Ebook Spatiotemporal Random Fields Second Edition Theory and Applications PDF Full Chapter PDF

Spatiotemporal Random Fields, Second
Edition: Theory and Applications -

eBook PDF
Visit to download the full and correct content document:
https://ebooksecure.com/download/spatiotemporal-random-fields-second-edition-theo
ry-and-applications-ebook-pdf/
Spatiotemporal Random
Fields
This page intentionally left blank
Spatiotemporal Random
Fields
Theory and Applications
Second Edition
George Christakos
Department of Geography, San Diego State University,
San Diego, California, USA;
Institute of Islands and Coastal Ecosystems, Ocean College,
Zhejiang University, Zhoushan, Zhejiang, China
Elsevier
Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
Copyright © 2017 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying, recording, or any information storage and retrieval system, without
permission in writing from the publisher. Details on how to seek permission, further information about the
Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance
Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the Publisher (other
than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience broaden our
understanding, changes in research methods, professional practices, or medical treatment may become
necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using
any information, methods, compounds, or experiments described herein. In using such information or methods
they should be mindful of their own safety and the safety of others, including parties for whom they have a
professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any
liability for any injury and/or damage to persons or property as a matter of products liability, negligence or
otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the
material herein.
Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the Library of Congress
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
ISBN: 978-0-12-803012-7
For Information on all Elsevier publications visit our

website at https://www.elsevier.com/books-and-journals
Publisher: Candice Janco

Acquisition Editor: Marisa LaFleur
Editorial Project Manager: Marisa LaFleur
Production Project Manager: Paul Prasad Chandramohan
Designer: Greg Harris
Typeset by TNQ Books and Journals
Dedicated
to
Yongmei
Contents
Preface ................................................................................................................................................. xv
CHAPTER I Space, Time, SpaceeTime, Randomness, and Probability .................. 1

1. Introduction ...................................................................................................................1
2. SpaceeTime Continuum and Kolmogorov Probability Space ....................................4
2.1 SpaceeTime Arguments: Points, Lags, Separations, and Metrics....................... 4
2.2 Transformations and Invariance in SpaceeTime ............................................... 17
2.3 SpaceeTime Interpretations................................................................................ 23
2.4 Functions of SpaceeTime Arguments................................................................ 26
3. Random Variables in SpaceeTime.............................................................................31
3.1 Kolmogorov’s Probability Theory ...................................................................... 31
3.2 Useful Inequalities............................................................................................... 35
3.3 Convergence of Random Variable Sequences .................................................... 37
CHAPTER II Spatiotemporal Random Fields........................................................... 39
1. Introduction .................................................................................................................40
1.1 The SpaceeTime Component ............................................................................. 41
1.2 The Randomness Component ............................................................................. 42
2. Characterization of Scalar Spatiotemporal Random Fields .......................................42
2.1 Probabilistic Structure ......................................................................................... 43
2.2 The Characteristic Function ................................................................................ 50
2.3 Spatiotemporal Variability Functions: Complete (or Full) and Partial .............. 51
2.4 Analysis in the Spectral Domain ........................................................................ 56
2.5 Data-Independent Spatiotemporal Variability Function ..................................... 57
2.6 Some Noticeable Special Cases of the Spatiotemporal Random
Field Theory ........................................................................................................ 59
2.7 SpaceeTime Separability.................................................................................... 60
3. Physical Insight Behind the Random Field Concept .................................................61
3.1 Random Field Realizations ................................................................................. 61
3.2 Probable Versus Actual ....................................................................................... 63
3.3 Probability and the Observation Effect............................................................... 64
3.4 Self-consistency and Physical Fidelity ............................................................... 65
4. Geometry of Spatiotemporal Random Fields.............................................................69
5. Vector Spatiotemporal Random Fields .......................................................................70
6. Complex Spatiotemporal Random Fields...................................................................73
7. Classifications of the Spatiotemporal Random Field Model .....................................73
7.1 First Classification: Discrete Versus Continuous Arguments............................. 74
7.2 Second Classification: Scalar Versus Vector Random Fields and Arguments... 74
vii
viii CONTENTS
7.3 Third Classification: Probability Law Shapes .................................................... 74

7.4 Fourth Classification: SpaceeTime Variability .................................................. 75
7.5 Fifth Classification: Spatiotemporal Random Field Memory Versus
Independence ....................................................................................................... 77
8. Closing Comments......................................................................................................78
8.1 The Methodological Importance of SpaceeTime .............................................. 78
8.2 A Conceptual Meeting Point for Modelers and Experimentalists ..................... 80
8.3 There Is No Assumptionless Modeling .............................................................. 81
CHAPTER III SpaceeTime Metrics ......................................................................... 83
1. Basic Notions ..............................................................................................................83
1.1 Formal and Physical Aspects of SpaceeTime Metric Determination ............... 84
1.2 SpaceeTime Metric Forms ................................................................................. 87
1.3 Derived SpaceeTime Metrics............................................................................. 92
1.4 SpaceeTime Metric Differentials ....................................................................... 93
1.5 Specifying SpaceeTime Relationships in the Covariance Function ................. 95
2. Covariance Differential Formulas.............................................................................100
3. SpaceeTime Metric Determination From Physical Considerations........................106
4. Examples ...................................................................................................................108
5. Concerning the Zeta Coefficients .............................................................................117
6. Closing Comments....................................................................................................118
CHAPTER IV SpaceeTime Correlation Theory ..................................................... 121
1. Focusing on SpaceeTime Variability Functions......................................................121
1.1 Basics of SpaceeTime Correlation Theory...................................................... 122
1.2 Physical Investigations Based on SpaceeTime Correlation Theory................ 123
2. SpaceeTime Variability Functions in Terms of Scalar SpaceeTime Statistics......124
2.1 Locality: One-Point SpaceeTime Variability Functions.................................. 125
2.2 Nonlocality: Omnidirectional Two-Point SpaceeTime Variability
Functions ........................................................................................................... 127
2.3 Nonlocality: Direction-Specific SpaceeTime Variability Functions ............... 133
2.4 Physical Considerations and Assumptions of SpaceeTime Variability
Functions ........................................................................................................... 133
2.5 Formal and Physical Covariance Permissibility ............................................... 137
3. Basic Properties of Covariance Functions................................................................139
4. CrosseSpaceeTime Variability Functions...............................................................141
5. Correlation of Gaussian and Related Spatiotemporal Random Fields ....................144
5.1 General Considerations ..................................................................................... 144
5.2 Gaussian Properties ........................................................................................... 144
CONTENTS ix
6. Correlation Theory of Complex Spatiotemporal Random Fields ............................146

6.1 Basic Notions .................................................................................................... 146
6.2 Other Types of Complex Covariance Functions............................................... 148
6.3 Gaussian Complex Spatiotemporal Random Fields ......................................... 151
6.4 Complex-Valued Versus Real-Valued Covariance Functions
of SpaceeTime Homostationary Random Fields ............................................. 152
6.5 Some Methodological Considerations .............................................................. 153
CHAPTER V Transformations of Spatiotemporal Random Fields......................... 155
1. Introduction ...............................................................................................................155
2. Fourier Transformation .............................................................................................157
2.1 Characteristic Functions .................................................................................... 157
2.2 Harmonizable Random Fields and Covariance Functions ............................... 158
2.3 Transfer Function and Evolutionary Mean Power............................................ 165
2.4 Fourier Transform of Vector Spatiotemporal Random Fields.......................... 167
3. Space Transformation ...............................................................................................168
3.1 Basic Notions .................................................................................................... 168
3.2 Space Transformation of Spatiotemporal Random Fields................................ 173
3.3 Space Transformation for Spatiotemporal Variability Functions..................... 175
3.4 Space Transformation in the Simulation of Spatiotemporal
Random Fields................................................................................................... 177
3.5 Space Transformation in the Solution of Stochastic Partial Differential
Equation............................................................................................................. 180
4. The Traveling Transformation ..................................................................................180
4.1 Basic Notions .................................................................................................... 181
4.2 Determination of the Traveling Vector ............................................................. 186
4.3 Traveling Transformation in Spatiotemporal Random Field Estimation:
The SpaceeTime Projection Technique ........................................................... 195
5. Closing Comments....................................................................................................201
CHAPTER VI Geometrical Properties of Spatiotemporal Random Fields ............ 203
1. Introduction ...............................................................................................................203
2. Stochastic Convergence ............................................................................................204
3. Stochastic Continuity ................................................................................................208
3.1 Basic Types of Stochastic Continuity ............................................................... 209
3.2 Equivalence, Modification, and Separability .................................................... 212
4. Stochastic Differentiation .........................................................................................216
4.1 Basic Notation and Definitions ......................................................................... 217
4.2 Covariances of Random Field Derivatives ....................................................... 223
4.3 Mean Squarely Differentiability Conditions..................................................... 227
4.4 Almost Surely Differentiability Conditions...................................................... 231
x CONTENTS
5. The Central Limit Theorem......................................................................................233

6. Stochastic Integration................................................................................................234
CHAPTER VII Auxiliary Hypotheses of Spatiotemporal Variation ....................... 239
1. Introduction ...............................................................................................................239
1.1 Hypothesis 1: Homostationarity........................................................................ 241
1.2 Hypothesis 2: Isostationarity............................................................................. 243
1.3 Hypothesis 3: Heterogeneity ............................................................................. 245
1.4 Hypothesis 4: Ergodicity................................................................................... 246
1.5 Hypothesis 5: Separability ................................................................................ 248
1.6 Hypothesis 6: Symmetry ................................................................................... 249
1.7 Hypothesis 7: Locational Divergence ............................................................... 249
2. SpaceeTime Homostationarity.................................................................................250
2.1 Omnidirectional Spatiotemporal Variability Functions .................................... 251
2.2 Direction-Specific Spatiotemporal Covariance Function ................................. 255
2.3 Anisotropic Features.......................................................................................... 255
2.4 Spatiotemporal Variogram and Structure Functions: Omnidirectional
and Direction Specific ....................................................................................... 257
3. Spectral Representations of SpaceeTime Homostationarity ...................................260
3.1 Spectral Functions of SpaceeTime Homostationary Random
Fields ................................................................................................................. 261
3.2 Properties of the Spectral Density Function ..................................................... 266
3.3 Partial Spectral Representations........................................................................ 268
3.4 More on Dispersion Relations........................................................................... 272
3.5 Separability Aspects .......................................................................................... 273
4. The Geometry of SpaceeTime Homostationarity ...................................................276
4.1 Differentiation Formulas: Physical and Spectral Domains .............................. 276
4.2 Stochastic Continuity and Differentiability ...................................................... 285
4.3 Spatiotemporal Random Field Integrability ..................................................... 296
5. Spectral Moments and Linear Random Field Transformations ...............................297
CHAPTER VIII Isostationary Scalar Spatiotemporal Random Fields ................... 303
1. Introduction ...............................................................................................................303
1.1 Basic Considerations ......................................................................................... 303
1.2 Power-Law Correlations.................................................................................... 309
1.3 Physical Considerations of Variogram Functions............................................. 313
2. Relationships Between Covariance Derivatives and SpaceeTime
Isostationarity ............................................................................................................314
3. Higher-Order Spatiotemporal Variogram and Structure Functions .........................319
4. Separable Classes of SpaceeTime Isostationary Covariance Models.....................320
5. A Survey of SpaceeTime Covariance Models ........................................................324
CONTENTS xi
6. Scales of Spatiotemporal Dependence and the Uncertainty Principle ....................329

6.1 Scales for Spatiotemporal Random Fields With Restricted SpaceeTime
Variability .......................................................................................................... 330
6.2 Relationships Between Physical and Spectral Domains: The Uncertainty
Principle............................................................................................................. 334
7. On the Ergodicity Hypotheses of Spatiotemporal Random Fields ..........................336
CHAPTER IX Vector and Multivariate Random Fields.......................................... 347
1. Introduction ...............................................................................................................347
2. Homostationary and Homostationarily Connected CrosseSpatiotemporal
Variability Functions and CrosseSpectral Density Functions.................................349
2.1 Basic Notions and Interpretations ..................................................................... 350
2.2 Geometry of Vector Spatiotemporal Random Fields ....................................... 355
3. Some Special Cases of Covariance Functions .........................................................356
4. Solenoidal and Potential Vector Spatiotemporal Random Fields ............................362
5. Partial Cross-Covariance and Cross-Spectral Functions ..........................................365
6. Higher-Order CrosseSpatiotemporal Variability Functions ....................................366
7. Isostationary Vector Spatiotemporal Random Fields ...............................................369
7.1 Direct (Lag-Based) SpaceeTime Isostationarity.............................................. 369
7.2 Composite Lag-FieldeBased SpaceeTime Isostationarity.............................. 372
7.3 Links With Solenoidal and Potential Spatiotemporal
Random Fields................................................................................................... 378
8. Effective Distances and Periods ...............................................................................381
CHAPTER X Special Classes of Spatiotemporal Random Fields......................... 383
1. Introduction ...............................................................................................................383
2. Frozen Spatiotemporal Random Fields and Taylor’s Hypothesis............................384
2.1 Basic Notions .................................................................................................... 385
2.2 Spectral Domain Analysis................................................................................. 389
2.3 Differential Equation Representations .............................................................. 391
2.4 Extensions of the Frozen Random Field Model............................................... 395
2.5 Integrals of Frozen Spatiotemporal Random Fields......................................... 399
2.6 Vector Frozen Spatiotemporal Random Fields ................................................. 399
3. Plane-Wave Spatiotemporal Random Fields ............................................................400
4. Lognormal Spatiotemporal Random Fields..............................................................402
5. Spherical Spatiotemporal Random Fields ................................................................402
6. Lagrangian Spatiotemporal Random Fields .............................................................407
CHAPTER XI Construction of Spatiotemporal Probability Laws .......................... 409
1. Introduction ...............................................................................................................409
2. Direct Probability Density Model Construction Techniques ...................................411
2.1 The Independency Techniques .......................................................................... 412
xii CONTENTS
2.2 The Spherical Symmetry Technique................................................................. 412

2.3 The Transformation Technique ......................................................................... 413
3. Factora-Based Probability Density Model Construction Techniques ......................414
4. Copula-Based Probability Density Model Construction Techniques.......................418
5. Stochastic Differential EquationeBased Probability Density Model
Construction Techniques...........................................................................................421
5.1 The Transformation of Variables Approach ..................................................... 422
5.2 The Characteristic Function Approach ............................................................. 425
5.3 The Functional Approach.................................................................................. 426
6. Bayesian Maximum EntropyeBased Multivariate Probability Density
Model Construction Techniques ...............................................................................428
7. Methodological and Technical Comments ...............................................................431
CHAPTER XII Spatiotemporal Random Functionals ............................................. 433
1. Continuous Linear Random Functionals in the SpaceeTime Domain ...................433
1.1 Basic Notions .................................................................................................... 433
1.2 Generalized Fourier Transform ......................................................................... 436
1.3 SpaceeTime Characteristic Functionals........................................................... 439
1.4 Functional Derivatives....................................................................................... 441
2. Gaussian Functionals ................................................................................................447
CHAPTER XIII Generalized Spatiotemporal Random Fields ................................ 455
1. Basic Notions ............................................................................................................455
1.1 The Notion of Generalized Spatiotemporal Random Field.............................. 456
1.2 Generalized Spatiotemporal Random Field Properties and Physical
Significance ....................................................................................................... 461
1.3 Homostationary Generalized Spatiotemporal Random Fields ......................... 464
2. Spatiotemporal Random Fields of Orders n/m..........................................................468
2.1 Departure From SpaceeTime Homostationarity .............................................. 468
2.2 SpaceeTime Detrending ................................................................................... 470
2.3 Ordinary Spatiotemporal Random Field-n/m Representations of the
Generalized Spatiotemporal Random Field-n/m ............................................... 474
2.4 Determination of the Operator Qn/m and Its Physical Significance .................. 475
3. The Correlation Structure of Spatiotemporal Random Field-n/m ............................477
3.1 SpaceeTime Functional Statistics .................................................................... 477
3.2 Generalized Spatiotemporal Covariance Functions.......................................... 479
3.3 Generalized Spectral Representations and Permissibility of Generalized
Covariances........................................................................................................ 481
3.4 Generalized Covariance Function Models........................................................ 484
CONTENTS xiii
4. Discrete Linear Representations of Spatiotemporal Random Fields .......................490

4.1 SpaceeTime Random Increments .................................................................... 490
4.2 SpaceeTime Variogram Analysis ..................................................................... 495
CHAPTER XIV Physical Considerations................................................................ 501
1. Spatiotemporal Variation and Laws of Change........................................................501
2. Empirical Algebraic Equations.................................................................................504
3. Physical Differential Equations ................................................................................506
4. Links Between Stochastic Partial Differential Equation and Generalized
Random Fields ..........................................................................................................512
4.1 Links in Terms of the Random Functional....................................................... 513
4.2 Links in Terms of the Detrending Operator ..................................................... 515
5. Physical Constraints in the Form of Integral Relationships, Domain
Restrictions, and Dispersion Equations ....................................................................518
CHAPTER XV Permissibility in SpaceeTime........................................................ 521
1. Concerning Permissibility.........................................................................................521
2. Bochnerian Analysis .................................................................................................522
2.1 Main Results...................................................................................................... 523
2.2 Limitations of Bochnerian Analysis ................................................................. 525
3. Metric Dependence ...................................................................................................528
4. Formal and Physical Permissibility Conditions for Covariance Functions .............529
4.1 Permissibility Conditions for SpaceeTime Homostationary Covariance
Functions ........................................................................................................... 530
4.2 Permissibility Conditions for SpaceeTime Isostationary Covariance
Functions ........................................................................................................... 532
4.3 Permissibility Conditions for Generalized Spatiotemporal Covariance
Functions ........................................................................................................... 535
4.4 Permissibility Conditions for Spatiotemporal Covariance Matrices ................ 537
5. More Consequences of Permissibility ......................................................................540
CHAPTER XVI Construction of Spatiotemporal Covariance Models ................... 543
1. Introduction.............................................................................................................543
2. Probability Density FunctioneBased and Related Techniques .............................545
2.1 Linking Directly Covariance Models and Probability Density
Functions ......................................................................................................... 545
2.2 Using Polynomial-Exponential Functions ...................................................... 548
2.3 Using Spectral Functions ................................................................................ 550
3. Delta and Related Techniques ................................................................................552
3.1 Basic Decomposition....................................................................................... 552
xiv CONTENTS
3.2 Normalized Angular Spectrum Decomposition.............................................. 554

3.3 Normalized Frequency Spectrum (or Coherency Function)
Decomposition................................................................................................. 555
4. Space Transformation Technique ...........................................................................557
5. Physical Equation Techniques ................................................................................560
5.1 Covariance Construction From Stochastic Partial Differential
Equation Representations ................................................................................ 560
5.2 Covariance Construction From Algebraic Empirical Relationships .............. 570
6. Closed-Form Techniques ........................................................................................572
7. Integral Representation Techniques........................................................................580
8. SpaceeTime Separation Techniques ......................................................................582
9. Dynamic Formation Technique ..............................................................................586
10. Entropic Technique .................................................................................................587
11. Attribute and Argument Transformation Techniques ............................................588
11.1 Attribute Transformation............................................................................... 588
11.2 Argument Transformation ............................................................................. 589
12. Cross-Covariance Model Construction Techniques ...............................................590
13. Revisiting the Role of Physical Constraints...........................................................593
14. Closing Comments..................................................................................................594
Exercises ............................................................................................................................................ 597
References.......................................................................................................................................... 643
Appendix............................................................................................................................................ 653
Index ................................................................................................................................................. 665
Preface
The study of real-world phenomena relies on theories of natural (physical, biological, etc.) sciences
that generally involve mathematical models. These models are usually defined by a set of equations
and supplemented by a set of logical reasoning rules for rigorously translating the quantitative analysis
results into meaningful statements about the phenomenon of interest. Additionally, and very impor-
tantly, the real-world study of a phenomenon is faced with various sources of uncertainty, ontic,
and epistemic (including phenomenal, technical, conceptual, and computational sources related to
quantitative modeling, data selection, and processing). As a result, exact deterministic model solutions
in terms of well-known analytic functions often turn out to be unrealistic and lack any visible means of
meaningful interpretation.
In light of the above considerations, in most real-world applications the mathematical models that
we currently use to describe the attributes of a natural phenomenon are stochastic in nature, i.e., these
attributes and the associated boundary/initial conditions are represented by random fields with argu-
ments in a composite spaceetime domain. In this domain, space represents the order of coexistence,
and time represents the order of successive existence of the attribute. Randomness manifests itself as
an ensemble of possible realizations regarding the attribute distribution, where the likelihood that each
one of these possible realizations occurs is expressed by the corresponding probability law. Thus,
spatiotemporal random model solutions are considerably more flexible and realistic than the determin-
istic single-valued solutions. Attribute distributions are well represented by theoretical probability
laws, and this permits us to calculate various spaceetime properties of these distributions with reason-
able accuracy.
The above considerations are the primary reasons for devoting this book to the spatiotemporal
random field theory and its potential applications in natural sciences. In this context, for any such the-
ory there is first the mathematical problem of analyzing, as far as possible, the stochastic model gov-
erning the relevant attributes together with the available data sets (hard or exact and soft or uncertain,
in general), and of finding as realistic and complete a solution as possible to the problem of interest that
maintains good contact with the real-world phenomenon in conditions of in situ uncertainty. Next
comes the interpretation (mathematical and physical) of the conclusions thus obtained, and their uti-
lization to make informative predictions. It should be pointed out that certain exact models and equa-
tions have played very important roles in the study of natural phenomena. It should also be noticed that
because many models and equations describing real-world phenomena are necessarily complicated
(multiparametered, highly nonlinear, and heterogeneous, whereas potentially critical features of the
phenomena remain unspecified), it is very useful to understand what qualitative features these models
and equations might possess, since they have been proven to offer an invaluable guide about the phe-
nomenon of interest.
Certainly, there are several important issues related to the distinction between theory and imple-
mentation. Concerning the in situ implementation of random field theory and techniques, one should
be aware that, although the fact is not always appreciated, the real-world complexities of the phenom-
ena mentioned above also mean that approximate techniques used as part of implementation could run
into hidden complications that have a tendency to distract attention from more useful issues. The sit-
uation may also be partially the fault of those of us working in this discipline, when we occasionally
xv
xvi PREFACE
propose abstract theories for the sake of greater “generality.” Yet this is not an excuse for the fact that,
as real-world experience shows, in the vast majority of cases the “ineffectiveness” is not a feature of
the theory or the modeling technique used but rather of the practitioners often attempting to use them
in a “black-box” manner. In this framework, exact models and equations that can be compared with
approximate or numerical results are very useful in checking the validity of approximation techniques
used in an application.
In addition to the above reasons for devoting this book to the theory of spatiotemporal random
fields, it should be noted that although much work has been done concerning the subject, it is often
not generally known because of the plethora of disciplines, journals, and mathematical terminologies
and notations in which it has appeared. It is hoped that one beneficial effect of the present effort will be
to save the interested readers from spending their time rediscovering already known results. And I
hope the present attempt to characterize the known results invariantly will help readers to identify
any new findings that may emerge. Throughout the book, for the basic issue regarding fundamental
concepts of probability, statistics, and random variables, I refer the reader to other texts, when
necessary.
Naturally, I begin by introducing the basic notions of the spaceetime continuum (points, lags, met-
rics, and geometrical transformations), mathematical field and related functions, probability, uncer-
tainty, and randomness (Chapter 1). Two chapters (Chapters 2 and 4) are devoted to the
presentation of standard results of the ordinary spatiotemporal random field theory, including much
of the terminology used later in the book. Among these two chapters I have interpolated one chapter
(Chapter 3) on spaceetime metrics. Its position is due to the fact that the spaceetime metric properties
can be used to elucidate the validity of certain random field issues introduced in the preceding two
chapters. This chapter also discusses the classification of spaceetime metrics for scalar and vector
random fields, and a physical lawebased metric determination technique is outlined and applied in
several cases. Intuitively, a natural attribute represented as a spatiotemporal random field is projected
on the physical (real) domain. Yet, there may be constructed other domains on which an attribute could
be projected. Such domains are the spectral domain, the reduced dimensionality, and the traveling do-
mains, which provide equivalent representations of the attribute defined on it. And while one’s intu-
ition may be better adapted to the physical domain, in certain cases it may be more convenient to
work in the alternative domains. So in Chapter 5 we discuss important concepts and methods associ-
ated with these alternative domains. Chapter 6 focuses on spatiotemporal random field geometry (con-
tinuity, differentiability, and integrability). This is one of the subjects that would warrant a book of its
own and, thus, I had to be very selective in the choice and manner of the material presented. Because of
its special physical and mathematical interest, the topic of homostationarity (space homogeneous/time
stationary) was given a chapter of its own (Chapter 7). Similar reasons are valid for devoting Chapter 8
to isostationary (space isotropic/time stationary) random fields. In this chapter a large number of
spatiotemporal variability functions (covariance, variogram, and structure functions of high order)
are presented. Chapter 9 deals with multivariate and vectorial random fields varying in the spaceetime
domain, including their main mathematical features and differences as regards their interpretation
(mathematical and physical). In Chapter 10, I discuss a selected group of spatiotemporal random fields
with special properties of particular interest to applications (this group includes the frozen random
field and its variations, the plane-wave, the lognormal, the spherical, and the Lagrangian random
fields). Chapter 11 focuses on techniques for constructing multivariate probability density functions
that offer a complete characterization of the spatiotemporal random field in stochastic terms. Due
PREFACE xvii
to their fundamental role in the study of spaceetime heterogeneous random fields, an entire chapter
(Chapter 12) is devoted to the theory of spatiotemporal random functionals. Indeed, the functional
description of randomness naturally involves more complex mathematics, but it has its rewards on
both theoretical and application grounds (e.g., many real-world phenomena and their measurements
need to be expressed in terms of random functionals). Chapter 13 provides a rather detailed account
of the theory of spaceetime heterogeneous (generalized) random fields that is useful in the case of
natural attributes characterized by complex variations and patterns (varying trends, fluctuations of
varying magnitude, coarse-grained measurements, etc.). Interestingly, since the first edition (1992)
of the present book, only certain limited aspects of this theory have been thoroughly discussed in
the literature. Chapter 14 emphasizes the importance of accounting for physical laws, scientific
models, and empirical relationships in the development of a spatiotemporal random field theory.
This valuable core knowledge concerning a phenomenon is usually quantitatively expressed in terms
of stochastic partial differential equations, several of which are reviewed in this chapter. Admittedly, I
only tangentially deal with the solution of these equations and relevant topics. The strongest reason for
excluding the omitted topics is that each would fill another book (I do, of course, give references to the
relevant literature). Chapter 15 presents a series of permissibility criteria for spaceetime covariance
functions (ordinary and generalized) that are widely used in applied stochastics. Certain of these
criteria are necessary and sufficient, whereas some others are only sufficient, but they have the advan-
tage that they refer directly to the covariance function itself. Further, some important practical impli-
cations of permissibility in different kinds of applications are discussed. Chapter 16 presents a rather
large number of techniques for constructing spaceetime covariance models, which can be used in a
variety of scientific applications. Formal and substantive model-building techniques are examined,
each of which has its own merits and limitations. There are many covariance model construction tech-
niques in use and they could not all be discussed in full: my choice of what to present in detail and what
to mention only as a reference simply reflects my personal taste and experience.
The book has benefited by the contributions in the field of my colleagues, collaborators, and stu-
dents during the last few decades. The second edition of the book was written mainly during my leave
of absence year at the Ocean College of Zhejiang University (China). I am grateful for the support of
Zhejiang University and of the CNSF (Grant no. 41671399). I am also grateful to my colleagues at the
Ocean College, particularly Professor Jiaping Wu, who did everything possible to create the right envi-
ronment for writing such a book. Last but not least, this work would not have been completed without
Yongmei’s infinite patience during the long process of writing the book, which is why to Yongmei this
book is dedicated.
George Christakos
CHAPTER
SPACE, TIME, SPACEeTIME,

RANDOMNESS, AND
PROBABILITY
I
CHAPTER OUTLINE
1. Introduction .......................................................................................................................................1
2. SpaceeTime Continuum and Kolmogorov Probability Space ..................................................................4
2.1 SpaceeTime Arguments: Points, Lags, Separations, and Metrics...........................................4
2.2 Transformations and Invariance in SpaceeTime.................................................................17
2.3 SpaceeTime Interpretations.............................................................................................23
2.4 Functions of SpaceeTime Arguments ...............................................................................26
3. Random Variables in SpaceeTime.....................................................................................................31
3.1 Kolmogorov’s Probability Theory .......................................................................................31
3.2 Useful Inequalities ..........................................................................................................35
3.3 Convergence of Random Variable Sequences .....................................................................37
1. INTRODUCTION
Due to its importance in almost any scientific discipline, random field theory is an active area of
ongoing research. Significant work has been done, indeed, in the theory of spatial random field, but
much less so in the theory and applications of spatiotemporal random field, where many important
topics still need to be studied and notions to be advanced. On the other hand, many practitioners argue
that random field remains a tough theory to work with, due to the difficulty of the “nondeterministic”
mathematics involved. This kind of mathematics is also known as stochastics, a term that generally
refers to the mathematical representation of phenomena that vary jointly in space and time under
conditions of in situ uncertainty. In a formal sense, deterministic mathematics can be viewed as a
special case of stochastics under the limiting and rare conditions that the phenomenon under study is
known with certainty. To phrase it in more words, stochastics deals with any topic covered by the
deterministic theory of functions, and, in addition, the presence of uncertainty (technically, sometimes
characterized as randomness) makes stochastics a much larger, considerably more complex and surely
more challenging subject than the deterministic theory of functions. Historically, the development of
stochastics can be traced back in the works of some of the world’s greatest scientists, such as Maxwell
(1860), Boltzmann (1868), Gibbs (1902), Einstein (1905), Langevin (1908), Wiener (1930), Heisen-
berg (1930) Khinchin (1934), Kolmogorov (1941), Chandrasekhar (1943), Lévy (1948), Ito (1954),
Gel’fand (1955), von Neumann (1955), Yaglom (1962), and Bohr (1963), among many others.
Spatiotemporal Random Fields. http://dx.doi.org/10.1016/B978-0-12-803012-7.00001-5

Copyright © 2017 Elsevier Inc. All rights reserved.
1
2 CHAPTER I SPACE, TIME, SPACEeTIME, RANDOMNESS, AND PROBABILITY
It should be pointed out that random field modeling is at the heart of many theoretical advances in
stochastics. It has led to the development of new mathematical concepts and techniques, and, also, it
has raised several interesting theoretical questions worthy of investigation. Computational random
field modeling, on the other hand, deals with computational and numerical aspects of the systematic
implementation of random field theory in the study of complex real-world phenomena, which covers
almost every scientific and engineering discipline. The term “computational” used here should not
create any confusion with computational mathematics and statistics: while computational mathematics
and statistics are concerned with numbers, computational random fields are concerned with physical
quantities.1
In applied sciences, random field modeling deals with spatiotemporal natural attributes, that is,
real-world attributes that develop simultaneously in space and time, and they are measurable or
observable. These natural attributes occur in nearly all the areas of applied sciences, such as ecology
and environment (e.g., concentrations of pollutants in environmental mediadwater/air/soil/biota),
climate predictions and meteorology (e.g., variations of atmospheric temperature, density, moisture
content, and velocity), hydrology (e.g., water vapor concentrations, soil moisture content, and pre-
cipitation data consisting of long time series at various locations in space), oil reservoir engineering
(e.g., porosities, permeabilities, and fluid saturations during the production phase), environmental
health (e.g., human exposure indicators and doseeeffect associations), and epidemiology (e.g., breast
cancer incidence, and Plague mortality). In all these cases, a central issue of random field modeling is
factual accuracy in the informational statements that describe what was observed and experienced.
For sure, the application of random fields in the study of real-world phenomena is not an un-
constrained theoretical exercise. It rather follows certain methodological criteria that involve the
identification of the bounds of the specific application, the evaluation of the context in making sense of
empirical data, a focus on probative evidence from diverse sources, an openness to inductive insights,
and an in-depth analysis justified by the generation of interpretable results. Induction, interpretation,
and abstraction are not competing objectives in this approach, but mutually reinforcing operations.
Random field modeling is concerned, although to varying extends, about both its internal and
external validity. Internal validity relates to whether the findings or results of the random field modeling
relate to and are caused by the phenomena under investigation, and not by other unaccounted for in-
fluences. On the other hand, external validity is assessed by the extent to which these findings or results
can be generalized, and thus applied to other real-world situations. While internal validity is the primary
concern of random field modeling, external validity is also a very important goal.
In this Chapter, I present the fundamentals regarding the conceptual and quantitative characterization
of spaceetime (or space/time, or spacetime) within which random fields will be defined in subsequent
chapters. Arguably, there are many issues surrounding the use and nature of the notion of “spaceetime”
in scientific modeling, and some of them are even controversial. Yet, “space” and “time” are funda-
mental concepts that were invented by humans in their effort to describe Nature, but the map is not the
territory. The formulation of spaceetime introduced in this chapter has the considerable merit of
maintaining close contact between mathematical description and physical reality. Among the central
goals of this formulation are to direct us toward a correct interpretation of spaceetime, and, to the extend
possible, to help us avoid asking the wrong questions and focus on the insignificant issues.
1
Surely, physical quantities are represented by numbers, but they also possess physical meaning and content (which are
ignored by computational mathematics and statistics), and they are also associated with spaceetime arguments.
1. INTRODUCTION 3
In this book, random quantities like the random variable, the random field, and the vector random
field will be studied in both the physical (real) and the frequency domains. Notationally, a random
variable is represented by lowercase Latin letters (x, y, etc.), a random field by uppercase Latin letters
(X, Y, etc.), and a vector random field by uppercase bold Latin letters (X, Y, etc.). Lowercase Greek
letters (c, j, etc.) denote random variable or random field values (realizations), and lowercase bold
Greek letters (c, j, etc.) denote vector random field values. The frequency domain counterparts of the
above random quantities will be denoted by inserting the symbol w above them (e.g., X). e The N
denotes the set of natural numbers (f0; 1; 2; .g or f1; 2; .g, depending on the context). The R1 (or R)
and C denote, respectively, the spaces of real and complex numbers. In the latter case, z ¼ c þ ij ˛ C,
pffiffiffiffiffiffiffi
where i ¼ 1, and the c h Re(z) ˛ R1 and j h Im(z) ˛ R1 denote, respectively, the real and
imaginary parts of z. The complex conjugate of z is denoted by z ¼ c ij ˛ C; also,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jzj ¼ jz j ¼ c2 þ j2 ˛R1þ;0 (i.e., the positive part of the real line including zero) is the modulus of z,
z. The symbol T is sometimes used to represent the time domain as a subset of R1 (T 4 R1), which is
in agreement with the physical irreversibility of most real-world phenomena. On the real line R1, I use
the convention for closed, open, and half-open intervals written as [c, j], (c, j), [c, j), and (c, j].
Also, Rn is the Euclidean space of dimension n 1.
Scalar, vector, and matrix notation will be used, noticing that scalars can be seen as tensors of rank
n
zero, vectors have rank Pnone, and matrices have rank two. A vector in R will be denoted as
s ¼ ðs1 ; .; sn Þ or s ¼ i¼1 si εi, where εi, i ¼ 1; .; n, are base vectors along the coordinate di-
rections. The simplest choice of an (orthonormal) basis is, of course, the set of unit length vectors εi,
where the ith component is 1 and all others 0. For any two vectors s and s0 ¼ s01 ; .; s0n , their scalar
P 1
product is defined as s$s0 ¼ ni¼1 si s0i . The length of the vector s is jsj ¼ ðs$sÞ2 , and
1
js s0 j ¼ ððs s0 Þ$ðs s0 ÞÞ2 is the distance between s and s0 in Rn. The spaceetime domain is
denoted as Rnþ1, or as Rn,1, if we want to explicitly distinguish space from time (for the same reason,
we may also denote the spaceetime domain as the Cartesian product Rn R1 or Rn T). I.e., in the
case of spaceetime, the domain dimensionality increases to n þ 1 by including the additional term s0
or t representing time. Table 1.1 lists some commonly used symbols. Table 1.2 gives a list of special
Table 1.1 Commonly Used Symbols

Symbol Mathematical Expression
R
Dirac delta dsdðsÞ ¼ 1;
dðss0Þ ¼ 0:

Kronecker delta 1 if i ¼ j;
dij ¼
0 otherwise.
8
< þ1 if ði; j; kÞ ¼ ð1; 2; 3Þ or ð2; 3; 1Þ or ð3; 1; 2Þ;
>
Levi-Chivita
εijk ¼ 1 if ði; j; kÞ ¼ ð3; 2; 1Þ or ð1; 3; 2Þ or ð2; 1; 3Þ;
>
:
0 if i ¼ j or j ¼ k or k ¼ i.
Kronecker/Levi-Chivita εijkεipq ¼ djpdkq djqdkp

relationship
Table 1.2 Special Functions and Polynomials

Name Notation
Gamma function. G(n)

Bessel function of 1st kind and vth order. Jv
Modified Bessel function of the 1st kind. In
Modified Bessel function of 2nd kind and vth order. Kv
Gegenbauer polynomials of degrees l, q. Gl,q
functions and polynomials that will be used in the mathematical expression of several results
throughout the book. For the readers’ convenience, the mathematical definitions and basic properties
of these special functions and polynomials are briefly reviewed in the book’s Appendix.
Although many of the theoretical results presented in each chapter of the book (in the form of
propositions and corollaries) will be repeatedly used in subsequent chapters, most proofs and other
details will not, so they will not be discussed. Instead, examples illustrating the most important
application-related aspects of these proofs will be presented when appropriate. A consistent effort is
made throughout the book to keep a balance between abstract mathematical rigor and real-world science.
In many cases, this means that by suppressing certain strict mathematical conditions, a more realistic
representation of the observed phenomenon is achieved, and, also, a richness of new material is produced
(this is, e.g., the case with spaceetime metric). The remaining of this chapter presents a review of the
basic concepts and principles (regarding space, time, field, uncertainty, and probability) around which
the spatiotemporal random field theory will be developed in the following chapters of the book.
2. SPACEeTIME CONTINUUM AND KOLMOGOROV PROBABILITY SPACE

This section presents an overview of the fundamental notions pertaining to the description of the
spaceetime domain shared by the natural phenomena and the mathematical constructions that
represent them. A general point to be stressed is that although on formal mathematics grounds many
aspects of the analysis in the n þ 1-dimensional (spaceetime) domain are similar to those in the
n-dimensional (spatial) domain, essential differences could exist on physical grounds. On the same
grounds, crucial links may emerge between space and time, in which case the deconstruction of the
concept of spaceetime into separate types of space and time may be an unnecessary conceptualization
often leading to unsatisfactory conclusions.
Having said that, I start with the introduction of the different geometrical notions and arguments
that play a central role in the spatiochronological specification of a phenomenon, an attribute or an
event occurring in physical spaceetime, highlighting similarities and differences as they emerge.
2.1 SPACEeTIME ARGUMENTS: POINTS, LAGS, SEPARATIONS, AND METRICS

Generally, a continuum E of spaceetime points is associated with the physical spaceetime domain,
and as such, it is assumed to be equipped with the corresponding metric that specifies “distances”
between points in spaceetime. Operationally, a point in E can be defined in different ways, depending
on the situation and the objectives of the analysis.
2. SPACEeTIME CONTINUUM AND KOLMOGOROV PROBABILITY SPACE 5
Definition 2.1
A point in E is denoted by a vector p, which can be defined, either as an element of the n þ 1-
dimensional domain,
p ¼ ðs1 ; .; sn ; s0 Þ ˛ Rnþ1 ; (2.1a)
where, in notational terms, s1 ; .; sn denote space coordinates and time is considered jointly with space
using the convention s0 ¼ t; or, as a pair of elements
p ¼ ðs; tÞ ˛ Rn;1 ; (2.1b)
where s ¼ ðs1 ; .; sn Þ ˛ Rn is the spatial location vector considered separately than the time instant
t ˛ R1. As noted earlier, instead of Rn,1 some authors use the Cartesian domain notation Rn T, i.e., the
time axis T (4R1) is indicated separately from space Rn.
The point vector p plays a key role in physical sciences, not only because it uniquely specifies a point in
the spaceetime domain of interest, but also because of the numerous functions of p (also known as fields,
see Section 2.4) encountered in physical sciences. In Definition 2.1, either the entire real line may be
considered as the time axis (referring, e.g., to situations in which the attribute of interest is reversible), or,
due to physical requirements (i.e., the attribute is irreversible in the real-world), the domain R1þ;f0g
(restricted to the positive part of the real line including zero) may be more appropriate to represent the time
axis. To phrase it in more words, time may be seen as a coordinate (s0) of the vector p in the Rnþ1 domain,
as in Eq. (2.1a). Intuitively, a spaceetime point in Rnþ1 is considered as a fusion of a space point and a
time point (e.g., the “here and now” exists as a unity not specifying the “here” and the “now” separately;
similar is the interpretation of the “there and then”). Alternatively, time may enter the analysis as a distinct
variable via the vectorescalar pair (s, t) of the Rn,1 domain, as in Eq. (2.1b). Both approaches have their
merits and uses, which will be discussed in various parts of the book. Yet, one should be aware of certain
noticeable consequences of the two approaches in applications, as illustrated in the examples below.
Example 2.1
From a modeling viewpoint, in many cases we may treat space and time on essentially the same formal
footing. By setting, e.g., s0 ¼ iat, the wave operator2 can be written as
v2 X X
n n;0
v2 v2
a2 ½$ 1 ½$; (2.2)
vt 2
i¼1
vsi
2
i¼1
vs2i
and we may view the solution of Poisson’s equation in two dimensions and that of the wave equation in
one dimension as analogous problems. The physical differences between space and time need to be
carefully taken into consideration, though, since certain formal analogies between space and time may
be deceptive. Indeed, the boundary conditions and the initial conditions may enter the problem in
different ways, even though the governing equation may look symmetric (as in the wave operator
above). Also, when we study the spaceetime variation of an air pollutant, the way pollution con-
centration changes across space (the distribution of spatial locations in which pollution exceeds a
critical threshold) can be essentially different than concentration changes as a function of time (fre-
quency of threshold exceedances at each location).
2
This operator expresses certain fundamental physical laws like the electromagnetic laws.
Our discussion so far of the issues surrounding the essence and use of the term “spaceetime” lead
to the first postulate.
Postulate 2.1
The vast majority of real-world data are interrelated both in space and time. This spaceetime
connection is ingrained through physical relations and is welcomed in scientific modeling because it
allows the representation of the spaceetime variation of a natural attribute from the limited number of
data usually available.
This postulate is supported by reality, including the fact that spaceetime coupling is known to
remove possibly unphysical divergences from the moments of the corresponding transport processes
(e.g., Shlesinger et al., 1993). Unfortunately, these crucial facts are often ignored in purely technical
treatments of spaceetime phenomena. Indeed, in statistical inferences (i.e., the inductive process of
inferring from a limited sample valid conclusions about the underlying yet unknown population) the
proper assessment of spaceetime correlations is often problematic, since most standard statistics tools of
data analysis and processing have been developed based on the key premise of independent (physical
relation-free) experiments.
Example 2.2
Working in the classical Newtonian conceptual framework, many practitioners find it tempting to
completely separate the space component s from the time component t. Although convenient, indeed,
this approach is often inadequate in real-world studies. A common example is the model-fitting
procedure in which a valid covariance function of space is fitted at any fixed time (or a valid
covariance function of time is fitted at any fixed distance). However, the resulting model is not
necessarily a valid spatiotemporal covariance model, as has been discussed in Ma (2003b). Further-
more, in ocean studies involving underwater acoustics propagation, the travel time t is related to the
horizontal (s1, s2) and vertical (s3) coordinates in R3,1 by means of
1
s21 þ s22 2 cos q þ s3 sin q yt ¼ 0; (2.3)
where q is the angle of a ray element in a refracting medium with sound speed y (Lurton, 2010). Hence,
the acoustics of the phenomenon imply that space s and time t are closely linked through the physical
relation of Eq. (2.5).
The perspective suggested by Eq. (2.1b) enables the introduction of alternative expressions of
spaceetime point determination, while still accounting for the spaceetime connection posited in
Postulate 2.1. These expressions are called conditional, and the reason for this will become obvious
below. Consider a spaceetime domain represented by the nodes i (i ¼ 1; .; m) of a grid or lattice. An
obvious expression of spaceetime at each grid node i is
ðsi ; ti Þ; i ¼ 1; .m; (2.4a)
denoting location si ¼ ðsi1 ; .; sin Þ and time ti. Eq. (2.4a) assigns the same subscript to space and time,
and, hence, it allows the consideration of a unique time instant at each node i. Sometimes, the physics
of the situation may require that more than one time instants need to be considered at each node, in
which case it is convenient to represent spaceetime at each grid node by

si ; tji ; i ¼ 1; .; m; and j ¼ 1; .; m0 ; (2.4b)
where now ji denotes the time instant considered given that we are at spatial position si (e.g., at a given
node 9, ji ¼ 19 means time instant 1 at node 9). Similarly, when more than one nodes need to be
considered during each instant, we can write,

sij ; tj ; i ¼ 1; .; m; and j ¼ 1; .; m0 ; (2.4c)
where now ij denotes the spatial position given that we are at time instant tj (e.g., at a given instant 9,
ij ¼ 19 means spatial node 1 at instant 9). In a sense, Eqs. (2.4b) and (2.4c) introduce conditional
expressions of the spaceetime point (i.e., time conditioned to space and space conditioned to time,
respectively).
In addition to the physical spaceetime domain one may refer, equivalently, to the location-instant
of a physical attribute occurring in the frequency spaceetime domain F .
Definition 2.2
A point in F is denoted by a vector w, which can be defined, either as an element of the n þ 1-
dimensional domain,
w ¼ ðw1 ; .; wn ; w0 Þ ˛ Rnþ1 ; (2.5a)
or, as a pair of elements of the ( n, 1)-dimensional domain,
w ¼ ðk; uÞ ˛ Rn;1 ; (2.5b)
where k ¼ ðk1 ; .; kn Þ denotes spatial frequency (wavevector) and u denotes temporal (wave scalar)
frequency.
In view of Definition 2.2, as was the case with the representations of Eqs. (2.1a) and (2.1b) being
formally equivalent, the same is valid for the representations of Eqs. (2.5a) and (2.5b), but they may be
both affected by physical context too. The attributes of interest in a real-world study satisfy certain
laws of change that may impose some intrinsic links between the location vector s and the time instant t
or, equivalently, between the wavevector k and the wave frequency u. Specifically, the matter of
spaceetime links is discussed in Section 3 of Chapter VII, in which it is shown that physical laws
representing wave phenomena impose a link between k and u, which are not independent but closely
linked by means of a physical dispersion relation.
Example 2.3
The acoustic pressure X(s, t) of a wave propagating in spaceetime is governed by the classical law

2 v
2
y V Xðs; tÞ ¼ 0;
2
(2.6a)
vt2
where y is the local sound speed. The dispersion relation of this underwater acoustics law in R3,1 that
links k and u is
k2 y2 u2 ¼ 0; (2.6b)
P
where k2 ¼ 3i¼1 ki2 .3 Similarly to Eq. (2.3), underwater acoustics laws representing wave phenomena
impose a strong link between k and u, which are not independent anymore, but they are closely
connected physically by the dispersion relation of Eq. (2.6b). As a result of this dependence, the
spaceetime frequency domain is restricted, thus affecting the form of the spaceetime covariance
function (more details in Chapter VII).
A real (or complex) function of the spaceetime vector p (including the random fields to be
considered, Chapter II) is termed continuous or discrete parameter according to whether its argument p
takes continuous or discrete values. Unless stated otherwise, in the following continuous-parameter
functions will be considered, which is most often the case in applied stochastics. Spaceetime varia-
tion analysis requires the simultaneous consideration of pairs of points defined by the vectors p and p0 ,
in which case the notion of spaceetime lag emerges naturally.
Definition 2.3
The spaceetime lag, Dp ¼ p0 p, between a pair of points p and p0 is defined, either as the single
vector
Dp ¼ ðDs1 ; .; Dsn ; Ds0 Þ ˛ Rnþ1 ; (2.7a)

where Dsi ¼ s0i si (i ¼ 1; .; n; 0) with the convention that the vector components with i ¼ 1; .; n
denote space lags and the component with i ¼ 0 denotes time separation; or, as a pair of vectorescalar
components
Dp ¼ ðh; sÞ ˛ Rn;1 ; (2.7b)

where the spatial vector lag h ¼ ðh1 ; .; hn Þ ˛ Rn , with hi ¼ s0i si (i ¼ 1; .; n), and the time scalar
lag, s ¼ t0 t, are represented separately.4
A direct correspondence can be established between the representations of Eqs. (2.7a) and (2.7b)
by observing that
Dsi ¼ hi ði ¼ 1; .; nÞ;
Ds0 ¼ h0 ¼ s ði ¼ 0Þ.
For simplicity, in the following the notations Dsi and hi (i ¼ 1; .; n; 0) will be used inter-
changeably. As in Definition 2.1, Eq. (2.7a) allows an intrinsic mixing of space and time lags, whereas
Eq. (2.7b) considers them explicitly. As we will see in various parts of the book, this distinction may
have considerable consequences in the study of certain aspects of random field theory and its physical
applications.
Remark 2.1
At this point, it is worth noticing that the above spaceetime notation is sometimes termed the Eulerian
spaceetime coordinate representation, where s and t are allowed to vary independently. Another
3
Here, the dispersion relation results from the requirement that X ðs; tÞ ¼ eiðk$sutÞ is a solution to the physical law.
4
The lag sign, i.e., the order of the spaceetime points p, p0 in the lag can affect mathematical manipulations and also can be
consequential in certain applications (Chapter VII).
possibility is the Lagrangian spaceetime coordinate representation, in which case s and t do not vary
independently, but, instead, s is considered a function of t, which is expressed as s(t). The Eulerian
notation has been used, e.g., in turbulence studies to represent correlations between physical attributes
(pressure or flow velocity) where s ¼ ðs1 ; .; sn Þ specifies a measurement location, s þ h denotes
locations at varying distances h from s, and s is the time increment between times t and t þ s. The
Lagrangian setting has been used to determine correlations for properties of fluid particles passing
through locations s at times t0 (different for each particle), traveling along certain trajectories and
arriving at locations s þ h(t0 þ s) at times t0 þ s. The displacement vector h(t0 þ s) is a random
variable describing the locations at times t þ s of the particles in the ensemble averaging with respect
to the initial locations s at times t0. In this book, Eulerian spaceetime coordinates will be predomi-
nantly considered, whereas the Lagrangian notation will be used only in special cases.
Spaceetime points and lags viewed as vectors, standard vector operations can be applied on them.
A partial list of such operations that are useful in the following is given in Table 2.1 (see also
Appendix). The spaceetime point vector p can be expressed in terms of the unit vectors along the
(orthogonal) coordinate directions (base vectors) εi, i ¼ 1; .; n. The notation for integer powers
involving spaceetime coordinates, pl, where l ¼ (l1,., ln, l0) is a multi-index of nonnegative
Table 2.1 Standard Vector Operations on SpaceeTime Points and Lags

P
n;0
p ¼ ðs1 ; .; sn ; s0 Þ ¼ s i εi
i¼1
Y
n;0 Y
n
pl ¼ sli i ¼ sli i tl0 ; l ¼ ðl1 ; .; ln ; l0 Þ
i¼1 i¼1
X
l
ðs þ hÞl ¼ k! k lk
Cl! s h ; k ¼ jkj; l ¼ jlj; k l
k¼0
X
l
ðt þ sÞl ¼ m! m lm
Cl! t s (2.8aeg)
m¼0

X n k
k 2 2
jsj ¼ si ðk ¼ 1; 2; .Þ
i¼1
vjsj
¼ si jsj1 ¼ zi
vsi
vzk
¼ ðdik zi zk Þjsj1
vsi
vp
¼ εi
vsi
vð p$p0 Þ vp 0 vp0
¼ $p þ p$ (2.9aec)
vsi vsi vsi
vð p p0 Þ vp vp0
¼ p0 þ p
vsi vsi vsi
Continued
Table 2.1 Standard Vector Operations on SpaceeTime Points and Lagsdcont’d

X
n;0
X
n;0
Dp ¼ p p0 ¼ si s0i εi ¼ h i εi
i¼1 i¼1
vDp (2.10aec)
¼ εi ði ¼ 1; .; n; 0Þ
vsi
vDp
¼ εi
vs0i
V$p ¼ V$Dp ¼ n þ 1
(2.11aeb)
V p ¼ V Dp ¼ 0
P
integers such that jlj ¼ n;0i¼1 li and l! ¼ Pi li (Table 2.1), is essential in the study of random fields
representing the spaceetime heterogeneous variation pattern of a physical attribute, and can be
combined with other functions, such as, e.g., pl eg$p (Chapter XIII).5 Also, spaceetime vector dif-
ferentiation operators in terms of p and Dp are useful in the study of derivative random fields, their
covariance, variogram, and structure functions (Chapters VI and VII), as well as the physical laws they
obey (usually expressed in terms of partial differential equations, PDE, Chapter XIV). In such and
similar cases, the proper representation of spaceetime geometry (in terms of coordinates, distances,
lags etc.) allows an efficient approach to random field modeling, which is the subject of this book.
More specifically, if the random field represents a spaceetime heterogeneous attribute distribution
(i.e., one exhibiting a spaceenonhomogeneous/timeenonstationary, or, simply, spaceetime hetero-
geneous pattern),6 the following space/time polynomial functions appear in the analysis (Chapter XIII)
m X
X n
r
hn=m ðs; tÞ ¼ cr;z s11 .srnn tz ; (2.12a)
z¼0 jrj¼0
where n and m are integers representing, respectively, the spatial and temporal orders of the phenomenon
heterogeneity, and cr,z are known coefficients. The notation used in Eq. (2.12a) shows explicitly the
degrees of all spatial and temporal monomials, but it is cumbersome, since it requires keeping track of all
the power exponents (ri, i ¼ 1; .; n, and z). An alternative notation, which turns out to be more effi-
cient, is based on the spaceetime monomial functions pl introduced earlier. In this setting, a spaceetime
polynomial of spaceetime degrees n=m can be expressed as (Christakos and Hristopulos, 1998)
ðn=mÞ
NnX
hn=m ðs; tÞ ¼ ca ha ðs; tÞ; (2.12b)
a¼1
P
n
In these applications, g ¼ (a, b), l ¼ (r, z), and g$p ¼
5
ai si þ bs0 .
i¼1
6
Etymologically, the words homogeneous and heterogeneous come from Ancient Greek. The Greek words mogεn 2 (homo-
genes) and sεrogεn 2 (heterogenes) come from m 2 (homos, “same”) and sεro2 (heteros, “other, different”) respectively,
followed by g no2 (genos, “kind”). The “-ous” is an adjectival suffix.
Table 2.2 Number of Monomials for n [ 2, 3, and Different Combinations

of Continuity Orders n=m
v m N2 (v/m) N3 (v/m)
0 0 1 1
1 0 3 4
0 1 2 2
1 1 6 8
2 0 6 10
2 1 12 20
1 2 9 12
2 2 18 30
n r
where ha ðs; tÞ ¼ pðr;zÞ ¼ sr tz ¼ P si i tz , and Nn ðn=mÞ is the number of monomials that depends on
i¼1
the spatial dimension n and the orders n, m. This notation is surely more compact than the one pre-
sented in Eq. (2.12a), and it involves only a single index a instead of the n þ 1 spaceetime indices
ðr1 ; .; rn ; zÞ. The Nn ðn=mÞ for any n and n=m can be then determined by
X n
ðjrj þ n 1Þ!
Nn ðn=mÞ ¼ ðm þ 1Þ ; (2.12c)
jrj¼0
jrj!ðn 1Þ!
i.e., for jrj n number of monomials it is equal to the permutations of n integers from f0; 1; .; ng that
add up to jrj. Table 2.2 displays the monomials for the n=m combinations that are most commonly used
in applications.
Example 2.4
For illustration, in this example, I chose to focus on the R2,1 domain, where (s1, s2) denote the Cartesian
coordinates associated with the locational vector s, and t is the time scalar. Also, I assume that n=m ¼
2=1 are the space/time orders of attribute nonhomogeneity/nonstationarity. Under these conditions, the
corresponding monomials (12 in number) are listed in Table 2.3.
In light of Definition 2.3, a relevant notion of spaceetime modeling is the spaceetime “distance”
or metric. In applied sciences, metrics are mathematical expressions of the concept of distance in the
spaceetime continuum E (an in depth treatment of the metric notion is presented in Chapter III). In
general, a spaceetime metric would be defined as a function of the pair ð p; p0 Þ. In many cases,
however, the value of the metric function is invariant for each given lag Dp ¼ p0 p, i.e., the metric
is defined as a function of the spaceetime lag (which is more restrictive, yet satisfactory in many
physical situations). These metrics are sometimes referred to as “lag-based metrics” and are denoted
by the absolute value of Dp, i.e., jDpj.7 Using the correct metric plays a crucial role in the study of
7
In a more conventional mathematical sense, jDpj is a real-valued function on R1 satisfying three basic conditions:
(1) jDpj 0 for all Dp (jDpj ¼ 0 only if Dp ¼ 0); (2) jaDpj ¼ jajjDpj for all Dp, a ˛ R1; and
(3) jDp1 þ Dp2j jDp1j þ jDp2j for all jDp1j, jDp2j.
Table 2.3 Monomials in R2,1 With n=m [ 2=1

r z Monomials
g1(s, t) 0 0 s01 s02 t0

g2(s, t) 1 0 s11 s02 t0
g3(s, t) 1 0 s01 s12 t0
g4(s, t) 0 1 s01 s02 t1
g5(s, t) 1 1 s11 s02 t1
g6(s, t) 1 1 s01 s12 t1
g7(s, t) 2 0 s21 s02 t0
g8(s, t) 2 0 s01 s22 t0
g9(s, t) 2 0 s11 s12 t0
g10(s, t) 2 1 s11 s12 t1
g11(s, t) 2 1 s11 s02 t1
g12(s, t) 2 1 s01 s22 t1
spatiotemporal phenomena, including the spatiotemporal variability analysis and the rigorous
assessment of spaceetime dependencies and correlations. In applied sciences, the quantitative no-
tions of spaceetime distances come from physical experience (laws, data, empirical support, etc.),
can be made definite only by reference to physical experience, and are subject to change if a
reconsideration of experience seems to warrant change. Hence, keeping the physical requirements of
applied sciences in mind, which go beyond mere abstract mathematical considerations, the
spaceetime metric may be defined as follows.
Definition 2.4
A spaceetime metric between pairs of points p and p0 is defined, either in Rnþ1 as a composite metric
on spaceetime lags
m ð p; p0 Þ ¼ jDpj ¼ gðDpÞ; (2.13a)
i.e., a scalar jDpj8 related to Dp via the function g (the shape of which depends on the phenomenon of
interest); or, in Rn,1 as a pair of separate space and time metrics
m ð p; p0 Þ ¼ ðjDsj; jDtjÞ ¼ ðg1 ðhÞ; g2 ðsÞÞ; (2.13b)
where Ds ¼ h ¼ ðh1 ; .; hn Þ and Dt ¼ s ¼ t0 t are as in Eq. (2.7b). 9
The metric mð p; p0 Þ in Eq. (2.13a) is expressed in terms of a single function g that accounts for
the physics of the composite spaceetime, whereas in Eq. (2.13b) the mð p; p0 Þ is expressed in terms
of two different functions g1 and g2 that account separately for the specifics of space and time.
Otherwise said, in the composite metric of Eq. (2.13a) space and time mix together, which is not the
case of the separate metric of Eq. (2.13b), where space and time are treated as two distinct entities (in
which case, s may be viewed as distance toward time dimension). For future reference, the class of
composite (single) spaceetime metrics, Eq. (2.13a), will be denoted by Mc, and the class of separate
Also, jDpj is called the magnitude of Dp.

8
For convenience, in the following we will usually assume that t0 > t, i.e., jsj ¼ s.
9
(pair) spaceetime metrics, Eq. (2.13b), by Ms. Class Ms turns out to be useful in applications of
practical interest in which we consider natural variations within a domain defined by the Cartesian
space time product. The generalization features of the Mc class may be not necessary in these
applications. The readers are also reminded that in applied sciences the metrics mð p; p0 Þ are
mathematical expressions that define substantively the notion of spaceetime distance in real-world
continua rather than abstract, purely mathematical constructs (Chapter III studies the subject of
spaceetime metrics in considerable detail; a discussion of technical and physical aspects of
spaceetime geometry can be also found in Christakos, 2000).
Remark 2.2
A notational comment may be appropriate here. An alternative way to denote the separate spaceetime
metric of Eq. (2.13b), which is sometimes favored in applied stochastics, is as follows,

m ðs; tÞ; s0 ; t0 ¼ dRn s; s0 ; dR1 t; t0 ;
where dRn ðs; s0 Þ denotes spatial distance, say, the standard Euclidean distance between locations s and s0 ,
and dR1 ðt; t0 Þ denotes the time separation between t and t0 .
Definition 2.5
A standard formal way to define a metric is as the following inner or dot vector product (in Rnþ1)
!12
1 Xn;0
jDpj ¼ ðDp$DpÞ 2 ¼ εij hi hj ; (2.14)
i;j¼1
where εij ¼ εi $εj (εi are suitable base vectors). The composite spaceetime metric of Eq. (2.14) is a
very general metric of the Mc class, which is useful in a large number of physical situations, where it is
known as the Riemann metric, sometimes denoted as rR.
The metric coefficients εij in Eq. (2.14) are themselves spaceetime dependent, in general, and
determined by the physics of the situation. The above metric includes two celebrated spaceetime
metrics: The Pythagorean spaceetime metric
!12
X
n;0
jDpj ¼ εii h2i ; (2.15a)
i¼1
denoted as rp, and the Minkowski or Einstein spaceetime metric

!1
Xn 2
jDpj ¼ c2 h20 h2i ; (2.15b)
i¼1
sometimes denoted as rMi or rEi, where ε00 ¼ c2 (c is a physical constant), εii ¼ 1, and εij ¼ 0 (i s j).
Moreover, using transformation h0 1h0 , hj 1ihj ð j ¼ 1; .; n; i is here the imaginary unitÞ on the
metric of Eq. (2.15b) we get the special case of rP with ε00 ¼ c2 and εii ¼ 1 ði ¼ 1; .; nÞ. And,
because imaginary numbers are involved in this transformation to a Euclidean domain some authors
talk about the “pseudo-Euclidean” domain.
The spaceetime metrics of Eqs. (2.15a) and (2.15b) belong to the Mc class too. A metric of the
quadratic form of Eq. (2.15a) may emerge even in rather simple spaceetime situations, as in the
following example.
Example 2.5
Consider an attribute (e.g., a pollutant distributed by atmospheric processes) that covers distance Ds ¼
1
P
n 2
Ds ¼ h2i within time h0 ¼ s, so that Ds ¼ yh0 , where y is the speed with which the attribute
i¼1

moves. We can then write Dp ¼ ðh1 ; .; hn ; h0 Þ ¼ h1 ; .; hn ; Ds
y , with unit base vectors (εii ¼ 1 and
εij ¼ 0, i s j, in suitable units), and the corresponding metric will be
X
n
Dp2 ¼ Dp$Dp ¼ y2 Ds2 þ h2i ;
i¼1
10
which is a quadratic form.
Eq. (2.15b) is a physical metric used in applied sciences, e.g., geodesy, radiophysics, optics. The
different signs of the time and space components denote that in this metric the coordinate time is not a
dimension just like the space dimension. Also, it is worth noticing that in physics the symbol Dp2 itself is
usually taken as a fundamental quantity, and not as the square of some other quantity Dp (Carroll, 2004).
Remark 2.3
One could comment that the numerical difference under the square root of Eq. (2.15b) might take a
negative value. Yet, in many physical situations to which this metric refers to the numerical difference
is always positive.11 Furthermore, it is not physically possible that h0 ¼ t0 t ¼ 0 and hi s 0 (i s 0),
since this will signify that an object moves in space while time stands still. For operational reasons, the
Dp of Eq. (2.15b) is often squared so that we get an additive quantity along spaceetime.
The analysis above further emphasizes the point made earlier concerning the difference between
strictly mathematical metrics versus physical spaceetime metrics. In addition, although Eqs.
(2.14)e(2.15aeb) introduce symmetric metrics, i.e., jDpj ¼ jDpj, asymmetric metrics may be also
considered in applied stochastics when physically justified. Some examples are given next.
Example 2.6
In many situations, the metric Dp is asymmetric because the reverse process, p0 to p, may be realis-
tically impossible (e.g., the evolution and radioactive decay processes are irreversible). Moreover, an
asymmetric metric is defined by

jp p0 j if p p0 ;
jDpj ¼ 0 (2.16)
1 þ 10jp pj otherwise.
Also, the Manhattan metric (also known as absolute or city-block metric; see Eq. (2.18b), and Chapter
III) in the real-world context of one-way streets may not satisfy the symmetry condition jDpj ¼ jDpj,
since a path from point p to point p0 may include a different set of streets than a path from p0 to p.
10
We notice that, due to physical considerations, in this metric, εii ¼ 1, whereas in that of Eq. (2.15b), εii ¼ 1. Also, in Eq.
(2.15b) it is formally possible that jDpj ¼ 0 (null case).
11
For example, in many physical situations in geodynamics, radiophysics etc., the c denotes the speed of light and in the real-
world nothing can travel faster than c.
Remark 2.4
The examples above stress the point that a distance cannot always be defined unambiguously in
spaceetime, i.e., it may not be possible to decide on purely formal grounds, without additional in-
formation, the appropriate form of the g functions in Definition 2.4. Moreover, we notice that real-life
metrics are combinations of physical distances and times that may not always satisfy all conditions of
the term “metric” considered in a strict mathematical sense (e.g., the symmetry condition may be
unnatural in real-world applications, where the best way to go from point p to point p0 , generally, may
not coincide with the best way to go from point p0 to point p). In sum, the definition of an appropriate
metric in the real-world does not depend solely on purely mathematical considerations, but on both the
intrinsic links of space and time as well as on physical constraints (natural laws of change, empirical
support, boundary, and initial conditions).
In view of the above considerations, an efficient approach favored in applied sciences is to consider
some general expression of the spaceetime metric, and then test if it fits the physical requirements of
the case of interest, i.e., if it is, indeed, a substantive metric on physical grounds. To phrase it in more
words, I introduce the next postulate.
Postulate 2.2
The definition of an adequate spaceetime metric that expresses the intrinsic links of space and time
should be made on the basis of physical laws and empirical support and not decided on purely formal
grounds or merely on computational convenience.
According to the postulate, e.g., it may be impossible to decide using only mathematical consid-
erations how the separation of two points p1 and p2 in a porous medium with a Euclidean spatial lag of
3 km and a time lag of 2 days compares with the separation of two points p01 and p02 with a 2-km spatial
lag and a 3-day time lag. In fact, as we will see in Example 2.18, the decision regarding which of the
two pairs of points has the larger separation needs to consider the underlying natural mechanisms. The
following examples further illustrate the perspective suggested by Postulate 2.2 and the interesting
issues it raises.
Example 2.7
Consider a spaceetime varying attribute X(p) that satisfies the physical law (in Rn,1),

v
þ y$V Xðs; tÞ ¼ 0; (2.17a)
vt
where y is a velocity vector. The solution is of the form X(s, t) ¼ X(s yt), in which case the asso-
ciated spaceetime metric is given by
" #1 " #1
X
n 2 X
n 2
jDpj ¼ jh ysj ¼ ðhi yi sÞ2 ¼ h2i 2yi hi s þ y2i s2 ; (2.17b)
i¼1 i¼1
(h0 ¼ s); i.e., the metric isP

of the general form (Eq. 2.14) with coefficients determined by the physical
law (Eq. 2.17a) as ε00 ¼ ni¼1 y2i , εii ¼ 1, and ε0i ¼ 2yi. This is a metric of the Mc class.
Returning to the formal definitions of spaceetime metrics of the Ms class, we notice that the
metric of Eq. (2.13b) allows the decomposition of the spaceetime metric into a spatial distance jhj and
a time lag s, which can be defined separately(always keeping in mind the link between space and time
specified by the physics of the phenomenon). Spaceetime separate metrics of this form include the
case in which the time interval is commonly defined as s ¼ t0 t > 0, whereas the spatial distance
may have a variety of forms. The following example presents two well-known metrics worth of our
attention.
Example 2.8
Commonly used spaceetime metrics of the Ms class (jhj, s) are the
0 1
!1
B X
n 2
C
@ εii h2i ; sA; (2.18a)
i¼1
where the spatial component is the Euclidean distance jhj ¼ rE ; and the12
!
Xn
εii jhi j; s ; (2.18b)
i¼1
where the spatial component is the Manhattan (or absolute) distance jhj ¼ rM . Both these metric forms
are symmetric. Also, particularly useful is the metric
rE
2r sin1 ;s ; (2.18c)
2r
where the spatial component is the arc length jhj ¼ rS determining the spatial distance between two
points on the surface of the earth (viewed as a sphere), rE is the (Euclidean) distance between the two
points, and r is the earth radius.
Although the above examples also stress the point that the spaceetime metric expressed by Eq.
(2.13a) is more general than that of Eq. (2.13b), yet, on occasion, the decomposition assumed in Eq.
(2.13b) may have some interesting implications in mathematical analysis that need to be kept in mind.
Example 2.9
To orient the readers, it is reminded that many applications (e.g., spaceetime statistics, and geo-
statistics) often assume a spaceetime metric defined as a pair of positive real numbers, i.e.,
0 1
!1
B X 2
n 2
C
ðjhj; sÞ ¼ rE jεii ¼1; s ¼ @ hi ; s A ; (2.19a)
i¼1
i.e., rE is as in Eq. (2.18a) with εii ¼ 1 (i ¼ 1; .; n). This separate spaceetime metric of the class Ms
should be distinguished from the composite spaceetime metric of the class Mc defined as the single
positive real number
!1
Xn;0 2
jDpj ¼ rP jεii ¼1 ¼ h2i ; (2.19b)
i¼1
12
The contextual introduction of these metrics is made in Chapter III.
i.e., rP as in Eq. (2.15a) with εii ¼ 1 (i ¼ 1; .; n; 0). Eq. (2.19b) is apparently different than the
spaceetime metric of Eq. (2.15b) that introduces a special physical partition of space and time. This
operational difference between the two spaceetime metrics has considerable consequences from a
physical interpretation viewpoint (e.g., as we will see below, Eq. (2.15b) is invariant under the Lor-
entzian transformation, whereas Eq. (2.19b) is not, in general).
The Riemann spaceetime metric, Eq. (2.14), and the Pythagorean spaceetime metric, Eq. (2.15a),
are often assumed in formal analysis in the Rnþ1 domain. The special case of the Pythagorean
spaceetime metric, Eq. (2.19b), plays an essential role in certain spaceetime geometry operations
(like random field continuity and differentiability, introduced in Chapter VI). Additional insight is
gained if the difference between the metrics of Eqs. (2.19a) and (2.19b) is viewed in the context of
spatiotemporal random field variability characterization. Specifically, as we shall see in Chapter VII,
composite spaceetime isotropy implies that the covariance function of the natural attribute depends on
the metric of Eq. (2.19b), whereas separate space isotropy/time stationarity (isostationarity) means that
the covariance is a function of the metric of Eq. (2.19a). Note that these metrics can be extended in the
context of the so-called geometrical anisotropy of spaceetime attribute variation (also, Chapter VII).
Remark 2.5
For future reference, it may be convenient to stress the point that we should distinguish between three
kinds of separability, namely, metric separability (as introduced above), and covariance separability
and sample path separability (that will be introduced in later chapters).
2.2 TRANSFORMATIONS AND INVARIANCE IN SPACEeTIME

In physical modeling, we often find it useful to apply some transformation of the original spaceetime
coordinates.13 In the context of group theory (Helgason, 1984), this leads to a so-called transformation
^
group involving a space S on which a group G acts transitively, i.e., for p; p ˛S, there exists a g ˛G
^
such that gp ¼ p ˛S. Not surprisingly, different hypotheses may be associated with different com-
binations of S and G. Commonly used groups of transformations that are useful in applied stochastics
are as follows.
Definition 2.6
The group of translation transformations g ¼ Ud for p ˛ S acts on S ¼ Rnþ1 or Rn,1 so that
^
p1Udp p ¼ p þ dp ¼ p ;
^ ^ (2.20aeb)
ðs; tÞ1Uh;s ðs; tÞ ¼ ðs þ h; t þ sÞ ¼ s; t ;
i.e., the translation of a vector p by dp or a vectorescalar pair (s, t) by (h, s), respectively. The group of
orthogonal transformations g ¼ Lt for p ˛ S acts on S ¼ Rnþ1 or Rn,1 so that
13
In modeling we commonly consider transformations applied on different entities. Here we consider transformations of
spaceetime coordinates (also, in Chapter V we study some interesting modeling implications of transformations of spacee
time attributes).
jLt ð pÞj ¼ j pj;

(2.21aeb)
jLt ðs; tÞj ¼ ðjsj; tÞ;
i.e., Lt is a linear transformation that preserves the length of a vector p or s, respectively.
In the stochastics context, the Ud and Lt transformations play a significant role in the determination
of vector or multivariate random field isostationarity (Chapter IX), which is why I introduce some basic
transformation results here. To start with, attractive features of the Ud transformation include the pos-
^
sibility of (1) a succession of translations, i.e., Udp Udp0 . p ¼ Udpþdp0 þ . p ¼ p þ dp þ dp0 þ . ¼ p ,
^ ^
and (2) unequal translations with equal distances, i.e., if Udp p1 ¼ p 1 and Udp0 p2 ¼ p 2 , then
^ ^
jp1 p2 j ¼ p 1 p 2 if dp ¼ dp0 .
Example 2.10
A Ud transformation of coordinates is defined by
^
Udp p ¼ ðs1 h; s2 ; s3 ; s0 þ h0 Þ ¼ p ;
^ ^
Uh;s ðs; tÞ ¼ ðs1 h; s2 ; s3 ; t þ sÞ ¼ s; t ;
where h and h0 (or s) are specified space and time lags, respectively.
Among the useful features of the Lt transformation are that (1) the inverse of Lt is orthogonal
and (2) the composition of Lt is orthogonal too. The most important cases of Lt are rotations,
^ ^
R p ¼ p , about a fixed point, a fixed axis etc., as well as reflections, Pp ¼ p , across any subspace
V 3 R . In the transformation context, when appropriate, I use the notation Rn,1 to emphasize the
n
rather obvious fact that, naturally, while Ud transformations apply in both the Rn and the T domains,
certain Lt transformations ðlike R and PÞ refer to the Rn domain only. Specifically, in R2,1 and in R3,1
the R can be specified using, respectively, polar and spherical coordinates. Combinations of R and P
are also possible. Relevant to the P transformation is that of symmetry. Symmetries, i.e., trans-
formations that preserve certain quantities, form a group in the sense that a composition of two
symmetries is also a symmetry, and the inverse transformation to a symmetry is also a symmetry. There
exist intimate relationships between random field geometry and spaceetime coordinate symmetry
transformations. The natural conservation laws are closely linked to the notion of the spaceetime
symmetry, whereas the relativity theory essentially elaborates the inherent symmetry of the spacee
time continuum.
Example 2.11
If the rotation is specified through an angle q (say, p is rotated in R2,1 about the origin by q), it is
^ ^
usually denoted as R q p ¼ p , in which case the components of p are functions of the components of
p and the angle q. A similar notational convention is valid for the P transformation. In R2,1, the R q
^
transformation that rotates points p around the origin by q (counterclockwise) is given by R q p ¼ p ,
where
2 3
cos q sin q 0
6 7
R q ¼ 4 sin q cos q 05
0 0 1
The P transformation that maps points p on to their reflected images about a line through the origin
^
that makes an angle q with the s1-axis is given by P q p ¼ p , where
2 3
cos 2q sin 2q 0
6 7
P q ¼ 4 sin 2q cos 2q 0 5.
0 0 1
In higher dimensions, more than one angles may be involved, say q, 4, and j, in which case we
^ ^
write R q4j p ¼ p or P q4j p ¼ p .
Invariance is one of the most important notions in mathematical and physical sciences. It can be a
conceptual tool that offers a deeper understanding of a physical phenomenon, or a computational tool
that solves complex systems of equations representing the phenomenon. Generally, the invariance of
an entity (e.g., a physical concept, a natural law, or an empirical model) under a certain transformation
means that the entity does not change by the transformation. For present purposes, invariance can be
described as follows.
Definition 2.7
If prior to a transformation g, the attribute of interest had the value F(p) for each p, and after the
transformation gp the entity has the value FðgpÞ, invariance means that the value of the attribute does
not change after g, i.e., FðgpÞ ¼ F(p) for all p.14
A natural system is translationally invariant (in space and/or time) with respect to a certain
attribute F if translation by certain characteristic spatial vectors and/or time lags, gp ¼ Udpp, does not
change F. For example, periodic systems such as crystals, planetary orbits, and clocks belong to this
category. Also, if F denotes energy, invariance means that the energy remains the same under the
corresponding transformation (e.g., rotation/shifting of electric charge configurations).15 It is often
desirable that the invariance definition does not depend on the choice of scale (e.g., units) or on the
choice of a coordinate system. Thus, coordinate invariance in physical sciences means that the entities
of interest (e.g., laws of physics or theoretical models) should be independent of the coordinate system
used. In the same context, if F denotes a spaceetime metric, an important metric property is formally
defined as follows.
Definition 2.8
A spaceetime metric jDpj is called invariant under a certain transformation g of coordinates, i.e.,
14
Otherwise said, invariance implies that the composition of F and g coincides with F.
15
On the basis of invariance, other useful properties may be derived. For example, as we shall see later, a system is called
homogeneous with respect to a specific attribute if this attribute has a uniform value in space.
g : jDpj1gjDpj; (2.22)
if it remains unchangeable when we move from one coordinate system to another using this
transformation.
For illustration, the first example is concerned with the most commonly assumed translation (Ud)
invariance.
Example 2.12
Consider the general spaceetime metric definition jDpj ¼ jp0 pj. If
we apply
0a Ud transformation

^ ^ ^
to the coordinate system, the spaceetime metric jDpj becomes D p ¼ p p ¼ Udp jDpj ¼
jðp0 þ dpÞ ð p þ dpÞj ¼ jp0 pj ¼ jDpj, i.e., the jDpj remained unchanged by the Ud trans-
formation. The point reflection (or central inversion) of coordinates defined as
^
p1 p ¼ R dp p ¼ p (2.23)
0
^ ^ ^
preserves the metric jDpj, i.e., D p ¼ p p ¼ R dp p0 R dp p ¼ jp0 þ pj ¼ jDpj.
The following examples present two transformations of principal importance in physical sciences,
namely, the Galilean and the Lorentz transformations.
Example 2.13
I start by considering the spaceetime metric of Eq. (2.19a) in the light of the Galilean transformation
of coordinates in Rn,1 defined as16
^ ^
ðs; tÞ1 s ; t ¼ ðG ds s; G dt tÞ ¼ ðs ytε1 ; tÞ ¼ ðs1 yt; s2 ; .; sn ; tÞ; (2.24)
where y is a physical constant, and ε1 is a unit vector along the direction s1. Let s and s0 be the spatial
^ ^
locations of two points at any given time t, and t ¼ t; t0 ¼ t0 be two different times associated with s
and s0 . By applying G d to s and s0 we can define the spatial lag
^
^0 ^
h ¼ s s ¼ jG ds s0 G ds sj ¼ jðs0 ytε1 Þ ðs ytε1 Þj ¼ js0 sj ¼ jhj.
Next, let t; t0 be two different times associated with s and s0 . By applying G d on t; t0 we get
^ ^ ^
^ ^ ^ ^
t ¼ t; t0 ¼ t0 , and the time lag is given by s ¼ t0 t ¼ t0 t ¼ s. Hence, h ; s ¼ ðjhj; sÞ i.e.,
the spaceetime metric of Eq. (2.19a) remained unchanged by the ðG ds ; G dt Þ transformation. For
comparison purposes, consider the Lorentz transformation of coordinates defined in Rnþ1 as
^ s0 yc2 s1 s1 ys0
p1 p ¼ Ldp p ¼ c2 ; ; s2 ; .; sn ; (2.25)
b b
16
Notice that the transformation is applied separately on the spatial and the time coordinates.
1
where y, and c are constant coefficients, and b ¼ 1 y2 c2 2 .17 By applying Ld on the Minkowski
metric of Eq. (2.15b), we find that

^
D p ¼ Ldp p0 Ldp p
" #1
s00 yc2 s01 s0 yc2 s1
2
s0 ys00 s1 ys0 2 X
n 2 2
¼ c2 1 s0i si
b b b b i¼2
!1
X
n 2
¼ c2 h20 h2i ¼ jDpj ¼ Ldp jDpj;
i¼1
i.e., the jDpj remained unchanged by Ld. In fact, the Riemann metric of Eq. (2.14) is also invariant
under the Ld transformation.
Next, I will discuss cases in which the metric is not invariant under the specified transformation.
Example 2.14
Straightforward algebraic manipulations show that the special Pythagorean metric of Eq. (2.19b) of
the Mc class is not invariant under the Ld transformation (Exercise I.5). Thus, operationally, the
physical spaceetime metric of Eq. (2.15b), although it also belongs to the Mc class, yet it differs
from the metric of Eq. (2.19b) in that the former is invariant under the Loretzian transformation,
whereas the latter is not, in general. On interpretational grounds, the invariance of Eq. (2.15b) means
that although space and time are relative, this metric introduces a certain mixture of space and time
that is absolute. Next, consider the spaceetime metric of Eq. (2.19b) under the anisotropic dilation
transformation
^
p ¼ Dl p ¼ ðl1 s1 ; .; ln sn ; l0 s0 Þ. (2.26)
Under the Dl transformation, the metric jDpj becomes
!1 !1
0 X
n;0
2 2 X
n;0 2
^ ^ ^
D p ¼ p p ¼ jDl p0 Dl pj ¼ l2i s0i si ¼ l2i h2i sjDpj;
i¼1 i¼1
i.e., the jDpj of Eq. (2.19b) changed by the Dl transformation.

Additional insight regarding metric invariance is gained if the difference between the two metric
groups discussed in Example 2.9 is viewed in the context of the so-called geometric anisotropy.
Example 2.15
The spatial geometric anisotropy transformation of coordinates, A a;0 is properly defined as
^ ^
h; s ¼ A a;0 ðs; sÞ ¼ ða$h; sÞ ¼ ða1 h1 þ . þ an hn ; sÞ (2.27)
17
The Ld transformation, in a sense, may be interpreted as rotation in Rnþ1.
(ai > 0, i ¼ 1; .; n), in which case the separate spaceetime metric of the form of Eq. (2.19a) is
given by
0 1
!1
^ X n 2
^ B C
h ; s ¼ @ a2i h2i ; sA. (2.28)
i¼1
On the other hand, the spatiotemporal geometric anisotropy transformation of coordinates, A a;a0 , is
such that
^
D p ¼ A a;a0 Dp ¼ ða; a0 Þ$Dp ¼ ða1 h1 þ . þ an hn þ a0 h0 Þ (2.29)
(a0 > 0), in which case for the corresponding composite spaceetime metric of the form of Eq. (2.19b)
it is valid that
!1
X
n;0 2
^
D p ¼ a2i h2i . (2.30)
i¼1
Clearly, none of the metrics of Example 2.9 is invariant to the geometric transformations of the co-
ordinates. On the other hand, as we shall see in Chapter VII, certain classes of attribute covariances
(i.e., functions assessing spaceetime attribute variability) may be justifiably defined on physical
grounds as functions of the metrics of Eqs. (2.28) and (2.30).
Galilean invariance is one of the most important properties of several fundamental physical
equations, like the NaviereStokes equations of fluid dynamics, because without this invariance it will
not be possible to compare fluid experiments performed in different parts of the world (i.e., if these
equations were not Galilean invariant, they could not describe fluid motion adequately). In line with the
Definition 2.7, when applied to a physical law Galilean invariance implies that if prior to the Galilean
8 0 D½Xð pÞ for each p, and after the transformation
transformation the physical equation had the form
> s ¼ G ds s ¼ s yt;
<
G dp;dX : s; t; X1 t0 ¼ G dt t ¼ t; (2.31)
>
: 0
X ¼ G dX X ¼ X þ y;

the equation has the form D G dp;dX XðpÞ , invariance means that D G dp;dX XðpÞ ¼ D½XðpÞ.
Example 2.16
To demonstrate the significance of Eq. (2.31), I consider Burgers’ equation in R1,1,

v v v2
þ Xðs; tÞ n 2 Xðs; tÞ ¼ 0; (2.32)
vt vs vs
where n is a diffusion coefficient (viscosity). Burgers’ equation is a PDE governing a random field
X(s, t) occurring in various applied stochastics areas (gas dynamics, fluid mechanics, acoustics, etc.).
By applying the Galilean18 transformation G dp;dX of Eq. (2.31) in R1,1, which is defined as
18
Generally, Galilean invariance is used to denote that measured physics must be the same in any nonaccelerating frame of
reference.
8 0
< s ¼ s yt;
>
G ds;dt;y : s; t; X1 t0 ¼ t;
>
: 0
X ¼ X þ y;
Eq. (2.32) leads to

v 0 0 v v2
þ ð X ð s ; t Þ þ y Þ n ð X ðs0 ; t0 Þ þ yÞ ¼ 0: (2.33)
vt0 vs0 vs0 2
Taking into account that y is a constant, the following derivative expressions hold,
v vXðs; tÞ vXðs; tÞ v
X ðs0 ; t0 Þ ¼ X ðs0 ðs; tÞ; t0 ðs; tÞÞ ¼ Xðsðs0 ; t0 Þ; tðs0 ; t0 ÞÞ; 0 X ðs0 ; t0 Þ ¼ y þ ; 0 X ðs0 ; t0 Þ
vt vs vt vs
0 ; t0
0 ; t0

vXðs; tÞ v 2
Xðs v 2
Xðs; t v 2
Xðs v 2
Xðs; t
¼ ; ¼ ; and ¼ ;
vs vs02 vs2 vt02 vt2
so that Eq. (2.33) becomes Eq. (2.32). That is, Burgers’ equation is invariant under the Galilean
transformation.
For completeness of presentation of coordinate transformations, I should notice that other commonly
used transformations are the standard geometrical coordinate transformations, where the standard
Cartesian set of coordinates is transformed to a new set in terms of polar, cylindrical, or spherical
coordinates. For illustration, I present here the spherical coordinate transformation, which is particularly
useful in the evaluation of n þ 1-dimensional integrals. This coordinate transformation is defined as
!
Y
i1 nY
1
ðs1 ; .; si ; .; sn ; tÞ1 s cos q1 ; .; s cos qi sin qj ; .; s sin qj ; t ; (2.34)
j¼1 j¼1
where i ¼ 2; 3; .; n 1, s ¼ jsj 0, qi ˛ [0, p] (i ¼ 1; .; n 2), and qn1 ˛ [0, 2p]. The detailed
presentation of these standard coordinate transformations can be found in many textbooks (also,
Appendix).
Remark 2.6
Before leaving this section, I would like to reiterate the following general points: (1) On physical
grounds, in certain cases modeling may be more straightforward and rigorous in the Rnþ1 domain,
whereas in some others the distinction of space and time introduced in the Rn,1 analysis may be
interpretationally more appropriate. (2) On logical grounds, it is often more meaningful and even
simpler to consider that a phenomenon occurs in a composite spaceetime continuum rather than to
divide our experience artificially according to unrelated spatial and temporal aspects (especially, since
it is not always clear how this distinction of the spatial and temporal aspects should be made).
2.3 SPACEeTIME INTERPRETATIONS

As a geometrical structure, spaceetime consists of a continuum E of spaceetime points p together
with a metric mð p; p0 Þ ¼ jDpj in the spaceetime composite domain Rnþ1, or with a metric
mð p; p0 Þ ¼ ðjDsj; sÞ in the spaceetime Cartesian domain Rn,1. I.e., the spaceetime structure is
formally characterized by the triplet ðE; p; mÞ. The vector Dp or the vectorescalar pair ðDs; sÞ would
count the embeddings of natural attributes and their dynamic relations in spaceetime, whereas the
form of the function g (expressed, e.g., by the values of the coefficients εij at every p in the case of the
general metric of Eq. (2.14)) enables us to determine the intrinsic geometrical properties of E.
Spaceetime point representations, Definition 2.1 or 2.2, may be affected by physical context when
the impacts of the laws of nature or of the composite spaceetime distances are considered.
Example 2.17
The traditional system of spaceetime coordinates p ˛ Rnþ1 used by continuum theories has the
following properties:
(a) The spaceetime coordinates p of a continuum E cannot be established in an absolute way, but
always relatively to some origin (there is no absolute scale in nature).
(b) The axes of the coordinate system must be defined (they may be rectilinear or, more generally,
curvilinear). In continuum theories, the coordinate system is assumed to be differentiable.
(c) Length and time intervals are relative quantities, in the sense that numerical values are assigned
to them only by comparison to some elementary intervals called units. This implies the existence
of a scale relativity that has significant ramifications for physical laws.
(d) Space and time coordinate measurements are always made with some finite resolution
corresponding to the minimal unit used to assign numbers to the length or time intervals. The
minimal unit may depend on the precision of the instrument or on some physical limitation.19
Hence, a key aspect of spaceetime analysis is that sciences are based on measurements of the
natural attributes represented as spaceetime fields. Their laws do not apply to natural phenomena by
themselves, but rather to the numerical results of measurements obtained regarding these phenomena.
Insofar as numerical measurement adequacy issues are concerned, two common strands emerge: first,
whether the means of measurement are accurate, and, second, whether they are actually measuring
what they are intended to measure.
The operational definition of a system of coordinates (such as in Definition 2.1) should include all
the relevant information that is necessary to describe these numerical results and to relate them in
terms of physical laws. The metric, in particular, is the central mathematical structure identified or
associated with spaceetime. Accordingly, some interesting observations are worth making about the
interpretation of a spaceetime metric. Ontologically, the metric may be interpreted:
(a) in a substantivist sense, as representing substantive space (i.e., it may be seen as the real
representor of spaceetime) or
(b) in a relationist sense, as a structural quality of spaceetime (i.e., spaceetime does not claim
existence on its own, but only in relation to the physical situation).
Both interpretations share two key features: the aforementioned mathematical structure ðE; p; mÞ
equated with the nature of spaceetime is identical in both cases, and if the metric is removed, spacee
time is removed too. Otherwise said, the substantivist and relationist interpretations of spaceetime
19
If, e.g., the resolution of a rod is 1 mm, it would have no physical meaning to express a measurement in mm, or to measure
the distance from Earth to Venus with a precision of 1 cm; and, according to Heisenberg’s principles, in microphysics the
results of measurements depend on the resolution of the instrument.
embody the same formal metric, but from the standpoint of diverse ontological perspectives of the same
underlying reality. In many real-world cases, what we need to know about spaceetime is its metric
structure (i.e., we focus on the crucial role of the metric used in the real-world case) and not necessarily
its interpretation. How mathematical spaceetime structures apply to reality, or are exemplified in the
real-world, is identical to the issue of how mathematical structures are exemplified in the real-world.
However, in most cases there is an important reciprocal relationship between the metric and the natural
attribute (Chapter III). Indeed, the relevance of empirical evidence in assessing spaceetime structure
should be emphasized, since the possible metric constructions are being judged not solely from a
mathematical perspective, but from a scientific and empirical standpoint as well.
In view of the above considerations, the following point should be stressed concerning the
spaceetime structure, ðE; p; mÞ: the distance cannot always be defined unambiguously in spacee
time, i.e., as was argued earlier, it may not be possible to compare on purely mathematical grounds (1)
the distance d12 between the pair of points p1 and p2 with a 3-km spatial lag and a 2-days time lag
0 between the pair of points p0 and p0 with a 2-km spatial lag and a 3-day time
versus (2) the distance d12 1 2
lag. As a matter of fact to decide which one of the two cases, (1) or (2), has the larger separation
distance, we may need to consider the outcome of a natural process.
Example 2.18
Consider an experiment during which a tracer is released inside a porous medium at points p1 and p01 .
If the tracer is detected at point p2 but not at point p02 , then we can claim that d12 < d12
0 with respect to
the particular experiment. This approach provides a way of ordering distances between spaceetime
points, but without further refinements a specific quantitative notion of spaceetime distances cannot be
obtained. Also, note that the distance defined above is not purely a geometric property of spaceetime,
but it also depends on the medium’s properties, and the particular phenomenon that we decide to use
in the measurement. That is, measuring distance by means of fluid tracer dispersion can lead to
very different results than measuring distance by means of electromagnetic propagation.
Experience and observation in the spaceetime domain Rn,1, in particular, ascribes certain funda-
mental properties to the concepts of space and time related to the concept of invariance (Definition
2.7). Space homogeneity means that one point of space is equivalent to any other point, in the sense that
a natural attribute will occur the same way under identical conditions, independent of the place in
which it occurs. Space homogeneity implies translational invariance of the properties of space in
which case the origin of a coordinate system can be arbitrarily chosen. Time homogeneity means that
one time instant is equivalent to any other instant, in the sense that an experimental setup will generate
the same result, whether performed at one time instant or another. Time homogeneity implies that
when we observe a physical phenomenon we can choose the zero of time at any time instant.20 Space
isotropy means that one direction in space is equivalent to any other direction. Rotational invariance of
the properties of space is assumed, and a specific experimental setup will yield the same result whether
the setup faces East or North. Space isotropy implies that we can arbitrarily orient the coordinate axes
in space. The case of time isotropy is a rather peculiar one. Intuitively, the time direction is forward,
20
This kind of space or time homogeneity should be distinguished from random field homogeneity or stationarity. As we will
see in Chapter VII, random field realizations do not exhibit such convenient invariance properties. Yet, it is often true that
ensemble moments (mean, covariance, variogram etc.) possess certain invariance properties, i.e., a random field is said to
exhibit stochastic translation invariance if its ensemble moments are translationally invariant.
from present to future, i.e., time has only one direction. Yet, in an analogous manner to space isotropy,
time isotropy would denote equivalence of time directions, forward and backward direction (from
present to past). In this sense, the question whether time is isotropic or not means whether time is
reversible or not. So, although the (Newtonian) laws of motion are time reversible, at least opera-
tionally, most natural processes in the real-world are time irreversible (described by the laws of
evolution, the second law of thermodynamics, etc.).
In sum, mathematical spaceetime structures ðE; p; mÞ may vary along with the postulated sci-
entific theories, a fact that explains why one may encounter a number of competing theories, and thus
different mathematical structures, compatible with the same evidence. The application of the math-
ematical spaceetime structure in such cases requires coordination with the physical hypotheses and
measurement/observation conventions (the choice of a hypothesis or a convention is normally based on
an assessment of its overall viability and consistency with scientific method). So, we should not
sanction otherwise undetectable attribute correlations and relations simply for the sake of using a
cherished metric (e.g., the Euclidean one). In the same context, a transformation may be useful that
captures invariance features of the spaceetime structure that we believe to be crucial in the physical
context of the phenomenon of interest.
2.4 FUNCTIONS OF SPACEeTIME ARGUMENTS

The point vector p plays a key role in physical sciences, not only because it uniquely specifies a point
in the spaceetime domain of interest, but also because of the numerous functions of p encountered in
physical sciences. Keeping the above considerations in mind, it is appropriate to continue with a
definition that brings us to the next level of stochastic modeling.
Definition 2.9
A field is generally a function of the spaceetime arguments (points, coordinates, lags, metrics) that,
generally, presupposes a continuum of spaceetime points and represents values of the physical
attribute at these points. A set of values at all points in the spaceetime continuum specifies a reali-
zation of the field.
A field may associate mathematical entities (such as a scalar, a vector, or a tensor) with specified
spaceetime points. For the field to obtain a physical meaning these entities must represent values of
some natural attribute (e.g., soil moisture, fluid velocity, cloud density, contaminant concentration, or
lung cancer mortality rate). Spaceetime plots of such fields render their behavior in the entire
spaceetime domain, in contrast with standard practice (e.g., varying space and keeping fixed time, or
vice versa). This being the case, the composite spaceetime domain (with varying both space and time)
can provide a deeper understanding of the field’s spaceetime variation than simple space or time field
plots allow. In this respect, of particular interest are certain major function spaces: the space C of all
real and continuous functions in spaceetime with compact support; the space C N of all real,
continuous, and infinitely differentiable functions with compact support; and the space C N 0 of all real,
continuous, and infinitely differentiable functions which, together with their derivatives of all orders,
approach zero rapidly at infinity (e.g., faster than jsj1 and t1 as jsj/N and t / N).21
21
Here, C N N
0 IC , as all functions in C
N
vanish identically outside a finite support, whereas those in C N
0 merely decrease
rapidly at infinity.
Another random document with
no related content on Scribd:
du balcon, sous les habits de Ginevra, je lui jetai l’échelle par
laquelle il montait souvent, et je ne m’aperçus de la fourberie que
lorsque tout le dommage en fut advenu.
« Pendant ce temps, le duc avait eu l’entretien suivant, ou à peu
près, avec Ariodant : — De grands amis qu’ils étaient auparavant, ils
étaient devenus ennemis à cause de leur rivalité pour Ginevra —
« Je m’étonne — commença mon amant — qu’ayant, entre tous mes
compagnons, toujours eu des égards et de l’amitié pour toi, tu m’en
aies si mal récompensé.
« Je suis certain que tu sais l’amour qui existe depuis longtemps
entre Ginevra et moi, et que tu connais mon espoir de l’obtenir de
mon seigneur comme légitime épouse. Pourquoi viens-tu me
troubler ? Pourquoi t’en viens-tu, sans résultat, lui offrir ton cœur ?
Par Dieu ! j’aurais pour toi plus d’égards, si j’étais à ta place et si tu
étais à la mienne. — »
« — Et moi — lui répondit Ariodant — je m’étonne bien plus
encore à ton sujet, car j’en suis devenu amoureux avant que tu l’aies
seulement vue. Et je sais que tu n’ignores pas combien est grand
notre amour à tous deux, et qu’il ne peut être plus ardent qu’il n’est.
Son intention, son seul désir est d’être ma femme, et je tiens pour
certain que tu sais qu’elle ne t’aime pas.
« Pourquoi donc n’as-tu pas pour moi, pour notre amitié, les
mêmes égards que tu prétends que je devrais avoir pour toi, et que
j’aurais, en effet, si tu étais plus avant que moi dans son affection ?
N’espère pas davantage l’avoir pour femme, bien que tu sois le plus
riche dans cette cour. Je ne suis pas moins que toi cher au roi, mais,
plus que toi, je suis aimé de sa fille. — »
« — Oh ! — lui dit le duc — grande est l’erreur qui t’a conduit à
un fol amour. Tu crois être plus aimé ; je crois la même chose. Mais
on peut en juger par le résultat. Dis-moi franchement ce que tu as
dans le cœur, et, moi, je te dirai mon secret en entier ; et celui de
nous qui paraîtra le moins favorisé, cédera au vainqueur et
cherchera à se pourvoir ailleurs.
« Et je n’hésite pas à te jurer que jamais je ne dirai mot de ce que
tu m’auras révélé ; de même, je désire que tu me donnes ta parole
que tu tiendras toujours secret ce que je t’aurai dit. — » Ils en
vinrent donc à un serment commun, la main posée sur les
Évangiles. Et après qu’ils se furent juré de se taire, Ariodant
commença le premier,
« Et dit, loyalement et droitement, comment entre Ginevra et lui
les choses s’étaient passées ; qu’elle lui avait juré, de vive voix et
par écrit, qu’elle ne serait jamais la femme d’un autre, mais bien la
sienne, et que, si le roi venait à s’y opposer, elle refuserait
constamment toutes les autres propositions de mariage, et vivrait
seule pendant tout le reste de ses jours ;
« Et que lui, Ariodant, grâce à la valeur qu’il avait montrée à plus
d’une reprise dans les combats, et qui avait tourné à la gloire, à
l’honneur et au bénéfice du roi et du royaume, avait l’espoir de s’être
assez avancé dans la bonne grâce de son seigneur, pour qu’il fût
jugé digne par lui d’avoir sa fille pour femme, puisque cela plaisait à
celle-ci.
« Puis il dit : « — J’en suis à ce point, et je ne crois pas que
personne ne me vienne supplanter. Je n’en cherche pas davantage,
et je ne désire pas avoir de témoignage plus marquant de son
amour. Et je ne voudrais plus rien, sinon ce qui par Dieu est permis
en légitime mariage. Du reste, demander plus serait vain, car je sais
qu’en sagesse elle surpasse tout le monde. — »
« Après qu’Ariodant eut exposé avec sincérité ce qu’il attendait
comme prix de ses soins, Polinesso, qui déjà s’était proposé de
rendre Ginevra odieuse à son amant, commença ainsi : « — Tu es
de beaucoup distancé par moi, et je veux que tu l’avoues toi-même,
et qu’après avoir vu la source de mon bonheur, tu confesses que
moi seul suis heureux.
« Elle dissimule avec toi ; elle ne t’aime ni ne t’estime, et tu te
repais d’espérance et de paroles. En outre, elle ne manque pas de
se railler de ton amour toutes les fois qu’elle s’entretient avec moi.
J’ai de sa tendresse pour moi une bien autre preuve que des
promesses ou de simples bagatelles. Et je te la dirai sous la foi du
secret, bien que je fisse mieux de me taire.
« Il ne se passe pas de mois, sans que trois, quatre, six et
quelquefois dix nuits, je ne me trouve nu dans ses bras, partageant
avec elle ce plaisir qu’on goûte dans une amoureuse ardeur. Par là,
tu peux voir si à mon bonheur doivent se comparer les babioles que
tu donnes comme des preuves. Cède-moi donc la place et pourvois-
toi ailleurs, puisque tu vois que tu m’es si inférieur. — »
« — En cela je ne veux pas te croire — lui répond Ariodant — et
je suis certain que tu mens. Tu as imaginé en toi-même tous ces
mensonges, afin de m’effrayer et de me détourner de mon
entreprise. Mais comme ils sont par trop injurieux pour Ginevra, il
faut que tu soutiennes ce que tu as dit. Et je veux te montrer sur
l’heure que non seulement tu es un menteur, mais encore un traître.
— »
« Le duc repartit : « — Il ne serait pas juste que nous en
vinssions à bataille pour une chose que je puis, quand il te plaira, te
faire voir de tes propres yeux. — » A ces mots, Ariodant reste
éperdu ; un frisson lui parcourt tout le corps ; il tremble, et s’il eût cru
complètement à ce qu’on lui avait dit, il en serait mort sur-le-champ.
« Le cœur brisé, le visage pâle, la voix tremblante et l’amertume
à la bouche, il répondit : « — Quand tu m’auras fait voir une si
étonnante aventure, je te promets de renoncer à celle qui t’est si
libérale et à moi si avare. Mais je ne veux pas te croire avant de
l’avoir vu de mes yeux. — »
« — Quand il en sera temps, je t’avertirai — répliqua Polinesso.
— » Et ils se séparèrent. Je crois qu’il ne se passa pas plus de deux
nuits sans que j’ordonnasse au duc de venir me voir. Afin donc de
déployer les lacs qu’il avait si secrètement préparés, il alla trouver
son rival, et lui dit de se cacher la nuit suivante parmi les maisons en
ruine, où jamais personne ne venait.
« Et il lui indiqua un endroit juste en face du balcon par lequel il
avait l’habitude de monter. Ariodant le soupçonnait de chercher à
l’attirer en un lieu où il aurait facilité de lui tendre un guet-apens et
de le faire tuer, sous prétexte de lui montrer ce qui, de la part de
Ginevra, lui paraissait impossible.
« Il résolut toutefois d’y aller, mais de façon à être aussi fort que
son rival, et, dans le cas où il serait assailli, de n’avoir pas à craindre
la mort. Il avait un frère prudent et courageux, le plus renommé de
toute la cour pour son adresse aux armes et nommé Lurcanio.
L’ayant avec lui, il était plus rassuré que s’il avait eu dix autres
compagnons.
« Il l’appelle, lui dit de prendre ses armes, et l’emmène avec lui,
sans lui avoir confié son secret, car il ne l’avait dit ni à lui ni à aucun
autre. Il le place à un jet de pierre loin de lui : « — Si tu m’entends
appeler — lui dit-il — accours ; mais si tu ne m’entends pas
t’appeler, ne bouge pas de là si tu m’aimes. — »
« — Va toujours et ne crains rien — dit son frère. — » Rassuré,
Ariodant s’en vient alors et se cache dans une maison solitaire,
située en face de mon balcon secret. D’un autre côté s’avance le
trompeur, le traître, tout joyeux de couvrir Ginevra d’infamie. Il me
fait le signe entre nous convenu d’avance, à moi qui de sa fourberie
étais tout à fait ignorante.
« Et, moi, avec une robe blanche ornée tout autour de la taille de
bandes d’or, ayant sur la tête un filet d’or surmonté de belles fleurs
vermeilles, — à la façon dont Ginevra seule avait coutume d’en
porter, — dès que j’eus entendu le signal, je parus sur le balcon qui
était placé de façon qu’on me découvrait en face et de tous côtés.
« Lurcanio, sur ces entrefaites, craignant que son frère ne soit en
péril, ou poussé par ce désir, commun à tous, de chercher à savoir
les affaires d’autrui, l’avait suivi tout doucement, se tenant dans
l’ombre et le chemin le plus obscur, et s’était caché à moins de dix
pas de lui, dans la même maison.
« Moi, qui ne savais rien de toutes ces choses, je vins au balcon,
sous les habits que j’ai déjà dits, ainsi que j’y étais déjà venue une
ou deux fois sans qu’il en fût rien résulté de fâcheux. Mes vêtements
se voyaient distinctement à la clarté de la lune, et comme je suis
d’aspect à peu près semblable à Ginevra, on pouvait facilement
nous prendre l’une pour l’autre ;
« D’autant plus qu’il y avait un grand espace entre l’endroit où
j’étais et les maisons en ruine. Il fut ainsi facile au duc de tromper les
deux frères qui se tenaient dans l’ombre. Or, tu penses dans quel
désespoir, dans quelle douleur tomba Ariodant. Polinesso s’avance,
s’approche de l’échelle que je lui lance, et monte sur le balcon.
« A peine est-il arrivé, je lui jette les bras au cou, car je ne
pensais pas être vue ; je l’embrasse sur la bouche et sur toute la
figure, comme j’avais coutume de le faire à chacune de ses visites.
Lui, plus que d’habitude, affecte de me combler de caresses, afin
d’aider à sa fraude. L’autre malheureux, conduit à un si douloureux
spectacle, voit tout de loin.
« Il tombe dans une telle douleur, qu’il veut s’arracher la vie. Il
pose à terre le pommeau de son épée, et va se jeter sur la pointe.
Lurcanio, qui avait vu avec un grand étonnement le duc monter
jusqu’à moi, mais sans reconnaître qui c’était, s’apercevant du
dessein de son frère, se précipite,
« Et l’empêche de se percer le cœur de sa propre main. S’il avait
tardé, ou s’il s’était trouvé un peu plus éloigné, il ne serait pas arrivé
à temps et n’aurait pu l’arrêter. « — Ah ! malheureux frère, frère
insensé, — s’écrie-t-il, — as-tu perdu l’esprit que, pour une femme,
tu songes à te tuer ? Qu’elles puissent toutes disparaître comme
neige au vent !
« Songe à la faire mourir, elle, et réserve ta mort pour une
occasion qui te fasse plus d’honneur. Tu as pu l’aimer, tant que sa
fourberie ne t’était point révélée ; maintenant elle doit t’être odieuse,
puisque tu as vu de tes yeux combien elle est coupable et de quelle
manière. Cette arme que tu tournais contre toi-même, conserve-la
pour rendre devant le roi un tel crime manifeste à tous. — »
« Quand Ariodant voit son frère près de lui, il abandonne son
sinistre dessein ; mais la résolution qu’il a prise de mourir n’en est
que peu écartée. Il s’éloigne, le cœur non pas blessé, mais déchiré
d’une suprême angoisse. Pourtant, devant son frère, il feint de ne
plus avoir au cœur la colère qu’il avait témoignée tout d’abord.
« Le lendemain matin, sans rien dire à son frère ni à personne, il
partit, conduit par un mortel désespoir, et de lui, pendant plusieurs
jours, on n’eut pas de nouvelles. Hormis le duc et son frère, tout le
monde ignorait la cause de son départ. Dans le palais du roi et par
toute l’Écosse, on tint à ce sujet les propos les plus divers.
« Au bout de huit jours ou à peu près, un voyageur se présente à
la cour devant Ginevra, et lui apporte une nouvelle d’une triste
nature. Ariodant s’était volontairement jeté à la mer pour y chercher
la mort, et n’y avait point été poussé par le vent ou la tempête. Du
haut d’un rocher qui faisait saillie sur la mer, il s’était précipité la tête
la première dans les flots.
« Ce voyageur ajoutait : « — Avant d’en venir là, il m’avait
rencontré par hasard sur son chemin et m’avait dit : « — Viens avec
moi, afin que Ginevra connaisse par toi ce qui m’est advenu. Dis-lui
que la cause de ce que tu vas voir m’arriver tout à l’heure consiste
en ce que j’ai trop vu. Heureux si j’eusse été privé de mes yeux ! —
»
« Nous étions alors près de Capobasso, qui, du côté de l’Irlande,
s’avance quelque peu dans la mer. Après qu’il m’eut ainsi parlé, je le
vis se précipiter tête baissée dans les ondes. Je l’ai laissé dans la
mer, et je suis venu en toute hâte t’apporter la nouvelle. — » A ce
récit, Ginevra, épouvantée, le visage couvert d’une pâleur livide,
resta à moitié morte.
« O Dieu ! que ne dit-elle pas, que ne fit-elle pas, quand elle se
retrouva seule sur sa couche fidèle ! Elle se frappe le sein, elle
déchire ses vêtements, elle dévaste sa belle chevelure d’or, répétant
à chaque instant les paroles qu’Ariodant avait dites à son heure
dernière : La cause de sa mort cruelle et douloureuse venait de ce
qu’il avait trop vu !
« La rumeur courut que c’était par désespoir qu’Ariodant s’était
donné la mort. Le roi ne put s’empêcher d’en verser des larmes,
ainsi que les chevaliers et les dames de la cour. Son frère se montra
le plus affligé de tous et s’abîma dans une douleur si forte, qu’à
l’exemple d’Ariodant, il fut sur le point de tourner sa main contre lui-
même pour le rejoindre.
« Et, se répétant toujours, à part soi, que c’était Ginevra qui était
cause de la perte de son frère, et que ce n’était pas autre chose que
l’action coupable dont il avait été témoin qui l’avait poussé à mourir,
il en vint à un tel désir de vengeance, que, vaincu par la colère et la
douleur, il ne craignit pas de perdre la bonne grâce du roi et de lui
devenir odieux, ainsi qu’à tout le pays.
« Et, devant le roi, choisissant le moment où le salon royal était le
plus rempli de courtisans, il s’en vint et dit : « — Sache, seigneur,
que de la folie qui a poussé mon frère à mourir, ta fille seule est
coupable, car il a eu l’âme traversée d’une douleur telle, pour l’avoir
vue oublier toute pudeur, que, plus que la vie, la mort lui fut chère.
« Il en était amoureux ; et comme ses intentions n’étaient point
déshonnêtes, je ne veux pas le cacher. Il espérait, par son mérite, et
grâce à ses fidèles services, l’obtenir de toi pour femme. Mais
pendant que le malheureux en respirait respectueusement de loin le
parfum, il a vu un autre monter sur l’arbre objet de son culte, et
cueillir le fruit si ardemment désiré. — »
« Et il continua, disant comment il avait vu Ginevra venir sur le
balcon, et comment elle jeta l’échelle par laquelle était monté jusqu’à
elle un amant dont il ne savait pas le nom, et qui avait, pour ne pas
être reconnu, changé ses vêtements et caché ses cheveux. Il ajouta
qu’il était résolu à prouver, par les armes, que tout ce qu’il avait dit
était vrai.
« Tu peux penser si le père de Ginevra fut atterré de douleur,
quand il entendit accuser sa fille. Il s’afflige non seulement d’avoir
appris d’elle ce qu’il n’aurait jamais soupçonné, et ce qui l’étonne
étrangement, mais aussi parce qu’il se voit dans la nécessité, si
aucun guerrier ne prend sa défense et ne convainc Lurcanio de
mensonge, de la condamner et de la faire mourir.
« Je ne pense pas, seigneur, que tu ignores que notre loi
condamne à mort toute dame ou damoiselle convaincue de s’être
livrée à un autre que son époux. Elle est mise à mort, si, au bout
d’un mois, il ne se trouve pas un chevalier assez vaillant pour
soutenir son innocence contre l’accusateur, et prouver qu’elle ne
mérite pas de mourir.
« Le roi, dans l’espoir de la sauver, a fait publier — car il croit que
sa fille est accusée à tort — que son intention est de la donner pour
femme, avec une grande dot, à qui la tirera de l’infamie dont elle est
victime. Mais on ne dit pas qu’aucun guerrier se soit encore présenté
pour elle. Tous se regardent les uns les autres, car ce Lurcanio est
tellement fort aux armes, qu’il semble que tout guerrier ait peur de
lui.
« Le sort cruel veut que Zerbin, frère de Ginevra, soit hors du
royaume. Depuis plusieurs mois déjà, il voyage, donnant de la valeur
de ses armes des preuves éclatantes. Si ce vaillant chevalier se
trouvait moins loin, et dans un lieu où il pût savoir à temps la
nouvelle, il ne manquerait pas de venir au secours de sa sœur.
« Entre-temps, le roi, qui cherche à savoir, au moyen d’autres
preuves que les armes, si ces accusations sont fausses ou vraies, si
sa fille est restée pure ou est devenue coupable, a fait arrêter
quelques-unes de ses suivantes, lesquelles, si la chose est vraie,
doivent le savoir. J’ai compris par là que si j’étais aussi arrêtée, trop
de périls en résulteraient pour le duc et pour moi.
« Et la nuit même je m’échappai de la cour et j’allai trouver le
duc. Je lui fis sentir combien il serait dangereux pour tous les deux
que je fusse arrêtée. Il m’approuva et me dit de ne rien craindre. Par
ses conseils, il m’engagea à me retirer dans une place forte qui lui
appartient près d’ici, et il me donna deux hommes à lui, pour me
servir d’escorte.
« Tu as vu, seigneur, quelles preuves de mon amour j’avais
données à Polinesso, et tu peux juger si je méritais ou non de lui
être chère. Or, écoute quelle récompense j’en ai reçue ; vois le prix
dont il a payé ma grande affection ; vois si, parce qu’elle aime
passionnément, une femme peut jamais espérer être aimée !
« Cet ingrat, ce perfide, ce cruel a fini par douter de ma foi ; il en
est venu à craindre que je révèle ses coupables ruses ourdies de si
loin. Il a feint, pour attendre que la colère du roi se soit apaisée, de
vouloir m’éloigner et me cacher dans une de ses places fortes, et il
avait résolu de m’envoyer droit à la mort.
« Car, en secret, il avait ordonné à mes guides, pour digne prix
de ma fidélité, de me tuer dans cette forêt où tu m’as soustraite à
leurs coups. Et son projet se fût accompli, si tu n’étais accouru à
mes cris. Vois comme Amour traite ceux qui lui sont soumis ! — »
Voilà ce que Dalinda raconta au paladin, pendant qu’ils
poursuivaient leur route.
Renaud fut charmé par-dessus tout d’avoir trouvé la donzelle qui
lui avait raconté toute l’histoire de l’innocence de la belle Ginevra. Et
s’il avait espéré la sauver quand elle paraissait accusée avec raison,
il se sentit une bien plus grande force en ayant la preuve évidente
qu’elle avait été calomniée.
Et vers la ville de Saint-André, où étaient le roi et toute sa famille,
et où devait se livrer le combat singulier pour la querelle de sa fille,
Renaud se dirigea aussi rapidement qu’il put, jusqu’à ce qu’il en fût
arrivé à quelques milles. Aux environs de la ville, il trouva un écuyer
qui lui apprit les plus fraîches nouvelles,
Et qu’un chevalier étranger était venu, qui s’était présenté pour
défendre Ginevra. Ce chevalier portait des insignes inaccoutumés,
et l’on n’avait pu le reconnaître, attendu qu’il se tenait le plus
souvent caché ; que, depuis son arrivée, personne n’avait encore vu
son visage à découvert, et que l’écuyer qui le servait disait en
jurant : « — Je ne sais pas qui c’est. — »
Ils ne chevauchèrent pas longtemps sans arriver sous les murs
de la ville, près de la porte. Dalinda avait peur d’aller plus avant ;
pourtant elle continue son chemin, réconfortée par Renaud. La porte
est fermée. A celui qui en avait la garde, Renaud demanda ce que
cela signifiait, et il lui fut répondu que c’était parce que toute la
population était sortie pour voir la bataille
Qui, entre Lurcanio et un chevalier étranger, se livrait de l’autre
côté de la ville, dans un pré spacieux et uni, et que déjà le combat
était commencé. La porte est ouverte au seigneur de Montauban, et
le portier la ferme aussitôt sur lui. Renaud traverse la cité vide, après
avoir tout d’abord laissé la donzelle dans une hôtellerie,
Et lui avoir dit de rester là en sûreté jusqu’à ce qu’il revienne vers
elle, ce qui ne tardera pas. Puis il se dirige rapidement vers le
champ de bataille, où les deux guerriers avaient déjà échangé de
nombreux coups et s’en portaient encore. Lurcanio avait le cœur mal
disposé contre Ginevra, et l’autre, pour sa défense, soutenait
vaillamment son entreprise volontaire.
Six chevaliers à pied, armés de cuirasses, se tenaient avec eux
dans la lice, ainsi que le duc d’Albanie, monté sur un puissant
coursier de bonne race. Comme grand connétable, la garde du
camp et de la place lui avait été confiée ; et de voir Ginevra en un si
grand danger, il avait le cœur joyeux et le regard plein d’orgueil.
Renaud s’avance à travers la foule, où le bon destrier Bayard se
fait ouvrir un large passage. Quiconque l’entend venir comme une
tempête n’est ni long ni boiteux à lui faire place. Renaud se
présente, dominant tout le monde et portant au visage la fleur de
toute vaillance. Puis il va s’arrêter devant la place où siège le roi.
Chacun s’approche pour entendre ce qu’il demande.
Renaud dit au roi : « — Grand prince, ne laisse pas la bataille se
poursuivre, car quel que soit celui de ces deux chevaliers qui meure,
sache que tu l’auras laissé mourir à tort. L’un croit avoir raison et est
induit en erreur ; il soutient le faux et ne sait pas qu’il ment. Cette
même erreur, qui a poussé son frère à la mort, lui met les armes aux
mains.
« L’autre ne sait s’il a tort ou raison ; mais il s’est exposé au péril
uniquement par courtoisie et par bonté, et pour ne pas laisser périr
tant de beauté. Moi, j’apporte le salut à celle qui est innocente et le
châtiment à qui a usé de fausseté. Mais, pour Dieu, arrête d’abord
ce combat ; puis donne-moi audience pour entendre ce que je vais
te raconter. — »
Le roi fut si ému du ton d’autorité d’un homme aussi digne que lui
paraissait être Renaud, qu’il fit un signe pour que le combat ne fût
pas poussé plus loin. Alors, en présence des barons du royaume,
des chevaliers et des autres spectateurs, Renaud dévoila toute la
fourberie que Polinesso avait ourdie contre Ginevra ;
Et il s’offrit à prouver par les armes que ce qu’il avait dit était vrai.
Il appela Polinesso, et celui-ci parut, mais le visage tout troublé.
Pourtant il commença à nier avec audace. Renaud dit : « — Nous
allons voir à l’épreuve. — » L’un et l’autre étaient armés, le champ
tout préparé, de telle sorte que sans retard ils en viennent aux
mains.
Oh ! comme le roi, comme son peuple font des vœux pour qu’il
soit prouvé que Ginevra est innocente ! Tous ont l’espérance que
Dieu montrera clairement qu’elle a été accusée injustement
d’impudicité. Polinesso avait la réputation d’un homme cruel,
orgueilleux, inique et trompeur, si bien qu’à personne il ne paraît
extraordinaire qu’une semblable fourberie ait été ourdie par lui.
L’air consterné, le cœur tremblant, le visage pâle, Polinesso
attend, et au troisième son de la trompette, il met sa lance en arrêt.
De son côté, Renaud se lance contre lui, et, désireux d’en finir, il le
vise de façon à lui transpercer le cœur avec sa lance. L’effet suit de
près le désir, car il lui plonge la moitié du fer dans la poitrine.
La lance fixée dans le corps, Polinesso est jeté à plus de six
brasses loin de son destrier. Renaud saute promptement à terre, et,
avant qu’il puisse se relever, lui saisit le casque et le délace. Mais
celui-ci, qui ne peut plus continuer le combat, lui demande merci
d’un air humble, et confesse, devant le roi et la cour qui l’entendent,
la fraude qui l’a conduit à la mort.
Il n’achève pas ; au milieu de ses aveux, la voix et la vie
l’abandonnent. Le roi, qui voit sa fille sauvée de la mort et de
l’infamie, joyeux et consolé, est plus heureux que si, après avoir
perdu sa couronne, il se la voyait rendre. Il glorifie uniquement
Renaud.
Puis, après l’avoir reconnu dès que celui-ci a ôté son casque —
car il l’avait vu plusieurs fois déjà — il lève les mains au ciel, et
remercie Dieu de lui avoir envoyé ainsi à temps un tel défenseur.
Quant à l’autre chevalier inconnu qui avait secouru Ginevra dans sa
triste situation, et avait combattu pour elle, il se tenait à l’écart,
attentif à tout ce qui venait de se passer.
Le roi le pria de dire son nom ou de se laisser voir au moins à
découvert, afin qu’il pût le remercier et lui offrir la récompense que
méritait sa bonne intention. Celui-ci, après qu’on l’eut prié
longuement, ôta son casque et se montra en plein jour. Je vous dirai
qui il était dans le chant qui va suivre, s’il vous est agréable de
l’entendre.
CHANT VI.
Argument. — On reconnaît que le chevalier inconnu est

Ariodant, l’amant de Ginevra. Le roi la lui donne pour femme et
pardonne à Dalinda. — Roger est porté par l’hippogriffe dans l’île
d’Alcine, où Astolphe, cousin de Bradamante, changé en myrte, lui
conseille de ne pas aller plus avant. Roger veut s’éloigner de l’île ;
divers monstres s’opposent en vain à sa fuite ; mais surviennent
plusieurs nymphes qui le font changer de résolution.
Malheur à celui qui, faisant le mal, s’imagine que son crime restera
toujours caché ! Alors que tous le tairaient, l’air et la terre elle-même
où est ensevelie sa victime le crieraient tout autour de lui. Et Dieu
fait souvent que le péché pousse le pécheur à le rendre lui-même
fortuitement manifeste, sans qu’il en soit accusé par personne, ou
après qu’il en a été absous.
Le misérable Polinesso avait cru cacher à tout jamais son crime
en faisant disparaître Dalinda qui le connaissait et pouvait seule le
dénoncer. En ajoutant un second crime au premier, il avança le
châtiment qu’il pouvait différer et éviter peut-être. Mais sa propre
précipitation le fit courir à la mort.
Et il perdit d’un seul coup ses amis, sa vie, son rang et, ce qui fut
bien pis encore, l’honneur. J’ai dit plus haut que le chevalier dont on
ne sait pas encore le nom fut longtemps prié de se faire connaître. Il
ôte enfin son casque, et montre aux yeux des assistants un visage
aimé et qu’ils ont vu plus d’une fois ; et il fit voir qu’il était Ariodant,
que l’Écosse entière pleurait ;
Ariodant, que Ginevra avait pleuré comme mort, que son frère
avait également pleuré, ainsi que le roi, la cour et tout le peuple, et
qui venait de faire éclater tant de bonté et de valeur. On vit alors que
le voyageur n’avait pas dit vrai dans ce qu’il avait raconté à son
sujet. Et pourtant il l’avait véritablement vu se jeter tête baissée dans
la mer du haut du rocher.
Mais — comme il arrive souvent au désespéré qui, de loin,
appelle et désire la mort, et la repousse quand il la voit près de lui,
tant elle lui paraît amère et cruelle — à peine Ariodant s’est-il
précipité dans la mer, qu’il se repent d’avoir voulu mourir. Et comme
il était fort, adroit et plus audacieux que n’importe qui, il se mit à
nager et regagna le rivage.
Et, traitant de folie le désir qu’il avait eu d’abandonner la vie, il se
mit en route, les vêtements imprégnés et amollis par l’eau, et arriva
à la demeure d’un ermite. Il y demeura secrètement, attendant de
savoir quel effet la nouvelle de sa mort avait fait sur Ginevra ; si elle
s’en était réjouie, ou si elle en avait été triste et affligée.
Il apprit d’abord que, dans sa grande douleur, elle avait failli
mourir — le bruit s’en était répandu rapidement dans toute l’île —
résultat tout à fait contraire à ce qu’il attendait, d’après ce que, à son
extrême chagrin, il croyait avoir vu. Il sut ensuite comment Lurcanio
avait accusé Ginevra auprès de son père.
Il ressentit autant de colère contre son frère, qu’il avait eu jadis
d’amour pour Ginevra. Cette action lui paraît trop impie et trop
cruelle, encore qu’elle ait été faite pour lui. Enfin il fut informé
qu’aucun chevalier ne s’était présenté pour défendre Ginevra, car
Lurcanio était si fort et si vaillant, que personne n’avait garde de se
mesurer à lui.
Et puis il était connu pour un homme discret, et si sage et si avisé
que, si ce qu’il avait raconté n’eût pas été vrai, il ne se serait pas
exposé à la mort pour le soutenir. C’est pourquoi la plupart hésitaient
à défendre une cause peut-être mauvaise. Ayant appris cela,
Ariodant, après s’être tenu à lui-même de grands discours, se
résolut à relever l’accusation de son frère.
« — Hélas ! je ne pourrais — disait-il en lui-même — la laisser
périr à cause de moi. Ma mort serait trop amère et trop misérable si,
avant moi, je la voyais mourir. Elle est toujours ma dame, ma
déesse ; elle est la lumière même de mes yeux. Je dois, qu’elle soit
innocente ou coupable, entreprendre de la délivrer et mourir sur le
champ du combat.
« Si j’entreprends une cause mauvaise, c’est à elle qu’en sera la
faute, et moi j’en mourrai ; et cela ne me décourage pas, car je sais
que ma mort entraînera la mort d’une si belle dame. Une seule
pensée me consolera en mourant, c’est qu’elle aura pu voir que ce
Polinesso, à qui elle a donné son amour, ne s’est pas même
présenté pour la défendre.
« Et moi qu’elle a si grandement offensé, elle m’aura vu courir à
la mort pour la sauver. Je me serai aussi par là vengé de mon frère
qui a allumé un tel feu. Et je le ferai gémir sur le résultat de sa
cruelle entreprise, quand il saura qu’en croyant venger son frère, il
lui a donné la mort de sa propre main. — »
Dès qu’il eut arrêté cela dans son esprit, il se procura de
nouvelles armes, un nouveau cheval, choisit une cotte de mailles et
un écu noirs, bordés de vert et de jaune. Et, ayant par aventure
trouvé un écuyer étranger au pays, il l’emmena avec lui. C’est alors
que, sans être connu, il se présenta, comme je l’ai déjà dit, contre
son frère qui attendait tout armé.
Je vous ai raconté l’issue du combat, et comment Ariodant fut
reconnu. Le roi n’en eut pas une moindre joie que lorsqu’il avait vu
sa fille délivrée. Il pensa en lui-même qu’elle ne pourrait jamais
trouver un plus fidèle, un plus sincère amant, puisqu’il l’avait
défendue contre son propre frère, après en avoir reçu une si grande
offense.
Et autant de sa propre inclination, car il l’aimait beaucoup, que
sur les prières de toute la cour et de Renaud, qui insistait plus que
les autres, il en fit l’époux de sa charmante fille. La duché d’Albanie,
qui retournait au roi après la mort de Polinesso, ne pouvait pas se
trouver vacante en meilleure circonstance ; c’est pourquoi il la donna
en dot à sa fille.
Renaud obtint la grâce de Dalinda qui, délivrée de sa funeste
erreur, rassasiée du monde, tourna son esprit vers Dieu et se
consacra à lui. Elle alla se faire religieuse en Dace, et quitta
immédiatement l’Écosse. Mais il est temps désormais de retrouver
Roger qui parcourt le ciel sur son léger cheval.
Bien que Roger soit d’un courage indomptable, et qu’il n’ait pas
changé de couleur, je ne puis croire que, dans sa poitrine, son cœur
ne tremble pas plus que la feuille. Il avait dépassé de beaucoup
l’Europe, et était parvenu bien au delà des bornes qu’Hercule avait
jadis imposées aux navigateurs.
L’hippogriffe, grand et étrange oiseau, l’emporte avec une telle
rapidité d’ailes, qu’il aurait laissé bien loin derrière lui le prompt
agent de la foudre. De tous les oiseaux qui vont, légers, par les airs,
aucun ne lui serait égal en vitesse. Je crois que c’est à peine si le
tonnerre et la flèche arrivent du ciel sur terre avec plus de
promptitude.
Après que le cheval-oiseau eut parcouru un grand espace en
ligne droite et sans jamais se détourner, fatigué d’aller dans les airs,
il commença à décrire de larges cercles et s’abattit sur une île. Elle
était semblable à celle où, pour éviter la longue poursuite de son
amant et se dérober à lui, la vierge Aréthuse se fraya en vain sous la
mer un chemin sombre et étrange [43] .
Le chevalier n’avait rien vu d’aussi beau ni d’aussi agréable dans
tout son voyage à travers les airs ; et, s’il avait cherché par le monde
entier, il n’aurait pas vu de plus joli pays que celui où, après avoir
plané un grand moment, le grand oiseau descendit avec Roger. Ce
n’était partout que plaines cultivées, collines charmantes, eaux
claires, rives ombreuses et prés moelleux.
De ravissants bosquets de lauriers odorants, de palmiers, de
myrtes gracieux, de cèdres et d’orangers qui portaient des fruits et
des fleurs et entrelaçaient leurs formes belles et variées, faisaient un
rempart contre les chaleurs ardentes des jours d’été, avec leurs
épaisses ramures en forme d’ombrelles. Et dans leurs rameaux
voltigeaient en sûreté et chantaient les rossignols.
Parmi les roses pourprées et les lis blancs, qu’une tiède brise
conserve toujours frais, on voyait les lièvres et les lapins courir sans
crainte, et les cerfs au front élevé et superbe, sans redouter d’être
pris et tués, paître l’herbe et ruminer en repos. Les daims et les
chèvres, agiles et pleins d’adresse, bondissaient en foule sous ces
bosquets champêtres.
Dès que l’hippogriffe est assez près de terre pour que l’on puisse
sauter sans trop de danger, Roger s’enlève rapidement de l’arçon et
se retrouve sur le gazon émaillé. Il serre toutefois les rênes dans sa
main, car il ne veut pas que le destrier s’envole de nouveau. Il
l’attache sur le rivage à un myrte verdoyant, entre un laurier et un
pin.
Puis, dans un endroit où jaillissait une fontaine couronnée de
cèdres et de palmiers touffus, il pose son écu, ôte son casque du
front, et se désarme les deux mains, Et, tourné tantôt vers la mer,
tantôt vers la montagne, il livre son visage aux brises fraîches et
suaves qui, avec de doux murmures, font trembler les hautes cimes
des hêtres et des sapins.
Il baigne dans l’onde claire et fraîche ses lèvres desséchées ; il
l’agite avec les mains, pour apaiser la chaleur qu’a allumée dans ses
veines le poids de sa cuirasse. Et il ne faut point s’étonner que cette
chaleur soit devenue si grande, car il a été loin de se tenir en une
même place ; au contraire, sans jamais se reposer et couvert de ses
armes, il est allé toujours courant pendant trois mille milles.
Pendant qu’il se repose en cet endroit, le destrier qu’il avait laissé
au plus épais du feuillage sous l’ombre fraîche, se cabre tout à coup,
comme s’il voulait fuir, épouvanté qu’il est par je ne sais quoi de
caché dans les branches. Et il secoue tellement le myrte auquel il
est attaché, qu’il encombre tout autour la terre de ses rameaux. Il
secoue le myrte au point d’en faire tomber les feuilles, mais sans
réussir à s’en détacher.
Comme fait parfois un tronc d’arbre à la moelle rare ou absente,
quand il est mis au feu, et que la grande chaleur consume l’air
humide qui le remplit et le fait résonner en dedans, jusqu’à ce qu’elle
se fraye un chemin au dehors avec un bouillonnement strident, ainsi
murmure, crie et se courrouce ce myrte blessé, et enfin ouvre son
écorce,
D’où, avec une voix triste et plaintive, sortent, distinctes et
claires, ces paroles : « — Si tu es courtois et accessible à la pitié,
comme le montre ta belle physionomie, éloigne cet animal de mon
arbre. Il suffit que je sois affligé de mon propre mal, sans qu’une
autre peine, sans qu’une autre douleur vienne encore du dehors
pour me tourmenter. — »
Au premier son de cette voix, Roger tourne les yeux et se lève
subitement. Et quand il s’aperçoit qu’elle sort de l’arbre, il reste plus
stupéfait que jamais. Il s’empresse d’écarter le destrier, et, la
rougeur sur les joues : « — Qui que tu sois — dit-il — pardonne-
moi, esprit humain ou déesse des bocages.
« Je ne savais pas que, sous ta rude écorce, se cachait un esprit
humain ; c’est pourquoi j’ai laissé endommager ton beau feuillage et
insulter à ton myrte vivace. Mais ne tarde pas à m’apprendre qui tu
es, toi qui, en un corps grossier et rugueux, vis et parles comme un
animal doué de raison. Que de l’orage le ciel te préserve toujours !
« Et si, maintenant ou jamais, je puis réparer par quelque service
le mal que je viens de te causer, je te promets, par la belle dame qui
possède la meilleure part de moi-même, de faire de telle sorte, par
mes paroles et par mes actes, que tu aies une juste raison de te
louer de moi. — » A peine Roger eut-il fini de parler, que le myrte
trembla de la tête au pied.
Puis on vit son écorce se couvrir de sueur, comme le bois
fraîchement tiré de la forêt, qui sent la violence du feu après lui avoir
en vain fait toute sorte de résistance. Et il commença : « — Ta
courtoisie me force à te découvrir en même temps qui j’ai d’abord
été, et ce qui m’a changé en myrte sur cette charmante plage.
« Mon nom fut Astolphe, et j’étais un paladin de France très
redouté dans les combats. J’étais cousin de Roland et de
Renaud [44] , dont la renommée n’a pas de bornes. Je devais, après
mon père Othon, régner sur toute l’Angleterre. J’étais si beau et si
bien fait, que plus d’une dame s’enflamma pour moi. Seul je me suis
perdu moi-même.
« Je revenais de ces îles lointaines qu’en Orient baigne la mer
des Indes, où Renaud et quelques autres avec moi avions été
retenus prisonniers dans un obscur et profond cachot, et d’où nous
avait délivrés la suprême vaillance du chevalier de Brava ; me
dirigeant vers le ponant, j’allais le long de la côte qui du vent du nord
éprouve la rage.
« Et comme si le destin cruel et trompeur nous eût poussés sur
ce chemin, nous arrivâmes un matin sur une belle plage où s’élève,
sur le bord de la mer, un château appartenant à la puissante Alcine.
Nous la trouvâmes sortie de son château, et qui se tenait sur le
rivage, attirant sur le bord, sans filets et sans amorce, tous les
poissons qu’elle voulait.
« Les dauphins rapides y accouraient, et les thons énormes à la
bouche ouverte ; les baleines et les veaux marins, troublés dans leur
lourd sommeil ; les mulets, les salpes, les saumons et les barbues
nageaient en troupes le plus vite qu’ils pouvaient. Les physitères, les
orques et les baleines montraient hors de la mer leurs monstrueuses
échines.
« Nous aperçûmes une baleine, la plus grande qui se soit jamais
vue sur toutes les mers. Onze pas et plus émergeaient hors des
ondes ses larges épaules. Et nous tombâmes tous dans une grande
erreur ; car, comme elle se tenait immobile et sans jamais bouger,
nous la prîmes pour une petite île, tellement ses deux extrémités
étaient distantes l’une de l’autre.
« Alcine faisait sortir les poissons de l’eau avec de simples
paroles et de simples enchantements. Avec la fée Morgane elle
reçut le jour ; mais je ne saurais dire si ce fut dans la même couche
ou avant, ou après. Alcine me regarda, et soudain mon aspect lui
plut, comme elle le montra sur son visage. Et il lui vint à la pensée
de m’enlever, par astuce et artifice, à mes compagnons. Son
dessein réussit.
« Elle vint à notre rencontre l’air souriant, avec des gestes
gracieux et prévenants, et dit : « — Chevaliers, qu’il vous plaise de
prendre aujourd’hui vos logements chez moi. Je vous ferai voir, dans
ma pêche, toutes sortes de poissons différents, les uns recouverts
d’écailles, les autres lisses, et d’autres tout poilus, et tous plus
nombreux qu’il n’y a d’étoiles au ciel.
« Et si nous voulons voir une sirène qui apaise la mer par son
doux chant, passons d’ici sur cette autre plage, où, à cette heure,
elle a toujours coutume de retourner. — » Et elle nous montra cette
grande baleine qui, comme je l’ai dit, paraissait être une île. Moi, qui
fus toujours trop entreprenant — et je m’en repens — j’allai sur ce
poisson.
« Renaud me faisait signe, ainsi que Dudon, de ne pas y aller,
mais cela servit peu. La fée Alcine, avec un visage riant, laissa les
deux autres et s’élança derrière moi. La baleine, à lui obéir diligente,
s’en alla, nageant à travers l’onde salée. Je ne tardai pas à me
repentir de ma sottise, mais je me trouvais trop éloigné du rivage.
« Renaud se jeta à la nage pour m’aider et faillit être englouti, car
un furieux vent du sud s’éleva, qui couvrit d’une ombre épaisse le
ciel et la mer. J’ignore ce qui lui est ensuite arrivé. Alcine s’efforçait
de me rassurer, et pendant tout ce jour et la nuit suivante elle me tint
sur ce monstre au milieu de la mer,
« Jusqu’à ce que nous arrivâmes à cette belle île, dont Alcine
possède une grande partie. Elle l’a usurpée sur une de ses sœurs, à
qui leur père l’avait entièrement laissée en héritage parce qu’elle
était sa seule enfant légitime. Les deux autres, à ce que m’a dit
depuis quelqu’un qui en était pleinement instruit, sont nées d’un
inceste.
« Et de même qu’elles sont iniques et pleines de scélératesse et
de vices infâmes, leur sœur, qui vit chaste, a dans son cœur toutes
les vertus. Les deux autres se sont liguées contre elle, et déjà plus

Ebook Spatiotemporal Random Fields Second Edition Theory and Applications PDF Full Chapter PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ebook Spatiotemporal Random Fields Second Edition Theory and Applications PDF Full Chapter PDF

Uploaded by

Copyright:

Available Formats

Spatiotemporal Random Fields, Second

Edition: Theory and Applications -

Copyright © 2017 Elsevier Inc. All rights reserved.

For Information on all Elsevier publications visit our

Publisher: Candice Janco

CHAPTER I Space, Time, SpaceeTime, Randomness, and Probability .................. 1

7.3 Third Classification: Probability Law Shapes .................................................... 74

6. Correlation Theory of Complex Spatiotemporal Random Fields ............................146

5. The Central Limit Theorem......................................................................................233

6. Scales of Spatiotemporal Dependence and the Uncertainty Principle ....................329

2.2 The Spherical Symmetry Technique................................................................. 412

4. Discrete Linear Representations of Spatiotemporal Random Fields .......................490

3.2 Normalized Angular Spectrum Decomposition.............................................. 554

SPACE, TIME, SPACEeTIME,

Spatiotemporal Random Fields. http://dx.doi.org/10.1016/B978-0-12-803012-7.00001-5

Table 1.1 Commonly Used Symbols

Kronecker/Levi-Chivita εijkεipq ¼ djpdkq djqdkp

Table 1.2 Special Functions and Polynomials

Gamma function. G(n)

2. SPACEeTIME CONTINUUM AND KOLMOGOROV PROBABILITY SPACE

2.1 SPACEeTIME ARGUMENTS: POINTS, LAGS, SEPARATIONS, AND METRICS

Dp ¼ ðDs1 ; .; Dsn ; Ds0 Þ ˛ Rnþ1 ; (2.7a)

Dp ¼ ðh; sÞ ˛ Rn;1 ; (2.7b)

Table 2.1 Standard Vector Operations on SpaceeTime Points and Lags

Table 2.1 Standard Vector Operations on SpaceeTime Points and Lagsdcont’d

Table 2.2 Number of Monomials for n [ 2, 3, and Different Combinations

Table 2.3 Monomials in R2,1 With n=m [ 2=1

g1(s, t) 0 0 s01 s02 t0

Also, jDpj is called the magnitude of Dp.

denoted as rp, and the Minkowski or Einstein spaceetime metric

(h0 ¼ s); i.e., the metric isP

2.2 TRANSFORMATIONS AND INVARIANCE IN SPACEeTIME

jLt ð pÞj ¼ j pj;

i.e., the jDpj of Eq. (2.19b) changed by the Dl transformation.

2.3 SPACEeTIME INTERPRETATIONS

2.4 FUNCTIONS OF SPACEeTIME ARGUMENTS

Argument. — On reconnaît que le chevalier inconnu est

You might also like