composition and structure of the grains and happens repeatedly during their lifetime. Dust is so important: • • • • • • because it weakens the visual light and thus limits our optical view in the Galaxy. About every 108 yr a density wave sweeps over a parcel of gas in the Milky Way and causes a drastic change in its physical state.

In the 1980s and 1990s. whether through coagulation of smaller units or by large-scale instabilities but the shocks in the spiral arms are probably an important ingredient in their creation. Although on maps in the CO line (see later) most molecular clouds have a sharp edge. in some loose way. Examples of well-studied nearby molecular clouds are the complexes in Taurus (distance D = 140 pc. J H Oort argued in 1932 that the absorbing particles cannot be larger than the wavelength of visible light as otherwise they would produce noticeable effects on the dynamics of stars perpendicular to the galactic plane. by mass. and the Orion GMC. on photographic plates. These GMCs are further structured into clumps which are the birth place of stars. .Molecular clouds 427 clouds. as the name suggests. In 1937 J S Hall established the extinction law in the visible and in 1949 Hall and W A Hiltner discovered interstellar polarization. the transition of the gas density to the intercloud medium is not abrupt. in Ophiuchus (similar parameters but containing much denser concentrations). τGMC . Molecular clouds trace the spiral arms of the Milky Way but are also found in the interarm medium. estimates for τGMC are several 107 yr. The breakthrough came when it was realized that the sky was littered with bright infrared sources. There. the near ones appear as holes in the sky that is otherwise littered by myriads of stars. Therefore. extension L = 20 pc). It is not clear how molecular clouds are made. one finds many molecular species and almost all hydrogen is in the form of H2 . on the star-formation efficiency. The biggest molecular clouds have masses greater than 105 M and are labeled giant molecular clouds (GMC). mass M ∼ 104 M . This has prompted their second name: dark clouds. The density contrast between clumps and their environment is about one order of magnitude.2 Molecular clouds Roughly half of the mass of the interstellar medium resides in molecular clouds. The average lifetime of a giant molecular cloud. The prevalence of molecules is due to the fact that the dust shields them from destructive ultraviolet radiation. the origin of the infrared emission is usually dust. There is always an intermediate layer with a dust optical depth AV ∼ 1 mag where CO is photo-dissociated. The likely destruction mechanism for GMCs is the internal formation of massive stars which produce strong winds and eventually supernova explosions. Their concentration towards the galactic plane could explain the zone of avoidance for globular clusters and galaxies. therefore. satellites mapped the total sky at infrared wavelengths not accessible from the ground. molecular clouds are enveloped by a halo of atomic hydrogen which. But the full relevance of dust for astronomy was grasped only with the advent of infrared detectors in the 1960s. is not always negligible with respect to the molecular cloud proper. Molecular clouds block the starlight and. 14. depends then.

2 Nomenclature of levels CO possesses rotational.1 The CO molecule 14.6 mm. We only discuss carbon monoxide (CO) which is the most widely observed molecule in the study of galactic and extragalactic molecular clouds because it possesses a number of favourable properties: • • • • It is very abundant.2. H2 CO and HCO+ . Where it exists. SiO. therefore. These molecules have very diverse structures and trace. . HCN. 3. 2. 13 CO. without being too precise.09 eV. . different physical conditions. All other molecules are at least one order of magnitude less frequent. H2 O.1 Why CO is the most frequently observed molecule Over 100 molecular species are known in space today and about ten of them are regularly used as probes of the interstellar medium. E el • E vib E rot . Average parameters of molecular clouds. 14. It is very stable with a dissociation energy of 11.5 14. We. OH. its number density relative to molecular hydrogen is [CO]/[H2 ] ∼10−4 . NH3 .428 Diffuse matter in the Milky Way Table 14.2. The lowest transition j = 1 → 0 has a wavelength λ = 2.1.1. CO. they are H2 .5 N(m) ∝ m −1. 1. . C18 O and C17 O spanning three orders of magnitude in abundance and likewise in column density and optical thickness. CS. Its lowest rotational levels are easily populated by collisions. . It has several observable isotopes: CO. Average gas density Average temperatures Mass fraction of interstellar medium Volume fraction in disk of Milky Way Scale height (in z-direction) Observable mainly through Mass spectrum of GMCs in Milky Way Mass spectrum of clumps within GMC n = 100 cm−3 Tgas = Tdust = 20 K f 50% v = a few percent H = 80 pc near Sun Molecular lines and IR dust emission N(M) ∝ M −1.1.2. . Rotational states have quantum numbers j = 0. in various lines. vibrational and electronic levels which require progressively higher energies for their excitation. exemplify the discussion of interstellar molecules by CO. The level j = 1 is only 5 K above the ground state and requires for its excitation a density of only n(H2 ) > 100 cm3 .

ψ = ψel · ψvib · ψrot . for = 1. . 3. . The letter X stands for the electronic ground state. the Q-branch is absent. The transition from the ground state to the next higher electronic state. the energy levels are 2 j ( j + 1) Ej = 2I and the angular momenta j ( j + 1). for a − . Lj = . E.3 Frequencies. . . so the quantum number = 0. . B. C. j = 1. Electronic states bear the names X. 1. The lowest CO state is denoted X 1 + v = 0 j = 0. j = 0 and R-branch.1 for v = 1 transitions under LTE at T = 1000 and 2000 K.Molecular clouds • • 429 Vibrational states are labeled v = 0.67 µm. 2. The change in rotational quantum number j in a radiative transition is subject to the selection rule j = j − j = 0. 14. The total eigenfunction ψ of the molecule is approximately the product of the electronic. energies and A-values of pure rotational transitions The rotational eigenstates for v = 0 are found from the Schr¨ dinger equation of a o rigid rotator. X 1 + → A 1 . vibrational and rotational wavefunctions. Such highly excited CO is found in accretion disks of young stars.1. ±1 with the further restriction that transitions between j = 0 and j = 0 are forbidden as well as transitions between -states with j = 0. is at ˚ 1544 A.and R-branches are displayed in figure 14. The P. The lowest v = 1 → 0 band is at 4. If I is the moment of inertia of the molecule about its rotation axis. . The plus superscript in X 1 + state means that the electronic state ψel does not change when all particles of the molecule are reflected at the origin (inversion). the Greek indicates that the total orbital angular momentum of the electrons vanishes. A. it would change sign. .2. Vibrational transitions split into branches depending on the changes in the rotational quantum number j : • • • P-branch. The superscript 1 at the letter shows the multiplicity. the state is denoted . j = −1 Q-branch. .

is missing. Inserting the eigenfunctions levels are E j /k of the rigid rotator into the Einstein coefficient (6.763 × 1010 s−1 and the energy 2.430 Diffuse matter in the Milky Way Figure 14.1. one finds for the j → j − 1 transition 512π 4 B 3 µ2 j 4 (14.4) A j. 3 → 2 and 4 → 3.3) For 12 CO. It gives the average emission per molecule over 4π. the rotational constant B = 5. j −1 = 3hc3 2j +1 . the lower curves refer to v = 2 → 1. The rotational levels are degenerate with statistical weight g j = 2 j + 1.41).6 × 10−14 erg s−1 . The upmost sequences of dots show the v = 1 → 0 bands. which reflects the potential energy of the atom. Defining the rotational constant B by B= it follows that h 8π 2 cI (14.2) (14. The intensity scale is linear. equal in both boxes.77 j ( j + 1) in the unit Kelvin. (14. and goes from zero to 4.1) Ej hB = j ( j + 1) k k and the frequency of the transition j → j − 1 is ν = 2B j. LTE line intensities of rot–vib transitions with v = 1. The eigenfunctions are similar to those of the hydrogen atom except that the function with the radial dependence.

2. Radiative transition probabilities of the CO molecule for pure rotational lines ( v = 0) in various vibrational levels v.12 × 10−19 cgs.112 Debye = 1. which assumes a bar of fixed length between the nuclei.Molecular clouds 431 Figure 14. where the dipole moment µ = 0. 14.2.3 the energies E v. The Avalues increase like ∼ j 3.0 = 7. for the lowest transition A1. The upmost sequence of points refers to v = 0. Figure 14.4 × 10−8 s−1 .2. With higher j -number. the internuclear distance increases and the rigid rotator model. the population n i of any level i is determined by (5. gi e−Ei / kT ni = .8). n Z (T ) . is not adequate.2. all for the electronic ground state.1 LTE population and critical density In local thermodynamic equilibrium at temperature T .2 plots the A-coefficients of rotational lines in various vibrational levels and figure 14.2 Population of levels in CO 14. j of the vib–rot levels.

respectively. When level populations are close to LTE.17. In molecular clouds. 1. 2. usually only the lowest rotational levels of CO are close to thermalization. vibrational and electronic part.7). 3. the higher ones are not. The temperature T computed from the Boltzmann equation (5. Z (T ) = Z rot · Z vib · Z el . in LTE at T = 10. Energies above ground of the CO molecule in the electronic ground state for vibrational quantum numbers v = 0.65). Energies E/k are expressed in Kelvin. the most densely populated levels are j = 1. The lowest sequence of dots is for v = 0. 5 as a function of the rotational level j . the molecule is in its vibrational and electronic ground state (Z vib = Z el = 1) and Z rot equals 3. Z (T ) can be split into its rotational.96.432 Diffuse matter in the Milky Way Figure 14. For example.56 and 11. 2. n is the total number of CO molecules per cm3 and Z (T ) the partition function of (5. To estimate the minimum or critical gas density n cr necessary to excite the lowest rotational line j = 1 → 0. If the molecule is not in LTE and the levels are not thermalized. This is always true in the limit of high densities.3. the ratio of level population n j /n i defines the excitation temperature Tx according to (13.8) is then the gas kinetic temperature Tkin . 7. respectively. we neglect photons so that n cr depends only on . At these temperatures. 20 and 30 K. the molecule is said to be thermalized.

2. 14. The solar value of the isotopic ratio [C]/[13 C] is 89. a nonzero energy E 0 = 1 hν in the ground level v = 0.98 Debye) require denser gas for their excitation.5.3 CO abundance and isotopic ratios From many pieces of indirect evidence.2. typical of molecular clouds. Because 13 CO is heavier than CO. temperature and velocity gradient of molecular clouds are shown in figure 14.1) when n cr = n(H2 ) ∼ A10 /C01 ∼ 100 cm−3 . Direct evidence for [CO]/[H2 ] can. Isotopic fractionation which favours 13 CO. Because A j. of course. By and large. At its root is the charge exchange reaction (14. in other places of the Galaxy it tends to be smaller. C10 3 × 10−11 cm3 s−1 when T is in the range from 10 to 50 K. j −1 ∝ j 3.75)). The balance of level population requires n(H2 )n 0 C01 = n 1 [ A10 + n(H2 )C10 ]. j −1 ∝ µ2 . and in the Galactic Center it is noticeably lower ([CO]/[13 CO] ∼ 20). E 0 (13 CO) < E 0 (CO) . Only in the diffuse medium have CO. somewhere between 40 and 80. The frequency ν decreases 2 with the molecular mass. The ratio is influenced by the folowing processes: • • Self-shielding in the dissociating lines which favours CO. be gained from simultaneous measurements of rot–vib absorption bands in CO and H2 towards strong background infrared continuum sources but this has not yet been done for molecular clouds. H and H2 absorption lines been observed in the far UV towards weakly reddened hot stars stars (AV ≤ 1 mag). molecules with a larger dipole moment (for example. 14. With C01 ∼ 3 × C10 .2.Molecular clouds 433 the collision rates C01 and C10 and on the Einstein A coefficient A10 (see equation (13.3). The value is much smaller in diffuse clouds because of photo-destruction.2 Population of rotational levels including radiative excitation Abundances of rotational levels for representative values of density. ν ∝ m −1/2 .2. and possibly also in dense cold clouds because CO freezes out onto dust grains. the upper transitions need higher densities and. Because A j. The vibrations of the molecules have. The values have been calculated including radiative excitation under the Sobolev approximation (section 13. the average abundance ratio [CO]/[H2 ] ∼ 10−4 . CS with µ = 1.4. like all harmonic oscillators. one finds that already 10% of the molecules are excited (n 1 /n 0 ∼ 0. in principle.5) CO + 13 C+ ⇔ 13 CO + C+ . This rough estimate indicates that the j = 1 → 0 line of CO is always seen in molecular clouds. For para-H2. also higher gas temperatures.

We see that for j ≥ 3. radiative excitation is more important than collisional excitation. b j .77) for a velocity gradient of dv/dr = 1 km s−1 pc−1 . The escape probability β is computed in the Sobolev approximation from (13. 104 cm−3 (squares) and 105 cm−3 (triangles).69).434 Diffuse matter in the Milky Way Figure 14. The deviation from LTE. is defined in (13. In the three vertical frames on the left.4. The CO abundance is [CO]/[H2 ] = 10−4 . . 20 and 40 K and for H2 densities 103 cm−3 (circles). those dots which are not connected by a line are calculated without a radiation field and thus without an escape probability. the number densities n j are normalized so that their sum equals one. Non-LTE level population of CO for T = 10.

5) is temperature-dependent and goes preferentially from left to right in cold clouds (T < 35 K). generally inconspicuous. There are two forms of molecular hydrogen: ortho-H2 with parallel and paraH2 with opposite proton spins. 14. Radiative transitions between ortho and para hydrogen are forbidden. The lowest + excited level of para-H2 (X 1 g v = 0 j = 2) lies 510 K above the ground ( j = 0). has a small Einstein coefficient (A = 3 × 10−11 s−1 ). as a quadrupole transition. by far the most abundant molecule. outside HII regions. gas temperatures and densities or shock velocities.2. Being homonuclear. which. rot–vib lines of H2 are seen. the isotope HD. spin exchange can occur by collisions with molecules and ions or on grain surfaces but the rates are low.8 × 10−10 s−1 ) is at 17 µm and even harder to excite.4 Formation of molecular hydrogen on dust surfaces The major destruction route of H2 in the diffuse gas. corresponding to the ratio of the statistical weights. only quadrupole rot–vib transitions are allowed (section 1. Mass loss from stars heavier than 1 M . is hardly excited under normal conditions of the interstellar medium and. with jets and copious UV radiation. The existence of the ortho and para forms is a consequence of the exclusion principle.4) subject to the selection rules j = ±2 or 0 (excluding 0 → 0 transitions). where the nuclei are not identical.3 Molecular hydrogen Molecular hydrogen. whereas u (ungerade) would imply asymmetry. the reaction (14. the ortho form is three times as abundant as the para form. is photo-dissociation. so these molecules are some five and 20 times less abundant than 13 CO. does not show this property. however.2. However. Wherever there is high-mass star formation. therefore. too high to be populated at normal temperatures. it is well observable in photo-dissociation regions and shocks and there its emission gives information about the ultraviolet radiation field. Therefore. The spacing of rotational levels is given by (14. + The electronic ground state of H2 is X 1 g (see nomenclature in section 14. even in external galaxies. enriches the interstellar medium more efficiently with 13 C than with 12 C.4. The j = 2 → 0 emission line from this level appears at 28 µm and. respectively.2. They are immediately followed by electric .Molecular clouds 435 • with an energy difference E 0 /k = 35 K. The standard isotopic ratio of C18 O and C17 O are 500 and 1800. Absorption of photons with wavelengths between 912 and + ˚ 1108 A leads to electronic transitions X 1 g → B 1 u (Lyman band) and + X 1 g → C 1 u (Werner band). it just supplies the mass for the interstellar gas.1). 14. When the temperature is high and many rotational levels are excited. A = 4. because of the specific preceding nucleosynthesis. The g (gerade) signifies that the the homonuclear molecule is symmetric with respect to swapping the two nuclei. The lowest ortho-line ( j = 3 → 1.3).

As the radiation penetrates into the cloud. The smallest grains should get the highest weight (see (7. n gr 1.16) their mean velocity and σgr = πa 2 the geometrical cross section of spherical grains of radius a. The only remaining alternative to create H2 is on grain surfaces. then. roughly speaking. is taken at the cloud surface. a (14. so a ∼ 10−6 cm seems appropriate. molecular hydrogen is dissociated for τV ≤ 1. R ∼ 10−17 n 2 T 1/2 cm−3 s−1 H and this should indicate the right order of magnitude.436 Diffuse matter in the Milky Way dipole decay ( A ∼ 108 s−1 ) to the electronic ground state. the chemistry of the major gas species is determined by photodissociation.5 eV is transmitted to the dust particle. τV . molecule formation and desorption. which has many modes of energy disposal. and how much is carried away by the H2 molecule in the form of excited rot–vib levels. Therefore R 7 × 10−23 η n(HI) n H T 1/2 . v from (5. On leaving the grain surface. three-body reactions are excluded as they require densities n ≥ 1012 cm−3 . This leads to the formation of a sequence of chemically distinct layers.5 g cm−3 . carbon is ionized for τV ≤ 2. If the final vibrational quantum number v is greater than 14.5 × 10−27 nH a3 where n H = n(HI) + n(H2 ) is the total hydrogen density. Radiative association H + H → H2 + hν is strongly forbidden. The reaction rate R (per cm3 and second) is obviously proportional to the number of H atoms impinging on a grain (n(HI) σgr v ) times the grain density n gr . 14. Its photo-dissociation can then be calculated and must be balanced by H2 formation. With a gas–to–dust of ratio 150 and a bulk density of 2.2.1 A few remarks on H2 and CO chemistry In the envelope of a cloud illuminated by ultraviolet light with wavelengths ˚ λ > 912 A. n(HI) n H .29)). and most gas-phase carbon . this happens in about one out of seven cases.6) In the diffuse medium. The efficiency η includes the probability for sticking. the state is unbound and the molecule dissociates. it becomes absorbed (mainly by dust) and the UV flux weakens.2. R = η n(HI) σgr v n gr . If η ∼ 0. Molecular hydrogen is observed in the diffuse medium if the ultraviolet radiation field is not too strong. If the zero point of the visual optical depth. n(HI) is the density of hydrogen atoms. other reactions are negligible.4. it is unclear how much of the formation energy of 4. finding another hydrogen atom by tunneling over the grain surface.

The line absorption leads to the effect of selfshielding.7) H+ + O 3 −→ −→ −→ OH+ + H2 H3 O+ + 2H H2 O + H When the UV shielding by dust is low (τV < 3). CO is formed by neutral–neutral reactions like CH + O → CO + H. . however. at a redshift z º 1000 when hydrogen was already neutral but before the appearance of the first stars.) that couples radiationless to a dissociating state. .Molecular clouds 437 is locked in CO for τV ≥ 3. For an isotope like 13 CO. . H+ . only hydrogen and helium. a small fraction becomes ionized by cosmic rays (CR) and is transformed into H+ . the dissociating lines are slightly shifted in frequency. it could be created via H + H+ H+ + H 2 or H+e H− + H −→ −→ H− + hν H2 + e −→ −→ H2 + H+ H+ + hν 2 . is reactive and relevant for the chemistry of the 3 interstellar medium as it can initiate the formation of important molecules like H2 O. via C+ + OH −→ CO + H+ . CO formation is promoted by ionized carbon. for example. 14. 13 CO is destroyed in a larger volume than 12 CO. H2 could not then form on grains. 3 H+ 2 The new species.2 The early universe In the early universe. there were no heavy elements and thus no dust.2. carbon is in the form of C+ and CO is absent. So at the transition between atomic and molecular hydrogen (τV 1). ˚ CO is then destroyed by line absorption (λ < 1118 A) from the ground state X 1 + to a bound level of a high electronic state (E. (14. 3 H2 + CR H+ + H2 2 −→ −→ H+ + H. Because the 13 CO column density is much smaller. Although molecular hydrogen is chemically rather inert.4. F. H+ + CO 3 −→ HCO+ + H2 . OH+ + 2H2 H3 O+ + e or remove carbon monoxide. When UV shielding is high.

Although the HI gas is very inhomogeneous.2 < τ < 1 and be destroyed radiatively or by collisions with H atoms. There also exists some hot ionized medium (HIM) with . Such clouds are seen in absorption and emission and make up what is called the cold neutral medium (CNM) of the diffuse gas because the hydrogen there is mostly neutral and relatively cold. The critical mass for gravitational instability of a gas cloud depends on temperature as T 3/2 (see (15. Its properties.3 Clouds of atomic hydrogen 14. τ < 0. it is useful to define a standard HI cloud with parameters as in table 14. They are. They contain only a few molecular species. . As hydrogen is mainly atomic there. which is a few times bigger than for CO and increases outside the solar circle. 14. .2. They consist mainly of atomic hydrogen.2.1 . . The scale height of the clouds at the locus of the Sun is about 250 pc. 10 cm−3 D ∼ 10 pc N(HI) ∼ 1020 cm−2 l ∼ 3 pc 50% of the HI gas and 25% of all gas a few percent of galactic disk v ∼ 30 km s−1 0. There are other components which are less dense and completely penetrated by UV radiation. more strongly ionized and much warmer. has a larger scale height and a very low optical depth in the 21 cm line. n = 0. Here molecular hydrogen could have acted as a coolant down to 100 K by emission of its lowest rotational lines. Typical parameters of HI clouds. it is much more voluminous. they led to a small fraction of molecular hydrogen which was important for the formation of the first stars. infrared dust emission and pulsar dispersion. In the warm ionized medium (WIM). are derived from observations of the famous 21 cm line but also from Hα and atomic fine structure lines.1 General properties of the diffuse gas About half of the mass of the interstellar medium is in diffuse clouds. to a large degree. this component is called a HI gas.3. Although the reactions were not efficient. therefore. 100 K Tdust ∼ 20 K n(HI) ∼ 1 .01. . Temperature Density Cloud diameter Column density Distance between clouds Mass fraction Volume fraction 21 cm linewidth 21 cm line optical thickness Tgas = 50 . . This component is only seen in emission.48)) and contraction is possible only if the release of potential energy can be radiated away. 1 cm−3 and gas temperatures are several thousand K. .438 Diffuse matter in the Milky Way Table 14.

3. F = 1 and the statistical weight is g1 = 3. It is a dipole field (see the electric analog in (3. (11. 14. In the lower state. Let F denote the quantum number of total angular momentum which is the sum of electron spin S and nuclear spin I . There is level splitting for the same reason as in the Zeeman effect. 14. only here the magnetic field is supplied by the nuclear magnetic dipole.793 times greater than a nuclear magneton µN (compare with (11. in the 1950s it played a dominant role in elucidating the structure of the interstellar medium of the Milky Way. µN = 2m p c The magnetic field in which the proton aligns is supplied by the electron. .10) for the Bohr magneton). the spins are anti-parallel. it is a triplet. Indeed. speaks of the 21 cm line.10)) of moment µB and decays with distance like r −3 . When one evaluates its strength at one Bohr radius a0 = 5.29×10−9 cm. is due to the study of the 21 cm line.793 · µN and is thus 2. probably most of the galactic disk but its mass fraction is negligible. e . This is the right order of magnitude but an underestimate because a0 is the mean distance and the positions where the electron is closer to the nucleus have a higher weight. O VI at 1030 A) towards hot stars. . electron and proton spin are parallel. density.16) which gives the potential energy of a magnetic dipole in a magnetic field. F = 0 and g0 = 1. In the upper state.16) yields a frequency ν = E/ h 530 MHz. kinematic behavior. 106 K.Clouds of atomic hydrogen 439 Tgas ∼ 105 . its temperature. very tenuous (n = 0. A very good number is obtained when one averages over the radius and takes the .2 The 21 cm line of atomic hydrogen In its electronic ground state. total mass and overall distribution.11 cm and one. it is a singlet. of course. As its propagation is not impeded by dust. The transition from an upper to the lower level occurs at ν = 1420 MHz or λ = 21.1 Frequency and transition probability of the line One obtains an estimate of the energy difference E between triplet and singlet from the classical formula (11. F = S + I. The dipole moment of the proton is given by µp = 2. It is.01 cm−3 ) and detected either by soft X-ray emission or through UV absorption lines of highly excited ˚ ions (for instance. therefore.2. most of what we know about the diffuse component. It occupies a lot of space. a hydrogen atom has two hyperfine sublevels because the spin of the electron couples to the spin of the proton.3.

c3 With the relations (6. (14. from a supernova remnant or the 3 K microwave background.869 × 10−15 s−1 ): first. because it is not an electric but a magnetic dipole transition which reduces A10 by a factor ∼106 .440 Diffuse matter in the Milky Way 1s wavefunction as the weight. for example. The value of A10 comes out very small (A10 = 2. nn 1 C10 + n 1 A10 + n 1 u ν B10 = nn 0 C01 + n 0 u ν B01 . The atomic levels change because of collisions as well as spontaneous and radiatively induced transitions. we have. 14. The excited state has an extremely long lifetime. hν hν (14.41) but this time µ is a magnetic dipole moment given by the sum µp + µB .41)). in standard notation. The Einstein A-coefficient of the line follows from the general formula (6. there is enough neutral hydrogen around to make the line easily observable.3 How the hyperfine levels of atomic hydrogen are excited 14. compared to optical transitions where τ ∼ 10−8 s.3. Nevertheless.8) The left-hand side populates state 0. As there are only two levels involved.40) between the Einstein A and B coefficients. the right-hand side depopulates it. which is often called spin temperature. τ = 1/A ∼ 107 yr. kT The excitation temperature Tx . Quantum mechanically. In equilibrium at kinetic temperature T . because the frequency is low ( A ∝ ν 3 after (6.3.39) and (6. Here n = n 1 +n 0 is the total number of hydrogen atoms per cm3 and C10 is the collision rate for de-excitation (F = 1 → 0). a radiation field u ν . and the radiation temperature TR are defined by g1 n1 = e−hν/ kTx n0 g0 8πν 2 kTR .1 The excitation or spin temperature In determining the population of the hyperfine levels. for generality. it is a number with 11 significant digits. and second. we include.9) and . It may come. 0 and 1. and hν αx ≡ kTx one finds uν = e−αx nC10 + A10 + kTR kTR A10 = nC10 e−α + A10 . g0C01 = g1 C10 e−α α≡ hν . the frequency has been determined to extremely high precision.3.

In the collision.05 cm−3 . After a linear expansion of the e-function. at very low frequencies (hν kT ).10) 14. which has T = 100 K and n ≥ 1 cm−3 . The statistical weights gi are gone.2. the usual formula n cr A10 /C01 is wrong by a factor ∼1000. is spin exchange. the electron spin of atom (1) can flip so that afterwards it is in the state F = 0. Then the levels are thermalized (Tx T ) above the critical density: n cr = y −1 = A10 kT . the proton of atom (1) is shielded by the electron cloud and its spin is unaffected. in view of the small Einstein coefficient.3. The cross section for collisional deexcitation (F = 1 → 0) in a gas of 100 K is σ10 ∼ 10−15 cm2 . rather surprisingly. equation (14. The brackets denote an average over the Maxwellian velocity distribution and u is the velocity.Clouds of atomic hydrogen 441 This is a general result for a two-level sysytem. Indeed. the sublevels are. C10 hν One often uses for the critical density the expression n cr ∼ A10/C01 . corresponding to a collision rate C10 = uσ10 ∼ 2 × 10−10 cm3 s−1 . the radiaton field is negligible. 14. so the population ratio n1 n0 3 and the total hydrogen density n is close to 4n 0 . in estimating n cr for the CO rotational levels in section 14. the spin temperature Tx can be directly expressed by the kinetic gas temperature T and the radiation temperature TR . For the 21 cm line. for example. The electron spin of the hitting atom .1. provided that α ≥ 1. A10 kT (14. In a standard HI cloud. thermalization requires a density that is kT / hν times larger. Tx = ynT + TR 1 + yn with y ≡ C10 hν . and T = 1000 K implying y there is. the exponents α and αx are much smaller than one.10)).3 Spin exchange by collisions among H atoms The physical process that determines the cross section. when there are only two levels and no radiation.9) implies Tx ∼ T when n = A10 /C01 . Because the temperature difference between the levels of the 21 cm line is only hν/k = 0. However. in a tenuous and hot 7 in equation (14. cloud (n = 0. a substantial excursion of the hyperfine levels from LTE and the ratio Tx /T is only one-quarter.3.068 K and the gas temperature T ∼ 100 K.3. σ10 . thermalized. We did so.3. Suppose a hydrogen atom labeled (1) with F = 1 and electron spin up is hit by another hydrogen atom. However.2 The critical density for excitation Under most circumstances.

As the potentials Vt and Vs are known. its proton spin is irrelevant. which coincide in Dirac theory. Because in the absence of collisions and radiatively induced transitions.3. σ10 changes only weakly with temperature (∝ T −1/4 ). the population of the hyperfine levels is determined by the Lα radiation field. the two potentials are equal. the complex is in the repulsive electronic triplet state (3 . i.3. The 2S1/2 and 2P1/2 states. the Lα optical depth is immense and each Lα photon is scattered many times. The state n = 2 consists of three sublevels. the phase between triplet and singlet changes by (Vt − Vs )dt/ . spin flip can also occur due to the interaction between the magnetic moments of the electrons but this process is several orders of magnitude weaker. the wavefunction of the complex is written as a sum of a symmetric and an anti-symmetric function with respect to the interchanging particles. The symmetric part is associated with the phase factor exp(−i Vtt/ ).e. the wavefunction stays the same. If the electron spin of the hitting atom is also up. it is not possible to specify the state of the atoms separately.442 Diffuse matter in the Milky Way is random and thus some linear combination of up and down. In quantum mechanics. At large separations. The radiation field in the Lα line therefore acquires the same shape (spectral slope) as the Planck function at that temperature.4 Excitation by electrons and Lα radiation Spin flip is also possible as a result of collisions with free electrons. one can semi– classically follow the collision over the time of the impact and thus calculate the probability that afterwards atom (1) is in state F = 0 [Pur56]. This yields σ10 . Only the 2P levels are allowed to combine with the ground state 1S1/2 . 14. because when the electrons are swapped. In principle. During the impact. the anti-symmetric part is associated with the phase factor exp(−i Vs t/ ). as in stable H2 molecules. the two atoms form a molecular complex. . Electron impact becomes important when the fractional ionization of hydrogen exceeds a few percent. Therefore the atoms exchange electron spins at a rate (Vt − Vs )/ . Here the cross sections are about half as large but electrons fly faster. So in a time interval dt. potential Vt ). They connect the n = 1 and n = 2 states and are created in HI clouds by cosmic-ray ionization and subsequent proton–electron recombination. As a resonance line. undergoes many collisions with H atoms at the gas kinetic temperature T . 2P1/2 2S1/2 2P3/2 each being double because of the hyperfine splitting. are actually separate due to the Lamb shift. the spin temperature Tx is also forced towards the kinetic gas temperature T . the complex is in the attractive electronic singlet state (1 . Instead. Another interesting way of exciting the hyperfine levels of hydrogen atoms is through Lα photons [Wou52]. potential Vs ). if it is down. so the collision rates are higher. Because the interaction potential is steep.

Tx (14. The brightness temperature Tb at velocity v observed towards a cloud in front of a background source of intensity Tc is. kTx Here n 0 is the number of atoms in the lower singlet state and K 0 is after (13. If N(HI) denotes the total column density of neutral hydrogen. But note that the intensity of optically thin line emission. N(HI) and Tx are measured in cm s−1 .74). So stimulated emission. It is often convenient to refer the profile function to the Doppler velocity v. again with P(v) dv = 1. Without it. dτν = K (ν) ds 443 where (ν) is the normalized profile function and the absorption coefficient K is from formula (13. Because hν kTx . (14. one can expand the e-function in (13. The factor hν/kTx is due to stimulated emission. which is usually negligible at optical wavelengths.11) Because of stimulated emission. the lower the optical depth. v = 0. Tb (v) = T0 · e−τv + Tx (1 − e−τv ).62). It is then denoted by P(v). one would have K n 0 K 0 . analogous to (14.4.82 × 1013.3.13) . (14.67) proportional to the Einstein coefficient A10 . the optical thickness becomes τν = N(HI) hν · K0 · · 4 kTx (ν).4 Gas density and temperature from the 21 cm line 14.11). from (13. the optical depth at velocity v is τv = C −1 N(HI)P(v) . Because P(v) dv = (ν) dν. and not to the frequency ν.68) and obtain hν K n0 K 0 .68). At the line center.12) When v. which is proportional to τν Tx . φ(ν) = c P(v) ν and. is independent of temperature. from (13.3. reduces the cross section in the 21 cm line by a factor of order 103 (!) at typical excitation temperatures Tx ∼ 100 K.Clouds of atomic hydrogen 14. cm−2 and K. C = 1. respectively. the hotter the gas.1 Optical depth and column density The optical depth element in the HI line is.

1 − e−τv (14. So far all attempts to detect the transition have failed because the isotope is so rare ([D]/[H] ∼ 10−5). the optical thickness need not be small.2 How to measure the optical depth and the spin temperature With the help of a strong background source. (14. one obtains. one can determine the optical depth τv as well as the spin temperature Tx of an HI cloud.13). it will be seen in absorption and equation (14. one has to assume that Tx is constant along the line of sight and that the cloud parameters do not change as one switches from the quasar to a nearby off-position. The magnetic moment of the neutron is µn = −1. integration over the observed line profile directly yields the HI column density: N(HI) = C Tb (v) dv. a quasar. with cosmological consequences because deuterium has been created during the Big Bang. also shows hyperfine splitting in its electronic ground state.3. at a position immediately adjacent to the quasar. As τv is known.3.13) reduces to Tb (v) = T0 e−τv . one performs a measurement towards the quasar at a frequency outside the line where the HI cloud is transparent (τv = 0).4. N(HI) = N(HI)P(v) dv = C Tb (v)τv dv.5 The deuterium hyperfine line Deuterium. as is often the case for the 21 cm transition. So the nuclear spin quantum number I of deuterium equals 1 and the spin flip of the electron corresponds to an F = 3/2 → 1/2 transition. Then one observes the quasar at various velocities in the line. The actual frequency changes are very small. First. Of course. the hydrogen isotope. the spins of proton and neutron are parallel.444 Diffuse matter in the Milky Way Without the background. from (14. such a measurement yields the spin temperature Tx . This gives τv . Tb = Tx (1 − e−τv ). when the optical depth is small everywhere in the line. 14. In the line core.14). If the quasar is much brighter than the foreground HI gas.12). One can now evaluate the column density N(HI) after (14.913 µN . The importance of this line rests in the fact that it could improve the estimate of the deuterium abundance of the interstellar medium. so T0 is practically constant.14) Therefore. say. In deuteron.15) 14. This provides the brightness T0 of the quasar after (14.

so W = mnuc · mel = gnuc gel I · S. When one aims at greater precision for µD . This is witnessed by its electric quadrupole moment. one must take into account that the deuteron nucleus possesses some orbital angular momentum (quantum number l). S = 1 2 F(F + 1) − I (I + 1) − S(S + 1) 2 . the deuterium hyperfine line has a frequency that is 2 a factor gH H /gD D 4. 0). This is in the inner parts of HII regions where hydrogen is ionized too. for the electron S = 1 and gel = 2. The magnetic moment of deuteron is quite close to the sum of the two. For deuterium.2). where g is the Land´ factor.857.6 cm.665 GHz (F = 0 → 1). which also gives rise to a magnetic dipole moment. I = 1. The square of the angular 2 momentum vector F = I + S is obviously F · F = (I + S) · (I + S) = I · I + S · S + 2I · S so that I·S= 1 2 F·F−I·I−S·S . one can only see it where helium is ionized. IH = 1 and 2 gH = 2 · 2. µD µn + µp 0. Another hyperfine line of cosmological relevance is emitted by 3 He+ at 8. we can write mnuc = gnucI for the e nucleus and mel = gel S for the electron. we have to translate the classical formula (11. But for a constant factor. Instead. 2 2 2 2 I = 1 . As the ionization zones of H and He are not co-extensive (section 14.793.857 µN .Clouds of atomic hydrogen 445 and directed opposite to the magnetic moment of the proton. F = 1. . for hydrogen. F = 3 .34 smaller corresponding to a wavelength of 91. Of course. The difference in these averages between upper and lower state is D = 3 for deuterium (S = 1 . deuteron is a mixture consisting to ∼95% of an S (l = 0) wavefunction and to ∼5% of a D (l = 2) wavefunction. Therefore. ID = 1 and gD = 0. In its ground state. To find the frequency of the deuterium hyperfine line relative to the hydrogen 21 cm line.4. we cannot simply scale with the ratio µD /µp . 1 ) and H = 2 for hydrogen (S = 1 . the derivation of the abundance ratio [3 He]/[H] along the line of sight is problematic. it therefore has the expectation value I. When I · S is turned into a quantum mechanical operator.16) for the potential energy of a magnetic dipole m in an outer field into an interaction operator W according to the rules of quantum mechanics. again very hard to observe.

1+ = vg c 2ω2 A signal from a pulsar at a distance L reaches us after a travel time τ = 0 L e2 1 ds L + = vg c 2πm e c ν 2 L 0 n e (s) ds. 2 ωp 1 1 . When one measures the arrival time of a pulse at two frequencies of which ν2 > ν1 . A pulse. are n e ∼ 0. approaches the Earth with a group velocity vg given by (1. via the dispersion measure DM. Typical values for the diffuse medium. (14. The travel time depends on the observing frequency ν: long waves are slower than short waves.1 The dispersion measure One can use the pulses emitted from rotating neutron stars to determine the electron density n e in the diffuse interstellar medium. outside HII regions. there is a delay τ between the two.126) yields for their difference in phase velocity vph = c n± vph vph = n e e2 eB 1 .3.6. When they are traveling parallel to a magnetic field B.6 Electron density and magnetic field in the diffuse gas 14. 2π 2 m e m e ν 3 vph c (14. we can write. approximately.6. equation (1. The quantity DM = 0 L n e (s) ds (14.01 cm−3 . which is a composite of many frequencies. the mean electron density.446 Diffuse matter in the Milky Way 14.2 The rotation measure A linearly polarized wave can be decomposed into two waves of opposite circular polarization.16) is called the dispersion measure and has the unit pc cm−3 .17) If the distance L is known. 14.3. τ= e2 2πm e c 1 2 ν1 − 1 2 ν2 DM. As the observing frequency ω is always much greater than the plasma frequency ωp = 4πe2 n e /m e (the latter is of order 1 kHz). a fundamental quantity of the interstellar medium.3.18) The relative phase changes per wavelength λ by 2π 2π . one thus finds.131).

The frequency shift between the components of opposite circular polarization is. but opposite to other σ -component (2) e B σ is right–hand circularly polarized π is missing σ is left–hand circularly polarized.19) = λ 0 c with the rotation measure RM defined by RM = e3 2πm 2 c4 e L n e B ds. one finds B ∼ 3 µG.6. The B vector lies preferentially parallel to the galactic plane. obviously. their ratio yields the mean magnetic field strength RM . therefore. To determine its value.Clouds of atomic hydrogen 447 and the rotation angle of the linearly polarized wave. one needs measurements at more than two wavelengths.3 The Zeeman effect in HI The strength of the interstellar magnetic field may be estimated from the Zeeman effect in neutral hydrogen.3. B ∝ DM From systematic observations in the Milky Way.2). it may be greater than 2π.80 2πm e c µG [Hz]. When B is perpendicular to the line-of-sight vector e. the degeneracy of the upper hyperfine state (the triplet with F = 1) is removed. ν = B eB = 2. 14. therefore. This is summarized here. not easy to observe. Measured values for B also cluster around 3 µG. all with different polarizations (section 6. together with the case when B is parallel to e. no frequency shift σ is linearly polarized perpendicular to B.20) Because the dispersion measure DM is equal to n e ds and the rotation measure RM proportional to Bn e ds. not only modulo 2π. Although the rotation angle is usually smaller than 90◦ .3. . by half this amount. It is much smaller than the linewidth (∼ 20 kHz) and. In the presence of a magnetic field B. in principle. 0 (14. The total change on the way from the source at distance L to the observer is π L vph ds = RMλ2 (14. (1) e ⊥ B • • • • • • σ is linearly polarized perpendicular to B π is linearly polarized parallel to B. one observes a central π-component and two symmetric σ -components.

1. or wavelengths ˚ λ ≤ λL = 912 A (see table 14. The number of Lyc photons emitted each second from a star of radius R∗ and effective temperature Teff is given by 2 NL = 4π R∗ ∞ νL Bν (Teff ) dν. HII regions fill about 1% of the volume of the galactic disk and their mass fraction of the interstellar medium is of the same order.3 lists NL .2 Relation between ionization and recombination coefficient In thermodynamic equilibrium.1. fine structure lines. (14.1 Ionization and recombination 14.56 eV. One often wishes to relate the Lyc flux NL of a main-sequence star to its mass m. The gas temperature is fairly constant there (Tgas ∼ 8000 K). By order of magnitude. a situation which does not prevail in an HII region. the atmospheres of OB stars are not blackbodies but are extended and non-LTE effects play an important role. hν ≥ hνL = χ(H) = 13. they ionize the surrounding hydrogen.82 s−1 .4. the number of electrons with velocities .448 Diffuse matter in the Milky Way 14. the spectral type and some other stellar parameters. hν (14. In reality.22) Table 14. Hence. Dust that survives the harsh conditions in the nebula is very warm (Tdust ∼ 50 K). the exact number depends foremost on the abundance of metals which produce cooling lines. HII regions can be observed in many ways: via free–free radiation (bremsstrahlung). radio and optical recombination lines. 14. The resulting nebula of hot plasma is called the HII region.13(log m)2 + 19. much warmer than the average dust in HI or molecular clouds.1 The Lyman continuum flux The hottest stars on the main sequence are of spectral type O and B.4. there is a detailed balance so that each process has its exact counterpart running in the opposite direction.4 for a list of symbols). Most HII regions are large and diffuse with number densities n ∼ 10 cm−3 but some are very compact (n > 105 cm−3 ) and these are always embedded in molecular clouds. we suggest from various sources of the literature the approximation log NL = −5. As their effective surface temperature Teff exceeds 20 000 K.4 HII regions 14. They have energies greater than the ionization potential of hydrogen.4. infrared emission of dust. The numbers are not precise but sufficient for many purposes.75 log m + 30.21) This is a simplistic formula. The ionization of hydrogen requires Lyman continuum (abbreviated Lyc) photons. If m is in solar units.

Lyman continuum flux. are calculated for an electron density n e = 1000 cm−3 .3.23) . cross section for ionization from the ground state [cm2 ] = Frequency average of K ν = Ionization potential from ground state = Ionization potential from level n = Gaunt factor (∼1) = Mean radiation intensity in the nebula = Emission of star [erg s−1 Hz−1 ] = Emission coefficient of gas [erg s−1 Hz−1 ster−1 ] in the interval [v. rS and MHII . protons and hydrogen atoms [cm−3 ] = Number density of neutral hydrogen atoms in level n [cm−3 ] = Recombination coefficient into level n = Recombination coefficient into all levels i ≥ n. α (2) = 3. bolometric luminosity.2 × 1048 5. np.5 × 105 4. List of symbols used in this section.3 × 106 2. The radius and mass of the HII region.6 × 1016 MHII [M ] 260 40 4 Table 14.ν Jν Lν ν = Threshhold frequency for ionizing hydrogen ˚ = c/νL = 912 A.4 rS [cm] 4. effective temperature. n e n p βn (v) f (v)v dv = Nn K n. mass and radius for main sequence stars of various spectral type. nH Nn αn α (i) βn K n.0 × 1018 3. hν (14. ν + dν].1 × 10−13 cm−3 s−1 = Cross section for electron capture into level n = Cross section for ionization from level n [cm2 ] = K 1.6 × 104 1. Spectral type.0 × 103 54 NL [s−1 ] 8.ν Kν K χ χn gn. νL λL NL Nc (r ) hνc ne.4 2.0 × 1043 M∗ [M ] 60 30 19 7.0 10. v + dv] captured by protons into level n equals the number of ionizations from level n by photons with frequencies [ν.HII regions 449 Table 14. SpT O4 O6 O9 B3 A0 Teff [K] 50000 42000 35000 18000 10000 L [L ] 1. α (1) = 4.4.6 2.2 × 1018 1.2 × 1049 1.0 6.0 × 1049 1.9 R∗ [R ] 15.6 × 10−13 . at 8000 K.ν 1 − e−hν/ kT cu ν dν. threshhold wavelength = Number of Lyc photons emitted by star [s−1 ] = Number of Lyc photons passing through shell of radius r [s−1 ] = Mean energy of Lyc photons = Number density of electrons.0 3.2 × 1018 2.ν .

2.22 × 10−6 x e E 1 (x). The brackets on the right contain the correction factor for induced emission.ν for photoionization.23).82 × 1029 gn.27) The radiation field has two components: one from the star (subscript s) which is directed radially outwards.26) It must also hold when the medium is not in thermodynamic equilibrium because βn (v) and K n.ν n5ν 3 (hν > χn ) (14.25) E 1 (x) is defined in (A. hν (14. Using the Saha equation.ν .40) with E n = −χn and substitutes K n.4. the rate of ionization of neutral hydrogen atoms by the radiation field of mean intensity Jν equals the rate of recombinations of electrons and protons to all quantum levels n ≥ 1.ν are atomic quantities and independent of the radiation density u ν and the particle velocity distribution f (v). frequencies and velocities are related through hν = χn + 1 m e v 2 .4. 2 In equation (14.23) yields the Milne relation βn (v) = 2n 2 h 2 ν 2 K n.ν = 26 π 4 m e e10 gn. u ν the blackbody radiation density of photons after (5. one finds after integration over all velocities and frequencies. and x = χn /kT .1 Ionization balance In each cm3 of a stationary and dust-free HII region of pure hydrogen.450 Diffuse matter in the Milky Way If χn is the energy necessary to ionize an atom in state n.ν 33/2 ch 6 n 5 ν 3 2.71).24) where the Gaunt factor gn. The latter (without proof) is K n. c2 m 2 v2 e (14. There is thus a remarkable link between the ionization coefficient K n. the total recombination coefficient αn (T ) ≡ 0 ∞ βn (v) f (v)v dv 3. (14. the other arising from recombinations within the HII .ν is close to one.ν from principal quantum number n and the cross section for electron capture βn (v) into that level. When one expresses in (14. If the corresponding recombination coefficient is denoted α (1) . T 3/2 n 3 (14. both at temperature T .2 Dust–free HII regions 14. βn (v) denotes the cross section for electron capture and K n. 14.24). f (v) is the Maxwellian velocity distribution (5.13) for electrons.20). then ∞ nH νL 4π Jν K ν (H) dν = n e n p α (1) . respectively.23) Nn /n e n p by the Saha equation (5.ν from (14.

which is the number of recombinations to the ground level only. by definition. Almost all neutral hydrogen atoms reside there because the probability for downward transitions from upper principal quantum numbers n is very high ( A-coefficients 104 . .29) The expression for the net flux of outwardly directed diffuse radiation Fdν is similar but contains additionally as a source term the emission coefficient dν arising from recombinations to the ground level n = 1. In spherical symmetry. 108 s−1 for L ± 1). and the second term equals α1 n e n p . The stellar flux Fdν declines with distance r from the star because it is geometrically diluted and weakened by absorption. The latter is diffuse (subscript d) and travels in all directions. The bracket on the left gives the number of Lyc photons that pass per s through a shell of radius r .31).30) Adding (14. one obtains d 4πr 2 dr Fsν + Fdν dν hν n H K ν (H) 4π Jsν + 4π Jdν dν − hν 4π dν dν . r 2 dr (14.28) n H K ν (H) ds is due to absorption of hydrogen atoms in their ground state 12 S. Here it is assumed that every diffuse Lyc photon arises from a recombination to n = 1 and is absorbed on the spot by a neutral H atom. The difference between . (14. which is the number of all recombinations per second and cm3 .29) and (14.28) can now be written as (see (13. dividing by hν and integrating over ν. hν (14.31) = −4πr 2 The integrals extend from νL to infinity.30).HII regions 451 region. altogether Jν = Jsν + Jdν . r2 r R∗ (14. 1 d 2 (r Fdν ) = −4π Jdν n H K ν (H) + 4π r 2 dr dν . the first term equals α (1) n e n p .26)) 1 d 2 (r Fsν ) = −4π Jsν n H K ν (H). Nc (r = R∗ ) = NL . . we denote this quantity by Nc (r ) (unit s−1 ). Fsν (r ) = The optical depth τ (ν) = 2 R∗ Fsν (R∗ )e−τ (ν). Equation (14. In the bracket on the right-hand side of (14.

is not determined by the Saha formula (5. Because rS o R∗ .37) • . one immediately finds. 3 S e 14.35) • For densities 103 cm−3 and K (H) ∼ 10−17 cm2 (table 14. 1 − y = n e /n H . (14. we may directly integrate equation (14.2. The transition from the HII region to the neutral gas is sharp and called ionization front. dr (14.452 Diffuse matter in the Milky Way both terms is α (2) n e n p . the fraction y = n H /n of hydrogen atoms follows from the quadratic equation Nc K (H)y = 4πr 2 α (2) n(1 − y)2 .4). whereas the diameter of the HII region may be greater than 1018 cm.33) n e is a very good approximation. hydrogen is almost fully ionized. If n ≡ n p + n H . (14. It is inversely proportional to the electron density.4. Note that the degree of ionization. ionization structure. it does not tell us the Although n H density of hydrogen atoms. a function of radius r .36) MHII = n e α (2) Per unit volume. (14.2 Str¨ mgren radius.38). the recombination rate n 2 α (1) equals the rate of ionization. the width dIF is only ∼1014 cm. therefore. Its width dIF is determined by the condition that the optical depth τ with respect to Lyc radiation is of order unity.32) from the stellar surface (r = R∗ ) to the edge of the HII region. where the coefficient α (2) = α (1) − α1 stands for all recombinations to levels n ≥ 2.34) • y is. NL m H .32) In an HII region. from the last equation. e An individual electron will. dIF = 1 . The mass of an HII region depends not only on the spectral type of the exciting star but also on the environment. We thus obtain d Nc (r ) = −4πr 2 α (2) n e n p . Therefore. recombination time o • For the Str¨ mgren radius. The radius there is denoted rS and called Str¨ mgren radius. n e α (1) (14. in the case of uniform plasma density NL = NL (r =R∗ ) = 4π 3 2 (2) r n α . n H . recombine after an interval τion = 1 . of course. that o rS = • 3NL 4πn 2 α (2) e 1/3 . n H ne = np. n H K (H) (14.

1 Ionization balance of a dusty HII regions The presence of dust reduces the number of Lyc photons available for ionization of the gas and.32) 4πα (2) nr 2 Nc 1 d Nc =− − .HII regions 453 • This is the characteristic time for recombination and ionization and also the time an HII region would go on shining after the star were switched off. The dusty HII region extends to the dusty Str¨ mgren radius rdS . If NL. (14.40) which gives Nc (τ ) directly as a function of the dust optical depth τ = r n K d . NL dτ NL K d NL Inserting rS from (14.He depends sensitively on the stellar effective temperature: For B0 stars.1). the Str¨ mgren radius rS (He) for helium becomes o rS (He) = 3NL.He is the number of Lyc photons with hν > χ(He).38) The flux NL.37) implies τion ∼ 104 yr. rS (He) = rS (H). the HeII zone is practically absent. the size of an HII region.3. So dτ = n K d dr . it competes with hydrogen for photon capture. Its Lα ˚ radiation from the singlet system of para–helium at λ = 584 A is also able to ionize hydrogen.3 Dusty HII regions 14.He 4πn e n(He+ )α (2) (He) 1/3 .4. As the degree of ionization within an HII region is close to one. where all Lyman o .4. Let τ be the optical depth in the Lyman continuum with respect to dust only.39) with the boundary condition Nc (τ =0) = NL . therefore.6 eV). 14. and K d the corresponding dust cross section per H atom. where n = n e + n H and helium has been negelected. Because helium ionization requires more energetic photons (χ(He) = 24. the HeII region is never greater than the HII region. it offers the analytical solution [Pet72] 3τ 2 − 6τ + 6(1 − e−τ ) Nc = e−τ − NL n K d rS (14. we get from (14. The equation may be solved numerically but if the density n is constant. When helium is present in the gas (typical galactic abundance by number [He]/[H] 0. For a plasma of moderate density (n e = 10 cm−3 ).33) leads to the ordinary differential equation 3 1 d Nc Nc =− τ2 − 3 K 3r 3 NL dτ NL n d S (14. equation (14. at best co-extensive. for stars of spectral type O6 or earlier.6 eV > χ(H) = 13.


Diffuse matter in the Milky Way

continuum photons have been absorbed either by gas or by dust. One finds rdS by putting (14.40) to zero and solving for τ . This yields the optical thickness τdS = rdS n K of a dusty HII region and thus rdS . For example, a region with optical depth τdS = 1 has a 56% smaller volume than one without dust where τdS = 0. There are no observations of the extinction curve in the Lyman continuum, ˚ the dust cross section K d is, therefore, quite uncertain for λ < 912 A. Of course, we have some guidance from the dielectric permeability ε of figure 7.19. According to it, when applying Mie theory, the cross section of an individual ˚ grain is at λ = 1000 A three times greater than in the visual and continues to rise ˚ towards shorter wavelengths up to a resonance at ∼700 A. 4 cm−3 ) around an O6 star (N For a compact HII region (n = 10 L = o 1049 s−1 ) the dust-free Str¨ mgren radius rS = 4.5 × 1017 cm. If the dust in the HII region had the same abundance and properties as in the diffuse ˚ interstellar medium, its cross section per hydrogen atom at 1000 A would be Kd 1.5 × 10−21 cm2 , and in the Lyman continuum greater still. This would imply a huge optical depth (τ = r n K d ∼ 10) so that the HII region would almost vanish. However, compact HII regions are being observed. Consequently, the dust must be modified, most likely it is depleted, partially destroyed by the strong and hard far UV field. In HII regions, dust absorption coefficients per gram of interstellar matter are probably reduced by a factor of four with respect to neutral gas clouds. The dust in an HII region is heated radiatively and its temperature Td is decoupled from the gas temperature. The radiation field Jν which the grains absorb includes stellar light, Lα photons and optical cooling lines. Td then follows from (8.7). In a compact HII region, more than two-thirds of the radiation is emitted by dust (disregarding the surrounding shell of neutral gas). Lyα photons of hydrogen Each Lyman continuum photon ionizing a hydrogen atom leads inevitably to a proton–electron recombination. No matter how the newly formed hydrogen atom cascades down from its excited state, eventually it produces a Lyα photon of energy hνα = 10.2 eV. Like all photons connecting to the ground state, they are trapped inside the HII region. Therefore their radiation density is very high. But it cannot build up indefinitely because of the following three destruction routes: • Absorption of Lyα photons on grains. This is by far the most important process and works independently of the amount of dust in the HII region. If there is very little dust because most has been destroyed, the few remaining grains receive all the Lyα flux and will be very hot. They absorb the fraction NL hνα /L ∗ of the stellar luminosity L ∗ , typically 25%. This energy then escapes as infrared radiation. If the HII region is not almost dust-free and the dust optical depth τdS not small, the infrared luminosity (without cooling

HII regions


fine structure lines) will be even greater than ∼25%. In addition, the grains then also directly absorb stellar photons of all wavelengths, as well as diffuse photons of the cooling lines. Decay of the metastable upper level 2 2 S (n = 2, l = 0, lifetime 0.12 s) into the ground state 1 2 S by emission of two photons (see end of section of total energy hνα . Averaged over many transitions, the decay forms a continuum. Diffusion of Lα photons into the optically thin line wings from where they can escape the HII region.

14.4.4 Bremsstrahlung In a hydrogen plasma of temperature T and electron density n e , the electrons have a mean velocity proportional to T 1/2 . The protons are much heavier and, therefore, almost at rest. As an electron passes a proton, it is accelerated and changes its direction, usually only a little. The acceleration of the electron leads after the dipole formula (1.97) to electromagnetic radiation. • The monochromatic emission coefficient is (without proof)


8 3

2πm e e6 gff n 2 e−hν/ kT . e 3kT m 2 c3 e


The emission spectrum is at low frequencies flat,

5.44 × 10−39 n 2 T −1/2 erg cm−3 s−1 Hz−1 ster−1 e

because exp(−hν/kT ) 1 and the Gaunt factor gff 1. In a plasma at 8000 K, the Boltzmann factor e−hν/ kT becomes important only at wavelengths λ < 10 µm. Denoting by gff the frequency average of the Gaunt factor, the total ¯ bremsstrahlung emission is 4π

dν =

2πkT 32πe6 gff n e n p ¯ m e 3hm e c3 √ 1.42 × 10−27 gff n 2 T erg cm−3 s−1 . ¯ e


The continuum absorption coefficient of the plasma follows from Kirchhoff’s law. At radio wavelengths Kν = Bν (T )

0.173 1 + 0.13 log

T 3/2 ν

nenp ν2



implying a continuum optical depth τc ∝ ν −2 . Defining the emission measure by EM = n 2 ds e (14.44)


Diffuse matter in the Milky Way
τc can be approximated by [Alt60] τc = 0.082a(ν, T )T −1.35 ν GHz



where the factor a(ν, T ) 1 and the emission measure is in the unit pc cm−6 . Below a turnover frequency νturn , the HII region is opaque, above transparent. When it is opaque, one receives a flux Sν that is proportional to the Planck function, Sν ∝ Bν ∝ ν 2 ; when it is transparent, Sν ∝ τν Bν ∝ ν −0.1 . Bright and compact HII regions have typical emission measures 107 pc cm−6 . At T = 8000 K, they become optically thin at frequencies above νturn ∼ 2 GHz. Measuring the integrated radio flux Sν of an optically thin source allows to determine the number of stellar ionizing photons Nc , Nc = 7.54 × 1046 ν GHz

T 104 K


D kpc


Sν Jy

s−1 . (14.46)

One can then infer the spectral type of the exciting star (table 14.3). In case of a compact source, one has to make allowance for the absorption of Lyc photons by dust. Because the integrated radio flux is Sν ∝ nenp d V

one can also derive from Sν the mean electron density if the volume V is estimated from the linear size of the HII region. A young HII region appears, first of all, as an infrared source whose spectral energy distribution is dominated by dust and peaks typically at 60 µm. With the help of table 14.3 or equation (14.22), one can predict from (14.46) the radio flux Sν that a star of certain stellar type or mass will produce. Because the dust emission drops with frequency like ν 4 , whereas the radio spectrum is flat, bremsstrahlung exceeds dust emission at long long wavelengths. The cross-over is usually at λ ∼ 3 mm.

14.4.5 Recombination lines Important information on HII regions is provided by recombination lines. They arise when electrons recombine with protons, or other ions, and cascade downwards. For hydrogen-like atoms, the frequencies follow from the Rydberg formula (n > m) 1 1 − 2 . ν = cRZ2 2 m n Z is the electric charge of the atomic nucleus (Z = 1 for H) and R the Rydberg constant (for infinite nuclear mass, R = 109 737.3 cm−1 ). Transitions

Mass estimates of interstellar clouds


n → n − 1 are called α-lines, n → n − 2 are β-lines, and so forth. Because the recombination and Einstein coefficients are known, the strength of the lines yields directly the number of ionizing photons in an HII region. Optical lines (Balmer lines, like Hα) can be very bright but in order to be interpreted they have to be corrected for dust extinction. De-reddening by means of an extinction curve (figure 7.8) requires observations of several lines at different wavelengths. The procedure is often ambiguous because one does not know how much dust resides within the HII region and how much is in the foreground. The problem is less severe in the near infrared (Brackett and higherorder lines). Interstellar dust extinction is irrelevant for transitions between high quantum states. Such radio recombination lines can be detected from very obscured and optically not accessible places. When combining the radial velocity with the rotation curve of the Milky Way, one obtains the (kinematic) distance of the source. The lines thus reveal the spiral structure of the Milky Way as traced by bright star-forming regions. We add three other interesting features of radio recombination lines: • As the Einstein coefficients are small, the levels have a long lifetime and they are reshuffled by collisions leading to a population that is close to thermodynamic equilibrium, at the temperature of the ionized gas (see (5.40)). Because electrons with high quantum numbers are far from the atomic nucleus, nuclei of the same charge but of different mass produce only slightly shifted frequencies. For example, the relative difference ν/ν between H und He radio recombination lines is 4.1 × 10−4 which allows us to observe both lines in one measurement, thus enhancing the accuracy of line and abundance ratios. The highest transitions observed are around n ∼ 700. The atomic radius r ˚ increases with quantum number like r = a0 n 2 (Bohr radius a0 = 0.529 A) and such highly excited atoms have macroscopic sizes (> 1 µm).

14.5 Mass estimates of interstellar clouds
The amount of interstellar gas, Mgas , that an astronomical object contains, be it a galaxy, a molecular cloud or a stellar disk, is one of its prime parameters, like luminosity or size. We summarize a few methods for estimating Mgas . None of them is very precise, so one employs as many methods as possible and takes an average. 14.5.1 From optically thin CO lines Consider rotational CO transitions j + 1 → j in a homogeneous cloud of temperature T and column density N j of molecules in level j . If (ν) is the


Diffuse matter in the Milky Way

Gaussian line profile from (13.59), the optical thickness τν becomes τν = N j g j +1 c2 b j +1 − hν A e kT 1− 2 j +1, j g j 8πν bj (ν) (14.47)

where b j = N j /N ∗ denotes the deviation of N j from the LTE value N ∗ (see j j (13.69)). Let ν = 2H be the half-width in frequency of the absorption profile and τc the optical depth in the line center. One then finds that in LTE (b j = b j +1 = 1), the column density of the upper level is given by N j +1 = τc hν/ kT e 3h v − 1 16π 3 µ2 π 2j +3 . ln 2 j + 1

µ is the dipole moment of the molecule and v = c ν/ν. The optical depth in 1) but for the isotope C18 O, the center of millimeter CO lines is always high (τc τc is often below one. Then v is also the observed linewidth and τc follows from the line temperature TL = τc T . If v is given in cm s−1 N j +1 = 6.81 × 109 v 2j +3 TL hν/ kT − 1 j + 1 T e cm−2 .

The column density of the upper level, N j +1 , is roughly proportional to the line area TL v. It depends only weakly on the kinetic gas temperature T and can be converted into the total CO column density N(CO) with the help of the Boltzmann distribution (5.8) and from there to the molecular hydrogen column density, N(H2 ), assuming the standard ratio [H2 ]/[CO] = 104 . 14.5.2 From the CO luminosity Most molecular clouds do not collapse under their own gravity but seem to be virialized. Their typical internal systematic velocity v is then determined by the cloud mass M (section 15.2.1). If R denotes the cloud radius, v The CO luminosity is defined by L CO = Abeam

G M/R.

Tmb (v) dv


where Abeam is the area of the beam at the distance of the source and Tmb is the observed main-beam-brightness temperature (section 13.1.4) of a CO rotational line, usually the (1–0) or (2–1) transition. It is customary to measure L CO in pc2 K km s−1 . The integral in (14.48) extends over the whole line. As long as the source is smaller than the beam, the product Abeam Tmb and thus L CO is constant,

Mass estimates of interstellar clouds


independent of the beam size. If the mass fraction of stars within the cloud is negligible such that M Mgas , then T L CO ∝ √ Mgas n (14.49)

where T is the cloud temperature and n the number density of the gas. If one √ assumes that the ratio T / n is constant for all clouds, the CO luminosity is directly proportional to the cloud mass, although the CO line may be optically very thick. Of course, the relation (14.49) must be calibrated. The assumed √ constancy of T / n implies that the gas temperature goes up when a cloud becomes more compact. By and large, this may be so, possibly as a result of increased star formation in dense regions but it is certainly not a rule. Nevertheless, the method seems to work, at least, better than we understand. The CO luminosity is also used to derive gas masses of galaxies. Gas motions there are large, of order 200 km s−1 , comparable to the rotational velocity of the Milky Way. The gas kinematics are entirely determined by the gravitational potential of the stars, and not by the gas. Nevertheless, equation (14.49) is applicable. When one makes an observation towards the center of a galaxy, the CO line will have a width of the same order, say, 200 km s−1 . As the beam usually covers a linear scale of a few hundred pc or more, at the distance of the galaxy, it samples many giant molecular clouds. Each GMC produces a relatively narrow line, say, of 10 km s−1 width. The observed line is then a superposition of many sublines that are well separated, either spatially because they come from different locations in the galaxy or in frequency space because they have different Doppler velocities. The contributions from the GMCs thus add up linearly to give the total gas mass of the galaxy. In the CO (2–1) line, for example, an average over a large sample of spiral galaxies yields L CO ∼ 4Mgas when Mgas is measured in M and L CO in pc2 K km s−1 . Of course, the masses from the CO luminosity refer only to the molecular gas. How to estimate column densities of the HI gas was discussed in section 14.3.4. Masses from dust emission (see later) include both components, HI and H2 . 14.5.3 From dust emission Because the dust absorption coefficient K λ drops quickly with wavelength (figure 12.10), almost all interstellar clouds are transparent in the submillimeter and millimeter region. The observed flux Sλ from the dust is then directly proportional to the dust mass Md . If the dust temperature Td is uniform over the source, K λ Bλ (Td )Md . (14.50) Sλ = D2 The distance D is usually known, Td has to be derived as a color temperature from the fluxes at, at least, two wavelengths (see (8.20) and discussion thereafter)


Diffuse matter in the Milky Way

or from molecular line data. At long wavelengths, when hc/λkT 1, the Planck function Bλ (Td ) is not very sensitive to the exact value of Td . A favourite observing wavelength is λ = 1.3 mm because the atmosphere there has a window of good transmission. The critical quantity in (14.50) is the absorption coefficient per gram of interstellar dust, K λ . At λ = 1.3 mm, one is in the Rayleigh limit (grain radius a λ), and the grain size distribution is fortunately irrelevant. One may, therefore, use the absorption efficiency (3.3), Q abs = λ 6nk 8πa . λ (n 2 − k 2 + 2)2 + 4n 2 k 2

In our dust model, whose optical constants are displayed in figure 7.19 and which is detailed in section 12.4, bare silicates have at this wavelength (n, k) (3.4, 0.04) and amorphous carbon (aC) has (n, k) (12, 4). These numbers imply Q 1.3mm = 0.86a 2.13a for silicate dust for amorphous carbon dust (14.51)

when a is measured in cm. The mass absorption coefficient per gram of silicate or carbon dust becomes K 1.3mm = 3πa 2 Q 1.3mm = 4πa 3ρ 0.26 cm2 0.64 cm2 per g of Si per g of aC . (14.52)

For the material density of the dust, we put for both components ρ = 2.5 g cm−3 . With a dust-to-gas mass ratio in the diffuse medium of Rd = 1 : 140, one obtains K 1.3mm = 0.0026 cm2 0.41 cm2 per g of interstellar matter per g of dust . (14.53)

Equation(14.53) may be extrapolated to far infrared wavelengths by assuming Q λ ∝ λ2 , K λ = 4.4 × 10−5 λ cm

cm2 per g of interstellar matter.


Fluffiness of the grains as well as ice mantles tend to increase K λ , so that in dense and cold clouds (see figure 12.10) it may be five to ten times bigger than that given in (14.54).

Chapter 15 Stars and their formation

To get an idea of how the elements that make up interstellar dust are created, section 15.1 deals with nulear burning, in particular, with the pp chain and the 3α process. We then present approximate formulae for the luminosity and lifetime of individual stars, discuss the initial mass f unction (IMF) which specifies how masses are distributed when stars are born, and show averages for star clusters displaying an IMF. In section 15.2, we delineate the structure of cloud clumps at the verge of gravitational instability and present Jeans’s criterion for collapse to actually occur. In section 15.3, we numerically analyse the equations for spherical, isothermal collapse and derive, for the earliest stages of protostellar evolution, the approximate density and velocity structure of the protostellar clump. Section 15.4 treats disks which inevitably form as the result of cloud rotation and which have a strong effect on the appearance and dynamical evolution of protostars, to say nothing about planet formation. The results of section 15.3 and 15.4 serve as the basis for models of dust and molecular line emission in chapter 16.

15.1 Stars on and beyond the main sequence
15.1.1 Nuclear burning and the creation of elements Stars form the most obvious component of the Milky Way. They are, by and large, the ultimate source of luminosity in galaxies. The energy is generated in the stellar cores by fusion of atomic nuclei. During the main-sequence phase, hydrogen is burnt into helium. In the post-main-sequence phase, when hydrogen has been exhausted, elements of atomic mass number greater than four are produced by compounding helium nuclei. The energy yield per gram of matter is then one order of magnitude smaller. Fusion ends with the generation of iron which has the lowest binding energy per nucleon; heavier nuclei release energy not in fusion but in fission. An example is uranium 235 U where fission, here on Earth, has very diverse technical applications: to supply electricity to our homes, or to remove 10 000 homes in one blow. 461


Stars and their formation

Table 15.1. Standard solar abuncance of selected elements and their origin from [And89] and [Tri89]. Element H He C N O Ne Mg Al Si S Fe Relative abundance 1 0.10 3.6 × 10−4 1.1 × 10−4 8.5 × 10−4 1.2 × 10−4 3.9 × 10−5 3.0 × 10−6 3.6 × 10−5 1.9 × 10−5 3.2 × 10−5 Origin Big bang Big bang, H-burning He-burning From carbon during H-burning He-burning C-burning Ne-burning Ne-burning O-burning O-burning Pre-supernovae

Nuclear burning not only supplies energy but also creates elements. We are mostly interested in the origin and abundance of carbon, nitrogen, oxygen, magnesium, silicon and iron as they are, together with hydrogen, the building blocks of interstellar dust and constitute the most frequently observed molecules. They are synthesized during the major burning phases and that qualitatively explains why they are so common. Estimates of their abundances are summarized in table 15.1, the burning phases are described later. Nuclear astrophysics is the branch where the synthesis of elements is studied. According to standard cosmological theories, none of the atoms heavier than helium can have been created in the Big Bang in any relevant amount. So they were made afterwards. But obviously, not all elements can have been synthesized by α-particles in nuclear burning. Nor does adding a proton to a helium core and then another and another usually work in building up heavier nuclei. For example, helium has more binding energy per nucleon than its heavier neighbors in the Periodic Table: lithium, beryllium and boron. So the latter three cannot be fused in this way. Instead, the light elements are probably made in cosmic-ray spallations. Most other elements (besides C, N, O, Ne, Mg, Si) and the plethora of isotopes are built by neutron capture. Neutrons are made in various side-branch reactions. They do not feel the Coulomb repulsion and easily penetrate an atomic nucleus. The resulting new atom can be a stable isotope or be unstable to β-decay. • • When the rate of neutron production and capture is slow compared to β-decay, as during the quiescent post-main-sequence burning stages, one speaks of s–processes. When it is rapid, as in supernova explosions and some late stellar phases,

Stars on and beyond the main sequence


more than one neutron is captured before the nulceus β-decays. In this case, one speaks of r–processes. 15.1.2 The binding energy of an atomic nucleus To understand nuclear burning and the resulting element abundances, we must know how tightly atomic nuclei are bound. The approximate energy of a nucleus containing N neutrons of mass m n and Z protons of mass m p , altogether A=N+Z particles, can be derived from the liquid drop model where one considers the binding of the nucleons in analogy to the chemical binding of molecules in a water drop (section 9.5). According to this model, the density in the nucleus is constant and there is mutual attraction only between neighboring particles. The total energy of the atomic nucleus consists of the energy equivalent of the mass of the free particles minus the binding energy E b > 0, W = Z m p c2 + Nm n c2 − E b . For A > 4, approximately Eb = α A − β A 3 − γ


δ e2 Z (Z − 1) (N − Z )2 ± . −η 1/3 A A A


The binding energy per nucleon, E b /A, is plotted in figure 15.1 as a function of atomic number Z . We discuss the individual terms in (15.2): • • The first is due to the strong force and its value per nucleon is constant, equal to α. The second term is a correction to the first because some nucleons, their number being proportional to A2/3, are on the surface of the nucleus and have fewer neighbors. It produces in figure 15.1 the decline at small Z . Term 1 and 2 correspond to equation (9.58) which gives the energy of an ordinary liquid drop. Term 3 comes from the Coulomb repulsion which acts not only between neighboring particles but everywhere and here over a mean distance proportional to the radius of the nucleus, or to A1/3. This also brings E b /A down at large Z so that the function E b /A obtains a maximum at intermediate Z . Term 4 follows from the Pauli principle and very much improves the approximation. When it is included, E b /A peaks near iron ( A = 56) at about 8.8 MeV. The term is zero if there is an equal number of neutrons and protons, N = Z . To derive it, one assumes that the energy levels available to the nucleons have a constant separation and that the same level can be occupied by a neutron–proton pair but not by two neutrons. It is then easy to


Stars and their formation

Figure 15.1. Binding energy E b /A per nucleon of the most abundant isotope (but including deuterium) of each element. Filled circles are derived from mass excess as tabulated by [Lan80] and are connected by a full line. The dotted curve is a fit after equation (15.1). It is quite satisfactory all the way from uranium down to lithium, except for nitrogen. Parameter values are given in the text. The insert is merely a magnification.

show that the binding energy decreases proportionally to (N − Z )2 when N > Z. The final term is relatively small and takes into account the effect of pairing among nucleons. It is positive when N and Z are both even, negative when both are odd, and zero otherwise. It introduces the zigzag and the local peaks when an atom consists of a multiple of α-particles (4 He, 12 C, 16 O, 20 Ne, 24 Mg, 28 Si, . . .).

The coefficients in equation (15.2) are determined by a combination of theory and experiment and are listed in the special literature. We used, in figure 15.1, α = 15.7, β = 17.8, γ = 0.71, η = 23.6 and δ = 132 MeV, with no attempt for optimization.

5 × 107 K. Because this process is at work in the Sun. the average kinetic energy of atoms is only 2 keV or a thousand times smaller.442 MeV include the annihilation energy of the positron e+ and the energy of the electron–neutrino νe .1. we discuss a few details.12 × 10−3 . The chain consists of three major steps: In the first step. The potential barrier between two protons separated by a distance r = 10−13 cm is E b = e2 /r ∼ 1.1 Energy yield in hydrogen burning 465 When stars are on the main sequence. The strong force is insensitive to the electric charge of the nucleons. PMaxw (v). The positrons unite with electrons and are annihilated producing γ -rays.73 MeV released in reaction (15. All γ -photons collide with other particles and deposit their energy in the form of heat. In other words.3.1.3) appear in the form of γ -rays.5 MeV. The strong force has a range of order ∼10−13 cm. it binds the nucleons but is not felt outside the nucleus. of finding in the tail of the Maxwellian . A helium nucleus (mass m He2+ ) is a bit lighter than four protons. 41 H −→ 4 He. The energy of 0.282 × 10−5 erg = 26. excluding gravitational collapse into a black hole. they burn hydrogen by melting four protons into helium. helium is created in the proton–proton or pp chain.2 The proton–proton chain In all low-mass stars. (15.Stars on and beyond the main sequence 15.3 Hydrogen burning 15. 15. neutrinos. Usually the Coulomb force prevents protons from coming close together but at very short distances the strong nuclear force sets in and overcomes electric repulsion.1.442 MeV. The chance. so it does not discern between protons and neutrons. The 26. The fusion of one helium nucleus thus liberates mc2 = 4.3) From burning 1 g of hydrogen into helium one receives 6. ∼0.263 MeV per fused He nucleus they carry away is lost. deuterium is generated according to 1 H + 1H −→ 2 D + e+ + νe + 1. (15. up to masses a little more than 1 M . positrons and as the kinetic energy of the reaction products.4) The 1. which has a central temperature of 1.3.40 × 1018 erg and this provides the greatest energy reservoir of a star. The mass difference m = m He2+ − 4m p is such that m/4m p = 7.73 MeV.7% of the mass is converted into energy. The neutrinos have an extremely small cross section and escape from the burning core without interacting with the stellar material. In the Sun.

up. down.4)).493 MeV.4) proceeds very slowly. the complex is bound. It is so much faster because it requires only the electromagnetic and not the weak force. a deuteron. using calculus. far beyond the range of the strong force. After the tunneling. As the protons get faster. The penetration probability. The reason is that the potential of the strong force depends on the spin direction.466 Stars and their formation velocity distribution atoms with a kinetic energy of 1. When the spins of the two nucleons are anti-parallel. Ppen . The photon energy is accounted for in the 5.6) . Ptun (v) increases. It is. A proton in the Sun has to wait ∼1010 yr before it is incorporated into a deuterium atom. Because the conversion of a proton into a neutron must take place in the short time while the proton traverses a distance of ∼10−13 cm. When we evaluate the velocity where Ppen (v) has its maximum. The appearance of a γ –quantum is necessary for the conservation of energy and momentum. a neutron of one up and two down quarks. Only then does one have a stable system. the reaction (15.23 MeV. is given by Ppen = PMaxw (v) · Ptun (v). We can estimate the tunneling probability Ptun (v) for protons of velocity v from (6.493 MeV (15. One of the protons must also be converted into a neutron (see reaction (15. The next step.859 MeV. however. the system is unbound. 2 D + 1H −→ 3 He + γ + 5. It is mediated by the weak nuclear force which has an extremely short range (∼10−15 cm) and is 1012 times weaker than the electromagnetic force. this state is excluded because the two protons are identical and as fermions their quantum numbers must not coincide. although this formula applies strictly only to square potentials in one dimension. we find that tunneling already occurs at a distance L ∼ 5 × 10−12 cm. there is an ephemeral pp complex but it is not stable: 2 He does not exist. Because a proton consists of two up and one down quark.5 MeV is negligible (see (5. therefore.57). with a binding energy of 2. The distance L at which tunneling sets in follows from 1 2 2 2 m p v = e /L. The only way protons can unite is by tunneling through the Coulomb potential barrier. (15.13)). when they are parallel. there is a transformation p = (up. down) −→ n = (up. not enough to have the two protons approach each other via tunneling to a nuclear distance. In the final step the build–up is completed by 3 He + 3 He −→ 4 He + 2 H1 + 12. while PMaxw (v) decreases. down).5) follows only seconds later.

.1. Be. In this way. symbolically 12 C −→ p 13 N −→ β 13 C −→ p 14 N −→ p 15 O −→ β 15 N −→ p 12 C + 4 He. Therefore.1. β stands for β-decay. then another α-particle is added to transform beryllium into carbon.7) The letter p above an arrow indicates proton capture. α+α −→ 8 Be + γ − 92 keV . electron and anti– neutrino.6) involving the elements Li. one builds up nuclei of subsequently higher mass number but eventually ends up again with a 12 C atom plus an α-particle. the temperatures have to be above 108 K. The beryllium atom resulting from the first step. The elements C. so the Coulomb repulsion is strong and as they are also rare. The 3 He nuclei are doubly charged. First. by far the slowest step being the formation of 15 O from 14 N which takes about 300 million years. It is made in red giants by fusing three α-particles.3 s). (15. and then another and another. equivalently. The CNO cycle dominates when the central stellar temperature is higer than ∼18 million K or. when the stellar mass is slightly above 1 M . The proton captures. The seemingly simple reaction 3 He + p → 4 He is irrelevant because it involves the weak force.4 The 3α process The next element. the reaction is fairly slow with a time scale of order 106 yr. 14 N will build up until an equilibrium abundance is reached. Indeed. Because of the strong Coulomb repulsion between α-particles.. N and O that give the burning its name take part in the reactions only as catalysts: in equilibrium. In the sun there are other channels besides (15. the cycle can only operate if initially one of the catalysts is present. Of course.Stars on and beyond the main sequence 467 For this reaction to proceed. however. The CNO cycle requires higher temperatures than the pp cycle because the Coulomb barriers are higher.3 The CNO cycle In all massive stars. there is an alternative mode of hydrogen burning: the CNO cycle.7) proceed quickly (102. The net result of the reaction chain is again the conversion of hydrogen into helium. their abundance does not change. the CNO cycle is the major source for nitrogen in the universe. It may be started by adding a proton to a 12 C atom. on about the same time scale that a free neutron disintegrates into a proton. 15. the preceding two must have occurred twice. The burning process consists of two steps. two α-particles unite to 8 Be. even if ones starts out with no nitrogen at all.3. after helium. B but they are far less common. are much slower. 15. It is also more sensitive to temperature: the energy production rate is at 18 × 106 K roughly proportional to T 18 versus T 4 in the pp chain. The β-decay reactions in (15. synthesized in stars by burning is carbon.

8) When the temperature in the stellar core reaches 1. unstable and α-decays.38). This makes a tremendous difference considering the extreme temperature sensitivity of the 3α process. and a temperature rise of only 5% makes the reaction ten times faster.65 MeV above the ground level but only E R = 0. Once formed. the stellar interior would have to be as hot as 2 × 108 K for helium to burn into carbon. 8 Be + α −→ 12 C + γ + 7. The formation of 12 C from the unstable compound (8 Be + α) proceeds through the resonant state C∗ . more frequent).8). C∗ . however. which lies = 7. For those atoms whose kinetic energy is close to E R . It allows the triple-α process to operate already at a temperature of 1. This reaction is endothermic requiring Q = 92 keV.37 MeV.2. the energy production rate is proportional to T 30 . in the form of 8 Be.2 × 108 K. The existence of the state C∗ speeds up the reaction 8 Be + α →12 C by a factor 107 .2).2 × 108 K.468 Stars and their formation C* 0. the cross section for creating C∗ has a sharp maximum (resonant reaction). the average 15 keV and in thermal equilibrium about kinetic energy per particle is 3 kT 2 1 out of 1010 atoms will be on the right-hand side of (15.65 MeV 12 C Figure 15. E∗ . is. The small fraction of beryllium nuclei reacts further to form carbon. (15. it goes in both ways. Without this. C∗ may then de-excite to the ground state 12 C (α-decay is. however. so we may write 2α ←→ 8 Be. At ∼108 K. Now it is important that carbon possesses an excited state. This follows from the Saha formula when one makes some obvious modifications in (5.288 Me above the complex 8 Be + α (figure 15.288 MeV 8 Be + α 7. The energy yield in the 3α process per gram of matter is ten times lower than in hydrogen burning.

They are based on computations of stellar evolution and adapted from [Mas88]. Their lifetime τ and luminosity L determine. and second. assuming that 20% of the mass of the Sun is in its hydrogen burning core and that the energy yield is 0. Most of this time. m.9)–(15. F. The stellar mass m is not an observational quantity and much less certain. the upper limit being rather uncertain. despite the fact that the post-main-sequence phase is short.5 Lifetime and luminosity of stars Stars are born with masses. because giants are so bright. the luminosity averaged over the total lifetime of a star. A. K. τms . the yield decreases as heavier elements are synthesized.16 MeV + 9. as a function of the stellar mass m. Formulae (15. . Stars are best classified by their line spectrum. and the luminosity the speed of its consumption. because 90% of the fuel reserve has already been used up.98 MeV.007c2 erg per gram of burnt matter (see (15.08 and 100 M . The subsequent evolution is very quick: first.32 MeV + 9. m .1. so the consumption rate is higher. a star spends on or near the main sequence. B. Therefore. between 0. Here are some reactions with their energy yield: 12 C + α 16 O 20 Ne −→ −→ −→ −→ 16 O + γ + 7. oxygen. However. because the star has become a giant and is generally much more luminous.3)). L ms and L av .13) and figure 15. and the luminosity class (roman numbers from I to VI) or surface gravity. 15. the total lifetime is not much longer than the main sequence lifetime. magnesium and so forth can be generated through the capture of α-particles. The analysis yields the spectral type (O. can be several times greater than the main sequence luminosity L ms . the stellar population and the optical and infrared appearance of the Milky Way. τ ∝ L The stellar mass fixes the amount of fuel. G. one gets a lifetime of 20 billion years. L av . to a large degree. It can be directly determined only in binary systems.Stars on and beyond the main sequence 469 Once carbon is present. neon.1 are approximations of the parameters τms . M) or effective temperature Teff .73 MeV 24 Mg + γ 28 Si + γ 24 Mg + α When taken per gram of stellar matter. The initial chemical composition of the star as well as its mass loss rate have some influence on the results but it is not important for our purposes. not bad as a first estimate. To first order. +α +α 20 Ne + γ + 4. For example.

6 ≤ m ≤ 1 m ≤ 0.1 × 1010 m −3. It is a white dwarf when the initial mass 0. almost simultaneously. log L ms = 4. .616 log m − 0. .e. (15. τms (15.2)1.65 ≤ m ≤ 2. L ms = • 1. m+dm. The constant 9.594 . If d N(m) stars are born in the mass range m .06313 log3 m.85 τms = 107 + 5 × 109m −2. (15.6557 log2 m − 0. 15. as in Orion. (15. stellar masses in M and luminosities in L .51m 3 0. then d N(m) ∝ ξ(m) dm (15.2 ≤ m ≤ m SN  0.77 × 1010 in (15.7  1. A fundamental problem concerns the distribution of masses of young stars. i.12) L av is expressed through the mass m rem of the stellar remnant.012m 4.2 m rem = (15.7 of the heavy elements (the ashes). It is a neutron star or a black hole when the progenitor explodes as a supernova.293 m < 2.4 0. either born in clusters.34 0.5 Main sequence luminosity: For m ≥ 1 M . one gets the following results • Main sequence lifetime:   3 × 106 + 1.65 ≤ m ≤ m SN :  m ≤ 0.989 × 10−3 + 4.14) .11) m > 7.2 × 109 m −1.77 × 1010 max m rem .105(m − 2.12) assumes an energy yield of 6 × 1018 erg per gram of burnt remnant matter with a mass fraction of 0.65 0. or elsewhere.10) For m < 1 M .6 The initial mass function Stars generally form in clusters. when the initial mass m > m SN = 6 M .594 ≤ m ≤ 7.65 m  0.13) 2.470 Stars and their formation When lifetimes are expressed in years.6.65  1.1m m > m SN .1.4 + 0. L ms .293 2. An outstanding example is the very compact cluster in Orion around the Trapezium which contains over 2000 members with a stellar density of more than 104 pc−3 .9) • Average luminosity (mean over lifetime of star including giant phase): L av = 9.

A much more direct approach to find ξ(m) is to count in a young stellar cluster. refer to 1 M of gas that has transformed into stars. for example. Mil78]. One way to derive the IMF observationally is to count the number of stars in the solar neighborhood per interval of apparent visual magnitude m V . . all of equal mass m and with coordinates x j . Also. . however. the number of stars per mass interval.3). m l and m u . Then one determines their distances to obtain the luminosity function φ(MV ). An outer pressure P0 is exerted on the cloud surface.206 ln2 m) m Salpeter IMF Miller–Scalo IMF.2. respectively but these numbers are debatable and depend on the environment.16) .1 Virialized clouds Consider an interstellar cloud of volume V consisting of N atoms. N (15.15) The functions are not normalized. m + dm.Clouds near gravitational equilibrium 471 defines the initial mass f unction ξ(m). Two analytical expressions for the IMF are in wide use [Sal55. prone to observational bias by the detection limit as young clusters are always heavily obscured by dust. When one restricts the counting to main sequence stars. MV can be uniquely translated into a stellar mass m (see figure 15.15). ¨ F j = mx j i = 1. Figure 15. where no member has yet evolved off the main sequence. . one often assumes 0.966 ln m − 0. Such a method is. The virial theorem asserts a relation between the various forms of energy when the cloud is in equilibrium so that all time derivatives with respect to the cloud as a whole but not with respect to an individual particle. the result from any one cluster is not representative for the Milky Way as a whole.4 illustrates how stars of different masses contribute to the total luminosity and thermal radio emission of a cluster with mass distributions as given in (15. ξ(m) = m −2. (15. 15. abbreviated IMF. It gives the number of stars per unit volume and interval of absolute visual magnitude. For the lower and upper mass limit of stars. .2 Clouds near gravitational equilibrium 15. vanish. This leads to a function ψ(m) such that ψ(m) dm is the present stellar density for the mass interval m . One may alternatively try to determine ξ(m) theoretically by making certain assumptions about the processes during star birth. The theorem follows immediately from the equation of motion. they may.1 M and 50 M . . One arrives at the initial mass function ξ(m) by deconvolving ψ(m) with the stellar lifetime τ (m) (see again figure 15.3) and assuming that the star-formation rate has been constant over the past. .35 1 exp(−0.

They also live much longer than 1010 yr. The latter never turn into giants. The former move after the exhaustion of hydrogen in the Hertzsprung–Russell diagram horizontally to the right. without increasing their luminosity. so their post-main-sequence evolution does not bother us. when one multplies by x j and sums over all particles. For stars of very high and very low mass.10) to (15. L ms L av . Because ¨ x·x = d ˙ ˙ (x · x) − x2 dt one finds for the time averages (we do not mark them explicitly) Fjxj = − j j ˙j m x2 . The top panel shows the lifetime of stars according to equations (15. The bottom one their luminosity after (15.13).9). dotted line shows L av which is an average over the whole stellar life including the giant phase.3. (15.472 Stars and their formation Figure 15.17) . The full curve gives main sequence luminosity L ms .

the gravitational (potential) energy E grav . is from (15. although they comprise 90% of the stellar mass. The other sum F j x j can be split into a term containing only those forces F j which act from outside. the other containing the inner forces. ˙j Without systematic motion. |xi − x j | An interstellar cloud fulfilling (15.Clouds near gravitational equilibrium 473 Figure 15. The end result links the thermal energy E therm = 3 NkT . In other words.18) is said to be virialized.10) and (15. Virialization is a special cloud property. L(m). and P0 V .4. one-tenth of the stars by mass accounts for almost all the luminosity (precisely. The inner forces come from mutual gravitational attraction of the atoms. To evaluate this sum. The dependence of the luminosity on mass. For a spherical cloud of uniform density. 2 P0 V = NkT + 1 E grav 3 with E grav = − i< j (15. stars with masses smaller than 10 M account for only 9% of the luminosity. for 91%).22). The cumulative mass M. radius R and . in a Salpeter IMF. all kinetic energy is thermal and m x2 = 3NkT . the Lyc flux NL (m) is from (14. luminosity L and Lyman continuum flux NL for the Salpeter (full curve) and Miller–Scalo (broken curve) IMF with lower and upper mass limit m l = 0. one has to integrate over the cloud surface and transform the expression into a volume integral according to Gauss’ theorem.18) Gm 2 . For example.1 M and m u = 50 M . The outer forces are due to the external pressure P0 which results in a force P0 d f that is perpendicular to each surface element d f and directed inwards.11).

22) and (15. when pressure and gravitational forces balance exactly everywhere. 1 d r2 d P = −4π Gρ. a low-mass protostar in its late stage of evolution moves towards the main sequence along the Hayashi track.20) m For example.20) also holds for planets and other objects whirling around the Sun if one replaces E therm by the average kinetic energy E kin over a revolution.23) Mr denotes the mass contained in a sphere of radius r .21) tKH = L 2m L where T is an average by mass over the star and m. which is of second order and called the Emden equation [Emd07].22) and (15. When accretion has ended and nuclear burning not yet started. Stars and their formation 3G M 2 . one switches off nuclear burning. (15. (15.2. as before. whereas the stellar radius and the luminosity shrink. G Mr dP = − 2 ρ dr r d Mr 2 = 4πr ρ. (15.474 mass M.18) yields 3kT M −E grav = 2E therm = . tKH is a thousand times smaller. It is not valid for comets that fly on a parabolic orbit of infinite extension. When the star is on the main sequence and.24) r 2 dr ρ dr We solve (15. is the mean particle mass. dr (15. The star is always in equilibrium and exactly half of the liberated gravitational energy goes according to the virial theorem into heating the gas. the luminosity would be sustained over the Kelvin– Helmholtz time scale 3kT M E therm (15. For the Sun. E grav = − 15.19) 5R When the outer pressure can be neglected with respect to the central pressure. for O stars. its surface temperature is then approximately constant. (15. tKH equals 30 million years.25) .23) subject to the boundary conditions ρ(0) = ρc dρ dr =0 r=0 (15.2 Isothermal cloud in pressure equilibrium In a gas sphere. in a thought experiment. Equation (15.23) can be combined into one. the other half is radiated away. the luminosity L is entirely due to gravitational contraction. Formulae (15.22) (15.

where r = R. For polytropes. which is true during the early protostellar phases.27) the isothermal sound speed. At the edge.e. dust and gas are thermally coupled by collisions between gas molecules and grains. the cloud is confined by an outer pressure P0 which is equal 2 to the pressure just below the cloud surface. P= kT 2 ρ = vs ρ m (15.23) are integrated from the center outwards. the heating rate per gram of interstellar matter is fairly uniform over the cloud and does not depend on density. especially of H2 O and CO). Cooling comes mainly from grains (although some is due to molecular lines. When the cloud collapses. The temperature does not enter but T is nevertheless not uniform over the cloud. Here we are interested in the structure and stability of an isothermal sphere which is described by a polytrope of index n = ∞.28) where K is some constant. At the prevailing high densities.2. we prescribe the central density ρc and its gradient. which has to be flat. One still has to add the equation of state. 15. the central density ρc being the parameter.26) where m denotes the mean mass of the gas particles and vs = ∂P = ∂ρ kT m (15. beneath its surface layer. the Pd V work of the gravitational compression is transferred from the gas to the dust and radiated away. The assumption of isothermality correctly reflects the conditions in protostellar clouds before they collapse. happens either through absorption of far infrared radiation by grains or by interaction of cosmic rays with the gas. the pressure P can be expressed solely by ρ. Heating of such a cloud. As the cloud temperatures are low.29) (15.Clouds near gravitational equilibrium 475 i. As long as the cloud is optically thin to dust radiation. In either case.22) and (15. so P0 = vs ρ(R). the collapse does not enhance the temperature. For every mass M.3 Structure and stability of Ebert–Bonnor spheres Equations (15. P = Kρ 1+ n 1 (15. General solutions of polytropes are discussed in [Cha39]. the maximum dust emission lies at submillimeter wavelengths and such photons can freely escape.30) . there is a family of solutions. By introducing the variables y = ρ/ρc x = r 4π Gmρc /kT (15.

However. Because obviously u /x → u (0) for x → 0. Thus the cloud is further compressed and finally collapses. they are called Ebert–Bonnor spheres. between A and C.7 depicts how the radius R of an Ebert–Bonnor sphere varies with the boundary pressure P0 . ∂ P0 /∂ R > 0. somewhere to the right of point A on the curve in figure 15. the cloud arrives at a new configuration where the actual outer pressure P0 surpasses the value required for equilibrium.31) with the boundary conditions y(0) = 1 and y (0) = 0. which is the point of minimum cloud radius.31) simplifies to u + 2u + eu = 0 x (15.6. A cloud in a low-pressure environment settles into an equilibrium configuration. When such a cloud is perturbed. In figure 15. which displays the I (x u ) = 0 xu y(x) x 2 d x (15. the instability continues.5.34) one can determine to which value of x u (or radius R) a sphere of given mass M. and ρ falls off in the envelope approximately like r −2 .32) with u(0) = 0 and u (0) = 0. The mass of an Ebert–Bonnor sphere is computed from M = 4π 0 R 1 r 2 ρ dr = √ 4πρc kT Gm 3/2 0 xu yx 2 d x (15. Although ∂ P0 /∂ R may again be negative for the cloud as a whole. If we further put y = eu . 3 The density is flat in the core.7. it follows that (u /x)x=0 = − 1 . (15. When somewhere on this part of the curve the radius is decreased by some inevitable fluctuation. The function P0 = f (R) and the ensuing stability criterion can be studied analytically but the treatment is laborious [Ebe55. Figure 15. Its solution describes the structure of isothermal gas balls in equilibrium.24) can be made dimensionless and brought into the form y − y2 2y + + y2 = 0 y x (15.32) behaves near zero. As one increases the central density and moves on the curve beyond the point C towards the eye of the spiral. we still have to know how u /x in (15. Collapse starts only when the Ebert–Bonnnor sphere becomes . With the help of figure 15. Its environment may vary but the cloud is stable and adjusts to a changing outside pressure P0 .7. interior parts are unstable.476 Stars and their formation equation (15. clouds from A over B to infinite radii are stable because ∂ P0 /∂ R < 0. it reverts to the configuration whence it started from. To solve this second-order differential equation by a Runge–Kutta method.33) where x u = R · integral 2 4π Gρc /vs . Bon56]. The result of the numerical integration is shown in figure 15. central density ρc and temperature T extends.

the cloud is stable only if the outer pressure is smaller than the critical value Pcrit and the radius greater than Rcrit . The radial coordinate x is 2 proportional to the cloud radius. P0 attains its maximum at x = 6. The structure of an (isothermal) Ebert–Bonnor sphere after (15.37) (15. kT (15. we write the outside pressure P0 and the cloud radius R as a function of the dimensionless variable x. y = ρ/ρc . where I 2 y/4π = 1.35) (15.Clouds near gravitational equilibrium 477 Figure 15.31). 2 P0 (x) = vs ρc y(x) = kT m 1/2 4 1 G3 M 2 I 2 (x)y(x) 4π (15. the broken curve shows for comparison the slope of a density distribution falling off like 1/r 2 . The density y is in units of the central density. Pcrit = 1.398 and x/I = 0. To determine the critical values.38) Rcrit = 0.450. Therefore. For a fixed cloud mass M and temperature T . In this logarithmic plot.4106. critical.5. kT I (x) The expressions I 2 y and x/I are readily evaluated numerically.411 .36) R=x kT 4π Gmρc = x Gm M .40 k4 T 4 G 3m 4 M 2 Gm M. x = r (4π Gρc /vs )1/2 .

multiplication of (15. A cloud in pressure equilibrium is also virialized and the virial theorem in the form (15.18) follows directly from the equation for hydrostatic equilibrium (15.22) with 4πr 3 and integration by parts over the cloud radius gives R 0 4πr 3 dP dr = 3P0 V − 3NkT = − dr R 0 G Mr 4πr 2 ρ(r ) dr = E grav . its density is almost constant and the relations between P0 and R are practically identical for an Ebert–Bonnor sphere and a virialized homogeneous sphere. become important and the curves diverge. For large values.45. r When the cloud has a large volume. pressure gradients. which are taken into account by the virial theorem in (15. the largest possible stable mass for a cold H2 cloud of 10 K is thus only ∼6 M . as in a tenuous HII region.34). Insert: The functions x/I (x) and y(x) I 2 (x)/4π near the critical value xcr = 6. 2 )1/2 marks the cloud boundary. say P0 = nkT ∼ 10−12 dyn cm−2 . Indeed. . roughly I (x ) ∝ x . The mass integral I (xu ) = yx 2 d x of equation (15.22). However. xu = R(4π Gρc vs u u 3 for small ones I (xu ) ∝ xu . when the cloud is compact.478 Stars and their formation Figure 15.18) only in an integral form.6. Assuming some reasonable pressure to prevail in the interstellar medium.

52) for unmagnetized spheres of constant density. ρc = 2. R = 1. at point B. ρc = 1. the full curve is from the virial relation (15.19 × 10−18 g cm−3 .Clouds near gravitational equilibrium 479 Figure 15.7.24) when R and P0 are normalized to the values of the critical sphere.2.2 from one point to the next as one follows the spiral of dots from right to left.4 Free-fall of a gas ball The ultimate simplification in the study of cloud collapse is to treat the contraction of a constant density sphere under the influence of gravity alone. The sequence of points shows the general relation between the radius R and the outer pressure P0 for an isothermal gas ball in hydrostatic equilibrium after (15. thus neglecting all other forces that arise from pressure gradients.33 × 1017 cm and P0 = 3. magnetic fields or rotation. at the critical point A. Consider a gas sphere of total mass M which at time t = 0 is at rest and has an outer radius R. Stable configurations are only possible from A over B to infinity. 15. For an H2 cloud of 1 M and T = 10 K.15 × 10−20 g cm−3 . t) denote the radius of an internal sphere that contains . The central density ρc increases by a constant factor of 1. Let r (m. For comparison.41 × 10−11 dyn cm−2 .

39) r2 together with the conditions that.41) tff depends only on the initial density ρ0 and is inversely proportional to its square root. the basic gas–dynamical equations of momentum and mass continuity for the time-dependent variables velocity v and density ρ are 1 ∂v + (v · grad)v = − grad P − grad φ ∂t ρ ∂ρ + div(ρv) = 0. like rotation or pressure gradients. clouds are generally not in the stage of collapse. =− r0 3 r Putting r/r0 = cos2 β. ∂t (15. i. at time t = 0. gas cooling. 3 (15. such as fragmentation of the cloud into subunits.2. Integration of (15. J Jeans was the first to give a quantitative answer to the question under what conditions a cloud collapses and the stability criterion bears his name. In a medium of constant temperature T . This happens for β = π/2. 15. 32Gρ0 (15.5 The critical mass for gravitational instability Although in any interstellar cloud there is always some gravitational pull which tries to compress it.40) β and thus also r/r0 do not depend on the mass m and the collapse therefore proceeds homologously. 0) = 0. when r = 0.42) (15. the star-formation rate would be orders of magnitude higher than what one observes.43) . 0) and ρ0 denotes the initial constant density. then obviously 3 4πρ0r0 /3 = m.e. are excluded. frosting of molecules onto dust particles. The collapse is then described by Gm (15. turbulence or outer forces are absent. If they were.39) yields r =− ¨ r ˙ 8π Gρ0 r0 −1 . grain coagulation and so forth. magnetic fields. ˙ When r0 ≡ r (m. r(m. the cloud is at rest. It is completed when the cloud has contracted to a point. one finds β+ 1 2 sin 2β = t 8π Gρ0 . The free-fall time gives the right order of magnitude for the duration of gravitational collapse but rather as a lower limit because all retarding effects. It is often useful to compare tff with the time scale of other processes. It is valid only when rotation.480 Stars and their formation the mass m ≤ M at time t. from which follows the free-fall time tff = 3π .

≈ 2.48) It is called the Jeans mass. If the initial values of density and velocity are ρ0 = 0 and v0 = constant.46) 4Gmρ For a cloud of molecular hydrogen (m = 2m H ).e. We introduce a small perturbation. Let the medium intitially be uniform and at rest. denoted by the subscript 1. (15. Therefore. when the wavenumber k is below some critical value kJ (the letter J stands for Jeans) given by kJ = π = LJ 4π Gmρ0 . ρ (15.47) where n is the number density of hydrogen molecules per cm3 .48) may be written as MJ L T.2 × 1019 T /n cm (15. (15. kT Do not confuse the symbol k for the wavenumber with Boltzmann’s constant k. (15.49) . For convenient comparison with observations. P = kρT /m = ρvs . Writing the density change ρ1 in the form of a plane wave. ρ1 ∝ ei(kz+ωt ) one obtains 2 ω2 = k2 vs − 4π Gρ0 . i.44) 2 and the gas pressure P by the equation of state (15. the maximum mass of an interstellar cloud of given density and temperature that does not collapse is MJ ≈ L 3 ρ = J πk 4Gm 3/2 T3 . L J ≈ 1.45) If the perturbation ρ1 increases exponentially.4 × 10−19 M cm (15. where m is the mean particle mass and vs the sound velocity of (15. So a cloud of density ρ and temperature T is unstable to collapse if it has a size greater than πkT LJ = . This occurs when ω is imaginary (ω2 < 0).27).Clouds near gravitational equilibrium The gravitational potential φ is given by the Poisson equation φ = 4π Gρ 481 (15.26). at some later time ρ = ρ0 + ρ1 v = v0 + v1 . and examine whether it grows or not. the cloud is unstable.

not be taken too literally. the collapse is isothermal because dust is a very effective coolant.6. ∇φ and thus φ would have to vanish but that is impossible because of the Poisson equation (15.2 Fragmentation and thermodynamic instability During cloud contraction. For example.4 × 10−7 T 4 M . Smaller subunits .482 Stars and their formation A gravitationally unstable clump of one solar mass has. therefore.46)– (15. However. formula (15.1 pc. Initially. When the optical depth is transformed into τ= πk 4Gm 2 Kd T 2 1 2 MJ one sees that clumps of very low mass cannot freely radiate the gravitational energy released in the collapse. we thus notice from (15. A homogeneous medium at rest cannot be in equilibrium.49) helps decide whether it should collapse or not. must be smaller than unity at wavelengths where dust emission is efficient. min(MJ ) ∼ 0.44). Continuing our crude analysis of cloud collapse. therefore. If it were. Using for K d the Planck mean absorption coefficent K P of equation (13.33). 15. linear size L and mass M of a suspected protostellar condensation.48) that a gravitationally unstable and contracting cloud has the tendency to break up into fragments because the critical mass MJ decreases proportionally with T 3 /ρ as the average cloud density ρ rises. For T = 20 K. gravitational energy is converted into heat. gas cooling is relatively inefficient. the minimum fragment mass that can undergo free collapse is (15. This derivation of the critical mass MJ from perturbation theory is logically not clean. 15.49) should. these numbers should not be taken at their face value but the remarkably strong dependence of MJ on the cloud temperature may explain the observational fact that in starburst galactic nuclei.2. the dust optical depth τ . from the cloud center to the surface.50) min(MJ ) = 2. the IMF is heavily biased towards massive stars and low-mass stars do not seem to form at all. where the dust is warm (Td ≥ 30 K). by measuring the temperature T .6 Implications of the Jeans criterion 15. so τ = 1 2 L J Kdρ < 1.1 The role of dust in gravitational stability Because an isothermal collapse of a molecular cloud requires radiation losses. Of course.6.04 M . All constants in the preceding equations (15. a size of ∼0. the Jeans criterion is qualitatively correct and very useful.2.2.

According to (15.51) Besides fragmentation. Isothermality implies γ = 1. If ttrav is shorter than the free-fall time tff of (15.6. called fragmentation. δ Q = dU + P d V = 0.3. the value γcr also figures in protostellar evolution (section 15. to a gas sphere of mass M. and the thermal energy by the energy of the turbulent motions. where the outer pressure is absent and the gravitational energy equals twice the thermal energy.3 Further remarks on the Jeans criterion • The Jeans instability criterion is also valid when the gas pressure is due to supersonic turbulence (because it just compares kinetic with gravitational energy).48). the gas behaves adiabatically. Compression then leads to heating and pressure gradients build up which stop the collapse: a hydrostatic core forms. radius R and with M/m gas particles. This process. So as the cloud shrinks.2). it cannot respond quickly enough to the density perturbation and collapse will ensue. When the gravitational energy is not radiated away. is fundamental. there is gravitational instability if γ falls below γcr = 4 .41). 5R m 2 . τ must at some moment exceed unity. also yields the Jeans mass or Jeans length. for inhomologous contraction. Suppose in some place of a cloud the gas is contracting. and T ∝ ρ γ −1 where γ = Cp /Cv is the ratio of specific heat at constant pressure to that at constant volume. in the corresponding formulae. one gets M 3 3G M 2 = 2 · kT.20). the dust optical depth τ grows proportionally to R −2 . The condition ttrav = R = vs 3π = tff 32Gρ • • together with (15. the sound velocity vs by the turbulent velocity. One must then replace.27) for the sound velocity. In a homogeneously contracting cloud. When one applies the virial theorem in the form of (15. The information about the contraction travels through the cloud in a time ttrav .2. 15.Clouds near gravitational equilibrium 483 of the mother cloud become unstable as long as the temperature does not rise faster than ρ 1/3 . Otherwise. the cloud will be stable. 3 (15. The isothermal phase ends when the cloud becomes opaque to the cooling radiation. for which R 3 ρ = constant. the rise is even steeper.

7).56)). from 5/3π = 0. which can obviously be written in two ways. for P. the critical mass for collapse is ∼1 M . 2 G 3/2 ρ 2 48π π G 1/2 (15.24 to 0.2. So the critical value γcr also holds for magnetic pressure Pmag .18) so that it incorporates magnetic forces. The radius of the critical Ebert–Bonnor sphere also has the same form as the Jeans length (see (15.18) is generalized to 4π R 3 P0 = 3NkT − 1 3G M 2 + R3 B 2. the √ proportionality factor drops. we are led to Mmag in (15. the magnetic pressure Pmag = B 2 /8π. As the terms in (15.53) The second expression for Mmag contains the flux = π R 2 B through the clump. One can approximately extend the virial equation (15. Indeed. dominates at the beginning. equation (15. it will do so throughout the subsequent evolution.52) If the field is frozen into the cloud.53). Mmag = B3 53/2 51/2 = .7.1 pc and B = 10 µG.1 Critical mass for magnetized clumps When a cloud is pervaded by a magnetic field.48) as MJ = (π/4G)3/2 P 3/2 /ρ 2 and insert. We conclude that for R = 0.13. We can thus extract from the Jeans criterion (15. but such refinements are academic as estimates based on the virial theorem are rough anyway. Indeed.48). again but for a factor of order unity.2.48) the critical mass Mmag in the presence of a magnetic field (see section 15. The cloud can. only collapse if 3G M 2 /5R > R 3 B 2/3. Mmag .7 Magnetic fields and ambipolar diffusion 15.52) have then the same 1/R-dependence and if one form of energy. This condition leads to a minimum mass for instability in the presence of a magnetic field. (15. one immediately recovers the formulae for L J and MJ . its net effect is to lend further support against collapse.18) represent global cloud energies. the magnetic flux is conserved and the relation B R 2 = constant holds no matter whether the cloud is contracting or expanding.484 Stars and their formation Within a factor of order unity. Any spherical cloud supported against its gravity by an internal pressure P becomes unstable when P ∝ ρ 4/3 .46) and (15. gravitational or magnetic. 5R 3 (15.38) and (15. when we rewrite equation (15.2.46)) and the two expressions are also practically equivalent. • 15. for a spherical cloud threaded by a uniform field of strength B. The gravitational and magnetic term in (15. When the analysis is extended and cloud flattening taken into account. . therefore. one expects the new term for the magnetic field to be equal to the total magnetic field energy (see (1. independently of the physical origin of P.

the diffusion time tad would exceed the cloud lifetime and collapse could not occur. The strength of the coupling between gas and field is expressed through a diffusion timescale td .2 Time scale of ambipolar diffusion Of course. only contract if its radius is smaller than ∼0. . 2 3 2 15. a more elaborate analysis yields 1 ≤ n ≤ 1 [Mou91]. in the 3 case of equipartition between thermal energy of the gas and magnetic energy. n = 1 . one assumes a relation B ∝ ρ n between gas density and field strength.1 pc. Assuming that the density of the neutrals is much larger than the ion density. to clump densities. flux freezing (B R 2 = consant) implies n = 2 . one ends up with a field strength of ∼10 µG or more. even in molecular cloud cores. for example. and this should be sufficient to effectively freeze in the magnetic field.7. (15. where measurements are easier and B is at least 1 µG. which we now define for a spherical cloud of mass M and radius R. Let the cloud be close to equilibrium so that the gravitational force is balanced by the magnetic field. the cross section for momentum transfer is large. Although the interstellar matter is mainly neutral.54) Because the slippage occurs between ions and neutrals. The ions are then kept at their position while the neutrals drift slowly across the field lines under the influence of gravity. the diffusion time tad is several million years. The force balance for 1 cm3 is thus GM nm = n i n σ u mwd R2 from which follows the ambipolar diffusion time tad = R ni 3 σu = wd n π Gm 7 × 1013 ni n yr. cosmic rays alone can probably keep one out of 107 gas particles ionized. Should the degree of ionization be much above 10−7 . The process is called ambipolar diffusion. For the extrapolation. because the medium is clumpy and stray UV photons manage to penetrate into the cloud interior. and that both kinds of particles have the same mass m. A 1 M cloud can. where wd is the drift speed between the two components and σ u an average over thermal velocity times cross section.2. it limits the pace at which stars can form. all this is correct only if the magnetic field is coupled to the gas. n n i . σ u ∼ 2 × 10−9 cm3 s−1 . For homogeneous contraction (R 3 ∝ ρ −1 ).Clouds near gravitational equilibrium 485 It is not easy to determine the strength of the magnetic field in molecular cloud cores but if one extrapolates from the lower density environment. All this suggests that a cloud must first get rid of its magnetic field through ambipolar diffusion before it can collapse. As ambipolar diffusion is a slow process compared to dynamical contraction. otherwise there is no interaction. With n i /n = 10−7 . for neutral–neutral collisions it would be a thousand times smaller. each ion imparts to the neutrals per second the momentum n σ u · mwd . therefore.

a reasonable mean value for the Sun. the Sun has a radius R = 7 × 1010 cm and a mean density 1. Half of the released gravitational energy.2 Rotational and magnetic energy A protostellar cloud rotates. The initial gravitational and thermal energy of the protosolar nebula are: E grav E therm G M2 R 3kT M = 2m 2 × 1042 erg 1 × 1042 erg. As a result.3. the gas has acquired an average temperature of ∼3 × 106 K. this gives a mean luminosity of ∼40 L . Evidence from molecular lines and dust radiation from low-mass protostellar clumps indicates a temperature T ∼ 10 K.55) (15.58) 2 d R0 .38) then yields a critical cloud radius of about R = 0. (15.1. the cloud radius has shrunk to ∼2R . (15.3. It must have had a mass of 1 M . at least.41) then becomes tff 1. corresponding to a molecular hydrogen density somewhat above 3 × 104 cm−3 .1 Gravitational and thermal energy To get a feeling for the parameters of a protostellar cloud.05 pc or 1. during some of its protostellar evolution more luminous than it is today.486 Stars and their formation 15. let us consider the presolar nebula. This is what the Milky Way does and what all other celestial objects do.1 The presolar nebula 15.3. ρ ¯ there are vast changes: the radius shrinks by a factor ∼107 .4 g cm−3 .3 × 10−19 g cm−3 . So very elementary considerations suggest that the Sun was. Equation (15.57) A = − R0 2 d R0 1 dω0 −1 −B = R0 + ω0 10 km s−1 kpc . Today. 15. possibly a bit more as some matter was dispersed and did not fall onto the star. namely G M 2 /4R ∼ 1048 erg.3 Gravitational collapse 15. and the cold neutral gas turns into a hot plasma of over 107 K with thermonuclear reactions.9 × 105 yr.56) By the time the accretion process is over. The free-fall time from (15. the density increases by 20 powers of ten.54 × 1017 cm and a density ρ ∼ 1. So on the way from the protostellar cloud to the main sequence. The galactic disk does not rotate rigidly but differentially and the angular frequency ω decreases with galactic radius R as described by Oort’s constants A and B: 1 dω0 15 km s−1 kpc−1 (15. The other half was radiated away. Over one free-fall time. about equal to the travel time of a sound wave through the cloud. has gone into heating. We know this from the position of T Tau stars in the Hertzsprung– Russell diagram.1.

(15.6 × 10−16 s−1 . ω ∼ 10−15 s−1 seems a fair guess for them. In more recent analyses. Rigid rotation of the disk would imply A = 0 and B = −ω0 . the initial rotational energy becomes E rot = 1 2 2 Iω ∼ 1037 erg. Their uncertainty is about 20%. one finds ω0 = A − B 8−16 s−1 corresponding to a rotation period of 250 million years. the magnetic energy is E mag = B 2 4π 3 R 8π 3 6 × 1040 erg. 15. Because . conservation of angular momentum I ω implies E rot ∝ R −2 . We may envisage a large cloud to revolve around its own axis at an angular frequency ω = |dv/d R0 | 1. In the differentially rotating galactic disk. so the rotational energy rises steeply as the radius shrinks. a few times smaller than ω0 .59) Here we have used dv = ω0 d R0 + R0 dω0 . the orbital speed of an interstellar cloud is higher on the near side facing the galactic center than on the far side.Gravitational collapse 487 They are determined observationally and refer to the locus of the Sun where ω = ω0 and R = R0 . however.58). Flux conservation implies for the contracting cloud B R 2 = constant and thus E mag ∝ R −1 . This. Note: the previously cited values of A and −B are not up-to-date but easy to remember. Although this is small compared to E grav or E therm . the numbers for both A and −B tend to converge around 12. They are based on sphisticated codes and carried out on computers as powerful as available at the time. A constant density sphere has a moment of inertia I = 2 M R 2 .2 Hydrodynamic collapse simulations Modern hydrodynamic calculations of protostellar evolution are multidimensional. is not what one observes. For a presolar 5 cloud. This raises a fundamental problem in the theory of star formation: How do we get rid of the angular momentum? When the protosolar nebula is pervaded by a magnetic field of strength B = 10 µG.57) and (15.3.5 km s−1 kpc−1 . The magnetic field is initially relatively weak but grows with R −2 and should thus be 1013 times stronger in the final star. The main mechanism by which a protostellar cloud rids itself of the magnetic field is probably ambipolar diffusion. this would reduce ω accordingly [Oll98]. From (15. Because of the differential rotation of the galactic disk. As protostellar clouds have undergone some contraction by which they were spun up. with two or three spatial coordinates. the orbital velocity v = ω0 R decreases towards the periphery like dω0 dv = R0 + ω0 = −A − B d R0 d R0 −5 km s−1 kpc−1 .

the non-trivial outer boundary condition reads as u = 0. The core becomes opaque when about 0. the assumption of isothermality has to be abandoned. flow velocities become quickly supersonic because the sound speed is only ∼1 km s−1 in a cloud of 10 K. ∂t r ∂r (15. so ∂ Mr − 4πr 2 ρ = 0.9.01 M of gas have been accumulated within a radius r ∼ 1014 cm. the cloud is made up mostly of H2 for which γ = 7/5 > γcr of (15. Here and in section 15.60) (15.488 Stars and their formation of the complexity of the numerical results.8). an example is presented in figure 15.60)–(15. and on time t. As one can read from figure 15. molecular hydrogen dissociates. the core warms up and settles into hydrostatic equilibrium.8 or 15. This second collapse is halted only when γ rises again above 4/3.6). The submillimeter dust radiation is then trapped. For the latter.62) ∂r During the early phases. The gas is driven by the gradient of the pressure P and by the gravitational attraction. therefore.60) the derivative ∂ P/∂r = vs ∂ρ/∂r . we.51).42) and (15. But when hydrogen dissociates. a naive understanding of what is happening gets lost.3.8. One must then add to (15. and as long as there are no shocks.62) into difference equations and solve them numerically.2. For a fixed cloud volume. (15. which happens at a central density of ∼10−3 g cm−3 . In this case. Initially. Inevitably. Mr is the mass of a sphere of radius r . sketch simplified scenarios which nevertheless catch important aspects of the collapse stages.43)) ∂u 1 ∂P G Mr ∂u +u + + 2 =0 ∂t ∂r ρ ∂r r 1 ∂ 2 ∂ρ + 2 (r ρu) = 0. it is straightforward to transform (15.3.29). The numerical solution of the full set of equations [Lar69] leads to a ρ ∝ r −2 density distribution even when starting at constant density (see figure 15. the collapse is isothermal. (15.62) an equation for the conservation of energy (first law of thermodynamics) and for the transport of radiation. When the gas temperature reaches ∼2000 K. they usually cannot be adequately presented in plots but have to be visualized in motion pictures. The released gravitational energy is then no longer used to increase T but to separate the hydrogen atoms. When a hydrostatic core has formed. The dynamical equations of a spherical cloud can be written as (cf.61) 2 The geometry is here one-dimensional: density ρ. The sound velocity vs is 2 then constant and in (15. one often uses a frequency–averaged (grey) approximation in the spirit of (13.60) to (15. γ falls below the critical value of 4/3 and the hydrostatic core begins to collapse. When the cloud becomes optically thick (see section 15. pressure P = vs ρ and velocity u depend only on radius r . the supersonically falling gas is stopped at the core .

86 × 105 yr. and the free-fall time tff = 1. The cloud is at temperature T = 10 K. The distribution of density ρ and velocity u during the isothermal contraction of a spherical and initially homogeneous H2 cloud of 1 M .Gravitational collapse 489 Figure 15. In the . its radius is fixed (u = 0 at the surface). surface giving rise to a shock. The collapse is very inhomologous. After 2. For an Ebert–Bonnor sphere (slightly beyond the brink of stability).8.60×105 yr.6 in the inner regions. so practically it is a discontinuity described by the Rankine–Hugoniot jump conditions and this poses a numerical challenge. If the density at that moment is approximated by a power law. the core (r < 1014 cm) turns opaque to its own far infrared radiation and the assumption of isothermality is no longer valid.3 in the envelope and −1. the exponent equals −2. The width of the shock is determined by the mean free path of the gas particles. the evolution would be similar.

invoke strong . the core has the tendency to split up. R∗ (15.68) with A = 2. to get the velocity u. u. For example.66) and (15. To get the radius r .9. shock front. for instance. If R∗ and M∗ are the radius and mass of the hydrostatic core. one multiplies the abscissa x by vs t0 .0001. from this figure.67) describing spherical isothermal collapse. the cloud structure in terms of r. the gas kinetic energy is converted into radiation. When the numerical effort to compute protostellar collapse is much enhanced and the physical processes are refined. becomes L shock = ˙ G M∗ M . The starting values at x = 10 are from (15. when the collapse calculations are carried out in two dimensions and cloud rotation is included.64)). One can determine. To evaluate it correctly. and to 2 obtain the density ρ. The time evolution goes along the lines from right to left. The solution of the ordinary coupled differential equations (15. new problems arise concerning the stability of the core. ρ of an initially (t = 0) 2 hydrostatic cloud of density ρ = vs /2π Gr 2 at any later time t0 > 0. one multiplies v by vs . To avoid fission of the core.63) ˙ The accretion rate M. a proper treatment of shocks is obviously necessary. one multiplies α by (4π Gt0 )−1 (see (15. one may. prior to nuclear burning and neglecting the contraction of the protostellar core. the protostellar luminosity.490 Stars and their formation Figure 15. the size and radius of the hydrostatic core are therefore the parameters that determine the protostellar luminosity.

t) and ρ(r. G (15.65) Now (15.61). t) = α(x) 4π Gt 2 M(r.60) and (15.3 Similarity solutions of collapse 15. x. x −v = 1 and α = 2/x. v = 0 and α = 2x −2 . can be discarded as unphysical because it implies a static mass distribution M(r. ρ and M into (15. and there are no more partial derivatives.66) (15. The second. It would guarantee an efficient outward transport of angular momentum and thus enable the formation of a single star [Tsc87].3. By definition. The similarity solutions can be found by introducing a dimensionless variable which expresses some relation between r and t.61) can be written as two ordinary differential equations: (x − v)2 − 1 v = α(x − v) − (x − v)2 − 1 2 (x − v) x (15. the dependent variables velocity. t) at a fixed moment t then differs from the spatial distribution at any other moment t only by scale factors that depend only on time.62) and into ∂M ∂M +u =0 ∂t ∂r which is an equivalent form of the continuity equation (15.64) Inserting these expression for u.3. although the velocity u is not zero. . t) = vs v(x) ρ(r. vs t Because of similarity. the spatial distribution of u(r. it is very instructive. one has only one. yields m = m (x − v) m = x 2α ⇒ m = αx 2 (x − v). This restricted class of solutions does not add anything new to the one-dimensional numerical calculations of [Lar69] but because of its simplicity. t) = 2a 2r/G.67) 2 α = α − (x − v) (x − v). The first. t) = 3 vs t m(x). 15. only derivatives with respect to x. density and mass are the product of a function of t times a function of x. (15.3. r and t.1 Equations for spherical collapse Similar solutions to the one-dimensional hydrodynamical equations for an isothermal cloud also exist [Shu77]. corresponds to a static sphere with a diverging central density. α x Instead of two independent variables. A fruitful ansatz is u(r. Two solutions are obvious. let us choose x= r .Gravitational collapse 491 turbulent friction (of speculative origin). which are denoted by a prime.

When t → 0 ⇒ x → ∞. Of course.68) The constant A has to be greater than 2 for v to be negative but only a tiny bit so that velocities are small. for other values of A slightly above two.66) and (15. 15.66) and (15. t) = 2 Avs r.67) v + v 2 /v + 2v /x = 0. .975.67) become v = α − 2x −2 α = α(α − 2)x −1 . the cloud is close to hydrostatic equilibrium and has a structure similar to the envelope of an Ebert–Bonnor sphere. From (15. if one takes the initial parameters from (15.3 Final cloud configuration There are also analytical expressions for the configuration towards which this cloud evolves as x → 0.9.3. t) = − 2m 0 vs t 2 r − 2 .66) one then finds v = α and from (15. therefore. As one can sketch the global aspects of its evolution with elementary mathematics. When A 2.3. Then v goes to zero and (15. This differential equation is satisfied by α→ m 0 /2x 3 v → − 2m 0 /x for x → 0 where m 0 = −(x 2 αv)x=0 is the limiting value of m(x). We can read from figure 15. m 0 = 0.492 Stars and their formation 15.3.0001 starting with x = 10 (that is large enough) is plotted in figure 15. One can integrate (15.2 Initial cloud structure Let us consider an infall scenario where initially (t = 0). A−2 x They yield. ρ(r.3. ρ(r. The initial structure (t → 0) of a singular isothermal sphere is. the cloud is at rest. t) = 0 M(r. for x → ∞. the asymptotic solutions α → Ax −2 v→− m → Ax. for A → 2.71) 3 u(r.69) The density falls off like r −2 .70) (15. The solution for A = 2. The final cloud structure for t → ∞ is. the absolute value of the dimensionless velocity |v| 1 and αx|v| 1. corresponding to t → ∞. the solutions are similar. t) = 3 m 0 vs − 1 − 3 t 2r 2 32π 2 G 2 1 1 (15. (15. therefore.67) from some large value of x by a Runge–Kutta method down to x = 0 (corresponding to t → ∞).68). the singular isothermal sphere with A 2 is often used to outline qualitatively protostellar collapse. G (15.9 that for x → 0. one may be skeptical whether such an artificial configuration is realized in nature and the results given here can in no way replace hydrodynamic calculations. t) = 2 vs A 4π Gr 2 u(r.

Inner and outer cloud boundaries are disregarded and should not influence the evolution of the flow.67 × 1017 cm. so this transition region is moving outwards with the speed of sound. Because the initial density declines like r −2 and the free-fall time tff of (15.6 × 105 yr. With the help of figure 15. ρ is still proportional to r −2 . At r = 1014 cm. one needs a starting model and boundary conditions at the edge and center of the cloud. only the parameters α and v are required for some value of the variable x ∝ r/vs t. is time-independent and given in the self–similarity analysis by 3 m 0 vs ˙ M = 4πr 2 ρu = .Gravitational collapse At a fixed time. for x < 1. the gas is still at rest.4 Remarks on the cloud evolution To solve the full equations (15.72) 15. from (15. velocities become supersonic and approach the free3 fall velocity towards a central object of mass m 0 vs t/G that is accreting at 3 a rate m 0 vs /G.3. for x > 1.2 km s−1 . the collapse begins at the center working its way outward. we can delineate the evolution of an initially 2 hydrostatic cloud of density distribution ρ = vs /2π Gr 2 . At the interface between free-falling core and static envelope. the density now changes like r −3/2 and the velocity u(r. • • In the envelope.60) and (15. G 493 ˙ The accretion rate M.0 × 10−15 g cm−3 and equals ∼0. the cloud stretches out to R = 1.61).3.3 M . After t = 5 × 1012 s. t) = − 2G M∗ (t)/r is the free-fall velocity towards the central mass M∗ (t) = 3 m 0 vs t. In the core. if the total cloud mass is 1 M and vs = 0. all gas interior to r = 1017 cm is collapsing and the mass accreted at the center 1. we note that as time proceeds the corresponding point of the similarity variable x = r/vs t moves along the curves α(x) and v(x) from right to left. which is the quantity that determines how long the collapse will last. x 1 and r = vs t. To follow the density and velocity at some fixed radius in time.9.41) decreases as ρ −1/2 . proportionally to t −1/2r −3/2 . • For example. G (15. The density decreases with time and radius. The density has not changed. In the similarity solution. the velocity u −9 × 10 .71). the density is ρ 5 cm s−1 . it does not yet know that the inner regions are collapsing. or 1.

The molecules of the liquid are always at rest on the walls of the vessel and on both sides of the plate. The gradient implies that the fluid particles change their relative positions. as it should be. (15. because of the term (∂v y /∂ x)2 .74) The force F does not depend on the velocity itself but on its gradient ∂v y /∂ x. F = νρ 2 A = νρ b ∂x (15.1 Viscous laminar flows The evolution and structure of an accretion disk is largely determined by internal friction. The total power dissipated in the vessel is Fu. these forces are absent. an initially rigidly rotating spherical cloud of angular velocity ω0 . We. v y .10). The friction also generates heat. therefore. In the gravitational field of a dominant central mass M∗ . vz ) = (0.4. a disk forms. after subtraction of gravity and buoyancy.494 Stars and their formation 15. v y . Eventually. the parcel can approach the center only to a distance rcenfug = j2 G M∗ (15. A gas parcel at radial distance r0 and under an angle θ0 to the rotation axis has the specific angular momentum 2 j = ω0r0 sin2 θ0 . the final equilibrium configuration in which they collect is a disk. When the width 2b of the vessel is small. Disks are widely observed in young stars and their presence is predicted by theory. one needs. be balanced by centrifugal forces. for example. whereas parallel to the rotation axis. One can get a feeling for the strength of the force and the dissipation rate by using numbers . the velocity increases linearly from the walls towards the sheet and ∂v y u 2 A. A flat piece of arbitrary material and total surface area 2 A is dipped into a vessel filled with a viscous fluid of density ρ and kinematic viscosity coefficient ν. begin with an elementary experiment illustrating Newton’s friction law (figure 15. Take. per second. The flow in between is plane parallel at velocity v = (vx .73) which depends on r0 and θ0 . so each cm3 of the liquid receives. a heat input F ∂v y Fu = ≥ 0. 0) but it has a gradient ∂v y /∂ x in the x-direction. 15. To pull it out at a constant velocity u.75) D= 2 Ab 2A ∂x It is positive. so tangential or shear forces appear and there is friction which causes the plate to resist being pulled out. a force F.4 Disks A rotating collapsing cloud inevitably flattens because the gravitational pull perpendicular to the rotation axis will. Because the gas parcels have different initial specific angular momenta. at some radius rcenfug .

in complete analogy to (15. dr Let us denote the derivative with respect to r by a prime and with respect to time by a dot. A torque τ then has to be applied to the middle cylinder in order to sustain the motion.75). Again the molecules are at rest at the cylinder walls and there is a shear flow in between. C2 .77) . we get. A rotational analog of this system consists of three coaxial cylinders of height h (figure 15. C3 decrease with radius. τ = νρr dω 4πr hr. the angular velocities of the cylinders C1 . from everyday’s life. D = τ ω/4πr hb or (15. The inner and outer cylinder are corotating.10. for air. C3 exerts on C2 the torque τ (r ) = νρr dω 2πr hr dr (15.12. To move a sheet at a constant speed through a viscous fluid. Because the work expended per second equals τ ω after (11. at 20 ◦ C. The thin layers between the cylinders are filled with liquid or gas. the kinematic viscosity ν is 15 times larger but the density ρ is 830 times smaller. For example. one needs a force F after (15. It equals force per unit area multiplied by area and by radius. water has ν = 10−2 cm2 s−1 and ρ = 1 g cm−3 .11). In the marginally modified experiment of figure 15.Disks 495 y u 2b x Figure 15. the middle one has a different angular velocity ω = ω0 .4).74). the dissipation rate per cm3 of gas.76) D(r ) = νρ(r ω )2 ≥ 0.

When we remove C2 in the experiment. 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 C3 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 ω 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 C2 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 ω +d ω 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 C1 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 r−dr 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 r+dr 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 r 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 1111111111111111111111111111111111111111111111111111111111111 0000000000000000000000000000000000000000000000000000000000000 Figure 15. the inner and outer one are corotating at angular velocity ω0 . ω −d ω trying to retard it. if the middle cylinder C2 has a thickness r . the one in the middle has ω = ω0 .11. The same considerations apply to C1 and C2 .496 Stars and their formation o 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 ω 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 ωo 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 r−b 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 r+b r 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111111111111111111111111111111111 00000000000000000000000000000000000000000000000000000000000000 11111111111111111111111111111111