L ECTURES ON

S TATIONARY S TOCHASTIC P ROCESSES

A COURSE FORP H D STUDENTS IN
M ATHEMATICAL S TATISTICS AND OTHER FIELDS

G EORG L INDGREN
October 2006

CENTRUM SCIENTIARUM MATHEMATICARUM

Faculty of Engineering
Centre for Mathematical Sciences
Mathematical Statistics
Contents

Foreword vii

1 Some probability and process background 1

1.1 Probability measures on sample spaces . . . . . . . . . . . . . . . 1
1.2 Events, probabilities and random variables . . . . . . . . . . . . . 3
1.2.1 Events and families of events . . . . . . . . . . . . . . . . 3
1.2.2 Probabilities . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Random variables and random sequences . . . . . . . . . 5
1.2.4 Conditional expectation . . . . . . . . . . . . . . . . . . . 7
1.3 Stochastic processes . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1 Stochastic processes and finite-dimensional distributions . 7
1.3.2 The distribution of a random sequence . . . . . . . . . . . 9
1.3.3 The continuous parameter case . . . . . . . . . . . . . . . 10
1.4 Stationary processes and fields . . . . . . . . . . . . . . . . . . . 13
1.4.1 Stationary processes . . . . . . . . . . . . . . . . . . . . . 13
1.4.2 Random fields . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Gaussian processes . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5.1 Multivariate normal distributions and Gaussian processes 15
1.5.2 Linear prediction and reconstruction . . . . . . . . . . . . 18
1.5.3 Some useful inequalities . . . . . . . . . . . . . . . . . . . 18
1.6 Some historical landmarks . . . . . . . . . . . . . . . . . . . . . . 19
1.6.1 Brownian motion and the Wiener process . . . . . . . . . 19
1.6.2 Rice and electronic noise . . . . . . . . . . . . . . . . . . . 21
1.6.3 Gaussian random wave models . . . . . . . . . . . . . . . 22
1.6.4 Detection theory and statistical inference . . . . . . . . . 24
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2 Stochastic analysis 27
2.1 Quadratic mean properties . . . . . . . . . . . . . . . . . . . . . 27
2.2 Sample function continuity . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Countable and uncountable events . . . . . . . . . . . . . 28
2.2.2 Conditions for sample function continuity . . . . . . . . . 30
2.2.3 Probability measures on C[0, 1] . . . . . . . . . . . . . . . 37
2.3 Derivatives, tangents, and other characteristics . . . . . . . . . . 37

i
ii Contents

2.3.1 Differentiability . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.2 Jump discontinuities and Hölder conditions . . . . . . . . 40
2.4 Quadratic mean properties a second time . . . . . . . . . . . . . 45
2.4.1 Quadratic mean continuity . . . . . . . . . . . . . . . . . 45
2.4.2 Quadratic mean differentiability . . . . . . . . . . . . . . 46
2.4.3 Higher order derivatives and their correlations . . . . . . 48
2.5 Summary of smoothness conditions . . . . . . . . . . . . . . . . . 49
2.6 Stochastic integration . . . . . . . . . . . . . . . . . . . . . . . . 49
2.7 An ergodic result . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3 Crossings 57
3.1 Level crossings and Rice’s formula . . . . . . . . . . . . . . . . . 57
3.1.1 Level crossings . . . . . . . . . . . . . . . . . . . . . . . . 57
3.1.2 Rice’s formula for absolutely continuous processes . . . . 58
3.1.3 Alternative proof of Rice’s formula . . . . . . . . . . . . . 60
3.1.4 Rice’s formula for differentiable Gaussian processes . . . . 62
3.2 Prediction from a random crossing time . . . . . . . . . . . . . . 63
3.2.1 Prediction from upcrossings . . . . . . . . . . . . . . . . . 64
3.2.2 The Slepian model . . . . . . . . . . . . . . . . . . . . . . 67
3.2.3 Excursions and related distributions . . . . . . . . . . . . 73
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4 Spectral- and other representations 79

4.1 Complex processes and their covariance functions . . . . . . . . . 79
4.1.1 Stationary processes . . . . . . . . . . . . . . . . . . . . . 79
4.1.2 Non-negative definite functions . . . . . . . . . . . . . . . 80
4.1.3 Strict and weak stationarity . . . . . . . . . . . . . . . . . 82
4.2 Bochner’s theorem and the spectral distribution . . . . . . . . . . 82
4.2.1 The spectral distribution . . . . . . . . . . . . . . . . . . 82
4.2.2 Properties of the spectral distribution . . . . . . . . . . . 86
4.2.3 Spectrum for stationary sequences . . . . . . . . . . . . . 88
4.3 Spectral representation of a stationary process . . . . . . . . . . 89
4.3.1 The spectral process . . . . . . . . . . . . . . . . . . . . . 89
4.3.2 The spectral theorem . . . . . . . . . . . . . . . . . . . . 90
4.3.3 More on the spectral representation . . . . . . . . . . . . 93
4.3.4 Spectral representation of stationary sequences . . . . . . 100
4.4 Linear filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.4.1 Projection and the linear prediction problem . . . . . . . 100
4.4.2 Linear filters and the spectral representation . . . . . . . 101
4.4.3 Linear filters and differential equations . . . . . . . . . . . 106
4.4.4 White noise in linear systems . . . . . . . . . . . . . . . . 112
4.4.5 The Hilbert transform and the envelope . . . . . . . . . . 115
4.4.6 The sampling theorem . . . . . . . . . . . . . . . . . . . . 119
Contents iii

4.5 Karhunen-Loève expansion . . . . . . . . . . . . . . . . . . . . . 121

4.5.1 Principal components . . . . . . . . . . . . . . . . . . . . 121
4.5.2 Expansion of a stationary process along eigenfunctions . . 122
4.5.3 The Karhunen-Loève theorem . . . . . . . . . . . . . . . . 124
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

5 Ergodic theory and mixing 133

2
5.1 The basic Ergodic theorem in L . . . . . . . . . . . . . . . . . . 133
5.2 Stationarity and transformations . . . . . . . . . . . . . . . . . . 134
5.2.1 Pseudo randomness and transformation of sample space . 134
5.2.2 Strict stationarity and measure preserving transformations135
5.3 The Ergodic theorem, transformation view . . . . . . . . . . . . . 137
5.3.1 Invariant sets and invariant random variables . . . . . . . 138
5.3.2 Ergodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.3.3 The Birkhoff Ergodic theorem . . . . . . . . . . . . . . . . 141
5.4 The Ergodic theorem, process view . . . . . . . . . . . . . . . . . 144
5.5 Ergodic Gaussian sequences and processes . . . . . . . . . . . . . 148
5.6 Mixing and asymptotic independence . . . . . . . . . . . . . . . . 150
5.6.1 Singularity and regularity . . . . . . . . . . . . . . . . . . 150
5.6.2 Asymptotic independence, regularity and singularity . . . 152
5.6.3 Uniform, strong, and weak mixing . . . . . . . . . . . . . 156
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

6 Vector processes and random fields 161

6.1 Cross-spectrum and spectral representation . . . . . . . . . . . . 161
6.1.1 Spectral distribution . . . . . . . . . . . . . . . . . . . . . 162
6.1.2 Spectral representation of x(t) . . . . . . . . . . . . . . . 163
6.2 Some random field theory . . . . . . . . . . . . . . . . . . . . . . 165
6.2.1 Homogeneous fields . . . . . . . . . . . . . . . . . . . . . . 166
6.2.2 Isotropic fields . . . . . . . . . . . . . . . . . . . . . . . . 167
6.2.3 Randomly moving surfaces . . . . . . . . . . . . . . . . . 169
6.2.4 Stochastic water waves . . . . . . . . . . . . . . . . . . . . 170
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

A The axioms of probability 175

A.1 The axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
A.2 Extension of a probability from field to σ -field . . . . . . . . . . 176
A.3 Kolmogorov’s extension to R∞ . . . . . . . . . . . . . . . . . . . 178
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

B Stochastic convergence 181

B.1 Criteria for convergence almost surely . . . . . . . . . . . . . . . 181
B.2 Criteria for convergence in quadratic mean . . . . . . . . . . . . 183
B.3 Criteria for convergence in probability . . . . . . . . . . . . . . . 184
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
iv Contents

C Hilbert space and random variables 187

C.1 Hilbert space and scalar products . . . . . . . . . . . . . . . . . . 187
C.2 Projections in Hilbert space . . . . . . . . . . . . . . . . . . . . . 189
C.3 Stochastic processes and Hilbert spaces . . . . . . . . . . . . . . 190

D Spectral simulation of random processes 193

D.1 The Fast Fourier Transform, FFT . . . . . . . . . . . . . . . . . . 193
D.2 Random phase and amplitude . . . . . . . . . . . . . . . . . . . . 194
D.3 Aliasing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
D.4 Simulation scheme . . . . . . . . . . . . . . . . . . . . . . . . . . 195
D.5 Difficulties and details . . . . . . . . . . . . . . . . . . . . . . . . 196
D.6 Simulation of the envelope . . . . . . . . . . . . . . . . . . . . . . 197
D.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

Literature 199

Index 202
Contents v
Foreword

The book Stationary and Related Stochastic Processes [9] appeared in 1967.
Written by Harald Cramér and M.R. Leadbetter, it drastically changed the
life of PhD students in Mathematical statistics with an interest in stochastic
processes and their applications, as well as that of students in many other fields
of science and engineering. By that book, they got access to tools and results for
stationary stochastic processes that until then had been available only in rather
advanced mathematical textbooks, or through specialized statistical journals.
The impact of the book can be judged from the fact that still in 1999, after
more than thirty years, it is a standard reference to stationary processes in PhD
theses and research articles.
Unfortunately, the book only appeared in a first edition and it is since long
out of print. Even if many of the more specialized results in the book now
have been superseded by more general results, and simpler proofs have been
found for some of the statements, the general attitude in the book makes it
enjoyable reading both for the student and for the teacher. It will remain a
definite source of reference for many standard results on sample function and
crossings properties of continuous time processes, in particular in the Gaussian
case.
These lecture notes are the results of a series of PhD courses on Stationary
stochastic processes which have been held at the Department of Mathemat-
ical Statistics, Lund University, during a sequence of years, all based on and
inspired by the book by Cramér and Leadbetter. The aim of the notes is to pro-
vide a reasonably condensed presentation of sample function properties, limit
theorems, and representation theorems for stationary processes, in the spirit of
[9]. It must be said, however, that it represents only a selection of the material,
and the reader who has found interest in the present course, should take the
time to read the original.
Even if the Cramér and Leadbetter book is the basic source of inspiration,
other texts have influenced these notes. The most important of these is the
now reprinted book on Probability [5] by Leo Breiman. The Ergodic chapter is
a mixture of the two approaches. The Karhunen-Loève expansion follows the
book by Wong [35]. Finally, the classical memoirs by S.O. Rice [27], have also
been a source of inspiration.
Some knowledge of the mathematical foundations of probability helps while

vii
viii

reading the text; I have included most of it in Appendices on the probability

axioms, together with the existence and basic convergence properties, as well
as some Hilbert space concepts. There is also an Appendix on how to simulate
stationary stochastic processes by spectral methods and the FFT algorithm.
I am grateful to the PhD-students Rikard Berthilsson, Jonas Brunskog,
Halfdan Grage, Peter Gustafsson, Pär Johannesson, Finn Lindgren, Karl-Ola
Lundberg, Dan Mattsson, Tord Rikte, Jesper Rydén, Martin Sköld, Martin
Svensson, Magnus Wiktorsson, in the 1998/99 course for many detailed com-
ments on the text and pertinent questions during the lectures, which hopefully
helped clarify some of the obscurities. They also helped to remove many of the
misprints.

Lund in May, 1999

Georg Lindgren

Printing September 2002 and April 2004

These printings of the Lecture notes differ considerably from the printing of May
1999. Many misprints have been corrected and there are also many additions,
mostly of previously left out details, but there is also some new aspects. I
am grateful to comments by Bengt Ringnér, which helped to remove unclear
statements and errors, the PhD students at the Lund 2000/01 course, Anastasia
Baxevanni, Torgny Lindström, Ulla Machado, Anders Malmberg, Sebastian
Rasmus, Mikael Signahl, and to Lars Holst and PhD student Henrik Hult at
KTH who used the material and made several suggestions.
Some new references have been added, in particular some standard text-
books on ergodic theory (K. Petersen: Ergodic Theory), real analysis (H.L.
Royden: Real Analysis), and the new probability book, Weighting the Odds by
D. Williams [25, 29, 37].
In the 2004 printing several more changes were made, as the result of nu-
merous valuable comment by Oskar Hagberg and Linda Werner.

Lund in September, 2002 and April 2004

Georg Lindgren

Printing October 2006

In this printing, several changes and additions have been made. I have included
some more elementary facts, together with historical and general examples in
Chapter 1, and expanded the section on Linear filters in Chapter 4 in order not
to rely too much on specific Lund courses.
To provide some background to the theory I have, in Chapter 1, highlighted
four remarkable research achievements that have helped to shape the theory of
stationary processes in general. The examples are specialized and based on real
 ix

demands, namely Albert Einsteins derivation of the physics and mathematics

behind the Brownian motion from 1905, Steve Rice 1944-45 broad invitation
to stochastic Gaussian noise, so important in communication theory, and the
equally important introduction of stochastic thinking in naval engineering by
StDenis and Pierson from 1953, and finally the link between stochastic processes
theory and statistical inference, by Ulf Grenander from 1950, which forms the
basis for present day signal processing.
The old Chapter 3, on prediction, has been transformed into a chapter on
crossing-related problems, including the form and use of the Slepian model
process. What remains of prediction has been moved to the chapter on Ergodic
theory. The section on random fields has been slightly expanded. The notation
λ for frequency has been changed to the standard ω .
Besides the additions, several errors and misprints have been corrected efter
comments from the PhD students in the 2004 course, Klas Bogsjö, Johan Lind-
ström, and Sofia Åberg. Timo Koski, who used the material in Linköping, has
also given me several suggestions and pointed at unclear points and misprints.
Last, but not least, the Cramér & Leadbetter book has finally reappeared,
reprinted by Dover Publications, 2004.

Lund in October, 2006

Georg Lindgren
x
Chapter 1

Some probability and process

background

This introductory chapter gives a brief summary of the probability theory

needed for the study of stochastic processes with discrete or continuous time.
It concentrates on the finite-dimensional distribution functions, which uniquely
define probabilities for a sufficiently rich family of events, namely events that
can be identified through process values at discrete sets of times. In particular,
they allow us to find conditions for sample function continuity, at least if we
restrict ourselves to a study of the process at a dense discrete time set.

1.1 Probability measures on sample spaces

Stochastic processes are often called random functions; these two notions put
emphasis on two different aspects of the theory, namely

• stochastic processes as families of infinitely many random variables on the

same sample space, usually equipped with a fixed probability measure,

• stochastic processes as a means to assign probabilities to sets of func-

tions, for example some specified sets of continuous functions, or sets of
piecewise constant functions with unit jumps.

These two aspects of stochastic processes can be illustrated as in Figure 1.1,

corresponding to an experiment where the outcomes are continuous functions.
The figure illustrates the three levels of abstraction and observability for
a random experiment. To be concrete, think of an experiment controlling the
steering of a ship. The general sample space Ω is an abstract set that contains
all the possible outcomes of the experiment that can conceivably happen – and
it may contain more. A probability measure P is defined on Ω that assigns
probabilities to all interesting subsets – we need only one single probability
measure to describe our whole world.

1
2 Some probability and process background Chapter 1

The abstract The function The finite-dimensional

sample space Ω sample space C co-ordinate space Rn

1 x(t, ω), t ∈ R; Px - x(t, ω), t = t , . . . , t ; P (n)

1 n x
outcome ω - y(t, ω), t ∈ R; Py - y(t, ω), t = t , . . . , t ; P (n)
1 n y
probability P
q z(t, ω), t ∈ R; Pz - z(t, ω), t = t , . . . , t ; P (n)
1 n z

This space contains This space contains This space contains

the possible the finite-dimensional
everything possible
function outcomes observations

Figure 1.1: Overview of the three types of worlds in which our processes live.

During the experiment one can record the time evolution of a number of
things, such as rudder angle, which we call {x(t), t ∈ R}, ship head angle,
called {y(t), t ∈ R}, and roll angle {z(t), t ∈ R}. Each observed function is an
observation of a continuous random process. In the figure, the randomness is
indicated by the dependence of the experiment outcome ω . The distributions
of the different processes are Px , Py , Pz – we need one probability measure for
each of the phenomena we have chosen to observe.1
In practice, the continuous functions are sampled in discrete time steps,
t = t1 , . . . , tn , resulting in a finite-dimensional observation vector, (x1 , . . . , xn ),
(n)
with an n-dimensional distribution, Px , etc. This is illustrated in the third
box in the figure.
Since we do not always want to specify a finite value for n, the natural
mathematical model for the practical situation is to replace the middle box,
the sample space C of continuous functions, by the sample space R∞ of infinite
sequences of real numbers (x0 , x1 , . . .). This is close, as we shall see later, really
very close, to the finite-dimensional space Rn , and mathematically not much
more complicated.
Warning: Taking the set C of continuous functions as a sample space and
assigning probabilities Px , etc, on it, is not as innocent as it may sound from
the description above. Chapter 2 deals with conditions that guarantee that a
stochastic process is continuous, i.e. has continuous sample functions. In fact,
these conditions are all on the finite-dimensional distributions.

Summary: The abstract sample space Ω contains everything that can conceiv-
ably happen and is therefore very complex and detailed. Each outcome ω ∈ Ω is
unique, and we need only one comprehensive probability measure P to describe
every outcome of experiment we can do. An experiment is a way to “observe
the world”.
1
The symbol ω is here used to represent then elementary experimental outcome, a practice
that is standard in probability theory. In most part of this book, ω will stand for (angular)
frequency; no confusion should arise from this.
Section 1.2 Events, probabilities and random variables 3

The function (sequence) sample space C (R∞ ) is simple. It can be used

as sample space for a specified experiment for which the result is a function or
sequence of numbers. We have to define a unique probability measure for each
experiment.

1.2 Events, probabilities and random variables

1.2.1 Events and families of events
A probability measure P assigns probabilities to certain events, i.e. subsets, in
the sample space Ω, in such a way that Kolmogorov’s probability axioms are
satisfied.2 Thus, if a subset A has a probability, then also its complement A∗
has a probability, and P (A∗ ) = 1 − P (A), and further, if A is disjoint with B ,
i.e. A ∩ B = ∅, and B has probability P (B), then also A ∪ B has a probability,
and P (A ∪ B) = P (A) + P (B). These requirements lead to the conclusion that
probabilities have to be defined at least on a certain minimal family of subsets
of Ω.

Definition 1:1 A family of subsets F0 to an arbitrary space Ω is called a

field if it contains the whole set Ω and is closed3 under the set operations
complement, A∗ , union, A ∪ B , and intersection, A ∩ B . It then also contains
all unions of finitely many sets A1 , . . . , An in F0 . A field F of subsets is
called a σ -field if it contains all countable unions and intersections of its sets.
The terms algebra and σ -algebra are also used, instead of field and σ -field.

To every collection A of subsets of Ω there is always a unique smallest field

F0 that contains all the sets in A. Similarly, there exists a (unique) smallest
σ -field F that contains all A-sets. That σ -field F is said to be generated by
A, and it is denoted F = σ(A).

Example 1:1 (Fields and σ -fields in R) The simplest useful field F0 of

subsets of the real line R consists of all finite half-open intervals, a < x ≤ b,
together with unions of a finite number of such intervals. In order for F0 to
be a field, it is required that it also contains the complement of such unions.
To anticipate the introduction of probabilities and random variables, we can
remark here that F0 is a natural family of sets, since distribution functions can
be used to assign probabilities to intervals.
The smallest σ -field that contains all half-open intervals is called the Borel
field in R. It is denoted B , and its sets are called Borel sets.

2
See Appendix A.
3
That is, if a set A is in the family F0 , then also the complement A∗ belongs to F0 , etc.
4 Some probability and process background Chapter 1

Example 1:2 (Fields and σ -fields in Rn ) Intervals in R correspond to n-

dimensional rectangles in Rn , and the smallest interesting field in Rn consists
of unions of finitely many half-open rectangles, with sides (ai , bi ], i = 1, . . . , n,
and the complements of such unions. As in R, the σ -field generated by F0 is
called the Borel field in Rn . It is denoted by Bn and its sets are the Borel sets
in Rn .
One could note here that it is possible to start with more general ”rect-
angles”, where the ”sides” are real Borel sets instead of intervals, i.e. sets
B1 × B2 × . . . × Bn , where the Bj are one-dimensional Borel sets. However,
even if these generalized rectangles form a richer class than the simple rectan-
gles, the smallest σ -field that contains all such generalized rectangles is exactly
equal to Bn .

1.2.2 Probabilities
Probabilities are defined for events, i.e. subsets of a sample space Ω. By a
probability measure is meant any function P defined for every event in a field
F0 , such that
0 ≤ P (A) ≤ 1, P (∅) = 0, P (Ω) = 1,

and such that, first of all, for any finite number of disjoint events A1 , . . . , An
in F0 one has

P (A1 ∪ . . . ∪ An ) = P (A1 ) + . . . + P (An ). (1.1)

That is, probabilities are finitely additive. In order to deal with limiting events
and the infinity, they are also required to be countably additive, i.e. equation
(1.1) holds for infinitely many disjoint events, i.e. it holds with n = ∞,
∞

P (∪∞
k=1 Ak ) = P (Ak )
k=1

for all disjoint events Ak ∈ F0 such that ∪∞

1 Ak ∈ F0 .
As remarked, it is easy to assign probabilities to intervals, and unions of
intervals, simply by taking

P ((a, b]) = F (b) − F (a),

for some distribution function F . By additivity and the property of fields, one
then also assigns probability to the field F0 of finite unions of intervals.
A natural question to ask is, whether this also produces probabilities to the
events in the σ -field F generated by F0 . In fact, it does, and that in a unique
way:
Section 1.2 Events, probabilities and random variables 5

Extension of probability measures: Every probability measue P , defined

and countably additive on a field F0 , can be extended to be defined for
every event in the σ -field F generated by F0 . This can be done in one
way only. This means that a probability measure on the real Borel sets
is uniquely determined by its values on the half-open intervals, i.e. it
depends only of the values of the function

F (x) = P ((−∞, x]), F (b) − F (a) = P ((a, b]).

Probabilities on the Borel sets in Rn are similarly uniquely determined

by the n-dimensional distribution function

F (x1 , . . . , xn ) = P ((−∞, x1 ] × . . . × (−∞, xn ]). (1.2)

For example, for a bivariate variable (x, x2 ),

P ((a1 , b1 ] × (a2 , b2 ]) = F (b1 , b2 ) − F (a1 , b2 ) − F (b1 , a2 ) + F (a1 , a2 )).

The probability measure P is defined on the measurable space (Ω, F),

and sets in the σ -field F are called the measurable sets. For a proof of
the existence of a unique extension, see Appendix A.

Remark 1:1 The completion of a probability measure P is obtained as follows.

Suppose P is defined on (Ω, F), i.e. it assigns a probability P (A) to every A in
F . Now, if there is an event B with P (B) = 0, then it seems natural to assign
probability 0 to any smaller set B  ⊂ B . Unfortunately, subsets of measurable
sets, are not necessarily measurable, so one can not immediately conclude that
P (B  ) = 0. However, no other choice is possible, and it is also easy to create
the σ -field that also contains all subsets of F -sets B with P (B) = 0. The
extended probability measure is called a complete probability measure.

1.2.3 Random variables and random sequences

1.2.3.1 A random variable and its distribution
A random variable is just a real-valued function x(ω), ω ∈ Ω, on a probability
space (Ω, F, P ), such that it is possible to talk about its distribution, i.e. the
probability
P (x ≤ a) = P ({ω; x(ω) ≤ a})
is defined for all real a. This means that the set (event)

Aa = x−1 ((−∞, a]) = {ω; x(ω) ≤ a}

is a member of the family F , for all a ∈ R. This is equivalent to the seemingly

more general statement that

x−1 (B) ∈ F for all Borel sets B ∈ B , (1.3)

6 Some probability and process background Chapter 1

and of course it holds that

P (x−1 (B)) = Prob(x ∈ B).

The requirement (1.3) is the formal definition of a random variable: a random

variable is a Borel measurable function.
If x is a random variable on (Ω, F, P ), then we write Px for the probability
measure on (R, B) that is defined by

Px (B) = P (x−1 (B)).

It is possible to define several random variables x1 , x2 , . . . , xn on the same

probability space (Ω, F, P ), and there is no difficulty to let n = ∞. In that
case we call the sequence {xn }∞
n=1 a stochastic process with discrete time, or a
random sequence.

1.2.3.2 The σ -field generated by random variables

When x is a random variable on (Ω, F, P ), the ω -set {ω ∈ Ω; x(ω) ≤ a}
belongs to F and hence, it has a probability Prob(x ≤ a). Furthermore, all
sets of the type x−1 (B), where B is a Borel set, belong to F . In fact, the family
of such Ω-sets is a σ -field, and it is denoted F(x) or σ(x). It is obvious that
F(x) ⊂ F and we already know that P assigns a probability to these sets. If
x were the only random variable of interest to us, we could have worked on the
probability space (Ω, F(x), P ). The reason for using a general, usually larger
σ -field F , is that is allows us perfect freedom to include any further random
variable without changing neither the σ -field, nor the probability measure.
Another characterization of F(x) is that it is the smallest σ -field on Ω that
makes the function x measurable, i.e. a random variable. The σ -field F(x) is
called the σ -field generated by the random variable x.
When there are several random variables x1 , . . . , xn , n ≤ ∞, there will
be a smallest σ -field, denoted F(x1 , . . . , xn ), that contains all the sub-σ -fields
F(xj ). It is the smallest σ -field that makes all the xj random variables.

Remark 1:2 When we have the σ -field generated by a random variable x we

have got our first opportunity to really construct a probability measure, in the
sense that we can define the values of P (A) for certain events A ∈ F . If F
is a distribution function4 on R and x is a Borel measurable function, i.e. a
random variable, then

P (Aa ) = P ({ω ∈ Ω; x(ω) ≤ a}) = F (a)

defines probabilities on the sub-class of events Aa , and that can be extended to

a probability measure on the σ -field F(x).
4
i.e. F is non-decreasing, right-continuous, with 0 ≤ F (x) ≤ 1 , and limx→−∞ F (x) = 0
and limx→∞ F (x) = 1 .
Section 1.3 Stochastic processes 7

1.2.4 Conditional expectation

Here we will give an elementary definition of the important concept conditional
expectation. A more general definition will be introduced in Section 5.6.1, but
here the simple definition is sufficient.
If x, y are two random variables, where y may be multivariate, with joint
density f (x, y), and with marginal y -density f (y) = u f (u, y) du, the condi-
tional expectation of x given y = v , is a random variable defined as a function
of y , for ω such that y(ω) = v , and f (v) = 0, as

f (u, v)
ϕ(v) = E(x | y = v) = u du. (1.4)
u f (v)

For outcomes such that f (y) = 0, ϕ(y) can be defined arbitrarily. We write
E(x | y) = ϕ(y).
It satisfies

E(x) = E(ϕ(y)) = E(E(x | y)) = ϕ(y)f (y) dy, (1.5)
y

V (x) = E(V (x | y)) + V (E(x | y)), (1.6)


where V (x | y) = x (x − ϕ(y))2 f (x | y) dx. The reader should show this, and
the following important theorem:

Theorem 1:1 The best predictor of x given y in least squares sense is given
by ϕ(y), i.e.
   
E (x − ϕ(y))2 ≤ E (x − ψ(y))2
for every function ψ(y).

1.3 Stochastic processes

1.3.1 Stochastic processes and finite-dimensional distributions
We are now ready to define stochastic processes in general. Remember that
we have already defined infinite sequences of random variables, y = {xn }∞ n=1 ,
defined on the same probability space (Ω, F, P ). Here, each xj is a real-valued
function on Ω, i.e. xj (ω) ∈ R.
There is no further difficulty in considering more than countably many ran-
dom variables at the same time, and letting t denote a general parameter taking
values in a parameter space T . Thus we can consider a family of functions,
{x(t, ω) ∈ R}t∈T , where each x(t) = x(t, ·) is a random variable, i.e. a mea-
surable function from Ω to R. Hence it has a distribution with a distribution
function on R, which we denote F (·; t), i.e.

F (a; t) = Prob(x(t) ≤ a).

8 Some probability and process background Chapter 1

Taking several variables, at times t1 , . . . , tn , one gets an n-variate random

variable
(x(t1 ), . . . , x(tn ))
with an n-variate distribution in Rn ,
F (a1 , . . . , an ; t1 , . . . , tn ) = Prob(x(t1 ) ≤ a1 , . . . , x(tn ) ≤ an ).
We write Ftn for the n-dimensional distribution function of any vector
(x(t1 ), . . . , x(tn )).

We summarize the terminology in a formal, but simple, definition.

Definition 1:2 Let T be a parameter set. A stochastic process {x(t)}t∈T in-

dexed by the parameter t ∈ T is a family of random variables x(t) defined on
one and the same probability space (Ω, F, P ). In other worlds, a stochastic
process is a function

T × Ω
(t, ω) → x(t, ω) ∈ R,

such that for fixed t = t0 , x(t0 , ·) is a random variable, i.e. a Borel measurable
function, Ω
ω → x(t0 , ω) ∈ R, and for fixed ω = ω0 , x(·, ω0 ) is a function
T
t → x(t, ω0 ) ∈ R.
The family {Ftn }∞n=1 of finite-dimensional distributions is the family of dis-
tribution functions

F (a1 , . . . , an ; t1 , . . . , tn ) = Prob(x1 ≤ a1 , . . . , xn ≤ an ); n = 1, 2, . . . ; tj ∈ T.
The finite-dimensional distributions in {Ftn }∞
n=1 of a stochastic process sat-
isfy some trivial conditions to make sure they are consistent with each other,
of the type
F (a1 , a2 ; t1 , t2 ) = F (a2 , a1 ; t2 , t1 )
F (a1 , ∞; t1 , t2 ) = F (a1 , t1 ).
By this definition we have the following concepts at our disposal in the three
scenes from Section 1.1:
sample space events probability
abstract space: Ω σ-field F P
continuous functions: C ??? ???
real sequences: R∞ ??? ???
real vectors: Rn Borel sets: Ptn from finite-dimensional dis-
Bn tribution functions Ftn
real line: R Borel sets: B P from a distribution function F
In the table, the ??? indicate what we yet have to define – or even show
existence of – to reach beyond the elementary probability theory, and into the
world of stochastic processes.
Section 1.3 Stochastic processes 9

1.3.2 The distribution of a random sequence

Our aim now is to find the events in R∞ (= all real sequences) and see how
one can define a probability measure for these events. When this is done, it is
legitimate to talk about the distribution of an infinite random sequence. It was
this step, from probabilities for one-dimensional or finite-dimensional real sets
and events, to probabilities in R∞ and probabilistic statements about infinite
sequences, that was made axiomatic in Kolmogorov’s celebrated Grundbegriffe
der Wahrscheinlichkeitsrechnung from 1933, [20].

Generalized rectangles, intervals, and the field I

The basic requirement on the events in R∞ is that they should not be sim-
pler than the events in the finite-dimensional spaces Rn , which means that if
an event Bn ∈ Bn is expressed by means of a finite set of random variables
x1 , . . . , xn , then it should be an event also in the space R∞ . Now, it can be
written

{y = (x1 , x2 , . . .) ∈ R∞ ; (x1 , x2 , . . . , xn ) ∈ Bn } = Bn × R × R × . . . = Bn × R∞ .

A set of this form is called a generalized rectangle in R∞ . Hence, we have

to require that the σ -field of events in R∞ contains at least all generalized
rectangles. The natural event field is exactly the smallest σ -field which contains
all such sets; cf. Example 1:2. This σ -field is denoted B∞ and is called the Borel
field. Symbolically, we can write

B∞ = σ (∪∞ ∞
n=1 (Bn × R )) .

A particularly simple form of rectangles are the intervals, which are sets of
the form
I = (a1 , b1 ] × (a2 , b2 ] × . . . × (an , bn ] × R∞ ,
where each (aj , bj ] is a half-open interval. Thus, the sequence x = (x1 , x2 , . . .)
belongs to the interval I if

a1 < x1 ≤ b1 , a2 < x2 ≤ b2 , . . . , an < xn ≤ bn . (1.7)

Sets which are unions of a finite number of intervals will be important later;
they form a field, which we denote I . The σ -field generated by I is exactly
B∞ , i.e.
σ(I) = B∞ .

Probabilities on R∞
The next step is to assign probabilities to the events in B∞ , and this can
be done in either of two ways, from the abstract side or from the observable,
finite-dimensional side:
10 Some probability and process background Chapter 1

from a random sequence: if the probability space (Ω, F, P ) is given a pri-

ori, and
y = {xn }∞n=1

is a random sequence, i.e. a function from Ω to R∞ , then a probability

measure Py is defined on (R∞ , B∞ ) by

Py (B) = P (y−1 (B)), for B ∈ B∞ .

Thus, each random sequence y produces a probability measure Py on

(R∞ , B∞ ).

from a family of finite-dimensional distributions: if a consistent family

of finite-dimensional distributions

F = {Ftn }∞
n=1

is given a priori, then one can define probabilities PF for all half-open
n-dimensional intervals in R∞ , by, for n = 1, 2, . . . , taking (cf. (1.2))

PF ((a1 , b1 ] × . . . × (an , bn ] × R∞ ) = Pn ((a1 , b1 ] × . . . × (an , bn ]).

Here the probability measure Pn on (Rn , Bn ) is uniquely defined by the

distribution functions in F. Now, it remains to show that this will give
us a countably additive probability measure on the field I of finite unions
of intervals. By the extension property of probability measures on fields,
one can then conclude that there is a unique probability measure P on
(R∞ , B∞ ) that has F as finite-dimensional distributions. The proof of
the countable additivity is a significant part of Kolmogorov’s existence
theorem for stochastic processes; see Appendix A.

By this, we have defined events in R∞ and know how to define proba-

bility measures in (R∞ , B∞ ). What to remember here is in particular, that
every probability measure on (R∞ , B∞ ) is uniquely determined by its finite-
dimensional distributions.

1.3.3 The continuous parameter case

Now we shall investigate stochastic processes with continuous, one-dimensional
time parameter, t in a real interval T . By definition, it is a family of random
variables {x(t)}t∈T , defined on the same probability space (Ω, F, P ), i.e. it is
a function of time t ∈ T and outcome ω ∈ Ω, measurable as a function of ω
for fixed t.
Even if this definition is simple and innocent – obviously such processes exist
– the practical application needs some care. The sample space Ω is an abstract
space and a mathematical construction, and the link to reality is provided by
the random variables. In an experiment, one can observe the values of one or
Section 1.3 Stochastic processes 11

more random variables, x1 , x2 , etc. and also find their distribution, by some
statistical procedure. There is no serious difficulty to allow the outcome to be
any real number, and to define probability distributions on R.
When the result of an experiment is a function with continuous parameter,
the situation is more complicated. In principle, all functions of t ∈ T are
potential outcomes, and the sample space of all functions on T is simply too
big to allow any sensible probabilistic structure. There are too many possible
realizations that ask for probability.
Here practice comes to our assistance. In an experiment one can only ob-
serve the values of x(t) at a finite number of times, t1 , t2 , . . . , tn ; with n = ∞
we allow an unlimited series of observations. The construction of processes with
continuous time is built on exactly this fact: the observable events are those
which can be defined by countably many x(tj ), j ∈ mN , and the probability
measure shall assign probabilities to only such events.
Write RT for the set of all real-valued functions of t ∈ T . By an interval in
R is meant any set of functions x(t) which are characterized by finitely many
T

inequalities of the same type as (1.7),

a1 < x(t1 ) ≤ b1 , a2 < x(t2 ) ≤ b2 , . . . , an < x(tn ) ≤ bn ,
the only difference being that now t1 , . . . , tn are any n time points in T . The
Borel field in RT is the smallest σ -field that contains all intervals,
BT = σ(I).

1.3.3.1 Sets with countable basis

One may wonder how far the Borel sets in RT are from the intervals. The
intervals were characterized by some restriction on function values at a finite
number of times. A set C ⊆ RT which is characterized by function values at a
countable set of times, T  = (t1 , t2 , . . .) is said to have a countable basis. More
precisely, C ⊆ RT has a countable basis T  if there is a Borel set B ⊂ R∞ ,
(with B ∈ B∞ ), such that
x∈C if and only if (x(t1 ), x(t2 ), . . .) ∈ B.
The Borel sets in RT are exactly those sets which have a countable basis, i.e.
BT = {C ⊂ RT ; C has a countable basis}.
We show this, as an example of a typical σ -field argument.
First, it is clear that if B is a Borel set in R∞ , then
C = {x ∈ RT ; (x(t1 ), x(t2 ), . . .) ∈ B}
is a Borel set in RT , since BT contains all intervals with base in T  , and hence
all sets in the σ -field generated by those intervals. This shows that
{C ⊂ RT ; C has a countable basis} ⊆ BT .
12 Some probability and process background Chapter 1

To show the other inclusion we show that the family of sets with countable
basis is a σ -field which contains the intervals, and then it must be at least as
large as the smallest σ -field that contains all interval, namely BT . First, we
note that taking complements still gives a set with countable basis. Then, take
a sequence C1 , C2 , . . ., of sets, all with countable basis, and let T1 , T2 , . . . , Tj =
{t11 , t12 , . . .}, . . . , {tj1 , tj2 , . . .} be the corresponding countable sets of time points,
so that
(j) (j)
Cj = {x ∈ RT ; (x(t1 ), x(t2 ), . . .) ∈ Bj }, with Bj ∈ B∞ .
Then T  = ∪j Tj is a countable set, T  = (t1 , t2 , . . .), and ∪∞
j=1 Cj is character-
ized by its values on T  .

Example 1:3 Here are some examples of function sets with and without count-
able basis, when T = [0, 1]:
• {x ∈ RT ; limn→∞ x(1/n) exists} ∈ BT ,

• {x ∈ RT ; limt→0 x(t) exists} ∈

/ BT ,

• {x ∈ RT ; x is a continuous function} ∈
/ BT ,

• {x ∈ RT ; x(t) ≤ 2 for all rational t} ∈ BT ,

1.3.3.2 Approximation by finite-dimensional events

The events in the σ -field B∞ in R∞ can be approximated in probability by
finite-dimensional sets. If (R∞ , B∞ , P ) is a probability space, and B ∈ B∞ ,
then for every  > 0, there is a finite n and an event Bn ∈ Bn such that
n ) ≤ ,
P (BΔB
where B n = {x ∈ R∞ ; (x1 , . . . , xn ) ∈ Bn } and AΔB = (A − B) ∪ (B − A).
Similarly, events in BT in RT can be approximated arbitrarily close by
events defined by the values of x(t) for a finite number of t-values: P (BΔB n ) ≤
, with
n = {x ∈ R∞ ; (x(t1 ), . . . , x(tn )) ∈ Bn }.
B
Remember that every probability measure on (R∞ , B∞ ) is uniquely de-
termined by its finite-dimensional distributions, which implies that also every
probability measure P on (RT , BT ) is determined by the finite-dimensional
distributions, {Ftn }∞
n=1 . In particular, the probability

P ( lim x(t0 + 1/n) exists and is equal to x(t0 ))

n→∞

is determined by the finite-dimensional distributions. Unfortunately, x(t0 +

1/n) → x(t0 ) as n → ∞, is almost, but not quite, the same as x(t) → x(t0 )
as t → t0 . To deal with sample function continuity we need more refined con-
struction of the probability measure from the finite-dimensional distributions.
Section 1.4 Stationary processes and fields 13

1.4 Stationary processes and fields

This section summarizes some elementary notation for stationary processes.
More details, properties, and proofs of the most important facts, will be given
in Chapter 4.
To avoid too cumbersome notation we from now on allow ourselves to talk
about “the process x(t)”, when we should have used the notation {x(t), t ∈ R}
for the process. If we mean “the random variable x(t) we will say so explicitly.

1.4.1 Stationary processes

A stochastic process x(t) is strictly stationary if all n-dimensional distributions
of
x(t1 + τ ), . . . , x(tn + τ )
are independent of τ . It is called weakly stationary (the term second order sta-
tionary is also used) if its mean is constant, E(x(t)) = m, and if its covariance
function
r(t) = Cov(x(s + t), x(s)),
is a function only of the time lag t. Every continuous covariance function has
a representation as a Fourier integral,
 ∞
r(t) = eiωt dF (ω), (1.8)
−∞

where the function F (ω) is called the spectral distribution function. It is char-
acterized by the properties:

• symmetry: dF (−ω) = dF (ω),

• monotonicity: ω ≤ ω  implies F (ω) ≤ F (ω  ),
• boundedness: F (+∞) − F (−∞) = r(0) < ∞.

As indicated by the way we write the three properties, F (ω) is defined only up
to an additive constant, and we usually take F (−∞) = 0. The spectral distri-
bution function is then equal to a cumulative distribution function multiplied
by a positive constant, equal to the variance of the process.
ω
If F (ω) is absolutely continuous with F (ω) = s=−∞ f (s) ds, then the
spectrum is said to be (absolutely) continuous, and f (ω) is the spectral density
function; see Section 5.6.2.1 for more discussion of absolute continuity.
The spectral moments are defined as
 ∞
ωk = |ω|k dF (ω).
−∞

Note that the odd spectral moments are defined as absolute moments. Since
F is symmetric around 0 the signed odd moments are always 0. Spectral
14 Some probability and process background Chapter 1

Simulation Spectrum Covariance

20 5
Narrow 5

0 10 0

-5
0 -5
0 50 100 150 200 250 0 1 2 0 50 100
20 5
5
Moderate

0 10 0

-5
0 -5
0 50 100 150 200 250 0 1 2 0 50 100
4 5
5
Low

0 2 0

-5
0 -5
0 50 100 150 200 250 0 1 2 0 50 100

Figure 1.2: Processes with narrow band spectrum, moderate width JONSWAP
wave spectrum, and low frequency white noise spectrum.

moments may be finite or infinite. As we shall see in the next chapter, the
finiteness of the spectral moments are coupled to the smoothness properties of
the process x(t). For example, the process is differentiable (in quadratic mean),
see Section 2.1, if ω2 = −r  (0) < ∞, and similarly for higher order derivatives.
As we shall see in later sections, ω is in a natural way interpreted as an
angular frequency, not to be confused with the elementary event ω in basic
probability theory.

Example 1:4 Here is a first example on the visual characteristics of spectrum,

covariance function, and sample function. Figure 1.2 illustrates one very nar-
row spectrum, one realistic water wave spectrum, and one “low frequency white
noise” spectrum. Figure 1.3 show the output of a linear oscillator driven by
white noise, and with different relative damping; see Section 4.4.3.3 and Exam-
ple 4:10.

1.4.2 Random fields

A random field is a stochastic process x(t) with multi-dimensional parameter
t = (t1 , . . . , tp ) ∈ T, which can be discrete or continuous. For example, if
Section 1.5 Gaussian processes 15

Harmonic random oscillator Spectrum Covariance

5
5 150

ζ = 0.01
100
0 0
50
-5
0 -5
0 50 100 150 200 250 0 1 2 0 50 100
5
5
15
ζ = 0.1

0 10 0
5
-5
0 -5
0 50 100 150 200 250 0 1 2 0 50 100
6 5
5
ζ = 0.5

4
0 0
2
-5
0 -5
0 50 100 150 200 250 0 1 2 0 50 100

Figure 1.3: Harmonic oscillator with different relative damping ζ .

t = (t1 , t2 ) is two-dimensional we can think of (t1 , t2 , x(t)), (t1 , t2 ) ∈ R2 , as a

random surface. The mean value and covariance functions are defined in the
natural way, m(t) = E(x(t)) and r(t, u) = C(x(t), x(u)).
A random field is called homogeneous if it has constant mean value m(t) =
m and the covariance function r(t, u) depends only on the vector t−u between
the two observation points, i.e. assuming m = 0,
r(t) = r(u + t, u) = E(x(u + t) · x(u)).
The covariance of the process values at two parameter points depends on dis-
tance as well as on direction of the vector between the two points.
If the covariance between x(u) and x(v) depends only on the distance
τ = u − v between the observation points and not on the direction, the field
is called isotropic. This requirement poses severe restrictions on the covariance
function, as we shall see in Chapter 6, where random fields are treated in more
detail.

1.5 Gaussian processes

1.5.1 Multivariate normal distributions and Gaussian processes
Definition 1:3 A vector ξ = (ξ1 , . . . , ξp ) of p random variables is said to
have a p-variate Gaussian (normal) distribution if every linear combination
16 Some probability and process background Chapter 1

of its components a · ξ = k ak ξk has a normal distribution. The variables
ξ1 , . . . , ξp are then said to be “jointly normal”.

With mean vector m = E(ξ) and covariance matrix

Σ = Cov(ξ; ξ) = E((ξ − m) · (ξ − m) ),

the variance of a · ξ is
V (a · ξ) = a Σa.
If the determinant of Σ is positive, the distribution of ξ is non-singular
and has a density

1 1  −1
fξ(x) = √ e− 2 (x−m) Σ (x−m)
.
(2π)p/2 Σ

If the determinant is zero, the distribution of ξ is concentrated to linear sub-

space of Rn and there exists at least one linear relationship between the com-
ponents, i.e. there is at least one a for which a · ξ is a constant.

Definition 1:4 A stochastic process {x(t), t ∈ R} is a Gaussian process if

every linear combination 
S= ak x(tk )
k

for real ak and tk ∈ R has a Gaussian distribution.

It is an easy consequence of the definition that the derivative of a Gaussian

process is also Gaussian (when it exists), since it is the limit of the Gaussian
variable zh = (x(t + h) − x(t))/h as h → 0. For a stationary Gaussian process
{x(t), t ∈ R} the mean of zh is 0 and it has variance V (zh ) = 2(r(0)−r(h))/h2 .
As we shall prove in Section 2.4.2 this converges to ω2 = ω 2 dF (ω ≤ ∞. The
derivative exists only if this limit is finite.
Also the integral of a Gaussian process is a Gaussian variable; conditions
for the existence will be given in Section 2.6.

1.5.1.1 Conditional normal distributions

The multivariate normal distribution has the very useful property that condi-
tioned on observations of a subset of variables, the unobserved variables are
also normal. Further, the conditional mean is a linear in the observations while
variances and covariances are independent of observations.
Let ξ = (ξ1 , . . . , ξn ) and η = (η1 , . . . , ηm ) be two jointly Gaussian vectors
with mean values
E(ξ) = mξ, E(η) = mη ,
and with covariance matrix (with Σξη = Σηξ )
Section 1.5 Gaussian processes 17

Σξξ Σξη
Σ = Cov ((ξ, η); (ξ, η)) = .
Σηξ Σηη

If the determinant of the covariance matrix Σ is positive, then the distri-

bution of (ξ, η) has a non-singular density

1 1 −1
(x−mξ ,y−mη )
fξη (x, y) = √ e− 2 (x−mξ ,y−mη )Σ .
(2π)(m+n)/2 det Σ

The density of η is

1 1 −1 
fη (y) =  e− 2 (y−mη )Σηη (y−mη ) .
(2π)m/2 det Σηη

and the conditional density fξ|η (x|y), defined as

fηξ (y, x)
fξ|η (x | y) = ,
fη (y)

is also Gaussian with conditional mean matrix

E(ξ | η = y) = 
ξ(y) = E(ξ) + Cov(ξ, η) Σ−1
ηη (y − E(η))


= mξ + Σξη Σ−1 
ηη (y − mη ) . (1.9)

(1.10)

The conditional covariance is

Σξξ|η = E((ξ − 
ξ(η)) · (ξ − 
ξ(η)) ) = Σξξ − Σξη Σ−1
ηη Σηξ . (1.11)

In two dimensions the formulas read

y − my
mx|y = mx + σx σy ρxy · ,
σy
2
σx|y = σx2 (1 − ρ2xy ),

with ρxy = Cov(x, y)/ V (x)V (y); thus the squared correlation ρ2xy gives the
relative reduction of the variability (uncertainty) in the random variable x
gained by observation of y .
Observe the mnemotechnical friendliness of these formulas. For example,
the covariance matrix Σξξ|η has dimension n × n and the configuration on
the right hand side of (1.11) is the only way to combine the matrices involved
that matches their dimensions – of course, you have to remember the general
structure.
18 Some probability and process background Chapter 1

1.5.2 Linear prediction and reconstruction

Prediction and reconstruction are two of the most important applications of
stationary process theory. Even though these problems are not main topics in
this work, we present one of the basic concepts here; in Section 5.6 we will deal
with the more philosophical sides of the prediction problem.
Suppose we have observed the outcomes of a set of random variables, η =
(η1 , . . . , ηm ) and that we want to give a statement ξ about the outcome of some
other variable ξ , either to be observed sometimes in the future, or perhaps a
missing observation in a time series. These two cases constitute the framework
of prediction and reconstruction, respectively. Also suppose that we want to
make the statement in the best possible way in the mean square sense, i.e. we
want E((ξ − ξ)  2 ) to be as small as possible.
Now we know from Theorem 1:1 that the best solution in mean square sense
is given by the conditional expectation, ξ = E(ξ | η) = φ(η). On the other
hand, if the variables are jointly Gaussian, then we know from Section 1.5,
formula 1.9 that the conditional expectation of ξ given η is linear in η , so that
for Gaussian variables the optimal solution is
ξ = E(ξ | η) = mξ + Σξη Σ−1
ηη (η − mη ), (1.12)
and expression that depends only on the the mean values and the second order
moments, i.e. variances and covariances.
We now look at the general case, without assuming normality, and restrict
ourselves to solutions that are linear functions of the observed variables. It is
clear that the solution that is optimal in the mean square sense only depends
on the mean values and variances/covariances of the variables. It therefore has
the same form for all variables with the same first and second order moments.
Thus, (1.12) gives the best linear predictor in mean square sense.

1.5.3 Some useful inequalities

We shall in the next chapter need the following inequality xfor the normal density
1 −x 2 /2
√
and distribution functions, φ(x) = 2π e , Φ(x) = −∞ φ(y) dy :

1 1 1
φ(x) − 3 ≤ 1 − Φ(x) ≤ φ(x) , (1.13)
x x x
for x > 0.
The following asymptotic expansion is useful as x → ∞,

1 1 1·3 1·3·5 k 1 · 3 · · · (2k − 1)

1 − Φ(x) ∼ φ(x) − + 5 − + − . . . (−1) .
x x3 x x7 x2k+1
Here the right hand side overestimates 1 − Φ(x) for x > 0 if k is even and
underestimates it if k is odd. More precisely, the difference between the left
and right hand side is of the same order as φ(x)/x2k+3 as x → ∞. (The sign
∼ means that the ratio between the left and the right hand side goes to 1.)
Section 1.6 Some historical landmarks 19

1.6 Some historical landmarks

This section contains a personal selection of research achievements that have
shaped the theory of stationary processes in general. The examples are cho-
sen, not only because they are important in their respective fields, but also
because they illustrate the necessity of exchange of ideas between probability
and statistics theory and applications.

1.6.1 Brownian motion and the Wiener process

The is no other Gaussian process with as wide applicability as the Wiener pro-
cess. Even if it a non-stationary process it appears repeatedly in the theory of
stationary processes, and we spend this section to describe some of its properties
and applications.

Definition 1:5 The Wiener process {w(t); t ≥ 0} is a Gaussian process with

w(0) = 0 such that E(w(t)) = 0, and the variance of the increment w(t + h) −
w(t) over any interval [t, t + h], h > 0, is proportional to the interval length,

V (w(t + h) − w(t)) = h σ 2 .

A Wiener process {w(t), t ∈ R} over the whole real line is a combination of

two independent Wiener processes w1 and w2 , so that w(t) = w1 (t) for t ≥ 0
and w(t) = w2 (−t) for t < 0.
It is an easy consequence of the definition that the increment w(t) − w(s) is
uncorrelated with w(s) for s < t,

V (w(t)) = V (w(s)) + V (w(t) − w(s)) + 2Cov(w(s), w(t) − w(s)),

and that therefore Cov(w(s), w(t)) = σ 2 min(s, t). Since the increments over
disjoint intervals are normal, by the definition of a normal process, they are
also independent.
A characteristic feature of the Wiener process is that its future changes are
statistically independent of its actual and previous values. It is intuitively clear
that a process with this property cannot be differentiable.
√ The increment over a
small time interval from t to t + h is of the order h, which is small enough to
make the process continuous, but it is too large to give a differentiable process.5
The sample functions are in fact objects that have fractal dimension, and the
process is self similar in the sense that when magnified with proper scales it
retains its statistical geometrical properties. More precisely, for each a > 0,
√
the process a w(t/a) has the same distributions as the original process w(t).
The Wiener process is commonly used to model phenomena where the local
changes are virtually independent. Symbolically, one use to write dw(t) for the
5
Chapter 2 gives conditions for continuity and differentiability of sample functions.
20 Some probability and process background Chapter 1

infinitesimal independent increments, or simply as a “derivative” w (t). The

Brownian motion is a good example of how one can use the Wiener process to
get models with more or less physical realism.
The Brownian motion, first described 1828 by the Scottish biologist Robert
Brown, is an erratic movement by small particles immersed in a fluid, for ex-
ample pollen particles in water as in Brown’s original experiment. Albert Ein-
stein’s presented 1905 a quantitative model for the Brownian movement in his
paper On the movements of small particles in a stationary liquid demanded by
the molecular-kinetic theory of heat, reprinted in [13], based on the assumption
that the movements are caused by independent impacts on the particle by the
molecules of the surrounding fluid medium.
In Einstein’s model the changes in location due to collisions over separate
time intervals are supposed to be independent. This requires however that the
particles have no mass, which physically wrong, but the model is still sufficiently
accurate for microscopic purposes. According to Einstein, the change in loca-
tion in any of the three directions, (x, y, z) over a time interval of length t is
random and normal with mean zero, which is not surprising, since they are the
results of a very large number of independent collisions. What made Einstein’s
contribution conclusive was that he derived an expression for the variance in
terms of other physical parameters, namely

4RT
V (x(t)) = V (y(t)) = V (z(t)) = t = t σ2, (1.14)
Nf

where T is the absolute temperature, R is the Boltzmann constant, and N

is Avogadro’s number, i.e. the number of molecules per mole of an ideal gas,
and the friction coefficient f depends on the shape and size of the particle and
on the viscosity of the fluid. Each coordinate are here independent Wiener
processes.
Observations of the Brownian movement and estimation of its variance
makes it possible to calculate any of the factors in σ 2 , for example N , from
the other ones. The French physicist J.B. Perrin estimated in a series of ex-
periments 1908-1911 Avogadro’s number in this way by observing suspended
rubber particles and found an estimate correct within about 10%.
In a more realistic model, one takes also particle mass and velocity into
account. If v(t) denotes the velocity at time t, the fluid offers a resistance
from the friction force, which is equal to f v(t), with the friction coefficient as
in (1.14). Further, the particle offers a resistance to changes in velocity propor-
tional to its mass m. Finally, one needs to model the independent collisions by
the fluid molecules, and here the Wiener process can be used, more precisely its
increments dw(t). This gives the Langevin equation for the particle velocity,

1
dv(t) + α v(t) dt = dw(t), (1.15)
m
Section 1.6 Some historical landmarks 21

where α = f /m, and w(t) is a standardized Wiener process. It is usually

written
dv(t) 1 
+ α v(t) = w (t). (1.16)
dt m
We will meet the Langevin equation in Example 4:3 on the Ornstein-Uhlenbeck
process in Section 4.3.3.6.

1.6.2 Rice and electronic noise

The two papers Mathematical analysis of random noise, by S.O. Rice, appeared
in Bell System Technical Journal, 1944-1945, [27]. They represent a landmark
in the history of stochastic processes in that they bring together and exhibit
the wide applicability of the spectral formulation of a stationary process as a
sum, or asymptotically an integral, of harmonic cosine functions with random
amplitudes and phases. Correlation functions and their Fourier transforms had
been studied at least since the early 1900s, and Rice’s work brought together
these results in a systematic way. But it also contained many new results, in
particular pertaining crossing related properties, on the statistical properties
of stationary processes “obtained by passing random noise through physical
devices”.
Rice uses the spectral representation of a stationary Gaussian process as a
sum over discrete positive frequencies ωn > 0,
 
x(t) = an cos ωn t + bn sin ωn t = cn cos(ωn t + φn ) (1.17)
n n
where the amplitudes an , bn and normal and independent with mean 0 and
E(a2n ) = E(b2n ) = σn2 , and φn uniformly distributed over (0, 2π), independent
of the amplitudes. As we shall see in Chapter2 such a process has covariance

r(t) = σn2 cos ωn t.
n
The spectral distributions function is a discrete distribution with point mass
σn2 /2 at the symmetricly located frequencies ±ωn .
The absolutely continuous, integral form of the spectral representation is
presented as limiting cases in Rice’s work. At about the same time, Cramér
gave a probabilistic formulation of the continuous spectral representation, in a
mathematically impeccable way; [6, 7].
Besides the previously known “Rice’s formula” for the expected number of
level crossings, Rice’s 1945 paper also analyzed crossing and excursion distri-
butions and investigated the joint occurrence of crossings of a fixed level at
two distinct points, necessary for calculation of the variance of the number of
crossings.
The flexibility and generality of Rice’s methods and examples, made cor-
relation and spectral theory fundamental ingredients in communication theory
and signal processing for decades to come. An example by Rice himself, is the
ingenious explanation of the intriguing click noise in analogue FM-radio, [28].
22 Some probability and process background Chapter 1

1.6.3 Gaussian random wave models

Steve Rice’s analysis of time dependent stationary processes had, as mentioned,
great influence on signal processing in the information sciences. Less well known
in the statistical world is the effect his work had in oceanography and naval
architecture.
It is well worth to cite in extenso (references deleted) the first two paragraphs
in Manley St. Denis and Willard J. Pierson’s paper: On the motion of ships
in confused seas, which came out 1954, [32].

History
Three years ago to first co-author of the present work collab-
orated with Weinblum in the writing of a paper entitled “On the
motion of ships at sea”. In that paper Lord Rayleigh was quoted
saying: “The basic law of the seaway is the apparent lack of any
law”. Having made this quotation, however, the authors then pro-
ceed to consider the seaway as being composed of “a regular train
of waves defined by simple equations”. This artificial substitution
of pattern for chaos was dictated by the necessity of reducing the
utterly confused reality to a simple form amenable to mathematical
treatment.
Yet at the same time and in other fields the challenging study
of confusion was being actively pursued. Thus in 1945 Rice was
writing on the mathematical analysis of random noise and in 1949
Tukey and Hamming were writing on the properties of stationary
time series and their power spectra in connection with colored noise.
In the same year Wiener published his now famous book on time
series. These works were written as contributions to the theory of
communication. Nevertheless the fundamental mathematical disci-
pline expounded therein can readily be extended to other fields of
scientific endeavor. Thus in 1952 the second co-author, inspired by
a contribution of Tukey, was able to apply the foregoing theories to
the study of actual ocean waves. As the result of analyses of actual
wave records, he succeeded in giving not only a logical explanation
as to why waves are irregular, but a statement as well of the laws
underlying the behavior of a seaway. There is indeed a basic law of
the seaway. Contrary to the obvious inference from the quotation
of Lord Rayleigh, the seaway can be described mathematically and
precisely, albeit in a statistical way.

If Rice’s work had been in the vein of generally accepted ideas in communi-
cation theory, the St Denis and Pierson paper represented a complete revolution
in common naval practice. Nevertheless, its treatment of irregular water waves
as, what now is called, a random field was almost immediately accepted, and
set a standard for much of naval architecture.
Section 1.6 Some historical landmarks 23

One possible reason for this can be that the authors succeeded to formulate
and analyze the motions of a ship that moved with constant speed through the
field in a rational way. The random sea could directly be used as input to a
linear (later also non-linear) filter representing the ship.
St. Denis and Pierson extended the one-dimensional description of a time
dependent process {x(t), t ∈ R}, useful for example to model the waves mea-
sured at a single point, to a random field x(t, (s1 , s2 )) with time and location
parameter (s1 , s2 ). They generalized the sum (1.17) to be a sum of a packet of
directed waves, with ω = (ω, κ1 , κ2 ),

Aω cos(ωt − κ1 s1 − κ2 s2 + φω ). (1.18)
ω

with random amplitude and phase.

For fixed t each element in (1.18) is a cosine-function in the plane, which is
zero along lines ωt − κ1 s1 − κ2 s2 + φω = π/2 + kπ , k integer. The parameters
κ1 and κ2 are called the wave numbers. For fixed (s1 , s2 ) it is a cosine wave
with (angular) frequency ω .
Water waves are special cases of homogeneous random fields, for which there
is a special relation between time and space frequencies (wave numbers). For
a one-dimensional time dependent Gaussian wave x(t, s), where s is distance
along an axis, the elementary waves have the form

Aω cos(ωt − κs + φω ).

By physical considerations one can derive an explicit relation, called the dis-
persion relation, between wave number κ and frequency ω . If h is the water
depth, then
ω 2 = κg tanh(hκ),
which for infinite depth reduces to ω 2 = κg . Here g is the constant of gravity.
The case of a two-dimensional time dependent Gaussian wave x(t, s1 , s2 ),
the elementary waves with frequency ω and direction θ becomes

Aω cos(ωt − κ(s1 cos θ + s2 sin θ) + φω ),

where κ is given by the dispersion relation.

The spectral distribution is often written in polar form, with spectral density

f (ω, θ) = f (ω)g(ω, θ),

 2π
where the spreading function g(ω, θ) has 0 g(ω, θ) dω = 1.
In their paper St. Denis and Pierson also laid out the theory for how the
wave spectrum should be transformed to a response spectrum for the motion of
a ship, and they also described how the spectrum is changed to an encountered
spectrum when a ship sails with constant speed through the waves.
24 Some probability and process background Chapter 1

1.6.4 Detection theory and statistical inference

The first three landmarks illustrated the relation between stochastic model
building and physical knowledge, in particular how the concepts of statistical
independence and dependence between signal and functions relate to the phys-
ical world. About the same time as Rice and StDenis & Pierson advanced
physically based stochastic modeling, the statistical inference methodology was
placed firmly into a theoretical mathematical framework, as documented by the
classical book by Harald Cramér, Mathematical methods of Statistics, 1945, [8].
A few years later, the connection between the theoretical basis bor statis-
tical inference and important engineering questions related to signal detection
was elegantly exploited by Ulf Grenander in his PhD thesis from Stockholm,
Stochastic processes and statistical inference, [15]. The classical problem in sig-
nal processing of deciding whether a deterministic signal of known shape s(t) is
present in an environment of Gaussian dependent, colored as opposed to white,
random noise, x(t) can be treated as an infinite dimensional decision problem,
testing an infinite dimensional statistical hypothesis; see also the classical book
on detection theory [33]
Suppose one observes a Gaussian stochastic process x(t), a ≤ t ≤ b, with
known correlation structure, but with unknown mean value function m(t). If
no signal is present, the mean value is 0, but with signal, the mean is equal to
the known function s(t). In statistical terms one has to test the following two
hypotheses against each other:

H0 : m(t) = 0,
H1 : m(t) = s(t).

 Grenander introduced a series of independent Gaussian observables, yk =

hk (t)x(t) dt, by chosing the filter functions hk as solutions to the integral
equation 
r(s, t)hk (t) dt = ck hk (s),

with c1 ≥ c2 ≥ . . ., ck → 0 as k → ∞. Under H1 the observables will have

mean ak = hk (t)m(t) dt and variance ck , while under H0 they will have mean
0, and the same variance. So instead of a continuous problem, we have gotten
a denumerable problem, in which one can make a Likelihood-Ratio test of the
two alternatives. We will return to this problem in Section 4.5, Example 4:12.
Section 1.6 Some historical landmarks 25

Exercises
1:1. Consider the sample space Ω = [0, 1] with uniform probability P , i.e.
P ([a, b]) = b − a, 0 ≤ a ≤ b ≤ 1. Construct a stochastic process y =
(x1 , x2 , . . .) on Ω such that the components are independent zero-one
variables, with P (xk = 0) = P (xk = 1). What is the distribution of
 ∞ k
k=1 xk /2 ?

1:2. Show that B(R) = the class of Borel sets in R is generated by

a) the open intervals,
b) the closed intervals.
1:3. A set A ⊂ {1, 2, . . .} is said to have asymptotic density θ if

lim n−1 |A ∩ {1, 2, . . . , n}| = θ.

n→∞

(Note, |B| denotes the number of elements in B .) Let A be the family

of sets for which the asymptotic density exists. Is A a field? A σ -field?
1:4. Let x1 , x2 , . . . be random variables with values in a countable set F , and
suppose there are real constants ak such that
∞
 ∞

P (xk = ak ) < ∞, ak < ∞.
k=1 k=1

Prove that the sum x = ∞ k=1 xk has a discrete distribution, i.e. there
exists a countable set D such thatP (x ∈ D) = 1. (Hint: Use the Borel-
Cantelli lemma, that says that if k P (Ak ) < ∞ then, with probability
one, only a finite number of the events Ak occur.
Show by example ∞ that it is possible for independent random variables xk
to have a sum k=1 xk with a continuous distribution, although all xk
are discrete variables with a common value space – obviously they can
not be identically distributed.
1:5. Take Rn and motivate that the family F0 whose elements are unions of
finitely many rectangles (ai , bj ] (with possibly infinite end points) is a
field.
Let T be an interval and convince yourself that the finite dimensional
rectangles in RT and unions of finitely many such rectangles, is a field.
1:6. Take T = [0, 1], and consider the set of functions which are continuous
on the rational numbers, i.e.

CQ = {x ∈ RT ; x(q) → x(q0 ) for all rational numbers q0 },

where the limit is taken as q tends to q0 through the rational numbers.

Show that CQ ∈ BT .
26 Some probability and process background Chapter 1

1:7. Prove Theorem 1:1.

1:8. Prove that the increments of a Wiener process, as defined as in Defini-

tion 1:5, are independent and normal.
Chapter 2

Stochastic analysis

This chapter is the stochastic equivalent of real analysis and integration. As in

its deterministic counterpart limiting concepts and conditions for the existence
of limits are fundamental. We repeat the basic stochastic limit definitions; a
summary of basic concepts and results on stochastic analysis is given in Ap-
pendix B.

Definition 2:1 Let {xn }∞ n=1 be a random sequence, with the random variables
x1 (ω), x2 (ω), . . . defined on the same probability space as a random variable
x = x(ω). Then, the convergence xn → x as n → ∞ can be defined in three
ways:
a.s.
• almost surely, with probability one (xn → x): P ({ω; xn → x}) = 1;

• in quadratic mean (xn → x): E |xn − x|2 → 0;
q.m.

P
• in probability (xn → x): for every  > 0, P (|xn − x| > ) → 0.
In Appendix B we give several conditions, necessary and sufficient, as well as
only sufficient, for convergence of a random sequence xn . The most useful of
these involve only conditions on the bivariate distributions of xm and xn . We
shall in this chapter examine such conditions for sample function continuity,
differentiability, and integrability. We shall also give conditions which guar-
antee that only simple discontinuities occur. In particular, we shall formulate
conditions in terms of bivariate distributions, which are easily checked for most
standard processes, such as the normal and the Poisson process.

2.1 Quadratic mean properties

We first repair some concepts and properties that may be well known from
previous courses in stochastic processes. We return to proofs and more details
in Section 2.4.

27
28 Stochastic analysis Chapter 2

A stochastic process {x(t), t ∈ R} is said to be continuous in quadratic

mean (or L2 -continuous) at time t if
q.m.
x(t + h) → x(t)

as h → 0, i.e. if E((x(t+h)−x(t))2 ) → 0. It is called differentiable in quadratic

mean with derivative y(t) if

x(t + h) − x(t) q.m.

→ y(t)
h
as h → 0. Of course, the process {y(t), t ∈ R} is called the (quadratic mean)
derivative of {x(t), t ∈ R} and is denoted x (t). A stationary process x(t) is
continuous in quadratic mean if its covariance function r(t) is continuous at
t = 0. It is differentiable if r(t) is twice differentiable and then the derivative
has covariance function
rx (t) = −r  (t).
Second and higher order derivatives are defined recursively: {x(t), t ∈ R}
is twice differentiable in quadratic mean if and only if its (quadratic mean)
derivative is (quadratic mean) differentiable, i.e. if r iv (t) exists. etc. The
covariance function of {x (t), t ∈ R} is rx (t) = r iv (t).
Expressed in terms of the spectral distribution function F (ω), the process
is differentiable in quadratic mean if and only if the second spectral moment is
finite, i.e.  ∞
ω2 = ω 2 dF (ω) < ∞;
−∞

for a proof, see Lemma 2.3, page 47. Since ω2 = −r  (0) = V (x (t)), the
finiteness of ω2 is necessary and sufficient for the existence of a quadratic
mean derivative. Analogous relations
 hold for higher derivatives of order k
and the spectral moments ω2k = ω 2k dF (ω). We will give some more details
on quadratic mean properties in Section 2.4.

2.2 Sample function continuity

2.2.1 Countable and uncountable events
The first problem that we encounter with sample function continuity is that
the sample event of interest, namely the set of continuous functions,

C = {x ∈ RT ; x(·) is a continuous function},

does not have a countable basis, and is not a Borel set, i.e. C ∈ BT . If {x(t); t ∈
T } is a stochastic process on a probability space (Ω, F, P ), then the probability
P (C) need not be defined – it depends on the structure of (Ω, F) and on how
complicated x is in itself, as a function on Ω. In particular, even if P (C) is
Section 2.2 Sample function continuity 29

defined, it is not uniquely determined by the finite-dimensional distributions.

To see this, take a process on a sufficiently rich sample space, i.e. one that
contains enough sample points ω , and suppose we have defined a stochastic
process {x(t); t ∈ R}, which has, with probability one, continuous sample paths.
Then x has a certain family of finite-dimensional distributions. Now, take
a random time, τ , independent of x, and with continuous distribution, for
example an exponential distribution.1 Then define a new process {y(t); t ∈ R},
such that

y(t) = x(t), for t = τ ,

y(τ ) = x(τ ) + 1.

Then y has the same finite-dimensional distributions as x but its sample func-
tions are always discontinuous at τ .

2.2.1.1 Equivalence
In the constructed example, the two processes x and y differ only at a single
point τ , and as we constructed τ to be random with continuous distribution,
we have
P (x(t) = y(t)) = 1, for all t. (2.1)
Two processes x and y which satisfy (2.1) are called equivalent. The sample
paths of two equivalent process always coincide, with probability one, when
looked at a fixed, pre-determined time point. (In the example above, the time
τ where they differed was random.)

2.2.1.2 Separability
The annoying fact that a stochastic process can fail to fulfill some natural
regularity condition, such as continuity, even if it by all natural standards should
be regular, can be partly neutralized by the concept of separability, introduced
by Doob. It uses the approximation by sets with countable basis mentioned
in Section 1.3.3. Loosely speaking, a process {x(t), t ∈ R} is separable in an
interval I if there exists a countable set of t-values T = {tk } ⊂ I such that
the process, with probability one, does not behave more irregularly on I than
it does already on T . An important consequence is that for all t in the interior
of I , there are sequences τ1 < τ2 < . . . τn ↑ t and τ1 > τ2 > . . . τn ↓ t such
that, with probability one,

lim inf n→∞ x(τn ) = lim inf τ ↑t x(τ ) ≤ lim supτ ↑t x(τ ) = lim supn→∞ x(τn ),

with a similar set of relations for the sequence τn . Hence, if the process is
continuous on any discrete set of points then it is continuous. Every process
has an equivalent separable version; see [11].
1
This is where it is necessary that Ω is rich enough so we can define an independent τ .
30 Stochastic analysis Chapter 2

2.2.2 Conditions for sample function continuity

The finite-dimensional distributions of any two equivalent processes x(t) and
y(t) are always the same – show that as an exercise. We shall now see under
what conditions, on the finite-dimensional distribution functions, we can assume
a stochastic process to have continuous sample paths. Conditions will be given
both in terms of the bivariate distributions directly, and in terms of probabilistic
bounds on the process increments. As we have seen, one has to be satisfied if,
among all equivalent processes, one can find one which has continuous sample
paths.

Theorem 2:1 Let {x(t); 0 ≤ t ≤ 1} be a given stochastic process. If there exist

two non-decreasing functions, g(h) and q(h), 0 ≤ h ≤ 1, such that

∞
 ∞

g(2−n ) < ∞ 2n q(2−n ) < ∞,
n=1 n=1

and, for all t < t + h in [0, 1],

P (|x(t + h) − x(t)| ≥ g(h)) ≤ q(h), (2.2)

then there exists an equivalent stochastic process y(t) whose sample paths are,
with probability one, continuous on [0, 1].

Proof: Start with the process x(t) with given finite-dimensional distributions.
Such a process exists, and what is questioned is whether it has continuous sam-
ple functions if its bivariate distributions satisfy the conditions in the theorem.
We shall now explicitly construct a process y(t), equivalent to x(t), and with
continuous sample paths. Then y(t) will automatically have the same finite-
dimensional distributions as x(t). The process y(t) shall be constructed as the
limit of a sequence of piecewise linear functions xn (t), which have the correct
distribution at the dyadic time points of order n,

tn(k) = k/2n , k = 0, 1, . . . , 2n ; n = 1, 2, . . . .

Define the process xn equal to x at the dyadic points,

(k)
xn (t) = x(t), for t = tn , k = 0, 1, . . . , 2n ,

and let it be linear between these points; see Figure 2.1.

Then we can estimate the maximal distance between two successive approx-
imations. As is obvious from the figure, the maximal difference between two
(k) (k+1)
successive approximations for t between tn and tn , occurs in the middle
Section 2.2 Sample function continuity 31

xn+1

xn

(k) (k+1)
tn tn
(2k) (2k+1) (2k+2)
= tn+1 tn+1 = tn+1

Figure 2.1: Successive approximations with piecewise linear functions.

of the interval, and hence

 
 (2k+1)
 1 (k) (k+1) 
|xn+1 (t) − xn (t)| ≤ x(tn+1 ) − x(tn ) + x(tn ) 
2
1  (2k+1)  1
(2k)   (2k+2)

(2k+1) 
≤ x(tn+1 ) − x(tn+1 ) + x(tn+1 ) − x(tn+1 )
2 2
1 1
= A + B, say.
2 2

The tail distribution of the maximal difference between two successive ap-
proximations,

1 1
Mn(k) = max |xn+1 (t) − xn (t)| ≤ A + B,
(k)
tn ≤t≤tn
(k+1) 2 2

can therefore be estimated by

P (Mn(k) ≥ c) ≤ P (A ≥ c) + P (B ≥ c),

(n)
since if Mn ≥ c, then either A ≥ c or B ≥ c, or both.
Now take c = g(2−n−1 ) and use the bound (2.2), to get

P (Mn(k) ≥ g(2−n−1 )) ≤ 2q(2−n−1 ),

32 Stochastic analysis Chapter 2

for each k = 0, 1, . . . , 2n − 1. By Boole’s inequality2 we get, since there are 2n

intervals,

−n−1
P max |xn+1 (t) − xn (t)| ≥ g(2 )
0≤t≤1
2n −1 

(k) −n−1
=P Mn ≥ g(2 ) ≤ 2n+1 q(2−n−1 ).
k=0
n+1 q(2−n−1 ) < ∞ by assumption, and then the Borel-Cantelli
Now n2
lemma (see Exercises in Appendix B), gives that, with probability one, only
finitely many of the events

max |xn+1 (t) − xn (t)| ≥ g(2−n )

0≤t≤1

occur. This means that there is a set Ω0 with P (Ω0 ) = 1, such that for every
outcome ω ∈ Ω0 , from some integer N (depending of the outcome, N = N (ω))
and onwards, (n ≥ N ),

max |xn+1 (t) − xn (t)| < g(2−n ).

0≤t≤1

First of all, this shows that there exists a limiting function y(t) for all
ω ∈ Ω0 ; the condition (B.4) for almost sure convergence, given in Appendix B,
says that limn→∞ xn (t) exists with probability one.
It also shows that the convergence is uniform: for ω ∈ Ω0 and n ≥ N ,
m > 0,

|xn+m (t) − xn (t)|

≤ |xn+1 (t) − xn (t)| + |xn+2 (t) − xn+1 (t)| + . . . + |xn+m (t) − xn+m−1 (t)|

m−1  ∞
≤ g(2−n−j ) ≤ g(2−n−j ).
j=0 j=0

Letting m → ∞, so that xn+m (t) → y(t), and observing that the inequalities
hold for all t ∈ [0, 1], we get that
∞
 ∞

max |y(t) − xn (t)| ≤ g(2−n−j ) = g(2−j ).
0≤t≤1
j=0 j=n

Since this bound tends to 0 as n → ∞, we have the uniform convergence, and

since all xn are continuous functions, we also have that y is continuous for all
ω ∈ Ω0 . For ω ∈ Ω0 , define y(t) ≡ 0, making y a continuous function for all
ω ∈ Ω.
2 P
P (∪Ak ) ≤ P (Ak ) .
Section 2.2 Sample function continuity 33

It remains to prove that x and y are equivalent, i.e. P (x(t) = y(t)) = 1,

for all t ∈ [0, 1]. For that sake, take any t ∈ [0, 1] and find a sequence of dyadic
(k )
numbers tn n → t such that

t(k
n
n)
≤ t < tn(kn ) + 2−n .

Since both g(h) and q(h) are non-decreasing, we have from (2.2),
     
   (kn ) 
P x(t(k
n
n)
) − x(t) ≥ g(2−n
) ≤ P  x(t n ) − x(t) ≥ g(t − t (kn )
n )
≤ q(t − t(k n)
n ) ≤ q(2
−n
).

Adding over n gives

∞
    
∞
 (kn )  −n
P x(tn ) − x(t) ≥ g(2 ) ≤ q(2−n ) < ∞,
n=1 n=1

and it follows from the Borel-Cantelli lemma that it can happen only finitely
(k )
many times that |x(tn n ) − x(t)| ≥ g(2−n ). Since g(2−n ) → 0 as n → ∞,
(k )
we have proved that x(tn n ) → x(t) with probability one. Further, since y(t)
(k ) (k ) (k )
is continuous, y(tn n ) → y(t). But x(tn n ) = y(tn n ), and therefore the two
limits are equal, with probability one, as was to be proved. 2

The theorem says that for each process x(t) that satisfies the conditions
there exists at least one other equivalent process y(t) with continuous sample
paths, and with exactly the same finite-dimensional distributions. Of course it
seems unnecessary to start with x(t) and immediately change to an equivalent
continuous process y(t). In the future we assume that we only have the contin-
uous version, whenever the sufficient conditions for sample function continuity
are satisfied.

2.2.2.1 Special conditions for continuity

Theorem 2:1 is simple to use, since it depends only on the distribution of the
increments of the process, and involves only bivariate distributions. For special
processes conditions that put bounds on the moments of the increments are
even simpler to use. One such is the following.

Corollary 2.1 If there exist constants C , and r > p > 0, such that for all
small enough h > 0,
|h|
E (|x(t + h) − x(t)|p ) ≤ C (2.3)
| log |h||1+r
then the condition in Theorem 2:1 is satisfied and the process has, with proba-
bility one, continuous sample paths.
34 Stochastic analysis Chapter 2

Note, that many processes satisfy a stronger inequality than (2.3), namely

E (|x(t + h) − x(t)|p ) ≤ C|h|1+c (2.4)

for some constants C , and c > 0, p > 0. Then (2.3) is automatically satisfied
with any r > p, and the process has, with probability one, continuous sample
paths.
Proof: Markov’s inequality, a generalization of Chebysjev’s inequality, states
that for all random variables U , P (|U | ≥ λ) ≤ E(|U |p )/λp . Apply the theorem
with g(h) = | log |h||−b , 1 < b < r/p. One gets,
C|h|
P (|x(t + h) − x(t)| > g(h)) ≤ .
| log |h||1+r−bp
  1
Since b > 1, one has g(2−n ) = (n log 2)b
< ∞, and, with

q(h) = C|h|/| log |h||1+r−bp ,

and 1 + r − bp > 1,
  C
2n q(2−n ) = < ∞,
(n log 2)1+r−bp
which proves the assertion. 2

Example 2:1 We show that the Wiener process W (t) has, with probability
one, continuous sample paths. In the standard Wiener process, the increment
W (t + h) − W (t), h > 0, is Gaussian with mean 0 and variance h. Thus,

E(|W (t + h) − W (t)|p ) = C|h|p/2 ,

with C = E(|U |p ) < ∞, for a standard normal variable U , giving the moment
bound
|h|
E(|W (t + h) − W (t)|4 ) = C|h|2 < ,
| log |h||6
for small h. We see that condition (2.3) in the corollary is satisfied with r =
5 > 4 = p. Condition (2.4) is satisfied with p = 3, c = 1/2.

2.2.2.2 Continuity of stationary processes

Stationary processes have constant mean and a covariance function

r(t) = Cov(x(s + t), x(s)),

which is a function only of the time lag t. Since the increments have variance

E((x(t + h) − x(t))2 ) = 2(r(0) − r(h)), (2.5)

Section 2.2 Sample function continuity 35

it is clear that continuity conditions can be formulated in terms of the covariance

function. Equivalent conditions can be formulated by means of the spectral
distribution function F (ω), which is such that
 ∞
r(t) = eiωt dF (ω).
−∞
q.m.
A first immediate consequence of (2.5) is that x(t + h) → x(t) as h → 0
if and only if the covariance function r(t) is continuous for t = 0. For sample
function continuity, a sufficient condition in terms of the covariance function
follows directly from Corollary 2.1.

Theorem 2:2 If r(t) is the covariance function of a stationary stochastic pro-

cess x(t), such that, as t → 0,

|t|
r(t) = r(0) − O , (2.6)
| log |t||q
for some q > 3, then x(t) has3 continuous sample functions.4

2.2.2.3 Continuity of stationary Gaussian processes

For stationary Gaussian processes the conditions for sample function continuity
can be considerably weakened, to require slightly more that just continuity of
the covariance function. We state the sufficient condition both in terms of the
covariance function and in terms of the spectrum, and to this end, we formulate
an analytic lemma, the proof of which can be found in [9, Sect. 9.3].

Lemma 2.1 a) If, for some a > 0,

 ∞
(log(1 + ω))a dF (ω) < ∞, (2.7)
0
then 
r(t) = r(0) − O | log |t||−b , as t → 0, (2.8)
for any b ≤ a.
b) If (2.8) holds for some b > 0, then (2.7) is satisfied for any a < b.

Theorem 2:3 A stationary Gaussian process x(t) has, with probability one,
continuous sample paths if, for some a > 3, any of the following conditions is
satisfied:
 
r(t) = r(0) − O | log |t||−a , as t → 0, (2.9)
 ∞
(log(1 + ω))a dF (ω) < ∞. (2.10)
0
3
Or rather ”Is equivalent to a process that has . . . ”
4
The notation f (x) = g(x) + 0(h(x)) as x → 0 means that |(f (x) − g(x))/h(x)| is bounded
by some finite constant C as x → 0 .
36 Stochastic analysis Chapter 2

Proof: In a stationary Gaussian process, x(t + h) − x(t) has a normal distri-

bution with mean zero and variance σh2 = 2(r(0) − r(h)), where by assumption
C
σh ≤
| log |h||a/2
x x 2
for some constant C > 0. Writing Φ(x) = −∞ φ(y) dy = √12π −∞ e−y /2 dy
for the standard normal distribution, we have


g(h)
P (|x(t + h) − x(t)| > g(h)) = 2 1 − Φ
σh
Now, take g(h) = | log |h|/ log 2|−b , where b is chosen so that 1 < b < (a−1)/2,
which is possible since a > 3 by assumption. From the bound (1.13) of the
normal distribution tail, 1 − Φ(x) ≤ φ(x)/x, we then get
  
g(h)| log |h||a/2
P (|x(t + h) − x(t)| > g(h)) ≤ 2 1 − Φ
C
 
2C g(h)| log |h||a/2
≤ φ
g(h)| log |h||a/2 C
= q(h), say.
The reader should complete the proof and show that
 
g(2−n ) < ∞ and 2n q(2−n ) < ∞,

and then apply Theorem 2:1 to see that (2.9) is sufficient.

Lemma 2.1 shows that also (2.10) is sufficient for sample function continuity
for a Gaussian process. 2

Example 2:2 Any stationary process with

r(t) = r(0) − C|t|α + o(|t|α ),
some C > 0, as t → 0,5 has continuous sample paths if 1 < α ≤ 2. If it
furthermore is a Gaussian process it is continuous if 0 < α ≤ 2.

Remark 2:1 The sufficient conditions for sample function continuity given in
the theorems are satisfied for almost all covariance functions that are encoun-
tered in applied probability. But for Gaussian processes, even the weak condition
(2.9) for a > 3, can be relaxed to require only that a > 1, which is very close
to being necessary; see [9, Sect. 9.5].
Gaussian stationary processes which are not continuous behave with neces-
sity very badly, and it can be shown that the sample functions are, with prob-
ability one, unbounded in any interval. This was shown by Belyaev [4] but is
also a consequence of a theorem by Dobrushin [10], see also [9, Ch. 9.5].
5
At this stage you should convince yourself that α > 2 impossible?
Section 2.3 Derivatives, tangents, and other characteristics 37

2.2.3 Probability measures on C[0, 1]

We can now complete the table at the end of Section 1.3.1 and define proba-
bilities on C[0, 1], the space of continuous functions on the interval [0, 1]. A
stochastic process {x(t)}0≤t≤1 on a probability space (Ω, F) has its realizations
in R[0,1] . If the finite-dimensional distributions of x satisfy any of the sufficient
conditions for sample function continuity, either x(t) or an equivalent process
y(t) has, with probability one, continuous sample functions, and hence has its
realizations in C[0, 1] ⊂ R[0,1] .
In the same way as the finite-dimensional distributions define a probability
measure on (R[0,1] , B[0,1] ), assigning probabilities to all Borel sets, we can now
define a probability measure on C[0, 1]. The question is, what is the σ -field of
events which get probability? In fact, we can take the simplest choice,
 
B = B[0,1] ∩ C[0, 1] = B ∩ C[0, 1]; B ∈ B[0,1] ,
i.e. take those parts of the Borel-sets which intersect C[0, 1].

Theorem 2:4 If {Ftn }∞ n=1 is a family of finite-dimensional distributions that

satisfy any of the sufficient conditions for sample functions continuity, then
there exists a probability measure on (C[0, 1], B) such that the co-ordinate pro-
cess
{x(t)}0≤t≤1
has the given finite-dimensional distributions.

2.2.3.1 Open sets in C[0, 1]

The family B can be described alternatively in terms of open sets in C[0, 1].
Take a continuous function x(t) ∈ C[0, 1]. By an -surrounding of x we mean
the set of functions which are in an -band around x,
 
y ∈ C[0, 1]; max |y(t) − x(t)| <  .
0≤t≤1

A set of functions A ⊆ C[0, 1] is called open if for every x ∈ A, there is an

-surrounding of x that is completely in A. The open sets in C[0, 1] generate
a σ -field, the smallest σ -field that contains all open sets, and that σ -field is
exactly B .

2.3 Derivatives, tangents, and other characteristics

2.3.1 Differentiability
2.3.1.1 General conditions

When is a continuous stochastic process differentiable in the sense that its

sample functions are continuously differentiable? The answer can be given
38 Stochastic analysis Chapter 2

as conditions similar to those for sample function continuity, but now with
bounds on the second order differences. By pasting together piecewise linear
approximations by means of smooth arcs, one can prove the following theorem;
see [9, Sect. 4.3].

Theorem 2:5 a) Suppose the stochastic process {x(t); 0 ≤ t ≤ 1} satisfies the

conditions for sample function continuity in Theorem 2:1. If, furthermore, for
all t − h and t + h in [0, 1],

P (|x(t + h) − 2x(t) + x(t − h)| ≥ g1 (h)) ≤ q1 (h), (2.11)

where g1 and q1 are two non-decreasing functions, such that

∞
 ∞

2n g1 (2−n ) < ∞ and 2n q1 (2−n ) < ∞,
n=1 n=1

then there exists an equivalent process {y(t); 0 ≤ t ≤ 1} with continuously

differentiable sample paths.

b) The sufficient condition in (a) is satisfied if

K|h|1+p
E(|x(t + h) − 2x(t) + x(t − h)|p ) ≤ , (2.12)
| log |h||1+r

for some constants p < r and K .

c) Many processes satisfy a stronger inequality than (2.12), namely

E (|x(t + h) − 2x(t) + x(t − h)|p ) ≤ C|h|1+p+c , (2.13)

for some constants C , and c > 0, p > 0. Then (2.12) is satisfied, and the
process has, with probability one, continuously differentiable sample paths.

In Section 2.1 we mentioned a condition for quadratic mean (q.m.) differ-

entiability of a stationary process. One may ask: What is the relation between
the q.m.-derivative and the sample function derivative? They are both limits
of the differential quotient (x(t + h) − x(t))/h as h → 0. Now, it is easy to
prove that if the limit exists in both quadratic mean and as a sample function
limit with probability one, then the two limits are equal (also with probability
one), and hence the two derivative processes are equivalent, and have the same
finite-dimensional distributions. It follows that the covariance function of the
derivative {x (t), t ∈ R} is rx (t) = −rx (t).
Section 2.3 Derivatives, tangents, and other characteristics 39

2.3.1.2 Differentiable Gaussian processes

In order that a stationary Gaussian process has continuously differentiable sam-
ple functions it is necessary that its covariance function has a smooth higher
order Taylor expansion; cf. condition (2.9).
As we shall see in the following theorem, demanding just slightly more
than a finite second spectral moment guarantees that a Gaussian process has
continuously differentiable sample paths. For a proof, the reader is referred to
[9, Sect. 9.3].

Theorem 2:6 a) A stationary Gaussian process is continuously differentiable6

if, for some a > 3, its covariance function has the expansion
 
ω 2 t2 t2
r(t) = r(0) − +O , (2.14)
2 | log |t||a

where ω2 = −r  (0) < ∞.

b) Condition (2.14) can be replaced by the condition that, for some a > 3,
 ∞
ω 2 (log(1 + ω))a dF (ω) < ∞. (2.15)
0

As for sample function continuity, it can be shown that for continuously

differentiable sample functions, it suffices that the constant a is greater than
1.

Example 2:3 Condition (2.14) is easy to use for Gaussian processes. Most
covariance functions used in practice have an expansion

ω 2 t2
r(t) = r(0) − + O(|t|a ),
2
where a is an integer, either 3 or 4. Then the process is continuously differ-
entiable. Processes with covariance function admitting an expansion r(t) =
r(0) − C|t|α + o(|t|α ) with α < 2 are not differentiable; they are not even dif-
ferentiable in quadratic mean. An example is the Ornstein-Uhlenbeck process
with r(t) = r(0)e−C|t| .

As a final example of covariance conditions, we encourage the reader to

prove a sufficient condition in terms of −r  (t), the covariance function of the
quadratic mean derivative.

Theorem 2:7 A stationary process x(t) is continuously differentiable if any

of the following two conditions hold,
6
As usual, this means that there exists an equivalent process that, with probability one,
has continuously differentiable sample functions.
40 Stochastic analysis Chapter 2

a) −r  (t) = −r  (0) − C|t|α + o(|t|α ) with 1 < α ≤ 2,

b) it is Gaussian and −r  (t) = −r  (0) − C|t|α + o(|t|α ) with 0 < α ≤ 2.

2.3.2 Jump discontinuities and Hölder conditions

What type of discontinuities are possible for stochastic processes which do not
have continuous sample functions? For example, how do we know that the
Poisson process has sample functions that increase only with jumps of size one?
We would like to have a condition that guarantees that only simple discontinu-
ities are possible, and such a condition exists, with restriction on the increment
over two adjacent interval. Similarly, for a process which is not continuously
differentiable, how far from differentiable are the sample functions.

2.3.2.1 Jump discontinuities

The proof of the following theorem is indicated in [9, Sec. 4.4].

Theorem 2:8 If there are positive constants C , p, r , such that for all s, t with
0 ≤ t < s < t + h ≤ 1,

E {|x(t + h) − x(s)|p · |x(s) − x(t)|p } ≤ C|h|1+r (2.16)

then the process {x(t); 0 ≤ t ≤ 1} has, with probability one,7 sample functions
with at most jump discontinuities, i.e.

lim x(t) and lim x(t)

t↓t0 t↑t0

exist for every t0 ∈ [0, 1].

Example 2:4 The Poisson process with intensity λ has independent incre-
ments and hence
 
E |x(t + h) − x(s)|2 · |x(s) − x(t)|2
= E(|x(t + h) − x(s)|2 ) · E(|x(s) − x(t)|2 )
= (λ(t + h − s) + (λ(t + h − s))2 ) · (λ(s − t) + (λ(s − t))2 ) ≤ Cλ2 h2 .

The conditions of the theorem are obviously satisfied.

7
As usual, this means that there exists an equivalent process with this property.
Section 2.3 Derivatives, tangents, and other characteristics 41

2.3.2.2 Hölder continuity and the continuity modulus

How large are the increments in a non-differentiable process? In fact, moment
bounds on the increments give precise estimates for the distribution of the
continuity modulus ωx (h) of the sample functions. This is defined as
ωx (h) = sup |x(s) − x(t)| (2.17)
|s−t|≤h

where the supremum is taken over 0 ≤ s, t ≤ 1. (For a continuous process the

supremum in (2.17) can be taken over the countable rationals, so it is a well
defined random variable.)
Functions for which there exist constants A, a such that for all t, t+h ∈ [0, 1]
|x(t + h) − x(t)| ≤ A|h|a ,
are said to be Hölder continuous of order a, and to satisfy a Lipschitz condition
of order a.
In this section we shall present stochastic estimates of the continuity modu-
lus for a stochastic process, and also give a sufficient condition for a stochastic
process to be Hölder continuous of order a < r/p ≤ 1.

Theorem 2:9 If there are constants C , p ≥ r > 0 such that

E {|x(t + h) − x(t)|p } ≤ C|h|1+r (2.18)
then there exists a random variable A with P (A < ∞) = 1, such that the
continuity modulus satisfies an inequality
ωx (h) ≤ A|h|a , for all h > 0,
for all 0 < a < r/p.
Proof: First examine the increments over the dyadic numbers
tn(k) = k/2n , k = 0, 1, . . . , 2n ; n = 1, 2, . . . .
Take an a < r/p and write δ = 2−a . Then, by Markov’s inequality,
(k+1) (k)
E(|x(tn ) − x(tn )|p )
P (|x(tn(k+1) ) − x(tn(k) )| > δn ) ≤
δnp
C(2−n )1+r C
−anp
≤ = n(1+r−ap) .
2 2
From Boole’s inequality, summing over n, we obtain in succession,
C2n
P ( maxn |x(tn(k+1) ) − x(tn(k) )| > δn ) ≤ n(1+r−ap)
0≤k≤2 −1 2
C
= ,
2n(r−ap)
∞
 ∞
 C
P( maxn |x(tn(k+1) ) − x(tn(k) )| >δ ) ≤
n
< ∞,
n=0
0≤k≤2 −1
n=0
2n(r−ap)
42 Stochastic analysis Chapter 2

since r −ap > 0. The Borel-Cantelli lemma gives that only finitely many events
An = { max |x(tn(k+1) ) − x(tn(k) )| > δn }
0≤k≤2n −1

occur, which means that there exists a random index ν such that for all n > ν ,
|x(tn(k+1) ) − x(tn(k) )| ≤ δn for all k = 0, 1, . . . , 2n − 1. (2.19)
(k)
Next we estimate the increment from a dyadic point tn to an arbitrary
(k) (k+1)
point t. To that end, take any t ∈ [tn , tn ), and consider its dyadic expan-
sion, (αm = 0 or 1),
∞
αm
t = tn(k) + .
2n+m
m=1
(k)
Summing all the inequalities (2.19), we obtain that the increment from tn to
t is bounded (for n > ν ),
∞
 δn+1
|x(t) − x(tn(k) )| ≤ δn+m = . (2.20)
m=1
1−δ

The final estimate relates t + h to the dyadic points. Let ν < ∞ be the
random index just found to exist. Then, suppose h < 2−ν and find n, k such
that 2−n ≤ h < 2−n+1 and k/2n ≤ t < (k + 1)/2n . We see that n > ν and
tn(k) ≤ t < tn(k+1) < t + h ≤ tn(k+) ,
where  is either 2 or 3. As above, we obtain
δn+1
|x(t + h) − x(tn(k+1) )| ≤ + δn . (2.21)
1−δ
Summing the three estimates (2.19-2.21), we see that
δn+1 δn+1 2 2
|x(t + h) − x(t)| ≤ δn + + + δn = (2−n )a ≤ ha ,
1−δ 1−δ 1−δ 1−δ
for 2−n ≤ h < 2−ν . For h ≥ 2−ν it is always true that
M a
|x(t + h) − x(t)| ≤ M ≤ h
2−ν
for some random M . If we take A = max(M 2ν , 2/(1 − δ)), we complete the
proof by combining the last two inequalities, to obtain |x(t + h) − x(t)| ≤ Aha .
2

Example 2:5 For the Wiener process,

E(|x(t + h) − x(t)|p ) = Cp |h|p/2 = Cp |h|1+(p/2−1) ,

so the Wiener process is Hölder continuous of order a < (p/2−1)/p = 1/2−1/p

for every p > 0. This means that it is Hölder continuous of all orders a < 1/2.
Section 2.3 Derivatives, tangents, and other characteristics 43

In the next section we shall investigate the existence of tangents of a prede-

termined level. Then we shall need a small lemma on the size of the continuity
modulus of a continuous stochastic process.

Lemma 2.2 Let x(t) be a stochastic process with continuous sample functions
in 0 ≤ t ≤ 1, and let ωx (h) be its (random) continuity modulus, defined by
(2.17). Then, to every  > 0 there is a (deterministic) function ω (h) such
that ω (h) ↓ 0 as h ↓ 0, and

P (ωx (h) < ω (h) for 0 < h ≤ 1) > 1 − .

Proof: The sample continuity of x(t) says that the continuity modulus tends
to 0 for h → 0,
lim P (ωx (h) < c) = 1
h→0

for every fixed c > 0. Take a sequence c1 > c2 . . . > cn ↓ 0. For a given  > 0
we can find a decreasing sequence hn ↓ 0 such that

P (ωx (hn ) < cn ) > 1 − /2n+1 .

Since ωx (h) is non-increasing as h decreases, then also

P (ωx (h) < cn for 0 < h ≤ hn ) > 1 − /2n+1 ,

for n = 1, 2, . . . . Summing the exceptions, we get that

∞

P (ωx (h) < cn for 0 < h ≤ hn and n = 1, 2, . . . ) > 1 − /2n+1 = 1 − /2.
n=1
(2.22)
Now we can define the deterministic function ω (h) from the sequences cn
and hn . Take

c0 for h1 < h ≤ 1
ω (h) =
cn for hn+1 < h ≤ hn , n = 1, 2, . . . .

If we take c0 large enough to make

P (ωx (h) < c0 for h1 < h ≤ 1) > 1 − /2,

and combine with (2.22) we get the desired estimate. 2

2.3.2.3 Tangencies
We start with a theorem due to E.V. Bulinskaya on the non-existence of tan-
gents of a pre-specified level.
44 Stochastic analysis Chapter 2

Theorem 2:10 Suppose the density ft (x) of x(t) is bounded for 0 ≤ t ≤ 1,

ft (x) ≤ c0 < ∞,

and that x(t) has, with probability one, continuously differentiable sample paths.
Then,
a) for any level u, the probability is zero that there exists a t ∈ [0, 1] such that
simultaneously x(t) = u and x (t) = 0, i.e. there exists no points where x(t)
has a tangent on the level u in [0, 1],
b) there are only finitely many t ∈ [0, 1] for which x(t) = u.
Proof: a) By assumption, x(t) has continuously differentiable sample paths.
We identify the location of those t-values for which x (t) = 0 and x(t) is close
to u. For that sake, take an integer n and a constant h > 0, let Hτ be the
event
Hτ = {x (τ ) = 0} ∩ {|x(τ ) − u| ≤ h},
and define, for k = 1, 2, . . . , n,

Ah = {Ht occurs for at least one t ∈ 0, 1]},

 
Ah (k, n) = {Hτ occurs for at least one τ ∈ k−1n , n },
k

Ah = ∪nk=1 Ah (k, n).

Now take a sample function that satisfies the conditions for Ah (k, n) and
let ωx be the continuity modulus of its derivative. For such a sample function,

x(k/n) = x(τ ) + (k/n − τ ) x (τ + θ(k/n − τ )),

for some θ, 0 ≤ θ ≤ 1, and hence, on Ah (k, n)

|x(k/n) − u| ≤ h + n−1 ωx (n−1 ). (2.23)

We now use Lemma 2.2 to bound ωx . If ω(t) ↓ 0 as t ↓ 0, let Bω denote the
sample functions for which ωx (t) ≤ ω(t) for all t in [0, 1]. By the lemma, given
 > 0, there exists at least one function ω (t) ↓ 0 such that P (Bω ) > 1 − /2.
For outcomes satisfying (2.23) we use the bound ω , and obtain

n
P (Ah ) ≤ P (Ah (k, n) ∩ Bω ) + (1 − P (Bω ))
k=1


n
≤ P (|x(k/n) − u| ≤ h + n−1 ω (n−1 )) + /2
k=1

≤ 2nc0 (h + n−1 ω (n−1 )) + /2,

where c0 is the bounding constant for the density ft (x).

Section 2.4 Quadratic mean properties a second time 45

Since ω (t) → 0 as t → 0, we can select first an n0 and then an h0 to make

P (Ah0 ) ≤ . But if there exists a time point t for which x (t) = 0 and x(t) = u
simultaneously, then certainly Ah has occurred for any h > 0 and the event
of interest has probability less than , which was arbitrary. The probability of
simultaneous occurrence is therefore 0 as stated.
b) To prove there are only a finite number of u-values in [0, 1], assume, on the
contrary, that there is an infinite sequence of points ti ∈ [0, 1] with x(ti ) = u.
There is then at least one limit point t0 ∈ [0, 1] of {ti } for which, by continuity,
x(t0 ) = u. Since the derivative of x(t) is assumed continuous, we must also
have x (t0 ) = 0, and we have found a point where simultaneously x(t0 ) = u,
x (t0 ) = 0. By (a), that event has probability 0. 2

2.4 Quadratic mean properties a second time

Continuous or differentiable sample paths are what we expect to encounter in
practice when we observe a stochastic process. To prove that a mathematical
model for a random phenomenon has continuous or differentiable sample paths
is a quite different matter. Much more simple is to base the stochastic analy-
sis on correlation properties, which could be checked against data, at least in
principle. Such second order properties are studied in quite some detail in ele-
mentary courses in stochastic processes, and we give here only some refinements
and extra comments in addition to those in Section 2.1. We assume throughout
in this section, as in most of the chapter, that the process x(t) has mean zero.
We first remind about the definition of convergence in quadratic mean of a
sequence of random variables {xn } with E(x2n ) < ∞ to a random variable x:
q.m.
xn → x if and only if E((xn − x)2 ) → 0,
as n → ∞; see Appendix B. We shall use the Loève criterion (B.8) for quadratic
mean convergence: the sequence xn converges in quadratic mean if and only
if
E(xm x̄n ) has a finite limit c, (2.24)
when m and n tend to infinity independently of each other.

2.4.1 Quadratic mean continuity

A stochastic process x(t) is continuous in quadratic mean (or L2 -continuous)
at t if
q.m.
x(t + h) → x(t)
when h → 0, i.e. if E(|x(t + h) − x(t)|2 ) → 0. We formulate the condition for
quadratic mean continuity in terms of the covariance function
r(s, t) = Cov(x(s), x(t)) = E(x(s) · x(t)),

for a, not necessarily stationary, process.

46 Stochastic analysis Chapter 2

Theorem 2:11 A stochastic process x(t) with mean zero is continuous in qua-
dratic mean at t0 if and only if the covariance function r(s, t) is continuous on
the diagonal point s = t = t0 .
Proof: If r(s, t) is continuous at s = t = t0 , then

E(|x(t0 + h) − x(t0 )|2 ) = E(|x(t0 + h)|2 ) + E(|x(t0 )|2 ) − 2E(x(t0 + h) · x(t0 ))

= r(t0 + h, t0 + h) − 2r(t0 + h, t0 ) + r(t0 , t0 ) → 0

as h → 0, which shows the ”if” part.

For the ”only if” part, expand

r(t0 + h, t0 + k) − r(t0 , t0 )
= E((x(t0 + h) − x(t0 )) · (x(t0 + k) − x(t0 )))
+ E((x(t0 + h) − x(t0 )) · x(t0 ))
+ E(x(t0 ) · (x(t0 + k) − x(t0 ))) = e1 + e2 + e3 , say.

Here

e1 ≤ E(|x(t0 + h) − x(t0 )|2 ) · E(|x(t0 + k) − x(t0 )|2 ) → 0,

e2 ≤ E(|x(t0 + h) − x(t0 )|2 ) · E(|x(t0 )|2 ) → 0,

e3 ≤ E(|x(t0 )|2 ) · E(|x(t0 + k) − x(t0 )|2 ) → 0,

so r(t0 + h, t0 + k) → r(t0 , t0 ) as h, k → 0. 2

2.4.2 Quadratic mean differentiability

A stochastic process x(t) is called differentiable in quadratic mean at t if there
exists a random variable, naturally denoted x (t), such that
x(t + h) − x(t) q.m. 
→ x (t),
h
as h → 0, i.e. if E(( x(t+h)−x(t)
h − x (t))2 ) → 0. If a process is differentiable
both in quadratic mean and in sample function meaning, with probability one,
then the two derivatives are equal with probability one. We shall now actually
prove the condition for quadratic mean differentiability of a stationary process,
which was stated in Section 2.1.

Theorem 2:12 A stationary process x(t) is quadratic mean differentiable if

and only if its covariance function r(t) is twice continuously differentiable in a
neighborhood of t = 0. The derivative process x (t) has covariance function

rx (t) = C(x (s + t), x (s)) = −r  (t).

Section 2.4 Quadratic mean properties a second time 47

Proof: For the ”if” part we use the Loève criterion, and show that, if h, k → 0
independently of each other, then

x(t + h) − x(t) x(t + k) − x(t) 1

E · = (r(h − k) − r(h) − r(−k) + r(0))
h k hk
(2.25)
has a finite limit c. Define

f (h, k) = r(h) − r(h − k),

∂
f1 (h, k) = f (h, k) = r  (h) − r  (h − k),
∂h
 ∂2
f12 (h, k) = f (h, k) = r  (h − k).
∂h∂k
By applying the mean value theorem we see that there exist θ1 , θ2 ∈ (0, 1) such
that (2.25) is equal to

f (h, k) − f (0, k) f  (θ1 h, k)

− =− 1
hk k
f1 (θ1 h, 0) + kf12
 (θ h, θ k)
1 2
=−
k

= −f12 (θ1 h, θ2 k) = −r  (θ1 h − θ2 k). (2.26)

Since r  (t) by assumption is continuous, this tends to −r  (0) as h, k → 0,

which is the required limit in the Loève criterion.
To prove the ”only if” we need a fact about Fourier integrals and the spectral
∞ ∞
representation r(t) = −∞ eiωt dF (ω) = −∞ cos ωt dF (ω).

2(r(0)−r(t)) ∞ 2 dF (ω)
Lemma 2.3 a) limt→0 t2
= ω2 = −∞ ω ≤ ∞.

b) If ω2 < ∞ then r  (0) = −ω2 and r  (t) exists for all t.

c) If r  (0) exists, finite, then ω2 < ∞ and then, by (b), r  (t) exists for all
t.

Proof of lemma: If ω2 < ∞, (a) follows from

 ∞
2(r(0) − r(t)) 1 − cos ωt
2
= ω2 dF (ω)
t −∞ ω 2 t2 /2
1−cos ωt
by dominated convergence, since 0 ≤ ω 2 t2 /2
≤ 1. If ω2 = ∞, the result
1−cos ωt
follows from Fatou’s lemma, since limt→0 ω2 t2 /2 = 1.
To prove (b), suppose ω2 < ∞.  ∞ Then it is possible to differentiate twice
under the integral sign in r(t) = −∞ cos ωt dF (ω) to obtain that r  (t) exists
and  ∞

−r (t) = ω 2 cos(ωt) dF (ω),
−∞
48 Stochastic analysis Chapter 2

and in particular −r  (0) = ω2 .

For (c), suppose r(t) has a finite second derivative at the origin, and im-
plicitly is differentiable near the origin. By the same argument as for (2.26),
with h = k = t, we see that
2(r(0) − r(t))
= −r  ((θ1 − θ2 )t) → −r  (0)
t2
as t → 0. Then part (a) shows that ω2 < ∞, and the lemma is proved. 2
Proof of the ”only if ” part of Theorem 2:12: If x(t) is quadratic mean
q.m.
differentiable, (x(t + h) − x(t))/h → x (t) with E(|x (t)|2 ) finite, and (see
further Appendix B.2),
 2 
 x(t + h) − x(t)  2(r(0) − r(h))
E(|x (t)|2 ) = lim E  
 = lim = ω2 < ∞.
h→0 h h→0 h2

Part (b) of the lemma shows that r  (t) exists.

To see that the covariance function of x (t) is equal to −r  (t), just take the
limit of
 
x(s + t + h) − x(s + t) x(s + k) − x(s)
E ·
h k
1
= (r(t + h − k) − r(t + h) − r(t − k) + r(t))
hk

= −f12 (t + θ1 h, θ2 k) = −r  (t + θ1 h − θ2 k) → −r  (t),

for some θ1 , θ2 as h, k → 0. 2

2.4.3 Higher order derivatives and their correlations

To obtain higher order derivatives one just has to examine the covariance func-
tion rx (t) = −rx (t) for the conditions in the existence theorem, etc.
The derivative of a (mean square) differentiable stationary process is also
stationary and has covariance function rx (t) = −rx (t), from Theorem 2:12.
One can easily derive the cross-covariance relations between the derivative and
the process (strictly, we need (B.7) from Appendix B).

Theorem 2:13 The cross-covariance between derivatives x(j) (s) and x(k) (t)
of a stationary process {x(t), t ∈ R} is

Cov(x(j) (s), x(k) (t)) = (−1)(k) rx(j+k)(s − t). (2.27)

In particular

Cov(x(s), x (t)) = −rx (s − t). (2.28)

Section 2.5 Summary of smoothness conditions 49

In Chapter 3 we will need the covariances between the the process and its
first two derivatives. The covariance matrix for x(t), x (t), x (t) is
⎛ ⎞
ω0 0 −ω2
⎝ 0 ω2 0 ⎠ ,
−ω2 0 ω4

where ω2k = (−1)k rx2k (0) = ω 2k dF (ω) are spectral moments. Thus, the slope
at a specified point is uncorrelated both with the process value at that point and
with the curvature, while process value and curvature have negative correlation.
We have, for example, V (x (t) | x(0), x (0)) = ω4 − ω22 /ω0 .

2.5 Summary of smoothness conditions

The following table summarizes some crude and simple sufficient conditions for
different types of quadratic mean and sample function smoothness.

Condition on r(t) as t → 0 further condition property

r(t) = r(0) − o(1) q.m. continuous
r(t) = r(0) − C|t|α + o(|t|α ) 1<α≤2 a.s. continuous
r(t) = r(0) − C|t|α + o(|t|α ) 0 < α ≤ 2 Gaussian a.s. continuous
r(t) = r(0) − ω2 t2 /2 + o(t2 ) q.m. differentiable
−r  (t) = ω2 − C|t|α + o(|t|α ) 1<α≤2 a.s. differentiable
−r  (t) = ω2 − C|t|α + o(|t|α ) 0 < α ≤ 2 aGaussian a.s. differentiable

2.6 Stochastic integration

In this section we shall define the two simplest types of stochastic integrals, of
the form
 b
J1 = g(t)x(t) dt,
a
 b
J2 = g(t) dx(t),
a

where g(t) is a deterministic function and x(t) a stochastic process with mean
0. The integrals can be defined either as quadratic mean limits of approximating
Riemann or Riemann-Stieltjes sums, and depending on the type of convergence
we require, the process x(t) has to satisfy suitable regularity conditions.
The two types of integrals are sufficient for our needs in these notes. A
third type of stochastic integrals, needed for stochastic differential equations,
are those of the form
 b
J3 = g(t, x(t)) dx(t),
a
50 Stochastic analysis Chapter 2

in which g also is random and dependent on the integrator x(t). These will
not be dealt with here.
The integrals are defined as limits in quadratic mean of the approximating
sums

n
J1 = lim g(tk )x(tk )(tk − tk−1 ),
n→∞
k=1
n
J2 = lim g(tk )(x(tk ) − x(tk−1 )),
n→∞
k=1

when a = t0 < t2 < . . . < tn = b, and max |tk − tk−1 | → 0 as n → ∞, provided

the limits exist, and are independent of the subdivision {tk }. To simplify the
writing we have suppressed the double index in the sequences of tk -values; there
is one sequence for each n.
The limits exist as quadratic mean limits if the corresponding integrals
of the covariance function r(s, t) = C(x(s), x(t)) = E(x(s) x(t)) are finite, as
formulated in the following theorem, in which we assume E(x(t)) = 0. Since we
shall mainly use the integrals with complex functions g(t) (in fact g(t) = eiωt ),
we formulate the theorem for complex random functions.

Theorem 2:14 a) If r(s, t) is continuous in [a, b] × [a, b], and g(t) is such
that the Riemann integral

Q1 = g(s)g(t)r(s, t) ds dt < ∞,
[a,b]×[a,b]

b
then J1 = a g(t)x(t) dt exists as a quadratic mean limit, and E(J1 ) = 0 and
E(|J1 |2 ) = Q1 .

b) If r(s, t) has bounded variation 8 in [a, b] × [a, b] and g(t) is such that the
Riemann-Stieltjes integral

Q2 = g(s)g(t) ds,t r(s, t) < ∞,
[a,b]×[a,b]

b
then J2 = a g(t) dx(t) exists as a quadratic mean limit, and E(J2 ) = 0 and
E(|J2 |2 ) = Q2 .

Proof: The simple proof uses the Loève criterion (B.8) for quadratic mean
convergence: take two sequences of partitions of [a, b] with points s0 , s2 , . . . , sm
8 P
That f (t) is of bounded variation in [a, b] means that sup |f (tk )−f (tk−1 )| is bounded,
with the sup taken over all possible partitions.
Section 2.6 Stochastic integration 51

and t0 , t1 , . . . , tn , respectively, and consider


m 
n
E(Sm Sn ) = g(sk )g(tj )r(sk , tj )(sk − sk−1 )(tj − tj−1 ).
k=1 j=1

If Q1 exists, then this converges to Q1 as the limits becomes infinitely fine.

This proves (a). The reader should complete the proof for (b). 2

Example 2:6 Take x(t) = w(t), as the Wiener process. Since rw (s, t) =
b
σ 2 min(s, t), we see that a g(t)w(t) dt exists for all integrable g(t).

Example 2:7 If g(t) is a continuously differentiable function and x(t) = w(t),

the Wiener process, then
 b  b
g(t) dw(t) = g(b)w(b) − g(a)w(a) − g (t)w(t) dt.
a a
To prove this, consider
m
S2 = g(tk )(w(tk ) − w(tk−1 ))
k=1

m
= g(tm )w(tm ) − g(t1 )w(t0 ) − (g(tk ) − g(tk−1 ))w(tk−1 ).
k=2
Since g(t) is continuously differentiable, there is a ρk such that g(tk )−g(tk−1 ) =
(tk − tk−1 )(g (tk ) + ρk ), and ρk → 0 uniformly in k = 1, . . . , m as m → ∞,
maxk |tk − tk−1 | → 0. Thus

m
S2 = g(tm )w(tm ) − g(t1 )w(t0 ) − g(tk ) w(tk−1 )(tk − tk−1 )
k=2

m
+ ρk w(tk−1 )(tk − tk−1 )
k=2
 b
→ g(b)w(b) − g(a)w(a) − g (t)w(t) dt.
a
The proofs of the following two theorems are left to the reader.

Theorem 2:15 If x(s) and y(t) are stochastic processes with cross-covariance
rx,y (s, t) = Cov(x(s), y(t)),
and if the conditions of Theorem 2:14 are satisfied, then
  d   b d
b
E g(s)x(s) ds · h(t)y(t) dt = g(s)h(t)rx,y (s, t) ds dt,(2.29)
a c a c
    b
b d d
E g(s) dx(s) · h(t) dy(t) = g(s)h(t) ds,t rx,y (s, t). (2.30)
a c a c
52 Stochastic analysis Chapter 2

Theorem 2:16 For the Wiener-process with rx,x (s, t) = min(s, t) one has

ds,t rx,x (s, t) = ds

for s = t and 0 otherwise, which gives, for a < c < b < d,

 b  d
 b
E g(s) dx(s) · h(t) dx(t) = g(t)h(t) dt.
a c c

Remark 2:2 A natural question is: are quadratic mean integrals and ordinary
integrals equal? If a stochastic process has a continuous covariance function,
and continuous sample paths, with probability one, and if g(t) is, for example,
b
continuous, then a g(t)x(t) dt exists both as a regular Riemann integral and
as a quadratic mean integral. Both integrals are random variables and they are
limits of the same approximating Riemann sum, the only difference being the
mode of convergence – with probability one, and in quadratic mean, respectively.
But then the limits are equivalent, i.e. equal with probability one.

2.7 An ergodic result

An ergodic theorem deals with convergence properties of time or sample func-
tion averages to ensemble averages i.e. to statistical expectation:

1 T
f (x(t)) dt → E(f (x(0))) as T → ∞,
T 0
for a function of a stationary stochastic process x(t). Such theorems will be
the theme of the entire Chapter 5, but we show already here a simple such
result based only on covariance properties. The process x(t) need not even
be stationary, but we assume E(x(t)) = 0 and that the covariance function
r(s, t) = Cov(x(s), x(t)) exists.

Theorem 2:17 a) If r(s, t) is continuous for all s, t and

 T T
1
r(s, t) ds dt → 0, as T → ∞, (2.31)
T2 0 0
then 
1 T q.m.
x(t) dt → 0.
T 0
b) If there exist constants K, α, β , such that 0 ≤ 2α < β < 1, and
s α + tα
|r(s, t)| ≤ K , for s, t ≥ 0 (2.32)
1 + |s − t|β
then  T
1 a.s.
x(t) dt → 0.
T 0
Section 2.7 An ergodic result 53

Proof: a) This is immediate from Theorem 2:15, since

 
2   T T
T
2 1 1
σT = E x(t) dt = 2 r(s, t) ds dt → 0,
T 0 T 0 0
by assumption.
b) Before we prove the almost sure convergence, note that the condition 2α <
β < 1 puts a limit on how fast E(x(t)2 ) = r(t, t) is allowed to increase as
t → ∞, and it limits the amount of dependence between x(s) and x(t) for
large |s − t|. If the dependence is too strong, it may very well happen that

1 T
T 0 x(t) dt converges, but not to 0 but to a (random) constant different from
0.
We show here only part of the theorem, and refer the reader to [9, p. 95]
for a completion. What we show is that there exists a subsequence of times,
Tn → ∞, such that T1n 0 n x(t) dt → 0.
T a.s.

First estimate σT2 from the proof of (a):

 
2   T T
2 1 T 1
σT = E x(t) dt = 2 r(s, t) ds dt
T 0 T 0 0
 
K T T s α + tα
≤ 2 ds dt
T 0 0 1 + |s − t|β
 T T
K 1 (s/T )α + (t/T )α
= β−α · 2−β ds dt
T T 0 0 1 + |s − t|β
 T
K 2 1
≤ β−α · 1−α β
du.
T T 0 1+u
T
Here (2/T 1−α ) 0 (1 + uβ )−1 du tends to a constant as T → ∞, which implies
that σT2 ≤ K  /T β−α for some constant K  .
Take the constant γ such that γ(β − α) > 1, which is possible from the
properties of α and β , and put Tn = nγ , with
∞
 ∞
 K
σT2 n ≤ < ∞. (2.33)
n=1 n=1
nγ(β−α)

That the sum (2.33) is finite implies, by the Borel-Cantelli lemma and the
T a.s.
Chebysjev inequality, see (B.3) in Appendix B, that Tn−1 0 n x(t) dt → 0, and
so we have showed the convergence for a special sequence of times.
To complete the proof, we have to show that
  T  Tn 
1 1  a.s.
sup  x(t) dt − x(t) dt  → 0,
 
Tn ≤T ≤Tn+1 T 0 Tn 0
as n → ∞; see [9, p. 95]. 2

For stationary processes, the theorem yields the following ergodic theorem
about the observed average.
54 Stochastic analysis Chapter 2
T
Theorem 2:18 a) If x(t) is stationary and T1 0 r(t) dt → 0 as T → ∞ then

1 T q.m.
T 0 x(t) dt → 0.
b) If moreover there is a constant K > 0 and a β > 0, such that |r(t)| ≤ |t|Kβ
T
as t → ∞, then T1 0 x(t) dt → 0.
a.s.
Section 2.7 An ergodic result 55

Exercises
2:1. Prove the following useful inequality valid for any non-negative, integer-
valued random variable N ,
1
E(N ) − E(N (N − 1)) ≤ P (N > 0) ≤ E(N ).
2
Generalize it to the following inequalities where

αi = E(N (N − 1) · . . . · (N − i + 1))

is the ith factorial moment:

2n−1 2n
1  i1 1  1
(−1) α(k+i) ≤ P (N = k) ≤ (−1)i α(k+i) .
k! i! k! i!
i=0 i=0

2:2. Let {x(t), t ∈ R} and {y(t), t ∈ R} be equivalent processes which both

have, with probability one, continuous sample paths. Prove that

P (x(t) = y(t), for all t ∈ R) = 1.

2:3. Find the values on the constants a and b that make a Gaussian process
twice continuously differentiable if its covariance function is

r(t) = e−|t| (1 + a|t| + bt2 ).

2:4. Complete the proof of Theorem 2:3 and show that, in the notation of the
proof,  
g(2−n ) < ∞, and 2n q(2−n ) < ∞.
n n

2:5. Show that the sample paths of the Wiener process have infinite variation,
a.s., by showing the stronger statement that if
n −1 
2


 k+1 k 
Yn = W −W
 2n 2n 
k=0
∞
then n=1 P (Yn < n) < ∞.

2:6. Show that a non-stationary process is continuous in quadratic mean at

t = t0 only if its mean value function m(t) = E(x(t)) is continuous at
t0 and its covariance function r(s, t) = Cov(x(s), x(t)) is continuous at
s = t = t0 .

2:7. Convince yourself of the “trivial” fact that a sequence of normal variables
{xn , n ∈ Z}, such that E(xn ) and V (xn ) have finite limits, then the
sequence converges in distribution to a normal variable.
56 Stochastic analysis Chapter 2

2:8. Give an example of a stationary process that violates the sufficient condi-
tions in Theorem 2:10 and for which the sample functions can be tangents
of the level u = 1.
2:9. Assume that sufficient conditions on r(s, t) = E(x(s) x(t)) are satisfied
so that the integral
 T
g(t)x(t) dt
0
exists for all T , both as a quadratic mean integral and as a sample function
integral. Show that, if
 ∞ 
|g(t)| r(t, t) dt < ∞,
0
∞
then the generalized integral 0 g(t)x(t) dt exists as a limit as T → ∞,
both in quadratic mean and with probability one.
2:10. Let (xn , yn ) have a bivariate Gaussian distribution with mean 0 , vari-
ance 1, and correlation coefficient ρn .
1
a) Show that P (xn < 0 < yn ) = 2π arccos ρn .
b) Calculate the conditional density functions for
(xn + yn ) | xn < 0 < yn , and (yn − xn ) | xn < 0 < yn .
c) Let zn and un be distributed with the in (b) derived density
 func-
tions and assume that ρn → 1 as n → ∞. Take cn = 1/ 2(1 − ρn ),
and show that the density functions for cn zn and cn un converge to
density functions f1 and f2 , respectively.
Hint: f2 (u) = u exp(−u2 /2), u > 0 is the Rayleigh density.
2:11. Let {x(t), t ∈ R} be a stationary Gaussian process with mean 0, and
with a covariance function that satisfies
−r  (t) = −r  (0) + o(|t|a ), t → 0,
for some a > 0. Define xn = x(0), yn = x(1/n), ρn = r(1/n) and use
the previous exercise to derive the asymptotic distribution of

x(1/n) − x(0) 
1/n  x(0) < 0 < x(1/n)
as n → ∞. What conclusion do you draw about the derivative at a
point with an upcrossing of the zero level? (Answer: it has a Rayleigh
distribution, not a half normal distribution.)
2:12. Find an example of two dependent normal random variables U and V
such that C(U, V ) = 0; obviously you cannot let (U, V ) have a bivariate
normal distribution.
2:13. Prove that Theorem 2:18 follows from Theorem 2:17.
Chapter 3

Crossings

3.1 Level crossings and Rice’s formula

In applications one encounters level crossings, and in particular the distribution
of the number of solutions to x(t) = u for a specified level u. This is a difficult
question in general and very few explicit results can be derived. There exists
however, a very famous formula, found by Marc Kac and Steve O. Rice [27], for
the expected number of upcrossings of a level u. We shall here state and prove
that formula. In Section 3.2 we shall use different forms of Rice’s formula to
investigate the conditional behavior of a stationary process when it is observed
in the neighborhood of a crossing of a predetermined level. Quantities like the
height and time extension of the excursion above the level will be analyzed by
means of an explicit stochastic model, the Slepian model, used first by David
Slepian 1962, [31], for a Gaussian process after zero crossings.

3.1.1 Level crossings

In practice, level crossing counting is often used as a means to describe the
variability and extremal behavior of a continuous stochastic process. For ex-
ample, the maximum of the process in an interval is equal to the lowest level
above which there exists no genuine level crossing, provided, of course, that
the process starts below that level. Since it is often easier to find the sta-
tistical properties of the number of level crossings than to find the maximum
distribution, crossing methods are of practical importance.
For sample functions of a continuous process {x(t), t ∈ R} we say that x(t)
has an upcrossing of the level u at t0 if, for some  > 0, x(t) ≤ u for all
t ∈ (t0 − , t0 ] and x(t) ≥ u for all t ∈ [t0 , t0 + ). For any interval I = [a, b],
write NI+ (x, u) for the number of upcrossings by x(t) in I ,
NI+ = NI+ (x, u) = the number of u-upcrossings by x(t), t ∈ I .
For continuous processes which have only a finite number of u-values, there
must be intervals to the left and to the right of any upcrossing point such that

57
58 Crossings Chapter 3

x(t) is strictly less than u immediately to the left and strictly greater than u
immediately to the right of the upcrossing point. Also define
NI = NI (x, u) = the number of t ∈ I such that x(t) = u.
By the intensity of upcrossings we mean any function μ+
t (u) such that

μ+ +
t (u) dt = E(NI (x, u)).
t∈I

Similarly, we define the intensity of crossings, as μt (u) if


μt (u) dt = E(NI (x, u)).
t∈I

For a stationary process, μ+ +

t (u) = μ (u) and μt (u) = μ(u) are independent
of t. In general, the intensity is the mean number of events per time unit,
calculated at time t.
In reliability applications of stochastic processes one may want to calculate
the distribution of the maximum of a continuous process x(t) in an interval I =
[0, T ]. The following approximation is then often useful, and also sufficiently
accurate for short intervals,
P ( max x(t) > u) = P ({x(0) ≤ u} ∩ {NI+ (x, u) ≥ 1}) + P (x(0) > u)
0≤t≤T

≤ P (NI+ (x, u) ≥ 1) + P (x(0) > u)

≤ E(NI+ (x, u)) + P (x(0) > u) = T · μ+ (u) + P (x(0) > u).
The upcrossing intensity μ+ (u) was found by Rice for Gaussian processes,
results which were later given strict proofs through counting methods developed
by Kac. The classical reference is [27]. We give first a general formulation and
then specialize to Gaussian processes; see [22, 30].

3.1.2 Rice’s formula for absolutely continuous processes

We present the simplest version and proof of Rice’s formula, valid for processes
{x(t), t ∈ R} with absolutely continuous sample paths1 and absolutely contin-
uous distribution with density fx(t) (u) = fx(0) (u), independent of t. For such
a process, the derivative x (t) exists almost everywhere, and the conditional
expectations
E(x (0)+ | x(0) = u) and E(|x (0)| | x(0) = u)
exist, (with x+ = max(0, x)).
1
A function x(t), t ∈ [a, b] is absolutely continuous if it is equal to the integral x(t) =
Rt
a
y(s) ds of an integrable function y(s) . This is equivalent to the requirement that for every
 > 0 there is a δ > 0 such that for every collection (a1 , b1 ), (a2 , b2 ), . . . , (an , bn ) of disjoint
Pn Pn
intervals in [a, b] with 1 (bk − ak ) < δ one has 1 |x(bk ) − x(ak )| <  . An absolutely
continuous function is always continuous and its derivative exists almost everywhere, x (t) =
y(t) , almost everywhere.
Section 3.1 Level crossings and Rice’s formula 59

Theorem 3:1 (Rice’s formula) For any stationary process {x(t), t ∈ R} with
density fx(0) (u), the crossings and up-crossings intensities are given by
 ∞
μ(u) = E(N[0,1] (x, u)) = |z|fx(0),x (0) (u, z) dz
−∞

= fx(0) (u)E(|x (0)| | x(0) = u), (3.1)

 ∞
+
μ+ (u) = E(N[0,1] (x, u)) = zfx(0),x (0) (u, z) dz
0
 +
= fx(0) (u)E(x (0) | x(0) = u). (3.2)

These expressions hold for almost any u, whenever the involved densities exist.

Before we state the short proof we shall review some facts about functions
of bounded variation, proved by Banach. To formulate the proof, write for any
continuous function f (t), t ∈ [0, 1], and interval I = [a, b] ⊂ [0, 1],

NI (f, u) = the number of t ∈ I such that f (t) = u.


Further, define the total variation of f (t), t ∈ I as sup |f (tk+1 ) − f (tk )|,
where the supremum is taken over all subdivisions a ≤ t0 < t1 < . . . tn ≤ b.

Lemma 3.1 (Banach) For any continuous function f (t), t ∈ I , the total vari-
ation is equal to
 ∞
NI (f, u) du.
−∞

Further, if f (t) is absolutely continuous with derivative f  (t) then

 ∞ 
NI (f, u) du = |f  (t)| dt.
−∞ I

Similarly, if A ⊆ R is any Borel measurable set, and 1A is its indicator func-

tion, then
 ∞ 
1A (u) NI (f, u) du = 1A (f (t)) |f  (t)| dt (3.3)
−∞ I

Proof of Rice’s formula We prove (3.1) by applying Banach’s theorem on

the stationary process {x(t), t ∈ R} with absolutely continuous, and hence
a.s. differentiable, sample paths. If x(t) has a.s. absolutely continuous sample
functions, then (3.3) holds for almost every realization, i.e.
 ∞ 
1A (u) NI (x, u) du = 1A (x(t)) |x (t)| dt
−∞ I
60 Crossings Chapter 3

Taking expectations and using Fubini’s theorem to change the order of integra-
tion and expectation, we get
  ∞
|I| μ(u) du = 1A (u) E(NI (x, u)) du
u∈A −∞



 
=E 1A (x(t)) |x (t)| dt = |I| E 1A (x(0)) |x (0)|
I

= |I| fx(0) (u) E(|x (0)| | x(0) = u) du;
u∈A

here we also used that {x(t), t ∈ R} is stationary.

Since A is an arbitrary measurable set, we get the desired result,

μ(u) = fx(0) (u) E(|x (0)| | x(0) = u)

for almost all u. The proof of (3.2) is similar. 2

3.1.3 Alternative proof of Rice’s formula

The elegant and general proof of Rice’s formula just given does not give an
intuitive argument for the presence of the factor z in the integral (3.2). The
following more pedestrian proof is closer to an explanation.

Theorem 3:2 For a stationary process {x(t), t ∈ R} with almost surely con-
tinuous sample paths, suppose x(0) and ζn = 2n (x(1/2n ) − x(0)) have a joint
density gn (u, z) which is continuous in u for all z and all sufficiently large n.
Also suppose gn (u, z) → p(u, z) uniformly in u for fixed z as n → ∞ and that
∞
gn (u, z) ≤ h(z) with 0 zh(z) dz < ∞.2 Then
 ∞
+
μ (u) = E(N[0,1] (x, u)) = zp(u, z) dz. (3.4)
0

Proof: We first device a counting technique for the upcrossings by dividing

the interval [0, 1] into dyadic subintervals, [(k − 1)/2n , k/2n ], k = 1, . . . , 2n ,
and checking the values at the endpoints. Let Nn denote the number of points
k/2n such that x((k − 1)/2n ) < u < x(k/2n ). Since x(t) has continuous
sample paths (a.s.), there is at least one u-upcrossing in every interval such
that x((k − 1)/2n ) < u < x(k/2n ), and hence

Nn ≤ N[0,1] (x, u).

Furthermore, since x(t) has a continuous distribution, we may assume that

x(k/2n ) = u for all n and k = 1, . . . , 2n . When n increases to n + 1 the
2
It is of course tempting to think of p(u, z) as the joint density of x(0), x (0) , but no
argument for this is involved in the proof.
Section 3.1 Level crossings and Rice’s formula 61

number of subintervals doubles, each interval contributing one upcrossing to

Nn will contribute at least one upcrossing to Nn+1 in at least one of the two
new subintervals. Hence Nn is increasing and it is easy to see that, regardless of
if N[0,1] (x, u) = ∞ or N[0,1] (x, u) < ∞, Nn ↑ N[0,1] (x, u) as n → ∞. Monotone
convergence implies that limn→∞ E(Nn ) = E(N[0,1] (x, u)).
Now, define
Jn (u) = 2n P (x(0) < u < x(1/2n )),
so, by stationarity,

E(N[0,1] (x, u)) = lim E(Nn ) = lim Jn (u).

n→∞ n→∞

By writing the event {x(0) < u < x(1/2n )} as

{x(0) < u < x(0) + ζn /2n } = {x(0) < u} ∩ {ζn > 2n (u − x(0))},

we have

Jn (u) = 2n P (x(0) < u, ζn > 2n (u − x(0)))

 u  ∞
n
=2 gn (x, y) dy dx.
x=−∞ y=2n (u−x)

By a change of variables, x = u − zv/2n , y = z, (v = 2n (u − x)/y), this is equal

to  ∞  1
z gn (u − zv/2n , z) dv dz,
z=0 v=0
where gn (u−zv/2n , z) tends pointwise to p(u, z) as n → ∞, by the assumptions
of the theorem. Since gn is dominated by the integrable h(z) it follows that
∞
the double integral tends to 0 zp(u, z) dz as n → ∞. 2

Remark 3:1 The proof of Rice’s formula as illustrated in Figure 3.1 shows the
relation to the Kac and Slepian horizontal window conditioning: one counts the
number of times the process passes through a small horizontal window; we will
dwell upon this concept in Section 3.2.1, page 65.

Remark 3:2 Rice’s formula can be extended to non-stationary processes, in

which case the crossings intensity is time dependent. Then the density function
fx(t),x (t) (u, z) in the integral depends on t:
 b ∞
E(N[a,b] (x, u)) = |z|fx(t),x (t) (u, z) dz dt.
a −∞

The integral  ∞
μt (u) = |z|fx(t),x (t) (u, z) dz
−∞
is the local crossings intensity at time t.
62 Crossings Chapter 3

x (t) = z
x(t + Δt) x (t) 6
zΔt
≈ x(t) + zΔt z ! "# $

x(t)

- -
t t + Δt u x(t)

Figure 3.1: Upcrossing occurs between t and t + Δt when x(t) is between

u − x (t)Δt and u. The probability is the integral of the joint density
fx(t),x (t) (u, z) over the dashed area.

3.1.4 Rice’s formula for differentiable Gaussian processes

For a Gaussian stationary process, Rice’s formula becomes particularly simple.
We know from elementary courses that for a stationary differentiable Gaussian
process {x(t), t ∈ R}, the process value, x(t), and the derivative, x (t) at
the same time point, are independent3 and Gaussian, with mean E(x(t)) =
m, E(x (t)) = 0, respectively, and variances given by the spectral moments,
V (x(t)) = r(0) = ω0 , V (x (t)) = −r  (0) = ω2 . Then
1 2 2
fx(0),x (0) (u, z) = √ e−(u−m) /2ω0 e−z /2ω2 . (3.5)
2π ω0 ω2

Simple integration of (3.1) and (3.2) gives that for Gaussian stationary pro-
cesses,
%
1 ω2 −(u−m)2 /2ω0
μ(u) = E(N[0,1] (x, u)) = e ,
π ω0
%
+ + 1 ω2 −(u−m)2 /2ω0
μ (u) = E(N[0,1] (x, u)) = e ,
2π ω0

which are the original forms of Rice’s formula. These formulas hold regardless
of whether ω2 is finite or not, so ω2 = ∞ if and only if the expected number
of crossings in any interval is infinite. This does not mean, however, that there
necessarily are infinitely many crossings, but if there is crossings, then there
may be infinitely many in the neighborhood.
3
If you have not seen this before, prove it by showing that x(t) and x (t) = limh→0 (x(t +
h) − x(t))/h are uncorrelated.
Section 3.2 Prediction from a random crossing time 63

Remark 3:3 The expected number of mean-level upcrossings per time unit in
a stationary Gaussian process is
&
1  1 ω 2 f (ω) dω
μ+ (m) = ω2 /ω0 =  ,
2π 2π f (ω) dω

and it is called the (root) mean square frequency of the process. The inverse
is equal to the long run average time distance between successive mean level

upcrossings, 1/μ+ (m) = 2π ω0 /ω2 , also called the mean period.
A local extreme, minimum or maximum, for a differentiable stochastic pro-
cess {x(t), t ∈ R} corresponds to, respectively, an upcrossing and a downcross-
ing of the zero level by the process derivative {x (t), t ∈ R}. Rice’s formula
applied to x (t) therefore gives the expected number of local extremes. For
a Gaussian process the formulas involve the fourth spectral moment ω4 =
V (x (t)) = ω 4 f (ω) dω . The general and Gaussian expressions are, respec-
tively,
 ∞ %
1 ω4
μmin = zfx ,x (0, z) dz = ,
0 2π ω2
 0 %
1 ω4
μmax = |z|fx ,x (0, z) dz = .
−∞ 2π ω2

If we combine this with Remark 3:3 we get the average number of local
maxima per mean level upcrossings,
1
 %
2π ω4 /ω2 ω0 ω4
1/α = 1 = .
2π ω2 /ω0 ω22

The parameter α is always bounded by 0 < α < 1, and it is used as an irregu-

larity measure: an α near 1 indicates a very regular process with approximately
one local maximum and minimum between mean level upcrossings. If α is near
zero one can expect many local extremes between the upcrossings.
Seen in relation to the spectrum, the parameter α can be seen as a measure
of spectral width. A spectrum with α near 1 is narrow banded, i.e. the spectral
density is concentrated to a small frequency band around a dominating center
frequency. A narrow banded process has very regular sample functions, with
slowly varying random amplitude; see Section 4.4.5 that deals with the envelope
process.

3.2 Prediction from a random crossing time

In the prediction theory briefly presented in Section 1.5.2 our concern was to
predict, as accurately as possible, the unknown future value x(t0 + τ ) of a
process {x(t), t ∈ R}, from what we know by observations available at time
64 Crossings Chapter 3

t0 . An implicit assumption has been that there is no stochastic dependence

between the choice of t0 and x(t0 + τ ).
For example, given that we have a complete record of all old values, the
best predictor, in the sense of smallest mean square error, is the conditional
expectation x(t0 + τ ) = E(x(t0 + τ ) | x(s), s ≤ t0 ). If the process is Gaussian
the predictor is linear in the observed value; for example, when we know only
the value of x(t0 ), the optimal solution that has smallest mean square error,
taken over all outcomes of x(t0 ), is
r(τ )
(t0 + τ ) = E(x(t0 + τ ) | x(t0 )) =
x x(t0 ). (3.6)
r(0)

3.2.1 Prediction from upcrossings

There are situations where the time point from which we want to predict the
future process is not a deterministic time point but a random time, determined
by the process itself. An example is an alert predictor of the water level in a
flood protection system: when the water level reaches a certain warning level
special actions are taken, such as special surveillance, more detailed prediction,
etc. Another example is a surveillance system in a medical intense care unit.
We assume throughout this section that the process {x(t), t ∈ R} has con-
tinuous sample functions. Prediction from an upcrossing time point t0 with
specified level u, shall be based on the conditional distributions, given that
an upcrossing of the level u has occurred at t0 . So, we need to find those
conditional distributions,4 in particular
u (t0 + τ ) = E(x(t0 + τ ) | x(t0 ) = u, upcrossing).
x (3.7)

3.2.1.1 A word about conditioning

A conditional expectation was
 defined in an elementary way in Section 1.2.4
as ϕ(v) = E(x | y = v) = u ufx,y (u, v)/fy (v) du. The main property of the
conditional expectation is relation (1.5), E(x) = y ϕ(y)f (y) dy and its more
refined version, 
y∈A ϕ(y)f (y) dy
E(x | y ∈ A) =  ,
y∈A f (y) dy
for every Borel set A. For example, when the density f (y) is continuous, with
A = [u − , u],
lim E(x | u −  ≤ y ≤ u) = E(x | y = u) = ϕ(u). (3.8)
→0

The meaning of this is that E(x | y = u) is the expected value of x for

those outcomes where the value of y is close to u; for more details, see [5, Ch.
4].
4
In Markov process theory one has introduced the strong Markov property to handle con-
ditioning from a time point that depends on the process.
Section 3.2 Prediction from a random crossing time 65

If {x(t), t ∈ R} is a stationary process, take y = x(t0 ) and x = x(t0 + τ ).

Then

P (x(t0 + τ ) ≤ v | x(t0 ) = u)
= lim P (X(t0 + τ ) ≤ v | u −  ≤ x(t0 ) ≤ u), (3.9)
→0

E(x(t0 + τ ) | x(t0 ) = u)
= lim E(X(t0 + τ ) | u −  ≤ x(t0 ) ≤ u), (3.10)
→0

calculated at time t0 , and (3.10) gives the best predictor of the future value
x(t0 + τ ) in the sense that it minimizes the squared error taken as an average
over all the possible outcomes of x(t0 ). By ”average” we then mean expected
value as well as an empirical average over many realizations observed at the
fixed predetermined time t0 , chosen independently of the process.

3.2.1.2 Empirical limits and horizontal window conditioning

We now consider prediction from the times of upcrossings of a fixed level u. This
differs from the previous type of conditioning in that the last observed value
of the process is known, and the time points are variable. The interpretation
of ”average future value” is then not clear at this moment and has to be made
precise. Obviously, what we should aim at is a prediction method that works
well on the average, in the long run, for all the u-level upcrossings we observe
in the process. Call these upcrossings time points tk > 0. To this end, define
the following distribution.

Definition 3:1 For a stationary process {x(t), t ∈ R}, the (long run, ergodic)
conditional distribution of x(t0 + ·) after u-upcrossing at t0 is defined as

P u (A) =  P (x(t0 + ·) ∈ A | x(t0 ) = u, upcrossing)

#{tk ; 0 < tk < T and x(tk + ·) ∈ A)}
= lim . (3.11)
T →∞ #{tk ; 0 < tk < T }

Thus, in P u (A) is counted all those u-upcrossings tk for which the process,
taken with tk as new origin, satisfies the condition given by A.

The definition makes sense only if the limit exists, but, as we shall prove
in Chapter 5, the limit exists for every stationary process {x(t), t ∈ R}, but it
may be random. If the process is ergodic the limit is non-random and it defines
a proper distribution on C for a (non-stationary) stochastic process.
The empirical long run distribution is related to Kac and Slepian’s hori-
zontal window conditioning, [19]. For a stationary process {x(t), t ∈ R}, the
(horizontal window) conditional distribution of x(t0 + ·) after u-upcrossing at
66 Crossings Chapter 3

5 Seven excursions above level u=1.5 ...

−5

0 20 40 60 80 100

5 ... translated to the origin.

−5

0 2 4 6 8 10

Figure 3.2: Excursions above u = 1.5 contribute to the distribution P 1.5 (·).

t0 is
P hw (A) =  P (x(t0 + ·) ∈ A | x(t0 ) = u, upcrossing in h.w.sense) (3.12)
= lim P (x(t0 + ·) ∈ A | x(s) = u, upcrossing, some s ∈ [t0 − , t0 ]).
→0

It is easy to show that the two distributions just defined are identical, P u (A) =
P hw (A), for every (ergodic) stationary process such that there are only a finite
number of u-upcrossings in any finite interval; see [9].
With point process terminology one can look at the upcrossings as a se-
quence of points in a staionary point process and the shape of the process
around the upcrossing as a mark, attached to the point. The conditional distri-
bution of the shape is then treated as a Palm distribution in the marked point
process; see [9].
The term, horizontal window condition, is natural, since the process has
to pass a horizontal window at level u somewhere near t0 . In analogy, the
condition in (3.9) is called vertical window condition, since the process has to
pass a vertical window near v exactly at time t0 .
The distribution P u can be found via its finite-dimensional distribution
functions. Take s = (s1 , . . . , sn ), v = (v1 , . . . , vn ), write
x(s) ≤ v for x(sj ) ≤ vj , j = 1, . . . , n,
and define
N[0,T ] (x, u; s, v) = #{tk ; 0 ≤ tk ≤ T, and x(tk + s) ≤ v},
as the number of u-upcrossings in [0, T ] which are such that the process, at
N ] (x,u;s,v)
each of the times sj after the upcrossing is less than vj . Thus [0,T
N (x,u) , v ∈
[0,T ]
Section 3.2 Prediction from a random crossing time 67

Rn is the empirical distribution function for the process at times sj after u-

upcrossings.

Theorem 3:3 If {x(t), t ∈ R} is ergodic, with A = {y ∈ C; y(s) ≤ v},

 
E N [0,1] (x, u; s, v)
P u (A) =   (3.13)
E N[0,1] (x, u)
∞
zfx(0),x (0) (u, z) P (x(s) ≤ v | x(0) = u, x (0) = z) dz
= z=0 ∞ . (3.14)
z=0 zfx(0),x (0) (u, z) dz

Proof: We need a result from Chapter 5, namely that for an ergodic process
(with probability one)
N[0,T ] (x, u)  
→ E N[0,1] (x, u)
T
N[0,T ] (x, u; s, v)  
→ E N[0,1] (x, u; s, v)
T
as T → ∞. This gives (3.13).
The proof of (3.14) is analogous to that of Theorem 3:2 and can be found
in [22, Ch. 10] under the (unnecessarily strict) condition that {x(t), t ∈ R} has
continuously differentiable sample paths. 2

Noting that
zfx(0),x (0) (u, z)
pu (z) =  ∞ , z ≥ 0, (3.15)
ζ=0 ζfx(0),x (0) (u, ζ) dζ

is a density function in z we can introduce a random variable ζ with this

density. Then (3.14) can be formulated as
 
P u (A) = Eζ P (x(s) ≤ v | x(0) = u, x (0) = ζ) . (3.16)

The interpretation is that the distribution of the process values at times s

after u-upcrossings is a mixture over the random slope ζ at the upcrossing
point, of the ordinary conditional distributions of x(s) given that x(0) = u
and x (0) = ζ , when ζ has density pu (z).

3.2.2 The Slepian model

Theorem 3:3 presents the conditional distribution of a stationary process in the
neighborhood of upcrossings of the fixed level u. The definition in the form
of integrals in (3.14) is not very informative as it is. However, for Gaussian
processes one can construct an explicit and simple process that has exactly this
distribution, and lends itself to easy simulation, numerical calculations, and
asymptotic expansions.
68 Crossings Chapter 3

Definition 3:2 Let {x(t), t ∈ R} be a stationary and ergodic stochastic process

and fix a level u, such that x(t) has a finite number of u-upcrossings in any
finite interval. A Slepian model process for {x(t), t ∈ R} after u-upcrossings
is any stochastic process {ξu (t), t ∈ R} that has its distributions given by P u
in (3.14). In particular, its finite-dimensional distribution are given by
 ∞
 
P (ξu (s) ≤ v) = pu (z) P x(s) ≤ v | x(0) = u, x (0) = z dz.
0

The Slepian model is a stochastic model for individual excursions after a

level upcrossing, and its distribution is equal to the distribution of the sample
functions illustrated in the lower diagram in Figure 3.2. More complex Slepian
models can be formulated for other crossing problems, for example the process
behavior after a local maximum or minimum, since these are defined as down-
crossing or upcrossing of the zero level by the derivative process {x (t), t ∈ R}.
Every Slepian model has two elements, which depend on the type of cross-
ing problem: the long run distribution of the gradient at the instances of the
crossings, and the conditional distribution of the process given its value and
the value of the gradient.
Typical problems that can be analyzed by a Slepian process are

• Prediction after crossing: What is the best predictor of the process

a time τ after one has observed a level u upcrossing?

• Excursion shape: How high above level u will an excursion extend,

and how long does it take before the process returns below the level?

• Crest shape: What is the shape of the process near its local maxima?

u (t0 + τ )
We can immediately solve the first problem: the best predictor x
after u-upcrossings is the expectation of the Slepian model:

u (t0 + τ ) = E (ξu (τ )) ,
x (3.17)

in the sense that the average of (x(tk + τ ) − a)2 , when tk runs over all u-
upcrossings, takes it minimum value when a = E (ξu (τ )).

3.2.2.1 An explicit Slepian model for crossings in a Gaussian process

The conditional distribution of {x(t), t ∈ R} after u-upcrossing is particularly
simple in the Gaussian case, and the Slepian model can be expressed in a very
explicit way.
We have to find the density pu (z) and the conditional distributions of x(s)
in (3.16). For Gaussian processes, formula (3.5) holds, and with mean μ = 0,
it says
1 2 2
fx(0),x (0) (u, z) = √ e−u /2ω0 e−z /2ω2 .
2π ω0 ω2
Section 3.2 Prediction from a random crossing time 69
2 /2ω
Canceling √ 1 e−u 0 in (3.15) we get
2πω0

z −z 2 /ω2
pu (z) = e , z ≥ 0, (3.18)
ω2

i.e. the slope at upcrossings has a Rayleigh distribution with parameter ω2 ,

regardless of the level u.5
Next, we need the conditional distribution of x(s) given x(0) = u, x (0) = z
and average over ζ = z with density pu (z). Since a conditional distribution in
a multivariate normal distribution is still a normal distribution, we need only
to find the conditional mean E(x(s) | x(0) = u, x (0) = z) and the conditional
covariances Cov(x(s1 ), x(s2 ) | x(0) = u, x (0) = z), and these were given in
Section 1.5.1, equations (1.9, 1.11).
Take ξ = (x(s1 ), x(s2 )), η = (x(0), x (0)), and calculate the joint covari-
ance matrix of (ξ, η) from the covariance function r(t) for {x(t), t ∈ R}. By
Theorem 2:13 it is
⎛ ⎞
r(0) r(s2 − s1 ) r(s1 ) −r  (s1 )

⎜r(s1 − s2 ) r(0) r(s ) −r  (s )⎟ Σ Σ

Σ=⎜ ⎝ r(s1 )
2 2 ⎟= ξξ ξη
.
r(s2 ) r(0) 0 ⎠ Σηξ Σηη
−r  (s1 ) −r  (s2 ) 0 −r  (0)

Use the notation ω0 = r(0), ω2 = −r  (0) and remember that mξ = mη =

0. Then we get the conditional expectation and covariance matrix given η =
y = (u, z) as
 
ur(s )/ω − zr  (s )/ω
1 0 1 2
E(ξ | η = y) = Σξη Σ−1 
ηη y = ,
ur(s2 )/ω0 − zr  (s2 )/ω2
 
−1 rκ (s1 , s1 ) rκ (s1 , s2 )
Σξξ|η = Σξξ − Σξη Σηη Σηξ = ,
rκ (s2 , s1 ) rκ (s2 , s2 )

say. Here

rκ (s1 , s2 ) = r(s2 − s1 ) − r(s1 )r(s2 )/ω0 − r  (s1 )r  (s2 )/ω2 (3.19)

is the covariance function for a non-stationary process.

Note the structure of rκ (s1 , s2 ): The first term is the unconditional co-
variance function of the process and the two other terms are the changes in
covariance that is obtained by the knowledge of the uncorrelated x(0) and
x (0). When s1 or s2 tend to infinity, the reduction terms go to 0 and the
influence of the conditioning vanishes.
We have now found the explicit structure of the Slepian model in the Gaus-
sian case; we formulate it as a theorem.
5
This is the solution to Exercise 2:11.
70 Crossings Chapter 3

Theorem 3:4 a) The Slepian model for a Gaussian process {x(t), t ∈ R} after
u-upcrossings has the form
ur(t) ζr  (t)
ξu (t) = − + κ(t), (3.20)
ω0 ω2
2
where ζ has the Rayleigh density pζ (z) = (z/ω2 )e−z /(2ω2 ) , and {κ(t), t ∈ R}
is a non-stationary Gaussian process, independent of ζ , with mean zero and
covariance function rκ (s1 , s2 ) given by (3.19).
b) In particular, the best prediction of x(tk + τ ) taken over all u-upcrossings

tk , is obtained by taking E(ζ) = πω2 /2 and κ (τ ) = 0 in (3.20) to get
%
ur(τ ) E(ζ)r  (τ ) u π 
 (t0 + τ ) =
x u
− = r(τ ) − r (τ ). (3.21)
ω0 ω2 ω0 2ω2
We have now found the correct way of taking the apparent positive slope at
a u-upcrossing into account in predicting the near future. Note that the simple
formula (3.6),
u
(t0 + τ ) = E(x(t0 + τ ) | x(t0 ) = u) =
x r(τ ),
ω0
in the Gaussian case, lacks the slope term. Of course, the slope at an upcrossing
is always positive, but it is perhaps intuitively obvious that the observed slopes
at the upcrossings are “more positive” than that. The slope = derivative of a
stationary Gaussian process is normal with mean zero, but we do not expect
slopes at fixed level upcrossings to have a half-normal distribution with a mode
(= most likely values) at 0. The r  -term in (3.21) tells us how we shall take
this “sample bias” into account.
The prediction of the slope is the expectation of the Rayleigh variable ζ . If
the slope at the u-upcrossing is observed and used in the prediction, then the
difference between the two approaches disappears; see [23].

Example 3:1 To illustrate the efficiency of the Slepian model, we shall analyse
the shape of an excursion above a very high level u in a Gaussian process, and
then expand the Slepian model ξu (t) in a Taylor series as u → ∞. It will turn
out that the length and height of the excursion will both be of the order u−1 ,
so we normalize the scales of ξu (t) by that factor. Using
t2 t
r(t/u) = ω0 − ω2 2
(1 + o(1)), r  (t/u) = −ω2 (1 + o(1)),
2u u
as t/u → 0, and that κ(t/u) = o(t/u), and omitting all o-terms, we get


r(t/u) r  (t/u) ω 2 t2
u {ξu (t/u) − u} = u u −1 −ζ + κ(t/u) ≈ ζt − .
ω0 ω2 2ω0
Thus, the excursion above a high leve u takes the form of a parabola with
ζ 2 ω0
height 2uω 2
and length 2ω 0ζ
uω2 . It is easy to check that the normalized height of
the excursion above u has an exponential distribution.
Section 3.2 Prediction from a random crossing time 71

Remark 3:4 One should be aware that a Slepian model as it is described here
represents the “marginal distribution” of the individual excursions above the
defined level u. Of course one would like to use it also to analyse the dependence
there may exist between successive excursions in the original process x(t), and
this is in fact possible. For example, suppose we want to find how often it
happens that two successive excursions both exceed a critical limit T0 in length.
Then, writing τ1 = inf{τ > 0; ξu (τ ) = u, upcrossing} for the first u-upcrosing
in ξu (t) strictly on the positive side, one can calculate

P (ξu (s) > u, for 0 < s < T0 , and ξu (τ1 + s) > u, for 0 < s < T0 ).

3.2.2.2 A Slepian model around local maxima in a Gaussian process

A local maximum for x(t) is a zero-downcrossing point for the derivative x (t),
and the second derivative x (t) at these local maxima has a negative Rayleigh
distribution. A Slepian model for the derivative after local maxima, therefore
has the same structure as the level crossing model, with rx (t) replaced by
rx (t) = −rx (t). The time difference between maximum and minimum can
therefore be calculated as previously.
If we want the distribution of the height difference between the maximum
and the following minimum, we need a more elaborate Slepian model, since
now also the height of the maximum is random, not only the curvature. The
reader is encouraged to prove the following theorem, copying Theorem 3:3 with
analogous notation, now with tk for the times of local maxima

Theorem 3:5 If {x(t), t ∈ R} is twice differentiable and ergodic the long run
empirical distribution of x(tk + s) around local maxima is equal to

P1max (A)
0
z=−∞ |z|fx (0),x (0) (0, z)P (x(s) ≤ v | 0, z) dz
= 0 ,
−∞ |z|fx (0),x (0) (0, z) dz

where A = {y ∈ C; y(s) ≤ v} and

P (x(s) ≤ v | 0, z) = P (x(s) ≤ v | x (0) = 0, x (0) = z).

An explicit representation of the model process for a Gaussian process is the

r  (t)
ξ1max (t) = ζ1max + Δ1 (t), (3.22)
ω4
where ζ1max has a negative Rayleigh distribution with density

|z| −z 2 /(2ω4 )
pmax
1 (z) = e , z < 0, (3.23)
ω4
72 Crossings Chapter 3

and the non-stationary Gaussian process Δ1 (t) is independent of ζ1max , has

mean 0 and the covariance function,
r  (s1 )r  (s2 ) r  (s1 )r  (s2 )
rΔ1 (s1 , s2 ) = r(s1 − s2 ) − − . (3.24)
ω2 ω4
Since Δ1 (0) is normal with mean 0 and variance rΔ1 (0, 0) = ω0 − ω22 /ω4 ,
we see in particular that the distribution of the height of a local maximum is
the same as the distribution of
ω2
+ Δ1 (0)
ξ max (0) = −ζ1max
ω4
√  
= ω0 1 − 2 · Rayleigh +  · Normal , (3.25)

with standard Rayleigh and normal variables, illustrating the relevance of the
√ 
spectral width parameter α = 1 − 2 = ω22 /(ω0 ω4 ).
Theorem 3:5 is the basis for numerical calculations of wave characteristic
distributions like height and time difference between local maxima and minima,
as they can be made by the routines in the Matlab package WAFO, [34].
The model (3.22) contains an explicit function r  (t)/ω4 with a simple random
factor, representing the Rayleigh distributed curvature at the maximum, plus
a continuous parameter Gaussian process. The numerical procedures work by
successively replacing the continuous parameter process by explicit functions
multiplied by random factors. We illustrate now the first steps in this procedure.

The model (3.22) contains the random curvature, and it is the simplest
form of the Slepian model after maximum. There is nothing that prevents us
to include also the random height of the local maximum in the model. We have
seen in (3.25) how the height and the curvature depend on each other, so we
can build an alternative Slepian model after maximum that explicitly includes
both the height of the maximum and the curvature.
To formulate the extended model we define three functions, A(t), B(t), C(t),
by

E(x(t) | x(0) = u, x (0) = y, x (0) = z)

= uA(t) + yB(t) + zC(t)
ω4 r(t) + ω0 ω2 r  (t) r  (t) ω2 r(t) + ω0 r  (t)
=u − y + z .
ω0 ω4 − ω22 ω2 ω0 ω4 − ω22

The conditional covariance between x(s1 ) and x(s2 ) are found from the same
theorem, and the explicit expression is given in the following Theorem 3:6.
As we have seen in Section 2.4.3 the derivative x (0) is uncorrelated with
both x(0) and x (0), but x(0) and x (0) are correlated. To formulate the
effect of observing a local maximum we will first introduce the crest height,
Section 3.2 Prediction from a random crossing time 73

x(0), and then find the conditional properties of x (0) given x(0) and x (0).
We use Theorem 2:13 and define the function
Cov(x(t), x (t) | x(0), x (0)) r  (t) + (ω2 /ω0 )r(t)
b(t) =  =  .
V (x (0) | x(0), x (0)) ω4 − ω22 /ω0

Theorem 3:6 If {x(t), t ∈ R} is twice differentiable and ergodic the long run
empirical distribution of x(tk + s) around local maxima is equal to

P2max (A)
∞ 0
u=−∞ z=−∞ |z|fx(0),x (0),x (0) (u, 0, z)P (x(s) ≤ v | u, 0, z) dz du
= ∞ 0 ,
u=−∞ −∞ |z|fx(0),x (0),x (0) (u, 0, z) dz du

where A = {y ∈ C; y(s) ≤ v} snd

P (x(s) ≤ v | u, 0, z) = P (x(s) ≤ v | x(0) = u, x (0) = 0, x (0) = z).

An explicit representation of the model process for a Gaussian process is the

ξ2max (t) = η2max A(t) + ζ2max C(t) + Δ2 (t), (3.26)

where the random (η2max , ζ2max ) has the two-dimensional density (with normal-
izing constant c),
 
ω0 z 2 + 2ω2 uz + ω4 u2
p2 (u, z) = c|z| exp −
max
, −∞ < u < ∞, z < 0.
2(ω0 ω4 − ω22 )

The process Δ2 (t) with mean zero is non-stationary Gaussian, independent of

(η2max , ζ2max ), and has covariance function

r(s)r(t) r  (s)r  (t)

rΔ2 (s, t) = r(s − t) − − − b(s)b(t).
ω0 ω2

3.2.3 Excursions and related distributions

What is the shape, height and extension, of an arbitrary excursion above a
fixed level u for a stationary process? An example of this type of problem is
when x(t) is an electrical potential which may not become too high, and should
 t +T
always stay below a certain level. The integral tkk k (x(t) − u) dt between an
excursion and the critical level represents the amount of extra electrical charges
that are transmitted between the upcrossing at tk and the next downcrossing
at tk + Tk .
One of the advantages of a Slepian model is that it lends itself to efficient
numerical calculations of important quantities related to crossings and maxima.
The structure of the model is such that the random variables that represent
slope, height, and curvature in the crossings and crest models are easy to handle
74 Crossings Chapter 3

numerically. The only problems are the residual processes which require infinite
dimensional probabilities to be calculated. To overcome this in a numerical
algorithm one can use a successive conditioning technique that first introduces
the value of the normal residual at a single point, say, κ(s1 ), and include that
as a separate term in the model. The residual process will be correspondingly
reduced and the procedure repeated.
For numerical calculations of interesting crossings probabilities one can trun-
cate the conditioning procedure when sufficient accuracy is attained. This ap-
proximation technique is called regression approximation in crossing theory.

3.2.3.1 Length of excursions

The Slepian process ξu (t) has a u-upcrossing at t = 0 and we denote by T
the time of its first downcrossing of the same level, so T is the length of an
excursion. Since T > t if and only if ξu (s) stays above the level u in the entire
interval 0 < s < t, we can express the probability P (T > t) by means of the
indicator function

zr  (s)
1, if ur(s)
ω0 − ω2 + κ(s) > u, for all s ∈ (0, t),
Iz (κ, t) =
0, otherwise.

The result is,

 ∞
P (T > t) = pu (z) · E(Iz (κ, t)) dz, (3.27)
z=0

where pu (z) is the Rayleigh density for the derivative at u-upcrossing, and the
expectation
 

ur(s) zr  (s)
E(Iz (κ, t)) = P inf − + κ(s) > u
0<s<t ω0 ω2

is an infinite dimensional normal probability. That probability has to be cal-

culated numerically by special software. By means of routines in the Matlab
package WAFO, [34], which is available on the departments homepage, one can
calculate the distribution with very high accuracy. Figure 3.3 shows the excur-
sion length densities for a realistic water wave process with a common Jonswap
North Sea spectrum.
The probability density function for the excursion time T is of course minus
the derivative of P (T > t). It can also be expressed by means of Durbin’s
formula,
 
fT (t) = fξu (t) (u) E I{ξu (s) > u, 0 < s < t} · (−ξu (t)− ) | ξu (t) = u ,

where ξu (t)− = min(0, ξu (t)) is the negative part of the derivative. The expec-
tation can be calculated by mean of algorithms from WAFO, by means of the
regression technique with successive conditioning on the residual process.
Section 3.2 Prediction from a random crossing time 75

Excusion time densities

0.4

0.35
Levels (from left to right):
0.3 2, 1, 0, −1

0.25

0.2

0.15

0.1

0.05

0
0 5 10 15
period [s]

Figure 3.3: Probability densities for excursions above u = −1, 0, 1, 2 for pro-
cess with North Sea wave spectrum Jonswap.

3.2.3.2 Wave shape

The distribution of wave characteristics such as drop in height and time dif-
ference between a local maximum and the next local minimum can be derived
from the Slepian models in Theorem 3:5 or Theorem 3:6.
First consider the model (3.22),

r  (t)
ξ1max (t) = ζ1max + Δ1 (t),
ω4
which completely describes the stochastic properties of the shape around maxi-
mum. The simplest, zero order, approximation is to delete the residual process

Δ1 (t) completely, only keeping the curvature dependent term ζ1max r ω(t) . By
 4
max
replacing ζ1 by its average − ω4 π/2. From this we can, for example, get
the average shape, as
√ r  (t)
ξ
max (t) = − ω α
0 .
ω2
The zero order approximation is usually too crude to be of any use. A
better approximation is obtained from the model (3.26), which also includes
the (random) height at the maximum point,

ξ2max (t) = η2max A(t) + ζ2max C(t) + Δ2 (t),

76 Crossings Chapter 3

and define the random variable T as the time of the first local minimum of
ξ2max (t), t > 0. The height drop is then H = ξ2max (0) − ξ2max (T ) and we ask for
the joint distribution of T and H .
Using the fact that A(0) = 1, C(0) = 0 and ξ2max (T ) = 0, we get the
following relations that need to be satisfied,

η2max A (T ) + ζ2max C  (T ) + Δ2 (T ) = 0,

η2max + Δ2 (0) − (η2max A(T ) + ζ2max C(T ) + Δ2 (T )) = H.

We now describe the regression approximation of order 1, which is obtained

by deleting all of the residual process terms. The relations will then be

η2max A (T r ) + ζ2max C  (T r ) = 0, (3.28)

η2max − (η2max A(T r ) + ζ2max C(T r )) =H ,r
(3.29)

where we write T r , H r for the approximative time and height variables.

To write the solution in a form that can be generalized to more complicated
problems, define

1 − A(t) C(t)
G(t) =
A (t) C  (t)
and write the equations (3.28) and (3.29) as ( T for matrix transpose),

G(T r )(η2max ζ2max )T = (H r T r )T .

If det G(T r ) = 0 we get from (η2max ζ2max )T = G(T r )−1 (H r 0)T that the vari-
ables with known distribution (η2max and ζ2max ) are simple functions of the
variables with unknown distribution,

η2max = H r p(T r ) q(T r ), ζ2max = H r q(T r ),

where
−C  (t) −A (t)
p(t) = , q(t) = .
A (t) (1 − A(t))C  (t) − A (t)C(t)
We want the density at the point T r = t, H r = h; let ξ(t, h), ζ(t, h) be
the corresponding solution and define the indicator function I(t, h) to be 1
if the approximating process ξ(t, h)A(s) + ζ(t, h)(s) is strictly decreasing for
0 < s < t.
The Jacobian for the transformation is J(t, h) = hp (t)q(t)2 , and therefore
the density of T r , H r is

fT r ,H r (t, h) = fξmax ,ζmax (hp(t)q(t), hq(t)) · |J(t, h)| I(t, h)

= const I(t, h) h2 |q 2 (t)3 p (t)|



1 2 2 4 2 2 ε2
× exp − 2 h q(t) (Tm /π) ((π/Tm ) p(t) + 1) + .
2ε 1 − ε2
Section 3.2 Prediction from a random crossing time 77

6 12

Level curves at:

4 10
0.01083
0.02707
0.05414
2 8 0.1083
0.2166
0.4331
0.5414
0 6 0.8121

−2 4

−4 2

−6 0
0 50 100 150 200 250 0 2 4 6 8

Figure 3.4: Probability density for T, H for process with North Sea wave spec-
trum Jonswap together with 343 observed cycles.

This form
 of the T, H distribution is common in the technical literature, where
Tm = π ω2 /ω4 is called the mean half wave period. Note that dependence
) on
ω2
the spectrum is only through the spectral width parameter  = 1 − ω0 ω2 4 =
√
1 − α2 .
This first order approximation of the T, H -density is not very accurate, but
it illustrates the basic principle of the regression approximation. The WAFO
toolbox, [34], contains algorithms for very accurate higher order approxima-
tions. Figure 3.4 shows the result for a process with a common North Sea
Jonswap spectrum.
78 Crossings Chapter 3

Exercises
3:1. Prove that κ(0) = κ (0) = 0 in the Slepian model after upcrossing.

3:2. Formulate conditions on the covariance function rx (t) that guarantee that
the residual process κ(t) has differentiable sample paths.

3:3. Complete the proof of Theorem 3:5.

Chapter 4

Spectral- and other

representations

This chapter deals with the spectral representation of weakly stationary pro-
cesses – stationary in the sense that the mean is constant and the covariance
Cov(x(s), x(t)) only depends on the time difference t − s. For real-valued
Gaussian processes, the mean and covariance function determines all finite-
dimensional distributions, and hence the entire process distribution. However,
the spectral representation requires complex-valued processes, and then one
needs to specify also the correlation structure between the real and the imag-
inary part of the process. We therefore start with a summary of the basic
properties of complex-valued processes, in general, and in the Gaussian case.
We remind the reader of the classical memoirs by S.O. Rice, [27], which can be
recommended to anyone with the slightest historical interest. That work also
contains many old references.

4.1 Complex processes and their covariance func-

tions
4.1.1 Stationary processes
A complex-valued process
x(t) = y(t) + iz(t)

is strictly stationary if all 2n-dimensional distributions of

y(t1 + τ ), z(t1 + τ ), . . . , y(tn + τ ), z(tn + τ )

are independent of τ . It is called weakly stationary or second order stationary

if E(x(t)) = m is constant, and
E(x(s) · x(t)) = r(s − t) + |m|2

79
80 Spectral- and other representations Chapter 4

only depends on the time difference s − t. The covariance function


r(τ ) = E (x(s + τ ) − m)(x(s) − m)

is Hermitian, i.e.
r(−τ ) = r(τ ).

For real-valued processes, the covariance function r(τ ) determines all co-
variances between x(t1 ), . . . , x(tn ),
⎛ ⎞
r(0) r(t1 − t2 ) . . . r(t1 − tn )
⎜ .. .. .. .. ⎟
Σ(t1 , . . . , tn ) = ⎝ . . . . ⎠ (4.1)
r(tn − t1 ) r(tn − t2 ) . . . r(0)
⎛ ⎞
V (x(t1 )) C(x(t1 ), x(t2 )) . . . C(x(t1 ), x(tn ))
⎜ .. .. .. .. ⎟
=⎝ . . . . ⎠.
C(x(tn ), x(t1 )) C(x(tn ), x(t2 )) . . . V (x(tn ))

4.1.2 Non-negative definite functions

It is a unique characteristic property of a covariance function that is it non-
negative definite in the following sense: Let t1 , . . . , tn be any finite set of time
points, and take arbitrary complex numbers a1 , . . . , an . Then, for simplicity
assuming E(x(t)) = 0,
⎛ ⎞
n n
aj ak r(tj − tk ) = E ⎝ aj x(tj ) ak x(tk )⎠ (4.2)
j,k j,k
 2
 n 
 

= E aj x(tj ) ≥ 0. (4.3)
 j=1 

Theorem 4:1 Every non-negative definite, possibly complex, function r(τ ) is

the covariance function for a strictly stationary Gaussian process. Thus, the
class of covariance functions is equal to the class of non-negative definite func-
tions.
Proof: We have to show that if r(τ ) is non-negative definite then there are
finite-dimensional distributions for a process x(t) with E(x(t)) = 0, such that

r(τ ) = E(x(s + τ )x(s)).

By Kolmogorov’s existence theorem, see Appendix A, we only have to show

that for every selection of time points t1 , . . . , tn , there is an n-dimensional
Section 4.1 Complex processes and their covariance functions 81

distribution with mean 0 and covariances given by Σ(t1 , . . . , tn ) as defined by

(4.1), and such that the obtained family of distributions forms a consistent
family, i.e. for example FX,Y (x, y) = FY,X (y, x) and FX,Y (x, ∞) = FX (x).
If r(t) is a real function and u = (u1 , . . . , un ) a real vector, consider the
non-negative quadratic form

Q(u) = uj uk r(tj − tk ).
j,k

Then we recognize exp(−Q(u)/2) as the characteristic function for an n-variate

normal distribution with covariance matrix Σ(t1 , . . . , tn ), and we have found a
distribution with the specified properties.
If r(t) is complex, with r(−t) = r(t), there are real functions p(t) = p(−t),
q(t) = −q(−t) such that
r(t) = p(t) + iq(t).
Take aj = uj −ivj and consider the non-negative quadratic form in real variables
(u, v) = (u1 , . . . , un , v1 , . . . , vn ),

Q(u, v) = aj ak r(tj − tk )
j,k

= (uj − ivj )(uk + ivk )(p(tj − tk ) + iq(tj − tk ))
j,k

= {p(tj − tk )(uj uk + vj vk ) − q(tj − tk )(uj vk − uk vj )} ;
j,k

note that the imaginary part vanishes since Q is assumed to be non-negative,

and hence real. Similarly, as in the real case,
⎛ ⎞

exp(−Q(u, v)/2) = E ⎝exp(i (uj yj + vj zj ))⎠
j

is the characteristic function of a 2n-dimensional normal variable

(y1 , . . . , yn , z1 , . . . , zn )
with the specified properties:
E(yj yk ) = E(zj zk ) = p(tj − tk )
E(yj zk ) = −E(yk zj ) = −q(tj − tk ).
√
With xj = (yj + izj )/ 2, we have E(xj ) = 0 and
E(xj xk ) = p(tj − tk ) + iq(tj − tk ) = r(tj − tk ),
as required. Since the Gaussian distribution of the y - and z -variables is de-
termined by the covariances, and these depend only on the time difference, the
process is strictly stationary. 2
82 Spectral- and other representations Chapter 4

4.1.3 Strict and weak stationarity

Since the first two moments determines a real normal distribution, it is clear
that each weakly (covariance) stationary normal process is strictly stationary.
For complex processes matters are not that easy, and one has to impose two
stationarity conditions in order to guarantee strict stationarity.

Theorem 4:2 A complex normal process x(t) = y(t) + iz(t) with mean zero is
strictly stationary if and only if the two functions

r(s, t) = E(x(s)x(t))

q(s, t) = E(x(s)x(t)),

only depend on t − s.
Proof: To prove the ”if” part, express r(s, t) and q(s, t) in terms of y and z ,
r(s, t) = E(y(s)y(t) + z(s)z(t)) + iE(z(s)y(t) − y(s)z(t)),
q(s, t) = E(y(s)y(t) − z(s)z(t)) + iE(z(s)y(t) + y(s)z(t)).
Since these only depend on t−s, the same is true for the sums and differences of
their real and imaginary parts, i.e. for E(y(s)y(t)), E(z(s)z(t)), E(z(s)y(t)),
E(y(s)z(t)). Therefore, the 2n-dimensional distribution of
y(t1 ), . . . , y(tn ), z(t1 ), . . . , z(tn )
only depends on time differences, and x(t) is strictly stationary. The converse
is trivial. 2

Example 4:1 If x(t) is a real and stationary normal process, and μ is a con-
stant, then
x∗ (t) = eiμt x(t)
is a weakly, but not strictly, stationary complex normal process,

E(x∗ (s)x∗ (t)) = eiμ(s−t) E(x(s)x(t)),

E(x∗ (s)x∗ (t)) = eiμ(s+t) E(x(s)x(t)).

4.2 Bochner’s theorem and the spectral distribution

4.2.1 The spectral distribution
We have seen that covariance functions for stationary processes are character-
ized by the property of being non-negative definite. From elementary courses
we also know that covariance functions are Fourier-transforms of their spectral
distributions. We shall now formulate this and prove this statement.
Section 4.2 Bochner’s theorem and the spectral distribution 83

Theorem 4:3 (Bochner’s theorem) A continuous function r(t) is non-negative

definite, and hence a covariance function, if and only if there exists a non-
decreasing, right continuous, and bounded real function F (ω) such that
 ∞
r(t) = eiωt dF (ω).
−∞

The function F (ω) is the spectral distribution function of the process, and it
has all the properties of a statistical distribution function except that F (+∞) −
F (−∞) = r(0) need not be equal to one. The function F (ω) is defined only up
to an additive constant, and one usually takes F (−∞) = 0.

Proof: The ”if” part is clear, since if r(t) = exp(iωt) dF (ω), then
  
zj zk r(tj − tk ) = zj zk eiωtj · e−iωtk dF (ω)
j,k j,k
 
= zj eiωtj zk eiωtk dF (ω)
j,k
 2
  
 
=  zj e  dF (ω) ≥ 0.
iωtj

 j 

For the ”only if” part we shall use some properties of characteristic func-
tions, which are proved elsewhere in the probability course. We shall show that,
given r(t), there exists a proper distribution function F∞ (ω) = F (ω)/F (∞)
such that

F∞ (∞) − F∞ (−∞) = 1,

r(t)
eiωt dF∞ (ω) = .
r(0)
To this end, take a real A > 0, and define
 A A
1
g(ω, A) = r(t − u) e−iω(t−u) dt du
2πA 0 0
1 
= lim r(tj − tk )e−iωtj e−iωtk Δtj Δtk
2πA
j,k

1 
= lim r(tj − tk )Δtj e−iωtj · Δtk e−iωtk ≥ 0,
2πA
j,k

since r(t) is non-negative definite. (Here, of course, the tj define a subdivision

of the interval [0, A].) Going to the limit, g will give the density of the desired
spectral distribution.
84 Spectral- and other representations Chapter 4

Before we proceed, we express g(ω, A) as

 A A
1
g(ω, A) = r(t − u)e−iω(t−u) dt du
2πA 0 0
 A
 ∞
1 |t| −iωt 1
= 1− r(t)e dt = μ(t/A)r(t)e−iωt dt,
2π −A A 2π −∞

where 
1 − |t| for |t| ≤ 1
μ(t) =
0 otherwise.

The proof will now proceed in three steps:

Step 1) Prove that g(ω, A) ≥ 0 is integrable, and


g(ω, A) dω = r(0), (4.4)
ω

so g(·, A)/r(0) is a regular statistical density function.

Step 2) Show that

 ∞
|t| r(t) g(ω, A) itω
1− = e dω, (4.5)
A r(0) −∞ r(0)

so the function 1 − |t|
A r(t)/r(0) for |t| ≤ A is the characteristic function for
the density g(ω, A)/r(0).
Step 3) Take limits as A → ∞,

|t|
lim 1 − r(t) = r(t). (4.6)
A→∞ A

Since the limit of a convergent sequence of characteristic functions is also a

characteristic function, provided it is continuous, we have shown that there
exists a statistical distribution such that r(t)/r(0) is its characteristic function.
We now show step (1) and (2). Multiply g(ω, A) by μ(ω/2M ), integrate, and
change the order of integration (since μ(ω/2M )μ(t/A)r(t)e−iωt is bounded and
has support in [−2M, 2M ] × [−A, A] Fubini’s theorem permits this):
 ∞  ∞  ∞
1
μ(ω/2M )g(ω, A) dω = μ(ω/2M ) μ(t/A)r(t)e−iωt dt dω
−∞ 2π −∞ −∞
 ∞  ∞
1
= μ(t/A)r(t) μ(ω/2M )e−iωt dω dt.
(4.7)
2π −∞ −∞
Section 4.2 Bochner’s theorem and the spectral distribution 85

Here,
 ∞  2M

−iωt |ω|
μ(ω/2M )e dω = 1− e−iωt dω
−∞ −2M 2M
 2M

2
|ω| sin M t
= 1− cos ωt dω = 2M ,
−2M 2M Mt

so (4.7) is equal to
 ∞
2  ∞
2
M sin M t 1 sin s
μ(t/A)r(t) dt = μ(s/M A)r(s/M ) ds
π −∞ Mt π −∞ s
 ∞
2
1 sin s
≤ r(0) ds = r(0).
π −∞ s

Now, μ(ω/2M )g(ω, A) ↑ g(ω, A) as M ↑ ∞, so

 ∞  ∞
g(ω, A) dω = lim μ(ω/2M )g(ω, A) dω ≤ r(0).
−∞ M →∞ −∞

We have now shown that g(ω, A) and μ(t/A)r(t) are both absolutely in-
tegrable over the whole real line. Since they form a Fourier transform pair,
i.e.  ∞
1
g(ω, A) = μ(t/A)r(t)e−iωt dt,
2π −∞
we can use the Fourier inversion theorem, which states that
 ∞
μ(t/A)r(t) = g(ω, A)eiωt dω,
−∞

which is step (2) in the proof.

By taking t = 0 we also get step (1), and fA (ω) = g(ω, A)/r(0) is a
probability density function for some distribution with characteristic function
 ∞
g(ω, A) iωt μ(t/A)
φA (t) = e dω = r(t).
−∞ r(0) r(0)

For step (3) we need one of the basic lemmas in probability theory, the
convergence properties of characteristic functions: if FA (x) is a family of dis-
tribution functions with characteristic functions φA (t), and φA (t) converges to
a continuous function φ(t), as A → ∞, then there exists a distribution func-
tion F (x) with characteristic function φ(t) and FA (x) → F (x), for all x where
F (x) is continuous.
Here the characteristic functions φA (t) = μ(t/A)
r(0) r(t) converge to φ(t) =
r(t)/r(0), and since we have assumed r(t) to be continuous, we know from the
86 Spectral- and other representations Chapter 4
x
basic lemma that FA (x) = −∞ fA (ω) dω converges to a distribution function
F∞ (x) as A → ∞, with characteristic function φ(t):
 ∞
r(t)
= eiωt dF∞ (ω).
r(0) −∞

We get the desired spectral representation with F (ω) = r(0)F∞ (ω). 2

4.2.2 Properties of the spectral distribution

4.2.2.1 The inversion theorem
The covariance function r(t) and the spectral density f (ω) form a Fourier
transform pair. In general, the spectral distribution is uniquely determined by
the covariance function but the precise relationship is somewhat complicated
if the spectrum is not absolutely continuous. To formulate a general inversion
theorem we need to identify those ω for which the spectral distribution function
is not continuous.1 Write ΔFω = F (ω) − F (ω − 0) ≥ 0 for the jump (possibly
0) at ω , and define F(ω) as the average between the left and right limits of
F (ω),
F (ω) + F (ω − 0) 1
F(ω) = = F (ω) − ΔFω . (4.8)
2 2
For a proof of the following theorem we refer to [12].

Theorem 4:4 a) If ω1 < ω2 , then we have

 T −iω2 t
1 e − e−iω1 t
F(ω2 ) − F(ω1 ) = lim r(t) dt. (4.9)
2π T →∞ −T −it

b) If the covariance function r(t), t ∈ R is absolutely integrable, i.e.

 ∞
|r(t)| dt < ∞,
−∞

then the spectrum is absolutely continuous and the Fourier inversion formula
holds,  ∞
1
f (ω) = e−iωt r(t) dt. (4.10)
2π −∞

Remark 4:1 The inversion formula (4.9) defines the spectral distribution for
all continuous covariance functions. One can also use (4.10) to calculate the
spectral density in case r(t) is absolutely integrable, but if it not, one may use
4.9) and take f(ω) = limh→0 (F(ω + h) − F(ω))/h. This is always possible, but
one has to be careful in case f (ω) is not continuous. Even when the limit f(ω)
exists it need not be equal to f (ω) as the following example shows. The limit,
which always exists, is called the Cauchy principal value.
1
Note that F (ω) can have only a denumerable number of discontinuity points.
Section 4.2 Bochner’s theorem and the spectral distribution 87

Example 4:2 We use (4.9) to find the spectral density of low frequency white
noise, with covariance function r(t) = sint t . We get
 T −i(ω+h)t
F(ω + h) − F(ω − h) 1 1 e − e−i(ω−h)t sin t
= lim dt
2h 2π 2h T →∞ −T −it t
 ∞
1 sin ht sin t
= e−iωt dt
2π −∞ ht t
⎧
⎪ 1
⎪
⎨ 2, for |ω| < 1 − h,
1
=
⎪ 4 (1 + (1 − |ω|)/h), for 1 − h < |ω| < 1 + h,
⎪
⎩ 0, for |ω| > 1 + h.
The limit as h → 0 is 1/2, 1/4, and 0, respectively, which gives as spectral
density ⎧
⎪
⎪ 1/2, for |ω| < 1,
⎨
f (ω) = 1/4, for |ω| = 1,
⎪
⎪
⎩ 0, for |ω| > 1.
Note that the Fourier inversion formula (4.10) gives 1/4 for ω = 1 as the
Cauchy principal value,
 T  T
1 −iωt 1 sin 2t
lim e r(t) dt = lim dt = 1/4.
T →∞ 2π −T T →∞ 2π −T 2t

4.2.2.2 The one-sided real form

A real stationary process has a symmetric covariance function and a symmetric
spectrum. In practical applications one often uses only the positive side of the
spectrum. The one-sided spectral distribution will be denoted by G(ω), with

0 for ω < 0,
G(ω) = (4.11)
F (ω) − F (−ω − 0) = 2F (ω) − r(0) for ω ≥ 0.

Then G(0−) = 0 and G(∞) = r(0). If F (ω) is discontinuous at ω = 0 then

G(ω) will have a jump F (0) − F (−0) = G(0+) at ω = 0. For discontinuity
points ω > 0 the jump of G(ω) will be twice that of F (ω).
The covariance function can be expressed as
 ∞  ∞
r(t) = cos ωt dG(ω) = cos ωt dF (ω),
0− −∞

and the inversion formula (4.9) gives that for any continuity points

2 ∞ sin ωt
G(ω) = F (ω) − F (−ω) = r(t) dt.
π 0 t
∞
Note that −∞ sin ωt dF (ω) = 0, since F is symmetric.
88 Spectral- and other representations Chapter 4

4.2.3 Spectrum for stationary sequences

If a stationary process {x(t),
 ∞t ∈ R}, with spectral distribution F (ω) and
covariance function r(t) = −∞ eiωt dF (ω), is observed only at integer time
points t ∈ Z one obtains a stationary sequence {xn , n ∈ Z} for which the
covariance function is the same as that of x(t) restricted to integer n = t. In
the spectral formula the factor eiωt = ei(ω+2kπ)t for all integer t and k , and
the spectrum may be restricted to the interval (−π, π]:
 ∞ ∞  2kπ+π
itω
r(t) = e dF (ω) = eiωt dF (ω)
−∞ k=−∞ 2kπ−π+0
 π ∞

= eiωt dF (ω + 2kπ).
−π+0 k=−∞

This means that all frequencies ω + 2kπ for k = 0 are lumped together with
the frequency ω and cannot be individually distinguished. This is the aliasing
or folding effect of sampling a continuous time process.
For a stationary sequence {xn , n ∈ Z}, the covariance function r(t) is
defined only for t ∈ Z. Instead of Bochner’s theorem we have the following
theorem, in the literature called Herglotz’ lemma.

Theorem 4:5 (Herglotz’ lemma) A function r(t), t ∈ Z, defined on the inte-

gers, is non-negative definite, and hence a covariance function for a stationary
sequence, if and only if there exists a non-decreasing, right-continuous, and
bounded real function F (ω) on (−π, π], such that
 π
r(t) = eiωt dF (ω). (4.12)
−π+0

Note that the spectrum is defined over the half-open interval to keep the right-
continuity of F (ω). It is possible to move half the spectral mass in π to −π
without changing the representation (4.12).

The inversion theorem states that if ∞ t=−∞ |r(t)| < ∞ then the spectrum
is absolutely continuous with spectral density given by
∞
1  −iωt
f (ω) = e r(t),
2π t=−∞

while in general, for −π < ω1 < ω2 ≤ π ,

1 
T
  1 e−iω2 t − e−iω1 t
F (ω2 ) − F (ω1 ) = r(0)(ω2 − ω1 ) + lim r(t) ,
2π T →∞ 2π
t=−T,
−it
t=0

where as before F(ω) is defined as the average of left and right hand side limits
of F (ω).
Section 4.3 Spectral representation of a stationary process 89

4.3 Spectral representation of a stationary process

4.3.1 The spectral process
In elementary courses one could have encountered processes of the form

x(t) = Ak cos(ωk t + φk ), (4.13)
k

where ωk > 0 are fixed frequencies, while Ak are random amplitudes, and
φk random phases, uniformly distributed in (0, 2π) and independent of the
Ak . The uniformly distributed phases make the process stationary, and its
spectrum is discrete, concentrated at {ωk }. The covariance function and one-
sided spectral distribution function are, respectively,

r(t) = E(A2k /2) cos ωk t,
k

G(ω) = E(A2k /2), ω > 0.
k;ωk ≤ω

The process (4.13) can also be defined as the real part of a complex process

x(t) =  Ak eiφk eiωk t ,
k

and it is in fact a special example of the general spectral representation of a

stationary process, which in takes the form of an integral
 ∞
x(t) = eiωt dZ(ω),
−∞

where {Z(ω); ω ∈ R} is a complex spectral process with E(Z(ω)) = 0 and

orthogonal increments, i.e.

E((Z(ω4 ) − Z(ω3 )) · (Z(ω2 ) − Z(ω1 ))) = 0,

for ω1 < ω2 < ω3 < ω4 . The variance of its increments is equal to the incre-
ments of the spectral distribution, i.e. for ω1 < ω2 ,

E(|Z(ω2 ) − Z(ω1 )|2 ) = F (ω2 ) − F (ω1 ).

One can summarize the relations between Z(ω) and F (ω) as


 dF (ω) if ω = μ,
E dZ(ω) · dZ(μ) = (4.14)
0 if ω = μ.

It follows that Z(ω) is continuous in quadratic mean if and only if the spectral
distribution function F is continuous. If F has a jump at a point ω0 ,

F (ω0 +) − F (ω0 −) = σ02 ,

90 Spectral- and other representations Chapter 4

then lim →0 (Z(ω0 + ) − Z(ω0 − )) exists and has variance σ02 .
Now, let us start with a spectral process {Z(ω); ω ∈ R}, a complex process
with E(Z(ω)) = 0 and with orthogonal increments, and define the function
F (ω) by 
E(|Z(ω) − Z(0)|2 ) for ω ≥ 0,
F (ω) =
−E(|Z(ω) − Z(0)|2 ) for ω < 0.

Since only the increments of Z(ω) are used in the theory, we can fix its value
at any point, and we take Z(0) = 0. Following the definition of a stochastic
integral in Section 2.6, we can define a stochastic process
 
x(t) = eiωt dZ(ω) = lim eiωk t (Z(ωk+1 ) − Z(ωk )),

where the limit is in quadratic mean. It is then easy to prove that E(x(t)) = 0
and that its covariance function is given by the Fourier-Stieltjes transform of
F (ω): use Theorem 2:14, and (4.14), to get
 
  
E eiωs dZ(ω) · eiμt dZ(μ) = ei(ωs−μt) E dZ(ω) · dZ(μ)

= eiω(s−t) dF (ω).

4.3.2 The spectral theorem

We shall now prove one of the central results in the theory, namely that every
L2 -continuous weakly
 ∞ stationary process {x(t), t ∈ R} has a spectral repre-
sentation, x(t) = −∞ eiωt dZ(ω), where Z(ω) ∈ H(x), i.e. Z(ω) is an element
in the Hilbert space which is spanned by limits of linear combinations of x(t)-
values. In fact, one can define Z(ω) explicitly for a continuity point ω of F ,

1 T
e−iωt − 1
Z(ω) = lim x(t) dt, (4.15)
T →∞ 2π −T −it

and prove that it has all the required properties. This is the technique used
in Yaglom’s classical book, [38]; see Exercise 5. We shall present a functional
analytic proof, as in [9], and find a relation between H(x) = S(x(t); t ∈ R) and

H(F ) = L2 (F ) = the set of all functions g(ω) with |g(ω)|2 dF (ω) < ∞. We
start by the definition of an isometry.

Definition 4:1 A linear one-to-one mapping f between two Hilbert spaces

X and Y is called an isometry if it conserves the inner product (u, v)X =
(f (u), f (v))Y . In particular u − vX = f (u) − f (v)Y , so distances are also
preserved.
Section 4.3 Spectral representation of a stationary process 91

Theorem 4:6 If {x(t), t ∈ R} is a zero mean continuous stationary process

with spectral distribution F (ω) there exists a complex-valued spectral process
{Z(ω), ω ∈ R} with orthogonal increments, such that

E |Z(ω2 ) − Z(ω1 )|2 = F (ω2 ) − F (ω1 ),

for ω1 < ω2 and 

x(t) = eiωt dZ(ω).

Proof: We shall build an isometry between the Hilbert space of random vari-
ables H(x) = S(x(t); t ∈ R) and the function Hilbert space H(F ) = L2 (F ),
with scalar products defined as

(u, v)H(x) = E(uv),


(g, h)H(F ) = g(ω)h(ω) dF (ω).

First consider the norms in the two spaces,

y2H(x) = E(|y|2 ),

g2H(F ) = |g(ω)|2 dF (ω),

and note that

x(t)2H(x) = E(|x(t)|2 ) = r(0),


i·t 2
e H(F ) = |eiωt |2 dF (ω) = r(0).

This just means that x(t) has the same length as an element of H(x) as has
ei·t as an element of H(F ), i.e. x(t)H(x) = ei·t H(F ) . Furthermore, scalar
products are preserved,

(x(s), x(t))H(x) = E(x(s)x(t)) = eiωs eiωt dF (ω) = (ei·s , ei·t )H(F ) .

This is the start of our isometry: x(t) and ei·t are the corresponding el-
ements of the two spaces. Instead of looking for random variables Z(ω0 ) in
H(x) we shall look for functions gω0 (·) in H(F ) with the same properties.
Step 1: Extend the correspondence to finite linear combinations of x(t) and
eiωt by letting

y = α1 x(t1 ) + . . . + αn x(tn ) (4.16)

g(ω) = α1 eiωt1 + . . . + αn eiωtn (4.17)
92 Spectral- and other representations Chapter 4

be corresponding elements. Check by yourself that scalar product is preserved,

i.e.
(y1 , y2 )H(x) = (g1 , g2 )H(F ) .

Step 2: Distances are preserved, i.e. y1 − y2 H(x) = g1 − g2 H(F ) , so y1 = y2

if and only if g1 = g2 where equality means equal with probability one, and
almost everywhere, respectively.
Step 3: Convergence in the two spaces means the same. If y1 , y2 , . . . converges
towards y in H(x), and g1 , g2 , . . . are the corresponding elements in H(F ),
then
yn − ym H(x) → 0 implies gn − gm H(F ) → 0,
and since H(F ) is complete there exists a limit element g ∈ H(F ) such that
yH(x) = gH(F ) . The reverse implication also holds. Thus we have extended
the correspondence between the two spaces to all limits of sums of x(t)-variables
and eiωt -functions.2
Step 4: The correspondence can then be extended to all of H(F ) and H(x).
The set H(x) consists by definition of limits of linear combinations of x(tk ),
and every function in L2 (F ) can be approximated by a polynomial in eiωtk for
different tk -s. This is the famous Stone-Weierstrass theorem. We have then
found the isometry between H(x) and H(F ): if u and v are elements in H(x)
and f (u) and f (v) the corresponding elements in H(F ), then u − vH(x) =
f (u) − f (v)H(F ) .
Step 5: The following function gω0 in H(F ) corresponds to Z(ω0 ),

1 for ω ≤ ω0 ,
gω0 (ω) =
0 for ω > ω0 .
  ω0
Obviously, gω0 2H(F ) = |gω0 (ω)|2 dF (ω) = −∞ dF (ω), and, with ω1 < ω2 ,

gω2 − gω1 2H(F ) = F (ω2 ) − F (ω1 ).

Step 6: Let Z(ω) be the elements in H(x) that correspond to gω (·) in H(F ). It
is easy to see that Z(ω) is a process with orthogonal increments and incremental
variance given by F (ω):

E (Z(ω4 ) − Z(ω3 )) · (Z(ω2 ) − Z(ω1 ))

= (gω4 (ω) − gω3 (ω))(gω2 (ω) − gω1 (ω)) dF (ω) = 0,

E(|Z(ω2 ) − Z(ω1 )|2 ) = F (ω2 ) − F (ω1 ),

2
Remember that in H(F ) all eiωt are functions of ω , and that we have one function for
every t . Similarly, in H(x) , x(t) = x(t, ω) is a function of ω .
Section 4.3 Spectral representation of a stationary process 93

for ω1 < ω2 < ω3 < ω4 .

Step 7: It remains to prove that Z(ω) is the spectral process to x(t), i.e. that

x(t) = eiωt dZ(ω)

= lim eiωk t (Z(ωk+1 ) − Z(ωk )) = lim S (n) (t),

for an increasingly dense subdivision {ωk } with ωk < ωk+1 . But we have that
x(t) ∈ H(x) and ei·t ∈ H(F ) are corresponding elements. Further,
 (n)
eiωt = lim eiωk t (gωk+1 (ω) − gωk (ω)) = lim gt (ω),

where the difference of g -functions within parentheses is equal to 1 for ωk <

(n)
ω ≤ ωk+1 and 0 otherwise. The limits are in H(F ), i.e. g(·) = ei·t = lim gt (·).
(n)
Since S (n) (t) corresponds to gt (·) and limits are preserved under the isometry,
we have that x(t) = lim S (n) (t), as was to be shown. 2

Corollary 4.1 Every y ∈ H(x) can be written


y = g(ω) dZ(ω),

for some function g(ω) ∈ H(F ).

Proof: Every y ∈ H(x) is the limit of a sequence of linear combinations,
 (n) (n)  
(n)
(n)
y = lim αk x(tk ) = lim αk eiωtk dZ(ω),
n n ω
k k

 (n) (n)
and g(n) (·) = αk ei·tk converges in H(F ) to some function g(·), and then
 
g(n) (ω) dZ(ω) → g(ω) dZ(ω)

in H(x) which was to be shown. 2

4.3.3 More on the spectral representation

4.3.3.1 Discrete spectrum
If the spectral distribution function F (ω) is piecewise constant with jumps of
height ΔFk at ωk , then Z(ω) is also piecewise constant
 with jumps of random
2
size ΔZk at ωk , and E(|ΔZk | ) = ΔFk , so x(t) = ΔZk e t . Note that the
iω k

covariance function then has the corresponding form, r(t) = ΔFk eiωk t .
In the general spectral representation, the complex Z(ω) defines a random
amplitude and phase for the different components eiωt . This fact is perhaps
94 Spectral- and other representations Chapter 4

difficult to appreciate in the integral form, but is easily understood for processes
with discrete spectrum. Take the polar form, ΔZk = |ΔZk |ei arg ΔZk = ρk eiθk .
Then,
  
x(t) = ρk ei(ωk t+θk ) = ρk cos(ωk t + θk ) + i ρk sin(ωk t + θk ).

For a real process, the imaginary part vanishes, and we have the form, well
known from elementary courses – see also later in this section –

x(t) = ρk cos(ωk t + φk ). (4.18)

If the phases φk are independent and uniformly distributed between 0 and 2π ,

then x(t) is also strictly stationary.
For discrete spectrum we also have the following important ways of recover-
ing the discrete components of F (ω) and of Z(ω); the proof of the properties
are part of the Fourier theory.

Theorem 4:7 If F (ω) is a step function, with jumps of size ΔFk at ωk , then

1 T
lim r(t)e−iωk t dt = ΔFk , (4.19)
T →∞ T 0
 
1 T
lim r(t)2 dt = (ΔFk )2 , (4.20)
T →∞ T 0
k
 T
1
lim x(t)e−iωk t dt = ΔZk . (4.21)
T →∞ T 0

4.3.3.2 Continuous spectrum

ω
If the spectrum is absolutely continuous, with F (ω) = −∞ f (x) dx, then one

can normalize the increments of Z(ω) by dividing by f (ω), at least for
f (ω) > 0, and use the spectral representation in the form
 ∞ 
x(t) = 
eiωt f (ω) dZ(ω), (4.22)
−∞
 

   2

with Z(ω) = √
dZ(x)
, and E dZ(ω) = dF (ω)
= dω . Even if
{x≤ω;f (x)>0} f (x) f (ω)

Z(ω) is not a true spectral process – it may for example have infinite incremental
variance – it is useful as a model for white noise. We will meet this “constant
spectral density” formulation several times in later sections.

4.3.3.3 One-sided spectral representation of a real process

For a real processes x(t), the complex spectral representation has to produce a
real integral. This of course requires Z(ω) to have certain symmetry properties,
Section 4.3 Spectral representation of a stationary process 95

which we shall now investigate. Write ΔZ0 for a possible Z -jump at ω = 0.

Then
 ∞
x(t) = eiωt dZ(ω)
−∞
 ∞  ∞
= ΔZ0 + eiωt
dZ(ω) + e−iωt dZ(−ω)
0+ 0+
 ∞
= ΔZ0 + cos ωt · (dZ(ω) + dZ(−ω))
0+
 ∞
+i sin ωt · (dZ(ω) − dZ(−ω)).
0+

For this to be real for all t it is necessary that ΔZ0 is real, and also that
dZ(ω)+dZ(−ω) is real, and dZ(ω)−dZ(−ω) is purely imaginary, which implies
dZ(−ω) = dZ(ω), i.e. arg Z(−ω) = − arg Z(ω) and |Z(−ω)| = |Z(ω)|. (These
properties also imply that x(t) is real.)
Now, introduce two real processes {u(λ), 0 ≤ λ < ∞} and {v(λ), 0 ≤ λ <
∞}, with mean zero, and with u(0−) = v(0−) = 0, du(0) = ΔZ0 , v(0+) = 0,
and such that, for ω > 0,

du(ω) = dZ(ω) + dZ(−ω) = 2 Re dZ(ω)

dv(ω) = i(dZ(ω) − dZ(−ω)) = −2 Im dZ(ω).

The real spectral representation of x(t) will then take the form
 ∞  ∞
x(t) = cos ωt du(ω) + sin ωt dv(ω)
0 0
 ∞  ∞
= cos ωt du(ω) + sin ωt dv(ω) + du(0). (4.23)
0+ 0

It is easily checked that with the one-sided spectral distribution function G(ω),
defined by (4.11),

E(du(ω) · dv(μ)) = 0, for all ω and μ, (4.24)


2 2dF (ω) = dG(ω), ω > 0,
E(du(ω) ) = (4.25)
dF (0) = dG(0), ω = 0,

2dF (ω) = dG(ω), ω > 0,
E(dv(ω)2 ) = (4.26)
dF (0) = dG(0), ω = 0.

In almost all applications, when a spectral density for a time process x(t) is
presented, it is the one-sided density g(ω) = 2f (ω) = dG(ω)/dω that is given.
96 Spectral- and other representations Chapter 4

4.3.3.4 Why negative frequencies?

One may ask why at all use negative frequencies ω < 0 in the spectral rep-
resentation of a real process. Since the two complex functions dZ(ω) eiωt and
dZ(−ω) e−iωt = dZ(ω) eiωt , which build up the spectral representation, circle
the origin in the counter clockwise and clockwise directions their contribution
to the total x(t)-process is real, and there seems to be no point in using the
complex formulation.
One reason for the complex approach is, besides from some mathematical
convenience, that negative frequencies are necessary when we want to build
models for simultaneous time and space processes, for example a random water
wave which moves with time t along a straight line with coordinate s. As
described in Section 1.6.3 on page 22, a random wave model can be built from
elementary harmonics Aω cos(ωt − κs + φω ) where ω is the frequency in radians
per time unit and κ is the wave number in radians per length unit. If ω and
κ have the same sign the elementary wave moves to the right with increasing t
and if they have opposite sign it moves to the left. In stochastic wave models
for infinite water depth the dispersion relation states that κ = ω 2 /g > 0, with
both positive and negative ω possible. The (average) “energy” attached to the
elementary wave Aω cos(ωt−κs+φω ) is A2ω /2 or, in the random case E(A2ω )/2.
If one observes the wave only at a single point s = s0 it is not possible
to determine in which direction the wave is moving, and one can divide the
elementary energy in an arbitrary way between ω and −ω . When we deal with
the spectral density for the time process we have chosen to divide it equally
between positive and negative frequencies.
If we have more than one observation point, perhaps a whole space interval
of observations, we can determine wave direction and see how the energy should
be divided between positive and negative ω . This can be done by splitting the
process in two independent components, one {x+ (t), t ∈ R} with only positive
frequencies, moving to the right, and one {− (t), t ∈ R} with only negative
frequencies, moving to the left. The spectra on the positive and negative side
need not be equal.
For a wave model with one time parameter and a two-dimensional space pa-
rameter (s1 , s2 ) the wave direction is taken care of by a two-dimensional wave
number and the spectrum defined by one part that defines the energy distribu-
tion over frequencies, and one directional spreading part that determines the
energy for different wave directions.

4.3.3.5 Gaussian processes

For Gaussian processes {x(t), t ∈ R}, the spectral process is complex Gaussian,
with independent real and imaginary parts. Since Z(ω) is an element in the
space H(x) of limits of linear combinations of x-variables, this is immediate
from the characterization of Gaussian processes as those processes for which
all linear combinations have a Gaussian distribution. Also the real spectral
Section 4.3 Spectral representation of a stationary process 97

processes {u(λ), 0 ≤ λ < ∞} and {v(λ), 0 ≤ λ < ∞} are Gaussian, and since
they have uncorrelated increments, they are Gaussian processes with indepen-
dent increments.
The sample paths of u(ω) and v(ω) can be continuous, or they could contain
jump discontinuities, which then are normal random variables. In the contin-
uous case, when there is a spectral density f (ω), they are almost like Wiener
processes, and they can be transformed into Wiener processes by normalizing
the incremental variance. In analogy with Z(ω)  in (4.22), define w1 (ω) and
w2 (ω) by
 
du(x) dv(x)
w1 (ω) =  , w2 (ω) =  , (4.27)
{x≤ω;f (x)>0} 2f (x) {x≤ω;f (x)>0} 2f (x)

to get, Theorem 2:16,

 ∞  ∞
x(t) = 2f (ω) cos ωt dw1 (ω) + 2f (ω) sin ωt dw2 (ω). (4.28)
0 0

Note that if f (ω) > 0 then E(dw1 (ω)2 ) = E(dw2 (ω)2 ) = dω .

The representation (4.28) is particularly useful for simulation of stationary
Gaussian processes, as described in detail in Appendix D. Then the continuous
spectrum is discretized to frequencies ωk = kΔ, k ∈ N, and the integrals (4.28)
replaced by sums. Since the increments in the Wiener processes are independent
normal variables, the approximative expressions become
   
x(t) = Uk 2ΔF (ωk ) cos ωk t + Vk 2ΔF (ωk ) sin ωk t, (4.29)
 
= Ak 2ΔF (ωk ) cos(ωk t + φk ), (4.30)

where Uk and Vk are independent standard normal variables, and

)
Ak = Uk2 + Vk2 , φk = − arg(Uk + iVk ).

Historically, the representation (4.29) was used explicitly already by Lord

Rayleigh in connection with heat radiation and by Einstein (1910) and others to
introduce Gaussian randomness. The form (4.30) appears to have come later,
at least according to S.O. Rice, who cites work written by W.R. Bennett in the
1930’s.

4.3.3.6 Gaussian white noise

The differentials dw1 (ω) and dw2 (ω) in (4.28) are examples of Gaussian white
noise. White noise in general is a common notion in stochastic process the-
ory when one needs a process in continuous time where all process values are
virtually independent, regardless of how close they are in time. Complete in-
dependence would require rx (t) = 0 for all t except t = 0, i.e. the covariance
98 Spectral- and other representations Chapter 4

function is not continuous and Bochner’s theorem, Theorem 4:3, is of no use

to find a corresponding spectrum. Fourier’s inversion formula (4.10) hints that
the spectrum should be independent of ω but f (ω) > 0 is not a spectral den-
sity. On the other hand, the δ -distribution, δ(ω), also called the Dirac delta
function, forms a Fourier transform pair together with the constant function
f (ω) = 1/2π . It is in fact possible to formulate a theory for “distribution
valued” stationary processes and covariance functions, but that theory is little
used in practical work and we do not go into any details on this; for a brief
introduction, see [38, Appendix I].
Instead we will use the two Wiener processes defined by (4.27) to illustrate
the common way to go around the problem with constant spectral density. We
used them as spectral processes in (4.28) without any difficulty; we only noted
that E(dw1 (ω)2 ) = E(dw2 (ω)2 ) = dω .
In the theory of stochastic differential equations, one often uses the notation
w (t) or dw(t) with the understanding that it is shorthand for a stochastic
integral of the form t0 g(u) dw(t), for g(t)2 dt < ∞. We will illustrate this
t

with the previously mentioned Langevin equation, (1.15), and deal with these
more in detail in Section 4.4.4.

Example 4:3 (The Ornstein-Uhlenbeck process) The Ornstein-Uhlenbeck pro-

cess is a Gaussian stationary process with covariance function r(t) = σ 2 e−α|t| ,
and spectral density
σ2 α
f (ω) = · 2 .
π α + ω2
We saw in Chapter 2 that a Gaussian process with this covariance function and
spectrum is continuous but not differentiable.
The process can be realized as a stochastic integral
√  t
x(t) = 2ασ 2 e−α(t−τ ) dw(τ ). (4.31)
−∞

As we will see in Section 4.4.4 the Ornstein-Uhlenbeck process is the solution

of the linear stochastic differential equation
√
αx(t) + x (t) = 2α σ w (t) (4.32)

with Gaussian white noise w (t). We met this equation in Section 1.6.1 under
the name Langevin’s equation.
For large α (α → ∞), the covariance function falls off very rapidly around
t = 0 and the correlation between x(s) and x(t) becomes negligible when
s = t. In the integral (4.31) each x(t) depends asymptotically only on the
increment dw(t) and are hence approximately independent. With increasing α,
the spectral density becomes increasingly flatter at the same time as f (ω) → 0.
In order to keep the variance of the process constant, not going to 0 or ∞, we
Section 4.3 Spectral representation of a stationary process 99

can let σ 2 → ∞ in such a way that σ 2 /α → C > 0. Therefore, the Ornstein-

Uhlenbeck process with large α and σ 2 /α = C can be used as an approximation
to Gaussian white noise.
As a stationary process the Ornstein-Uhlenbeck has a spectral representa-
tion of the type (4.28), and one may ask what connection there is between the
two integral representations.
To see the analogy, take w1 (ω) and w2 (ω) from (4.28) and define wC (ω)
for −∞ < ω < ∞, as

w1 (ω) + iw2 (ω), ω > 0,
wC (ω) =
w1 (−ω) − iw2 (−ω), ω < 0,

to get  ∞ 
x(t) = eiωt f (ω) dwC (ω). (4.33)
−∞

We then do some formal calculation with white noise: w (t) is the formal
derivative of the Wiener process, and it is a stationary process with constant
spectral density equal to 1/2π over the whole real line, i.e. by (4.33),
 ∞
1
w (t) = √ eiωt dwC (ω),
2π −∞

for some complex Wiener process wC (ω). Inserting this in (4.31), we obtain

√  t
x(t) = 2ασ 2 e−α(t−τ ) w (τ ) dτ
−∞
√   ∞ 
2ασ 2 t −α(t−τ )
= √ e iωτ
e dwC (ω) dτ
2π τ =−∞ ω=−∞
√   t 
2ασ 2 ∞ −(α+iω)(t−τ )
= √ e dτ eiωt dwC (ω)
2π ω=−∞ τ =−∞
√ 
2ασ 2 ∞ 1
= √ eiωt dwC (ω)
2π ω=−∞ α + iω
√ 
2ασ 2 ∞ 1
= √ √ ei(− arg(α+iω)+ωt) dwC (ω)
2π ω=−∞ α + ω 2 2
 ∞ 
= ei(ωt+γ(ω)) f (ω) dwC (ω),
ω=−∞

with γ(−ω) = −γ(ω). The same Wiener process wC (ω) which works in the
spectral representation of the white noise in (4.31) can be used as spectral
process in (4.33) after correction of the phase.
100 Spectral- and other representations Chapter 4

4.3.4 Spectral representation of stationary sequences

A stationary sequence {x(t), t ∈ Z} can be thought of as a stationary process
which is observed only at integer times t. The spectral representation can then
be restricted to ω -values only in (−π, π] as for the spectral distribution. In the
formula  π
x(t) = eiωt dZ(ω), (4.34)
−π+
there is now an explicit expression for the spectral process,
⎧ ⎫
1 ⎨  e−iωk ⎬
Z(ω) = ωx(0) − x(k).
2π ⎩ ik ⎭
k=0

4.4 Linear filters

4.4.1 Projection and the linear prediction problem
One of the most useful instruments in the theory of stochastic processes is
the linear prediction device, by which we ”predict” or approximate a random
variable x by a linear combination of a set of observed random variables, or by
a limit of such linear combinations. The general formulation in the theory of
stochastic processes is the linear filtering problem in which one seeks a linear
filter h(u) such that the linearly filtered process
 ∞
ŷ(t) = h(u)x(t − u) du
u=−∞

approximates some interesting random quantity y(t), dependent on x(s), s ∈ R.

If the impulse response function h(u) is zero for u < 0 we talk about linear
prediction, otherwise we call it linear reconstruction. The impulse response
may
 contain δ -functions δτk , which act as time delays; for example y(t) =
δτ0 (u)x(t − u) u = x(t − τ0 ).
The projection theorem in Hilbert spaces states that if M is a closed linear
subspace of a Hilbert space H , and x is a point in H not in M, then there is
a unique element y in M closest to x, and then z = x − y is orthogonal to
M; see Appendix C.
Formulated in statistical terms, if x is a random variable and y1 , . . . , yn
is a finite set of random variables, then there is a unique linear combination
 = c1 y1 + . . . + cn yn that is closest to x in the  · -norm, i.e. such that
x
1  12 
1 1
1x − cj yj 1 = E(|x − cj yj |2 )

is minimal. This
 linear combination is characterized by the requirement that the
residual x − cj yj is orthogonal, i.e. uncorrelated with all the yj -variables.
This is the least squares solution to the common linear regression problem.
Section 4.4 Linear filters 101

Expressed in terms of covariances, the coefficients in the optimal predictor

 = c1 y1 + . . . + cn yn satisfy the linear equation system
x

Cov(x, yj ) = c1 Cov(y1 , yj ) + . . . + cn Cov(yn , yj ), j = 1, . . . , n, (4.35)


which follows from Cov(x − k ck yk , yj ) = 0.
Note that the projection theorem says that the random variable y = x  is
unique, in the sense that it y is another random variable that minimizes the
prediction error, i.e. E(|x − y|2 ) = E(|x − y|2 ) then E(|y − y|2 ) = 0 and P (y =
) = 1. This does not mean that the coefficients in the linear combination
y
cj yj are unique; if the variables y1 , . . . , yn are linearly dependent then many
combinations produce the same best predictor.

Example 4:4 For the MA(1)-process, x(t) = e(t) + b1 e(t − 1), we found in
Example C:1 that
 ∞
k=0 (−b1 ) x(t − k), if |b1 | < 1,
k
e(t) = n
limn→∞ k=0 (1 − nk )x(t − k), for b1 = −1.

Thus, x(t + 1) = e(t + 1) + b1 e(t) has been written as the sum of one variable
e(t + 1) ⊥ H(x, t) and one variable b1 e(t) ∈ H(x, t). The projection theorem
implies that the best linear prediction of x(t + 1) based on x(s), s ≤ t, is
 ∞
k=0 (−b1 ) x(t − k), if |b1 | < 1,
b1 k
t (t + 1) = b1 e(t) =
x 
limn→∞ − nk=0 (1 − nk )x(t − k), for b1 = −1.

Note that H(e, t) = H(x, t).

Example 4:5 We can extend the previous example to have H(x, t) ⊂ H(e, t)
with strict inclusion. Take a series of variables e∗ (t) and a random variable U
with E(U ) = 0 and V (U ) < ∞, everything uncorrelated, and set

e(t) = U + e∗ (t).

Then x(t) = e(t) − e(t − 1) = e∗ (t) − e∗ (t − 1), and H(e, t) = H(U ) ⊕ H(e∗ , t)
with H(U ) and H(e∗ , t) orthogonal, and H(e, t) ⊇ H(e∗ , t) = H(x, t).

4.4.2 Linear filters and the spectral representation

4.4.2.1 Frequency response
In the previous section we formulated the prediction solution as a linear fil-
ter with an impulse response functions. Here we take slightly more abstract
approach and use a frequency formulation. A linear time-invariant filter is a
102 Spectral- and other representations Chapter 4

transformation S that takes a stationary process x(t) = eiωt dZ(ω) into a
new stationary process y(t) so that,
 ∞
y(t) = g(ω)eiωt dZ(ω), (4.36)
−∞

where g(ω)
 is the transfer function (also called frequency function). It has to
satisfy |g(ω)|2 dF (ω) < ∞. That the filter is linear and time-invariant means
that, for any (complex) constants a1 , a2 and time delay τ ,

S(a1 x1 + a2 x2 ) = a1 S(x1 ) + a2 S(x2 ),

S(x(· + τ )) = S(x)(· + τ ).

As an alternative to the impulse function approach in Section 4.4.1 we may

take (4.36) as the definition of a linear time-invariant filter.
The process y(t) is also stationary, and it has covariance function given by
  ∞
E y(s + t)y(s) = |g(ω)|2 eiωt dF (ω).
−∞

In particular, if x(t) has spectral density fx (ω) then the spectral density of
y(t) is
fy (ω) = |g(ω)|2 fx (ω). (4.37)

Many of the interesting processes we have studied in previous sections, were

obtained as linear combinations of x(t)-variables, or, more commonly, as limits
of linear combinations. To formulate the spectral forms of these operations, we
need the following property, cf. Step 3, in the proof of Theorem 4:6.


Lemma 4.1 If gn → g in H(F ), i.e. |gn (ω) − g(ω)|2 dF (ω) → 0, then
 
iωt
gn (ω) e dZ(ω) → g(ω) eiωt dZ(ω)

in H(x).

Proof: Use the isometry,

1  12
1 1
1 gn (ω) eiωt dZ(ω) − g(ω) eiωt dZ(ω)1
1 1
H(x)

 
= gn (ω)eiωt − g(ω)eiωt 2 dF (ω) = gn − g2
H(F ) .

2
Section 4.4 Linear filters 103

Example 4:6 For example, the linear operation ”derivation” of a stationary

process is the limit of
 ∞ ihω
x(t + h) − x(t) e − 1 iωt
= e dZ(ω).
h −∞ h

If x(t) satisfies the condition ω 2 dF (ω) < ∞ for quadratic mean differentia-
bility, (eiωh − 1)/h → iω in H(F ) as h → 0, and hence
 
x (t) = iω eiωt dZ(ω) = ω ei(ωt+π/2) dZ(ω).

The frequency function for derivation is therefore g(ω) = iω , and the spectral
density of the derivative is fx (ω) = ω 2 fx (ω).

In general, writing y(t) = |g(ω)| ei(ωt+arg g(ω)) dZ(ω), we see how the filter
amplifies the amplitude of dZ(ω) by a factor |g(ω)| and adds arg g(ω) to the
phase. For the derivative, the phase increases by π/2, while the amplitude
increases by a frequency dependent factor ω .

4.4.2.2 A practical rule

The spectral formulation of linear filters gives us an easy-to-use tool to find
covariance and cross-covariance functions between stationary processes. If the
process {x(t), t ∈ R} is stationary with spectral distribution function Fx (ω),
and {u(t), t ∈ R} and {v(t), t ∈ R} are generated from {x(t), t ∈ R} by linear
filters,

x(t) = eiωt dZ(ω),

u(t) = g(ω)eiωt dZ(ω),

v(t) = h(ω)eiωt dZ(ω),

then, by (4.14),
  
Cov(x(s), u(t)) = eiωs e−iμt g(μ) E dZ(ω) · dZ(μ)
ω μ

= ei(s−t)ω g(ω) dFx (ω),

and similarly,

Cov(u(s), v(t)) = g(ω)h(ω) ei(s−t)ω dFx (ω). (4.38)
104 Spectral- and other representations Chapter 4

4.4.2.3 Impulse response and frequency response

Suppose a linear filter is defined by its impulse response function h(t), as in
Section 4.4.1,
 ∞  ∞
y(t) = h(u)x(t − u) du = h(t − u)x(u) du.
−∞ −∞

Inserting the spectral representation of x(t) and changing the order of integra-
tion, we obtain a filter in frequency response form,
 ∞  ∞   ∞
y(t) = e h(t − u) du dZ(ω) =
iωu
g(ω) eiωt dZ(ω),
ω=−∞ u=−∞ −∞

with  ∞
g(ω) = e−iωu h(u) du, (4.39)
u=−∞

if |h(u)| du < ∞. 
Conversely, if h(u) is absolutely integrable,
 |h(u)| du < ∞, then g(ω),
defined by (4.39), is bounded and hence |g(ω)|2 dF (ω) < ∞. Therefore

y(t) = g(ω) eiωt dZ(ω)

defines a linear filter with frequency function g(ω) as in (4.36). Inserting the
expression for g(ω) and changing the order of integration we get the impulse
response form,
  
iωt −iωu
y(t) = e e h(u) du dZ(ω)
   
= h(u) eiω(t−u)
dZ(ω) du = h(u)x(t − u) du.

The impulse response and frequency response function form a Fourier transform
pair, and  ∞
1
h(u) = eiωu g(ω) dω. (4.40)
2π ω=−∞
If h(u) = 0 for u < 0 the  ∞filter is called causal or physically realizable,
indicating that then y(t) = u=0 h(u)x(t − u) du depends only on x(s) for
s ≤ t, i.e. the output from the filter at time t depends on the past and not on
the future.

4.4.2.4 Linear processes

A stationary sequence xt or a stationary process x(t) is called linear if it is
the output of a linear time invariant filter acting on a sequence of orthogonal
Section 4.4 Linear filters 105

random variables, i.e.

∞

xt = ht−k yk , (4.41)
k=−∞
 ∞
x(t) = h(t − u) dY (u), (4.42)
u=−∞

where yk are uncorrelated with mean 0, and E(|yk |2 ) = 1, and {Y (t), t ∈ R}

is a stationary process with orthogonal increments, E(dY (u) · dY (v)) is equal
to 0 for u = v and equal to du for u = v . The term infinite moving average is
also used for processes of this type.

Theorem 4:8 a) A stationary sequence {xt ; t ∈ Z} is an infinite moving av-

erage
∞
xt = ht−k yk ,
k=−∞

with orthonormal yk and k |hk |2 < ∞, if and only if its spectrum is absolutely
ω
continuous, F (ω) = −π f (x) dx.
b) A stationary process {x(t), t ∈ R} is an infinite continuous moving average
∞
x(t) = u=−∞ h(t − u) dY (u), with an orthogonal increment process Y (ω),

and u |h(u)|2 du < ∞, if and only if its spectrum is absolutely continuous,
ω
F (ω) = −∞ f (x) dx.

Proof: We πshow part a); part b) is quite similar. For the ”only if” part, use
that yk = −π e dZ(ω), where E(|dZ(ω)|2 ) = dω
iωk
2π . Then

   
 π π 
xt = ht−k eiωk dZ(ω) = eiωt ht−k e−iω(t−k) dZ(ω)
k −π −π k
 π
= eiωt g(ω) dZ(ω),
−π

with g(ω) = k hk e−iωk . Thus, the spectral distribution of xk has

dω
dF (ω) = E(|g(ω) dZ(ω)|2 ) = |g(ω)|2 ,
2π
1
with spectral density f (ω) =2π |g(ω)|2 .
ω 1 2
For the ”if” part,F (ω) = −∞ f (x) dx, write f (ω) = 2π |g(ω)| , and expand
|g(ω)| in a Fourier series,

|g(ω)| = ck eiωk .
k
106 Spectral- and other representations Chapter 4

From the normalized spectral representation (4.22),

 π 
xt = 
eiωt f (ω) dZ(ω), 
with E(|dZ(ω)|2
) = dω,
−π

we then get
  
π
1  
xt = eiωt √ ck eiωk dZ(ω)
−π 2π k
 ck  π  
= √ 
eiω(t+k) dZ(ω) = ck et+k = ht−k ek ,
k
2π −π k k
π

with ek = √12π −π eiωk dZ(ω), 
and hk = ck+t . Since the Z(ω) has constant
incremental variance, the ek -variables are uncorrelated and normalized as re-
quired. 2

4.4.3 Linear filters and differential equations

Linear filters expressed in terms of differential equations are common in the
engineering sciences. The linear oscillator, also called the harmonic oscillator,
is the basic element in mechanical systems which exhibit resonant periodic
movements. Its counterpart in electronic systems is the resonance circuit. We
shall describe both of these as examples of a general technique, common in the
theory of ordinary differential equations.
To illustrate the general ideas we start with the exponential smoothing filter,
also called the RC-filter, with a term borrowed from electrical engineering.

4.4.3.1 The RC-filter and exponential smoothing

Consider the electrical circuit in Figure 4.1 with potential difference x(t) on
the left hand side and potential difference y(t) on the right hand side. The
circuit consists of a resistance R and a capacitance C . Regarding x(t) as the
driving process and y(t) as the resulting process we will see that this device
acts as a smoother that reduces rapid high frequency variations in x(t). The
relation between the input x(t) and the output y(t) is

RCy  (t) + y(t) = x(t), (4.43)

and the equations has the (deterministic) solution

 t
1
y(t) = e−(t−u)/(RC) x(u) du.
RC −∞
Thus, the impulse response of the RC-filter is
1 −u/(RC)
h(u) = e , for u > 0,
RC
Section 4.4 Linear filters 107
A1 B1
R
6 6
x(t) C y(t)

? ?

A2 B2
Figure 4.1: Input x(t) and output y(t) in an exponential smoother (RC-
filter).

with frequency response

 ∞
1 −u/(RC) 1
g(ω) = e−iωu e du = .
0 RC 1 + iωRC
Applying the relation (4.37) we get the spectral density relation between input
and output,
fx (ω)
fy (ω) = . (4.44)
(ωRC)2 + 1
The depreciation of high frequencies in the spectrum explains the use of the
RC-filter as a smoother.
To precede the general results for covariance function relations we also make
the following elementary observation about the covariance functions, where we
use the cross-covariances from Theorem 2:13:
rx (τ ) = Cov(RCy  (t) + y(t), RCy  (t + τ ) + y(t + τ ))
= (RC)2 ry (τ ) + RC ry,y (t, t + τ ) + RC ry ,y (t, t + τ ) + ry (τ )
= (RC)2 ry (τ ) + RC ry (τ ) + RC ry (−τ ) + ry (τ )
= (RC)2 ry (τ ) + ry (τ ).
Using the spectral density ω 2 fy (ω) for {y  (t), t ∈ R}, according to Example 4:6,
we find
 
2 iωt 2
rx (t) = (RC) e ω fy (ω) dω + eiωt fy (ω) dω

= eiωt {(ωRC)2 + 1}fy (ω) dω,

and get the spectral density for {x(t), t ∈ R},

fx (ω) = {(ωRC)2 + 1} fy (ω), (4.45)
in accordance with (4.44).
As a final observation we note that the impulse response function satisfies
the differential equation
RCh (u) + h(u) = 0, (4.46)
for u > 0 with the initial condition h(0) = 1/(RC).
108 Spectral- and other representations Chapter 4

4.4.3.2 Linear stochastic differential equations

Suppose we have a stationary process {x(t), t ∈ R}, sufficiently differentiable,
and assume that the process {y(t), t ∈ R} is a solution to an ordinary linear
differential equation with constant coefficients,

p
ap−k y (k) (t) = x(t). (4.47)
k=0

or, seemingly more generally,


p 
q
ap−k y (k) (t) = bq−j x(j) (t). (4.48)
k=0 j=0

By “solution” we mean either that (almost all) sample functions satisfy the
equations or that there exists a process {y(t), t ∈ R} such that the two sides are
equivalent. Note that (4.48) is only marginally more general than (4.47), since
both right hand sides are stationary processes without any further assumption.
What can then be said about the solution to these equations: when does it
exist and when is it a stationary process; and in that case, what is its spectrum
and covariance function?
For the linear differential equation (4.47),
a0 y (p) (t) + a1 y (p−1) (t) + . . . + ap−1 y  (t) + ap y(t) = x(t). (4.49)
we define the generating function,
A(r) = a0 + a1 r + . . . + ap r p ,
and the corresponding characteristic equation
r p A(r −1 ) = a0 r p + a1 r p−1 + . . . + ap−1 r + ap = 0. (4.50)
The existence of a stationary process solution depends on the solutions to the
characteristic equation. The differential equation (4.49) is called stable if the
roots of the characteristic equation all have negative real part.
One can work with (4.49) as a special case of a multivariate first order
differential equation. Dividing both sides by a0 the form is
y = Ay + x, (4.51)
with y(t) = (y(t), y  (t), · · · , y (p−1) (t)) , x(t) = (0, 0, · · · , x(t)) , and
⎛ ⎞
0 1 0 ... 0
⎜ 0 0 1 ··· 0 ⎟
⎜ ⎟
A=⎜ . . .. . . ⎟.
⎝ . . .
. . . 1 ⎠
−ap −ap−1 ap−2 · · · −a1
This is the formulation which is common in linear and non-linear systems the-
ory; cf. for example [14, Ch. 8], to which we refer for part of the following
theorem.
Section 4.4 Linear filters 109

Theorem 4:9 a) If the differential equation (4.49) is stable, and the right hand
side {x(t), t ∈ R} is a stationary process, then there exists a stationary process
{y(t), t ∈ R} that solves the equation. The solution can be written as the output
of a linear filter  t
y(t) = h(t − u) x(u) du, (4.52)
−∞

where the function h(u) solves the equation

a0 h(p) (t) + a1 h(p−1) (t) + . . . + ap−1 h (t) + ap h(t) = 0, (4.53)

with initial conditions h(0) = h (0) = . . . = h(p−2) (0) = 0, h(p−1) (0) = 1/ap .
∞
Further −∞ |h(u)| du < ∞.
b) If {x(t), t ∈ R} is a p times differentiable stationary process with spectral
p (j)
density fx (ω), then also 0 aj x (t) is a stationary process, and it has the
spectral density
 p 2
 
 j
 aj (iω)  fx (ω). (4.54)
 
0

c) If {x(t), t ∈ R} and {y(t), t ∈ R} are two stationary processes that solve

the differential equation (4.48), then their spectral densities obey the relation
 2  2
   
 ak (iω)k  fy (ω) =  bj (iω)j  fx (ω). (4.55)

Proof: a) If {x(t), t ∈ R} has q times differentiable sample paths (with

probability one), we use a standard result in ordinary differential equations to
get a solution for almost every sample path; see [14, Ch. 8].
If we work only with second order properties one can take (4.52) as the def-
inition of a process {y(t), t ∈ R} and then show that it is p times differentiable
(in quadratic mean) and that the two sides of (4.49) are equivalent.
Part b) and c) are easy consequences of the spectral process property (4.14).
Just write the differential form in the right hand sides by means of the spectral
representation and perform the integration. 2

4.4.3.3 The linear oscillator

The linear random oscillator is the basic ingredient in many engineering ap-
plications of stationary processes. We will examine two formulations from me-
chanical and electrical engineering, respectively.

Example 4:7 First consider a spring-and-damper system as in Figure 4.2, with

mass m, stiffness k and damping coefficient c. When the mass is subject to
a regular or irregular varying force x(t) if moves more or less periodically, and
we denote the displacement from the equilibrium by Y (t); see Figure 4.2.
110 Spectral- and other representations Chapter 4

x(t)
y(t)
⇑ 6
m

c k

Figure 4.2: A simple damped oscillator.

The relation between the force X(t) and the resulting displacement is de-
scribed by the following differential equation,

my  (t) + cy  (t) + ky(t) = x(t). (4.56)


Here ω0 = k/m is called the response frequency or eigenfrequency, and
ζ = 2√cmk the relative damping. Expressed in terms of the damping and eigen-
frequency the fundamental equation is

y  (t) + 2ζω0 y  (t) + ω02 y(t) = m−1 x(t).

This equation can be solved just like an ordinary differential equation with a
continuous x(t) and, from Theorem 4:9, it has the solution
 t
y(t) = h(t − u)x(u) du,
−∞

expressed with the impulse response function

0−1 e−αu sin(

h(u) = m−1 ω ω u), u ≥ 0,

with the constants

α = ζω0 ,
0 = ω0 (1 − ζ 2 )1/2 .
ω

To find the frequency function g(ω) for the linear oscillator we consider
each term on the left hand side in (4.56). Since differentiation has frequency
function iω , and hence, repeated differentiation has frequency function −ω 2 ,
we see that g(ω) satisfies the equation
 
−mω 2 + ciω + k · g(ω) = 1,
Section 4.4 Linear filters 111

and hence
1
g(ω) = . (4.57)
−mω 2 + icω + k
Since
1
|g(ω)|2 = ,
(k − mω 2 )2 + c2 ω 2
the spectral density for the output signal y(t) is
fx (ω) fx (ω)/m2
fx (ω) = = . (4.58)
(k − mω 2 )2 + c2 ω 2 (ω02 − ω 2 )2 + 4α2 ω 2

Example 4:8 A resonance circuit with one inductance, one resistance, and one
capacitance in series is an electronic counterpart to the harmonic mechanical
oscillator; see Figure 4.3.
A1 B1

6 6
L
x(t) R y(t)
C
? ?

A2 B2
Figure 4.3: A resonance circuit.

If the input potential between A1 and A2 is x(t), the current I(t) through
the circuit obeys the equation

 1 t
LI (t) + RI(t) + I(s) ds = x(t).
C −∞
The output potential between B1 and B2 , which is y(t) = RI(t), therefore
follows the same equation (4.56) as the linear mechanical oscillator,
1
Ly  (t) + Ry  (t) + y(t) = Rx (t), (4.59)
C
but this time with x (t) as driving force. The frequency function for the filter
between x(t) and y(t) is (cf. (4.57)),
iω
g(ω) = .
−(L/R)ω 2 + iω + 1/(RC)
√
The response frequency ω0 = 1/ LC is here called the resonance frequency.
The relative damping ζ corresponds to the relative bandwidth 1/Q = 2ζ , where

1/Q = Δω/ω0 = R C/L,
√
and Δω = ω2 − ω1 is such that |g(ω1 )| = |g(ω2 )| = |g(ω0 )|/ 2.
112 Spectral- and other representations Chapter 4

Example 4:9 As a final example we consider

y  (t) + 2y  (t) + y(t) = x(t),

with the stationary process {x(t), t ∈ R} as input. The frequency function for
the filter in (4.52) is the solution to h (u) + 2h (u) + h(u) = 0, and is of the
form
h(u) = e−u (C1 + C2 u),

where the boundary conditions give C1 = 0, C2 = 1. The solution

 t
y(t) = (t − u)e−(t−u) x(t) du,
−∞

has the spectral density

fx (ω) fx (ω)
fy (ω) = = .
|1 + 2(iω) + (iω)2 |2 (1 + ω 2 )2

4.4.4 White noise in linear systems

4.4.4.1 White noise in a linear differential equation
The Wiener process was constructed as a mathematical model for the Brownian
motion of particles suspended in a viscous fluid in which the erratic particle
movements are the results of bombardment by the fluid molecules. The Wiener
process model requires that the fluid has zero viscosity and infinite mass.
A more realistic model gives room also for the viscosity and particle mass.
If x(t) denotes the force acting on the particle and y(t) is the velocity, we get
the Ornstein-Uhlenbeck differential equation (4.32) from Example 4:3,

a0 y  (t) + a1 y(t) = x(t), (4.60)

where a1 depends on the viscosity and a0 is the particle mass. If the force x(t)
is caused by collisions from independent molecules it is reasonable that different
x(t) be independent. Adding the assumption that they are Gaussian leads us
to take x(t) = σ w (t) as the “derivative of a Wiener process”, i.e. Gaussian
white noise,
a0 y  (t) + a1 y(t) = σ w (t).
This equation can be solved as an ordinary differential equation by
 t  t
1 −α(t−u)  1
y(t) = e w (u) du = e−α(t−u) dw(u). (4.61)
a0 −∞ a0 −∞

Here the last integral is well defined, Example 2:7 on page 51, even if the
differential equation we started out from is not.
Section 4.4 Linear filters 113

By carrying out the integration it is easy to see that the process y(t) defined
by (4.61) satisfies
 t
a1 y(u) du = a0 (y(t) − y(t0 )) + σ(w(t) − w(t0 )),
u=t0

which means that, instead of equation (4.60), we could have used the integral
equation
 t
a0 (y(t) − y(t0 )) + a1 y(u) du = σ (w(t) − w(t0 )), (4.62)
u=t0

to describe the increments of y(t).

The general differential equation

a0 y (p) (t) + a1 y (p−1) (t) + . . . + ap−1 y  (t) + ap y(t) = σw (t)

can be solved in a similar way, and expressed as a stochastic integral,

 t
y(t) = σ h(t − u) dw(u), (4.63)
−∞

The impulse response function h(u) is the solution to

a0 h(p) (t) + a1 h(p−1) (t) + . . . + ap−1 h (t) + ap h(t) = 0,

as in Theorem 4:9. The formal differential equation can be replaced by the well
defined differential-integral equation

a0 (y (p−1) (t) − y (p−1) (t0 )) + a1 (y (p−2) (t) − y (p−2) (t0 ))+

 t
+ . . . + ap−1 (y(t) − y(t0 )) + ap y(u) du
t0
= σ (w(t) − w(t0 )).

Stochastic differential equations involving Gaussian white noise are often

written as
a0 dy(t) dt + a1 y(t) dt = σ dw(t),
or more generally as

dy(t) = a(t)y(t) dt + σ(t) dw(t),

with variable deterministic coefficients.

In the most general form, with random coefficients,

dy(t) = a(y(t), t) y(t) dt + σ(y(t), t) dw(t),

a complete new theory is needed, namely stochastic calculus; the reader is re-
ferred to [39] for a good introduction.
114 Spectral- and other representations Chapter 4

4.4.4.2 White noise and constant spectral density

A linear systems equation defines the relation between an input signal x(t)
and a response process y(t). The linear system acts as a frequency dependent
amplifier and phase modifier on the input. Of special importance is the case
when the input is white noise. This idealized type of process is strictly defined
only in the context of linear systems. The characteristic feature of white noise,
it may be denoted n(t) or, if it is Gaussian, w (t), is that all frequencies are
represented in equal amount, i.e. it has constant spectral density

σ2
fn (ω) = , −∞ < ω < ∞.
2π
Strictly, this is not a proper spectral density of a stationary process since it
has infinite integral, but used as input in a linear system with a impulse re-
sponse function that satisfies |h(u)|2 du < ∞, it produces a stationary output
process.

Theorem 4:10 a) The stationary process {x(t), t ∈ R}, defined as a stochastic

integral from a standard Wiener process {w(t), t ∈ R} by
 ∞
x(t) = h(t − u) dw(u),
−∞

has covariance function

 ∞
rx (t) = h(t − u)h(−u) du
−∞

and spectral density

|g(ω)|2
fx (ω) = ,
2π
∞
where g(ω) = −∞ e−iωu h(u) du is the frequency response function correspond-
ing to the impulse response h(u).
t
b) If x(t) = −∞ h(t − u) dw(u) is a solution to a stable stochastic differential
equation

a0 y (p) (t) + a1 y (p−1) (t) + . . . + ap−1 y  (t) + ap y(t) = σw (t), (4.64)

then its spectral density is

σ2 1
fx (ω) = ·   .
2π  p p−k 2
k=0 ak (iω)

Proof: a) The covariance function is a direct consequence of Theorem 2:16.

Now the right hand side of the integral expression for rx (t) is the convolution of
h(u) with h(−v) and their Fourier transforms are g(ω) and g(ω) , respectively.
Section 4.4 Linear filters 115

Since convolution corresponds to multiplication of the Fourier transforms, the

spectral density of rx (t) is, as stated,

fx (ω) = g(ω)g(ω)/(2π) = |g(ω)|2 /(2π).

b) The relation between the impulse response and the frequency response func-
tion g(ω) is a property of the systems equation (4.64) and does not depend on
any stochastic property. One can therefore use the established relation
1
g(ω) = p p−k
k=0 ak (iω)

to get the result. 2

Part (b) of the theorem finally confirms our claim that the Gaussian white
noise σw (t) can be treated as if it has constant spectral density σ 2 /(2π).
C
A stationary process with spectral density of the form |P (iω)| 2 where P (ω)

is a complex polynomial can be generated as the output from a stable linear

system with white noise input; this is a very convenient way to produce a
stationary process with suitable spectral properties.

Example 4:10 The linear oscillator

y  (t) + 2ζω0 y  (t) + ω02 y(t) = n(t)

σ2
where the white noise input n(t) has constant spectral density, fn (ω) = 2π ,
has spectral density (cf. (4.58)),

σ2 1
fy (ω) = · 2 .
2π (ω0 − ω )2 + 4α2 ω 2
2

The covariance function is found, for example  by residue calculus, from ry (t) =
 iωt
e fy (ω) dω . With α = ζω0 and ω0 = ω0 1 − ζ 2 one gets the covariance
function

σ 2 −α|t| α
ry (t) = e 0 t +
cos ω 0 |t| .
sin ω (4.65)
4αω02 0
ω

4.4.5 The Hilbert transform and the envelope

4.4.5.1 The Hilbert transform
By the spectral representation we have expressed a stationary process x(t) in
terms of random cosine and sine functions with positive and negative frequen-
cies. The complex form of the spectral representation,
 ∞
x(t) = eiωt dZ(ω),
−∞
116 Spectral- and other representations Chapter 4

yielded a real process by the requirement dZ(−ω) = dZ(ω). In fact, x(t) is

then expressed as the sum of two complex processes, of which one is the complex
conjugate of the other.
If we take only the half spectral representation,
 ∞
x∗ (t) = 2 eiωt dZ(ω) + ΔZ(0),
0+

where ΔZ(0) is the jump of Z(ω) at the origin, we obtain a particularly useful
linear transform of x(t). One can obtain x∗ (t) as the limit, as h ↓ 0 through
continuity points of F (ω), of the linear operation with frequency function
⎧
⎨ 0, ω < −h,
⎪
gh (ω) = 1, |ω| ≤ h,
⎪
⎩
2, ω > h.

(t), defined by
The process x

x∗ (t) = x(t) + i
x(t),

is the result of a linear filter on x(t) with frequency function

⎧
⎪
⎨ i, ω < 0,
g(ω) = 0, ω = 0,
⎪
⎩
−i, ω > 0.

It is called the Hilbert transform of x(t).3

When x(t) is real, with dZ(−ω) = dZ(ω), with the real spectral represen-
tation (4.23),
 ∞  ∞
x(t) = cos ωt du(ω) + sin ωt dv(ω) + du(0),
0+ 0

(t) = i(x(t) − x∗ (t)) is real, and that it is given by

it follows that also x
 ∞  ∞
(t) =
x sin ωt du(ω) − cos ωt dv(ω). (4.66)
0 0+

Thus, x∗ (t) is a complex process with x(t) as real part and x (t) as imaginary
part. All involved processes can be generated by the same real spectral processes
{u(λ), 0 ≤ λ < ∞} and {v(λ), 0 ≤ λ < ∞}.

Theorem 4:11 Let {x(t), t ∈ R} be stationary and real, with mean 0, co-
variance function r(t) and spectral distribution function F (ω), with a possible
jump ΔF (0) at ω = 0. Denote the Hilbert transform of x(t) by x (t). Then,
with G(ω) denoting the one-sided spectrum,
3
Matlab, Signal processing toolbox, contains a routine for making Hilbert transforms.
Try it!
Section 4.4 Linear filters 117

a) {
x(t), t ∈ R} is stationary and real, with mean 0, and covariance function
 ∞  ∞
r(t) = r(t) − ΔF (0) = eiωt dF (ω) − ΔF (0) = cos ωt dG(ω).
−∞ 0+

(t) has the same spectrum F (ω) as x(t), except that any
b) The process x
jump at ω = 0 has been removed.

(t) is
c) The cross-covariance function between x(t)and x
 ∞
∗
r (t) = E(x(s) · x
(s + t)) = sin ωt dG(ω).
0

(t) are uncorrelated, taken at the same time

In particular, x(t) and x
instant.

Proof: Part (a) and (c) follow from (4.23), (4.66), and the correlation prop-
erties (4.24-4.26) of the real spectral processes. Then part (b) is immediate.
2

4.4.5.2 The envelope

Assume that F (ω) is continuous at ω = 0 so there is no ”constant” random
component in x(t), and consider the joint behavior of x(t) and its Hilbert
(t). Also assume x(t), and hence x
transform x (t), to be Gaussian processes,
with common covariance function r(t), and consider the complex process
 ∞
∗
x (t) = x(t) + i
x(t) = 2 eiωt dZ(ω).
0+

The envelope R(t) of x(t) is the absolute value of x∗ (t),


R(t) = x(t)2 + x (t)2 .

In particular, |x(t)| ≤ R(t), with equality when x (t) = 0.

Since, for Gaussian processes, x(t) and x(t) are independent with the same
Gaussian distribution, the envelope has a Rayleigh distribution with density
r −r2 /2σ2
fR (r) = e , r ≥ 0.
σ2
The envelope, as defined here, does always exist, and it always has the
stated statistical properties. The physical meaning of an envelope is however
not clear from the mathematical definition, and we must turn to special types
of processes before we can identify any particularly interesting properties of
R(t).
118 Spectral- and other representations Chapter 4

4.4.5.3 The envelope of a narrow band process

For processes with spectrum concentrated to a narrow frequency band, the sam-
ple function have a characteristic ”fading” look, as a wave with one dominating
frequency. Then the envelope represents the slowly varying random amplitude.
We consider a stationary process x(t) with spectral density f (ω) concen-
trated around some frequency ω0 ,
1 1
f (ω) = f0 (ω − ω0 ) + f0 (ω + ω0 ),
2 2
for some function f0 (ω) such that f0 (ω) = f0 (−ω), and f0 (ω) = 0 for |ω| > d,
some d < ω0 . Express x(t) in spectral form, using the normalized form (4.22),
to obtain
 ∞ 
x(t) = eiωt f (ω) dWC (ω)
−∞
 ω0 +d 
= f0 (ω − ω0 )/2 eiωt dWC (ω)
ω0 −d
 −ω0 +d 
+ f0 (ω + ω0 )/2 eiωt dWC (ω)
−ω0 −d

= I1 (t) + I2 (t), say.

By change of variables in I1 (t) and I2 (t) this gives,

 d
iω0 t
I1 (t) = e f0 (ω)/2 eiωt dω WC (ω + ω0 ) = eiω0 t Y (t),
−d
 d 
−iω0 t
I2 (t) = e f0 (ω)/2 eiωt dω WC (ω − ω0 )
−d
 d 
= e−iω0 t f0 (ω)/2 e−iωt dω WC (−ω − ω0 ) = I1 (t),
−d

and, in combination,
x(t) = 2 ReY (t) eiω0 t .
Here, Y (t) is a complex process
 d 
Y (t) = Y1 (t) + iY2 (t) = f0 (ω)/2 eiωt dω WC (ω + ω0 )
−d

with only low frequencies. With R(t) = 2|Y (t)| and Θ(t) = arg Y (t), we obtain

x(t) = R(t) Re ei(ω0 t+Θ(t)) = R(t) cos(ω0 t + Θ(t)).

The envelope R(t) has here a real physical meaning as the slowly varying am-
plitude of the narrow band process.
Section 4.4 Linear filters 119
A narrowband wave, with Hilbert transform and envelope
A Pierson−Moskowitz wave, with Hilbert transform and envelope
2.5
3
2

1.5 2

1
1
0.5

0 0

−0.5
−1
−1

−1.5 −2

−2 −3
−2.5
0 10 20 30 40 50 0 50 100 150 200

Figure 4.4: Gaussian processes and their Hilbert transforms and envelopes.
Left: Pierson-Moskowitz waves, Right: process with triangular spectrum
over (0.8, 1.2).

4.4.6 The sampling theorem

The spectral form expresses a stationary process as an integral, in quadratic
mean, of elementary cosine functions with random amplitude and phase. If all
these amplitudes and phases are known, the process can be reconstructed. For
band limited processes, this is particularly simple. A process is band limited to
frequency ω0 if
F (−ω0 +) − F (−∞) = F (∞) − F (ω0 −) = 0,

i.e. its spectrum is restricted to the interval [−ω0 , ω0 ]. We require that there
is no spectral mass at the points ±ω0 .

Theorem 4:12 If the stationary process {x(t), t ∈ R} is band limited to ω0 ,

then it is perfectly specified by its values at discrete time points spaced t0 = π/ω0
apart. More specifically, with probability one,
∞
 sin ω0 (t − α − kt0 )
x(t) = x(α + kt0 ) · , (4.67)
ω0 (t − α − kt0 )
k=−∞

where α is an arbitrary constant.

Proof: The spectral representation says that
 ω0 −
x(t) = eiωt dZ(ω).
−ω0 +

For a fixed t, the function gt (ω) = eiωt · I[−ω0,ω0 ] is square integrable over
(−ω0 , ω0 ), and from the theory of Fourier series, it can be expanded as

N
sin ω0 (t − kt0 )
e iωt
= lim eiωkt0 · ,
N →∞ ω0 (t − kt0 )
k=−N
120 Spectral- and other representations Chapter 4

with convergence in H(F ), i.e.

  2
ω0  
N
− 
 iωt sin ω 0 (t kt 0 
)
e − eiωkt0 ·  dF (ω) → 0.
−ω0  k=−N
ω0 (t − kt0 ) 

The convergence is also uniform for |ω| < ω0 = π/t0 . For ω = ±ω0 it converges
to
eiω0 t + e−iω0 t
= cos ω0 t.
2
Therefore, if dF (±ω0 ) = 0, then


N
sin ω0 (t − kt0 )
eiωkt0 · → eiωt , (4.68)
ω0 (t − kt0 )
k=−N

in H(F ) as N → ∞.
Inserting this expansion in the spectral representation of {x(t), t ∈ R} we
obtain,
⎛ 2 ⎞
 N
sin ω0 (t − kt0 )  ⎠

E ⎝x(t) − x(kt0 ) ·  (4.69)
 ω0 (t − kt0 ) 
k=−N
⎛   2 ⎞
 ω0 
N
sin ω0 (t − kt0 ) 
 
= E ⎝ eiωt − eiωkt0 · dZ(ω) ⎠
 −ω0 ω0 (t − kt0 ) 
k=−N

 ω0  
N
2

 iωt sin ω 0 (t − kt 0 
)
≤ e − eiωkt0 ·  dF (ω).
−ω0  k=−N
ω0 (t − kt0 ) 

By (4.68), this goes to 0 as N → ∞, i.e.


N
sin ω0 (t − kt0 )
x(t) = lim x(kt0 ) · ,
N →∞ ω0 (t − kt0 )
k=−N

in H(x), which is the statement of the theorem for α = 0. For arbitrary α,

apply the just proved result to y(t) = x(t + α). 2

Remark 4:2 If there is spectral mass F + = dF (ω0 ), F − = dF (ω0 ), at the

endpoints ±ω0 , then (4.69) would tend to

sin2 ω0 t(F + + F − ),

failing the sampling representation.

Section 4.5 Karhunen-Loève expansion 121

An example of this is the simple random cosine process, x(t) = cos(ω0 t+φ),
which has covariance function r(t) = 12 cos ω0 t, and spectrum concentrated at
±ω0 . Then
x(α + kt0 ) = (−1)k x(α),

which means that for every t, the sum

 sin ω0 (t − α − kt0 )
x(α + kt0 ) ·
ω0 (t − α − kt0 )
k

is proportional to x(α). On the other hand x(α + t0 /2) is uncorrelated with

x(α) and cannot be represented by the sampling theorem.

4.5 Karhunen-Loève expansion

4.5.1 Principal components
In a multivariate distribution of a random vector the components may be more
or less statistically dependent. If there is strong dependence between the com-
ponents it is possible that it suffices to specify a few (random) values in order
to specify almost the entire outcome of the full random vector. The formal tool
to generate such a common behavior is the concept of principal components,
which is a tool to approximately reduce the dimensionality of a variation space.

Let x = (x1 , . . . , xn ) be a vector of n random variables with mean zero

and a covariance matrix Σ, which is symmetric and non-negative definite by
construction. The covariance matrix has n eigenvalues ωk with corresponding
orthonormal eigenvectors pk , decreasingly ordered as ω1 ≥ ω2 ≥ . . . ≥ ωn ,
such that
Σpk = ωk pk .
The transformation
1
zk = √ pk x
ωk
gives us n new standardized random variables

pk Σpk p pk ωk
V (zk ) = = k = 1.
ωk ωk

Furthermore, they are uncorrelated, for j = k ,

1   1
Cov(zj .zk ) = E(zj zk ) = √ E pj x x pk = √ p Σ pk = 0,
ωj ωk ωj ωk j

since the eigenvectors pj and pk are orthogonal.

122 Spectral- and other representations Chapter 4
√
The random variables yk = ωk zk , k = 1, . . . , n, are called the principal
components of the vector x. In matrix language, with P = (p1 , . . . , pn ) as the
matrix with the eigenvectors as columns,
y = Px with inverse x = P y ,
is a vector of uncorrelated variables with decreasing variances ωk .
Since the matrix P is orthogonal, P−1 = P , the original x-variables can
be expressed as a linear combination of the uncorrelated variables yk and zk ,

n
√
xk = pk y= ωj pk zj . (4.70)
j=1

In practice, when one wants to simulate a large vector of many correlated

Gaussian variables, one can use (4.70) to generate successively the most im-
portant variational modes and truncate the sum when, for example it describes
99% of the variation of the x-variables.

4.5.2 Expansion of a stationary process along eigenfunctions

We shall now generalize the finite-dimensional formulation in the previous sec-
tion to continuous
 parameter stochastic processes. By the spectral representa-
tion, x(t) = eiωt dZ(ω), every stationary process can be expressed by means
of uncountably many orthogonal variables dZ(ω). For processes with discrete
spectrum
 with jumps at ±ωk , Z(ω) has countable many jumps ΔZ(ωk ) and
iω
x(t) = e Δk Z .k t

In fact, every quadratic mean continuous process, stationary or not, can

be expressed on a finite interval [a, b] as a sum of deterministic functions with
random orthogonal coefficients,

n
x(t) = lim ck (t)zk .
n→∞
0

The convergence is uniform for a ≤ t ≤ b, in the sense that

⎛ 2 ⎞
 n 
 
E ⎝x(t) − ck (t)zk  ⎠ → 0,
 
0

uniformly in t, as n → ∞.
The functions ck (t) depend on the choice of observation interval [a, b], and
the random variables zk are elements in the Hilbert space spanned by x(t); t ∈
[a, b],
zk ∈ H(x(t); t ∈ [a, b]).
Let us first investigate what properties such an expansion should have, if it
exists. Write H(x) instead of H(x(t); t ∈ [a, b]). Suppose there exists zk with
zk 2H(x) = 1, (zj , zk ) = E(zj zk ) = 0, j = k,
Section 4.5 Karhunen-Loève expansion 123

and assume that the family {zk } is complete for H(x), i.e. the zk form a basis
for H(x). In particular this means that for every U ∈ H(x),

U ⊥ zk , ∀k ⇒ U = 0.

Now take any y ∈ H(x), and define ck = (y, zk ). Then, by the orthogonal-
ity,
⎛ 2 ⎞
 
n  
n
 
E ⎝y − ck zk  ⎠ = . . . = E(|y|2 ) − |ck |2 ,
 
0 0

n ∞
so 0 |ck |2 ≤ y2H(x) for all n, and hence 2 2
0 |ck | ≤ yH(x) . This means
∞
that 0 ck zk exists as a limit in quadratic mean, and also that

∞

y− ck zk ⊥ zn
0

 ∞
for all n. But since {zk } is a complete family, y = ∞0 ck zk = 0 (y, zk )zk .
Now replace y by a fixed time observation of x(t). Then,  naturally, the ck
will depend on the time t andare functions ck (t), x(t) = ck (t)zk . For the
covariance function of x(t) = k ck (t)zk we have, by the orthogonality of zk ,
 
r(s, t) = E(x(s)x(t)) = cj (s) ck (t) E(zj zk ) = ck (s)ck (t).
j,k k

Thus, we shall investigate the existence and properties of the following pair
of expansions,

x(t) = ck (t)zk , (4.71)
k

r(s, t) = ck (s)ck (t). (4.72)
k

Not only can the zk be taken as uncorrelated but also the functions ck (t)
can be chosen as orthogonal, i.e.
 b
cj (t) ck (t) dt = 0, j = k
a
 b
|ck (t)|2 dt = ωk ≥ 0.
a

As a final check on the consequences of an expansion (4.72) and the ortho-

124 Spectral- and other representations Chapter 4

gonality of the functions ck (t), we observe the following arguments4 ,

 b  b 
∞

r(s, t)cj (t) dt = ck (s) ck (t) cj (t) dt
a a 0
∞
  b ∞

= ck (s) cj (t) ck (t) dt = ck (s) ωk δj−k = ωj cj (s).
0 a 0

Thus, the functions cj (t) are eigenfunctions with eigenvalues ωj to the

covariance operator r(s, t),
 b
c(·) → r(·, t) c(t) dt.
a

Call the normalized eigenfunctions φj (t) = √1 cj (t) if ωj > 0, making φk (t) a

ωj
family of orthonormal eigenfunctions.

4.5.3 The Karhunen-Loève theorem

Theorem 4:13 Let {x(t); a ≤ t ≤ b} be continuous in quadratic mean with
mean zero and covariance function r(s, t) = E(x(s)x(t)). Then there exist
orthonormal eigenfunctions φk (t), k = 0, 1, . . . , N ≤ ∞, for a ≤ t ≤ b, with
eigenvalues ωk ≥ 0, to the equation
 b
r(s, t) φ(t) dt = ω φ(s),
a

such that the random variables

 b
1
zk = √ φk (t) x(t) dt
ωk a

are uncorrelated, and can represent x(t)as

∞
 √
x(t) = ωk φk (t) zk . (4.73)
0
n √
The sum is a limit in quadratic mean and x(t) − 0 ωk φk (t) zk 2H(x) → 0
uniformly for t ∈ [a, b].
The variables zk are sometimes called observables and they can be used
for example to make statistical inference about the distribution of the process
x(t). Note, that if x(t) is a normal process, then the zk are uncorrelated normal
variables, and hence independent, making inference simple.
Before we sketch the proof of the theorem, we present an explicit construc-
tion of the Wiener process as an example of the Karhunen-Loève theorem.
4 P∞
The termwise integration is allowed, since sups,t∈[a,b] | 0 ck (s)ck (t)| ≤ supt r(t, t) < ∞ .
Prove this as an exercise.
Section 4.5 Karhunen-Loève expansion 125

Example 4:11 The standard Wiener process w(t), observed over [0, T ] has
covariance function r(s, t) = min(s, t), and eigenfunctions can be found explic-
itly: from
 T
min(s, t) φ(t) dt = ωφ(s),
0
it follows by twice differentiation,
 s  T
tφ(t) dt + sφ(t) dt = ωφ(s) (4.74)
0 s
 T
sφ(s) − sφ(s) + φ(t) dt = ωφ (s) (4.75)
s

−φ(s) = ωφ (s). (4.76)

The initial conditions, φ(0) = 0, φ (T ) = 0, obtained from (4.74) and (4.75),

imply the solution φ(t) = A sin √tω , with cos √Tω = 0. Thus, the positive
eigenvalues ωk satisfy

T π
√ = + kπ, k = 0, 1, 2, . . . .
ωk 2

The normalized eigenfunctions are

%
2 (k + 12 )π t
φk (t) = sin ,
T T
with eigenvalues
1
ωk = T 2 .
π 2 (k + 12 )2
With
  %
1 T π(k + 12 ) T
2 tπ(k + 12 )
zk = √ φk (t) w(t) dt = · sin · w(t) dt,
ωk 0 T 0 T T

we have that the Wiener process can be defined as the infinite (uniformly con-
vergent in quadratic mean) sum
% ∞ πt(k+ ) 1
2  sin T 2
w(t) = zk ,
T π(k+ 12 )
k=0 T

with independent standard normal variables zk .

The reader should simulate a Wiener process w(t), find the variables zk
for k = 0, 1, . . . , n < ∞ by numerical integration, and reproduce w(t) as a
truncated sum.
126 Spectral- and other representations Chapter 4

Proof: (of Theorem 4:13) We only indicate the steps in the proof, following
the outline in [35]. One has to show: the mathematical facts about existence
and properties of eigenvalues and eigenfunctions, the convergence of the series
(4.73), and finally the stochastic properties of the variables zk . This is done in
a series of steps.
b
(i) If a r(s, t) φ(t) dt = ωφ(s), then ω is real and non-negative. This follows
from   
b b b
0≤ r(s, t) φ(s) φ(t) ds dt = ω |φ(s)|2 ds.
a a a

(ii) There is at least one non-zero eigenvalue. The largest eigenvalue is

 b b
ω0 = max r(s, t) φ(s) φ(t) ds dt,
φ;φ=1 a a

b
where the maximum is taken over φ2 = a |φ(t)|2 dt = 1. As stated in [35],
”this is not easily proved”. The corresponding eigenfunction is denoted by
φ0 (t), and it is continuous.
(iii) The function r1 (s, t) = r(s, t) − ω0 φ0 (s)φ0 (t) is a continuous covariance
function, namely for the process
 b
x1 (t) = x(t) − φ0 (t) φ0 (s) x(s) ds,
a

and it holds that  b

r1 (s, t) φ0 (t) dt = 0. (4.77)
a

Repeating step (ii) with r1 (s, t) instead of r(s, t) we get a new eigenvalue
b
ω1 ≥ 0 with eigenfunction φ1 (t). Since a r1 (s, t) φ1 (t) dt = ω1 φ1 (s), we have
 b  b  b 
1
φ1 (s) φ0 (t) dt = φ0 (s) r1 (s, t) φ1 (t) dt ds
a ω1 a a
 b  b 
1
= φ1 (t) r1 (s, t) φ0 (s) ds dt = 0,
ω1 a a

according to (4.77) since r1 (s, t) is real. Thus φ0 and φ1 are orthogonal.

It also follows that φ1 is an eigenfunction to r(s, t),
 b  b  b
r(s, t) φ(t) dt = r1 (s, t) φ1 (t) dt + ω0 φ0 (s) φ0 (t)φ1 (t) dt
a a a

= ω1 φ1 (s) + 0.

(iv) Repeat (ii) and (iii) as long as there is anything remaining of

Section 4.5 Karhunen-Loève expansion 127


n
rn (s, t) = r(s, t) − ωk φk (s)φk (t).
k=0

Then, either there is a finite n such that rn ≡ 0,

or there is an infinite decreasing
sequence of positive eigenvalues ωk ↓ 0 with k ωk < ∞. Show as an exercise
 b
that k ωk ≤ a r(s, s) dt.
(v) If there is an infinite number of eigenvalues, then
 
 
n 
 
sup r(s, t) − ωk φk (s)φk (t) → 0
a≤s,t≤b  
k=0

as n → ∞, i.e.
∞

r(s, t) = ωk φk (s)φk (t),
k=0
with uniform convergence. (This is Mercer’s theorem from 1909.)
(vi) For the representation (4.73) we have
 2
 
n
√  
n
  2
E x(t) − ωk φk (t) zk  = E(|x(t)| ) − ωk |φk (t)|2
 
0 0


n
= r(t, t) − ωk |φk (t)|2 → 0,
0

uniformly in a ≤ t ≤ b, according to (v), as n → ∞.

(vii) The properties of zk follow from the orthogonality of the eigenfunctions.
2

Example 4:12 As a classical engineering problem about signal detection we

shall illustrate the use of the Karhunen-Loève expansion we shall show how one
can make hypothesis testing on the mean value function for a Gaussian process
x(t) with known covariance function r(s, t) = Cov(x(s), x(t)) but unknown
mean value function m(t) = E(x(t)). Following Grenander, [15], suppose we
have observed x(t)for a ≤ t ≤ b, and that we have two alternative hypotheses
about m(t),

H0 : m(t) = m0 (t),

H1 : m(t) = m1 (t)

wanting to ascertain which one is most likely true.

We calculate the independent N (0, 1)-variables zk ,
∞
 √
x(t) = m(t) + (t),
ωk φk (t) zk = m(t) + x
0
128 Spectral- and other representations Chapter 4
b
where zk = √1ωk a φk (t) x
(t) dt. The zk are not observable, since they require
(t) = x(t) − m(t), and m(t) is unknown, but we can introduce observable
x
independent Gaussian variables yk by a similar procedure working on x(t),
 b  b
√
yk = φk (t) x(t) dt = φk (t) m(t) dt + zk ωk .
a a

Writing
 b
ak = φk (t) m(t) dt
a
 b
aik = φk (t) mi (t) dt, i = 0, 1,
a
√
we have that yk ∈ N (ak , ωk ) and the hypotheses are transformed into hy-
potheses about ak :

H0 : ak = a0k , k = 0, 1, . . .

H1 : ak = a1k , k = 0, 1, . . .

Testing hypotheses about an infinite number of independent Gaussian vari-

ables in no more difficult than for finitely many. The likelihood ratio (LR) test
can be used in any case. Let the parameters be T0 = (a00 , a01 , . . . , a0n , . . .)
and T1 = (a10 , a11 , . . . , a1n , . . .), respectively. The LR-test based on y0 , . . . , yn
rejects H0 if the likelihood ratio pn (y) = f1 (y)/f0 (y) > c, where the constant
c determines the significance level. In this case
2n −(yk −a1k )2 /2ωk
√ 1
fy0 ,...,yn (y0 , . . . , yn ; T1 ) k=0 2πωk e
pn (y) = = 2n
fy0 ,...,yn (y0 , . . . , yn ; T0 ) √ 1 −(yk −a0k )2 /2ωk
k=0 2πωk e
 n   n 
 yk (a0k − a1k ) 1  n
a20k − a21k 
= exp − + = exp − Uk ,
ωk 2 ωk
k=0 k=0 k=0
n
say. The LR-test thus rejects H0 if 0 Uk < cα .
We now let n → ∞, and examines the limit of the test quantity. If
∞
 ∞

a2 0k a2 1k
<∞ and < ∞,
ωk ωk
0 0
∞
then the sums converge; in particular 0 Uk converges in quadratic mean to
a normal random variable.
Since ⎧ 2
⎨ + 1 (a1k −a0k ) if H0 is true
2 ωk
E(Uk ) =
⎩ − 1 (a1k −a0k )2 if H is true
2 ωk 1
Section 4.5 Karhunen-Loève expansion 129

and
(a1k − a0k )2
V (Uk ) =
ωk
∞
under both H0 and H1 , we have that 0 Uk is normal with mean
∞  ⎧  2
 ⎨ m0 = ∞ 1 (a1k −a0k ) if H0 is true
0 2 ωk
E Uk = 
⎩ m = −m = − ∞ 1 (a1k −a0k ) if H is true
2
0 1 0 0 2 ωk 1

and with variance

∞
V( Uk ) = 2(m0 − m1 ) = 4m0 .
0
∞ √
Thus, H0 is rejected if 0 Uk < m0 − ωα 2 m0 , where ωα is the upper normal
α-quantile.
∞ (a0k −a1k )2
If 0 ωk2
< ∞, the test can be expressed in a simple way by the
observation that
∞
  b

m0 (t) + m1 (t)
Uk = f (t) x(t) − dt,
a 2
0

where
∞
 a0k − a1k
f (t) = φk (t).
ωk
0
130 Spectral- and other representations Chapter 4

Exercises
4:1. Let x(t) be a stationary Gaussian process with E(x(t)) = 0, covariance
function rx (t) and spectral density fx (ω). Calculate the covariance func-
tion for the process
y(t) = x2 (t) − rx (0),
and show that it has the spectral density
 ∞
fy (ω) = 2 fx (μ)fx (ω − μ) dμ.
−∞

4:2. Derive the spectral density for u(t) = 2x(t)x (t) if x(t) is a differentiable
stationary Gaussian process with spectral density fx (ω).
4:3. Let et , t = 0, ±1, ±2, . . . be independent N (0, 1)-variables and define the
stationary processes

t
xt = θxt−1 + et = θ t−n en ,
n=−∞
yt = et + ψet−1 ,

with |θ| < 1. Find the expressions for the spectral processes Zx (ω)
and Zy (ω) in terms of the spectral process Ze (ω), and derive the cross
spectrum between xt and yt . (Perhaps you should read Chapter 6 first.)
4:4. Let un and vn be two sequences of independent, identically distributed
variables with zero mean and let the stationary sequences xn and yn be
defined by

yn = a1 + b1 xn−1 + un
xn = a2 − b2 yn + vn .

Express the spectral processes dZx and dZy as functions of un and vn ,

and derive the spectral densities for xn and yn and their cross spectrum.
4:5. Use the limit
⎧
 T ⎨ 1 for ω > 0,
1 sin ωt
lim dt = 0 for ω = 0,
T →∞ π −T t ⎩
−1 for ω < 0.
for the following alternative derivation of the spectral representation of a
stationary process x(t) with spectral distribution function F (ω).

a) First show that the following integral and limit exists in quadratic
mean:  T −iωt
1 e −1
Z(ω) = lim x(t) dt.
T →∞ 2π −T −it
Section 4.5 Karhunen-Loève expansion 131

b) Then show that the process Z(ω), −∞ < ω < ∞, has orthogonal
increments and that

E|Z(ω2 ) − Z(ω1 )|2 = F (ω2 ) − F (ω1 )

for ω1 < ω2 .
c) Finally, show that the integral
 ∞ 
eiωt dZ(ω) = lim eiωk t (Z(ωk+1 ) − Z(ωk ))
−∞


exists, and that E|x(t) − eiωt dZ(ω)|2 = 0.

4:6. Complete the proof of Lemma 4.1 on page 102.

4:7. Consider the covariance function

σ 2 −α|t| α
ry (t) = e 0 t +
cos ω 0 |t| ,
sin ω
4αω02 0
ω

of the linear oscillator in Example 4:10 on page 115.

The covariance function contains some |t|; show that the covariance func-
tions fulfil a condition for sample function differentiability, but not for
twice differentiability.
Find the relation between the relative damping zeta and the spectral
√
width parameter α = ω2 / omega0 ω4 .

4:8. Prove that sups,t∈[a,b] | ∞
0 ck (s)ck (t)| ≤ supt r(t, t) < ∞ in the expan-
sions (4.71) and (4.72).

4:9. Prove that in step (iv) in the proof of Theorem 4:13,

 
ωk = |φk (t)|2 dt ≤ max r(s, s) · (b − a)2 .
a≤s≤b
k k
132 Spectral- and other representations Chapter 4
Chapter 5

Ergodic theory and mixing

The concept of ergodicity is one of the most fundamental in probability, since

it links the mathematical theory of probability to what can be observed in a
deterministic mechanical world. It also plays an important role in statistical
physics and in fact, the term ergodic was coined by Ludwig Boltzmann in 1887
in his study of the time development of mechanical particle systems. The
term itself stems from the Greek ergos = ”work” and hodos = ”path”, possibly
meaning that ergodic theory is about how the energy in a system evolves with
time. Ergodicity in itself is not a probabilistic concept and it can be studied
within a purely deterministic framework, but it is only in terms of probabilities
that a sensible interpretation can be given to the basic ergodic results. The main
result in this chapter is the Birkhoff ergodic theorem, which in 1931 settled the
question of convergence of dynamical systems. For an account of the parallel
development in statistical physics and probability theory, see the interesting
historical work by von Plato [26]. The account in this chapter is based on [5]
and [9]. More results on ergodic behavior of random and non-random sequences
can be found in [12], and for general stochastic aspects of dynamical systems,
[21].

5.1 The basic Ergodic theorem in L2

We met our first ergodic theorem in Section 2.7, Theorem
T 2:17, stating covari-
ance conditions under which the time average T −1 0 x(t) dt tends, in quadratic
mean and with probability one, to the expectation of the stationary process
{x(t), t ∈ R}. In the special case when x(t) is stationary with covariance
function r(t) = Cov(x(s), x(s + t)), the quadratic mean convergence becomes
particularly simple. If E(x(t)) = 0 then

 T  T
1 1 q.m.
r(t) dt → 0 implies x(t) dt → 0,
T 0 T 0

133
134 Ergodic theory and mixing Chapter 5

as T → ∞. This was proven by elementary calculation in Theorem 2:17(a).

Note that it follows from the statements in Theorem 4:7 in Section 4.3.3, that
the relation is satisfied if and only if the spectral distribution function is con-
tinuous at the origin.

By means of the spectral representation x(t) = eiωt dZ(ω), we can formu-
late a more precise theorem and see what happens if this sufficient condition
does not hold. In Section 4.3.2 we stated an explicit expression (4.15) for the
spectral process Z(ω). In fact, if ω1 and ω2 are continuity points of the spectral
distribution function F (ω), then we have the parallel expressions

1 T
e−iω2 t − e−iω1 t
F (ω2 ) − F (ω1 ) = lim r(t) dt,
2π T →∞ −T −it

1 T
e−iω2 t − e−iω1 t
Z(ω2 ) − Z(ω1 ) = lim x(t) dt,
2π T →∞ −T −it

the latter convergence being in quadratic mean.

We repeat the statement from Theorem 4:7, that when the spectral distri-
bution F is a step function, then
 T
1
lim r(t)e−iωk t dt = ΔFk ,
T →∞ T 0
 T 
1
lim |r(t)|2 dt = (ΔFk )2 ,
T →∞ T 0 k
 T
1
lim x(t)e−iωk t dt = ΔZk .
T →∞ T 0

5.2 Stationarity and transformations

5.2.1 Pseudo randomness and transformation of sample space

For strictly stationary processes one can obtain limit theorems of quite different
character from those valid for processes which satisfy a covariance stationarity
condition. These theorems also require much deeper conditions than simple
covariance conditions. Remember, however, that a Gaussian (weakly) station-
ary process is also strictly stationary, so the general ergodicity properties of
Gaussian processes can be inferred already from its covariance function. We
start by giving all results for stationary sequences {xn ; n ∈ Z}.
For a strictly stationary sequence the location of the origin is unessential for
the stochastic properties, i.e. P ((x1 , x2 , . . .) ∈ B) = P ((xk+1 , xk+2 , . . .) ∈ B)
for every Borel set B ∈ B∞ ; see Section 1.3.2. This also means that we can
assume the sequence to be double ended, and to have started in the remote
Section 5.2 Stationarity and transformations 135

past.1 From now on in this chapter, by a stationary process we mean a process

that is strictly stationary.
How do we – or nature – construct stationary sequences? Obviously, first
we need something (call it ”a game”) that can go on forever, and second, we
need a game where the rules remain the same forever.

Example 5:1 (The irrational modulo game) A simple game, that can go on
forever and has almost all interesting properties of a stationary process, is the
adding of an irrational number. Take a random x0 with uniform distribution
between 0 and 1, and let θ be an irrational number. Define

xk+1 = xk + θ mod 1,

i.e. xk+1 is the fractional part of xk + θ . It is easy to see that xk is a stationary

sequence. We shall soon see why this game is more interesting with an irrational
θ than with a rational one. If computers could handle irrational numbers, this
type of pseudorandom number generator would be even more useful in Monte
Carlo simulations than it is.

Example 5:2 One can define other stationary sequences by applying any time
invariant rule to a stationary sequence xn , e.g. with 0 ≤ x0 ≤ 1, yn = 1 if
xn+1 > x2n and yn = 0 otherwise. A more complicated rule is

yn = xn + max e−k xn+k .

k>0

The well-known quadratic transformation,

xk+1 = cxk (1 − xk ),

is an example of an interesting transformation giving rise to a stochastic se-

quence when x0 is random; in Exercise 5:5 you are asked to find its stationary
distribution.

5.2.2 Strict stationarity and measure preserving transforma-

tions
Deterministic explicit rules, like xk+1 = xk + θ mod 1, are just examples of
transformations of a sample space which can produce stationary sequences for
certain probability distributions. We need to define a general concept of a mea-
sure preserving transformation, which makes the resulting process stationary.
1
More precisely, for every strictly stationary sequence {xn , n ∈ N} there exists a strictly
stationary sequence {e xn ; n ∈ Z} such that x
en ; n ≥ 0 have the same finite-dimensional distri-
butions as xn ; n ≥ 0 ; prove this in Exercise 5:4.
136 Ergodic theory and mixing Chapter 5

Definition 5:1 a) Consider a probability space (Ω, F, P ). A measurable trans-

formation2 T on (Ω, F, P ) is called measure preserving if

P (T −1 A) = P (A) for all A ∈ F . (5.1)

b) Given a measurable space (Ω, F) and a measurable transformation T on

(Ω, F), a probability measure P is called invariant if (5.1) holds.

In statistical physics or dynamical systems theory, a probability space is

regarded as a model for the ”universe”, with the outcomes ω representing all
its different ”states”, e.g. the location and velocity of all its particles. The
measure P defines the probabilities for the universe to be in such and such a
state that certain events occur. A transformation T ω is just a law of nature that
changes the state of the universe; that a transformation is measure preserving
means that the events occur with the same probability before and after the
transformation.

5.2.2.1 From a measure preserving transformation to a stationary

sequence

Every measure preserving transformation can generate a stationary sequence.

To see how, take a random variable x(ω) on (Ω, F, P ), i.e. a ”measurement” on
the state of the universe, for example its temperature. Then x(T ω) is the same
measurement taken after the transformation. Define the random sequence

x1 (ω) = x(ω), x2 (ω) = x(T ω), . . . , xn (ω) = x(T n−1 ω), . . .

Since T is measure preserving, this sequence is strictly stationary: With B ∈

B∞ and

A = {ω; (x1 (ω), x2 (ω), . . .) ∈ B} = {ω; (x(ω), x(T ω), . . .) ∈ B},

P (A) = P ((x1 , x2 , . . .) ∈ B). Further,

T −1 A = {ω; T ω ∈ A} = {ω; (x(T ω), x(T 2 ω), . . .) ∈ B}

= {ω; (x2 (ω), x3 (ω), . . .) ∈ B},

and thus P (T −1 A) = P ((x2 , x3 , . . .) ∈ B). Since T is measure preserving,

P (A) = P (T −1 A), and hence (x1 , x2 , . . .) and (x2 , x3 , . . .) have the same dis-
tribution, that is, xn is a stationary sequence.
2
A measurable transformation T on a measurable space (Ω, F) is a function defined on
Ω such that the inverse images under T of all sets in F are again in F ; that is T −1 A =
{ω; T ω ∈ A} ∈ F for all A ∈ F .
Section 5.3 The Ergodic theorem, transformation view 137

5.2.2.2 From a stationary process to a measure preserving transfor-

mation
We have just seen how to construct a stationary sequence from a measure
preserving transformation. Conversely, every stationary sequence generates a
measure preserving transformation on R∞ , the space of realizations for the
stationary sequence.
Take the probability space (R∞ , B∞ , P ) with outcomes
ω = (x1 , x2 , . . .),
and define the shift transformation T by
T ω = (x2 , x3 , . . .).
As an example, take the set A = {ω; x1 < x2 } to be the outcomes for which
the first coordinate is smaller than the second one. Then T −1 A = {ω; T ω ∈ A}
is the set {(x1 , x2 , . . .); (x2 , x3 , . . .) ∈ A} = {(x1 , x2 , . . .); x2 < x3 }, that is, the
second coordinate is smaller than the third. The transformation just shifts the
event criterion one step to the right.
Take the coordinate process x(ω) = ω = (x1 , x2 , . . .), for which the nth vari-
able in the sequence is equal to the nth coordinate in the outcome, and assume
that P is such that x is a stationary sequence. Then the shift transformation
T is measure preserving – the shifted sequence has the same distribution as the
original one.

Remark 5:1 The property that a transformation is measure preserving, of

course depends on the probability measure P and not only on the transfor-
mation itself. In probability theory, where the probability measure is often given
a priori, it is natural to put the request on the transformation.
In the mathematical study of dynamical systems, one often starts with a
transformation T and seeks a measure under which T is measure preserving.
Such a measure is called invariant. Thus, invariant measures in the dynamical
systems theory are equivalent to the strictly stationary processes in probability
theory.

5.3 The Ergodic theorem, transformation view

The classical ergodic theory for dynamical systems, deals with what happens in
the long run when one observes some characteristic of the system, i.e. when one
takes observation x(ω). The state of the system changes by the transformation
T , and our interest is with the time average of the sequence of measurements
x(ω), x(T ω), x(T 2 ω), . . . , taken for a fixed initial outcome ω ,
1
n
x(T k−1 ω)
n
1
138 Ergodic theory and mixing Chapter 5

as n → ∞. Think of Ω as all the possible states our universe can be in,

and, to be concrete, think of x(ω) as the temperature at one specific location.
The universe changes states from day to day, and if ω is the state of the
1 n
universe today, and T ω its state tomorrow, then n 1 x(T k−1 ω) is the average
temperature observed over an n-day period.
In this section we shall take the transformation view on ergodic theory, and
prove the ergodic theorem in that vein. In the next section we shall do exactly
the same thing, but take a probabilistic strictly stationary process view, and
prove the ergodic theorem in terms of random variables and expectations.

5.3.1 Invariant sets and invariant random variables

To motivate the introduction of invariant sets and invariant random variables,
we consider the possible limits of the average of a sequence of random variables,

1
n
Sn /n = xk .
n
1

If Sn /n converges, as n → ∞, what are the possible limits? Say, if Sn /n → y ,

a random variable, then obviously xn /n = Sn /n − Sn−1 /n → y − y = 0, and3

x2 + x3 + . . . + xn+1 x1 − xn+1
y = lim + lim
n n
x2 + x3 + . . . + xn+1 xn+1
= lim − lim .
n n

We see that the limit of Sn /n is the same for the sequence x1 , x2 , . . . as it

would be for the shifted sequence x2 , x3 , . . . .

Definition 5:2 Consider a probability space (Ω, F, P ) and a measure preserv-

ing transformation T of Ω onto itself.

(a) A random variable x on (Ω, F, P ) is called invariant under T if x(ω) =

x(T ω), for almost every ω ∈ Ω.

(b) A set A ∈ F is called invariant under T if T −1 A = A.

Example 5:3 The limit of Sn /n is invariant (when it exists) under the shift
transformation (x1 , x2 , . . .) → (x2 , x3 , . . .) of R∞ . The random variables y =
lim sup xn and lim sup Sn /n are always invariant.

3
Show as an exercise, if you have not done so already, that xn is a stationary sequence
with E(|xn |) < ∞ , then P (xn /n → 0, as n → ∞ ) = 1 ; Exercise 5:6.
Section 5.3 The Ergodic theorem, transformation view 139

Example 5:4 Let xn be a Markov chain with transition matrix

⎛ ⎞
1/2 1/2 0 0
⎜1/3 2/3 0 0 ⎟
⎜ ⎟
P=⎜ ⎟,
⎝ 0 0 2/3 1/3⎠
0 0 1/2 1/2

with starting distribution p(0) = (2p/5, 3p/5, 3q/5, 2q/5) is stationary for every
p + q = 1. The sets A1 = {ω = (x1 , x2 , . . .); xk ∈ {1, 2}} and A2 = {ω =
(x1 , x2 , . . .); xk ∈ {3, 4}} are both invariant under the shift transformation.

The proof of the following simple theorem is left to the reader.

Theorem 5:1 (a) The family of invariant sets,

J = {invariant sets A ∈ F },

is a σ -field.
(b) A random variable y is invariant if and only if it is measurable with respect
to the family J of invariant sets.

5.3.2 Ergodicity
The fundamental property of ergodic systems ω → T ω with a stationary (or
invariant) distribution P , is that the T k ω , with increasing k , visits every corner
of the state space, exactly with the correct frequency as required by P . Another
way of saying this is that the ”histogram”, counting the number of visits to any
neighborhood of states, converges to a limiting ”density”, namely the density
for the invariant distribution P over the state space. If we make a measurement
x(ω) on the system, then the expected value E(x) is the ”ensemble average”
with respect to the measure P ,

E(x) = x(ω) dP (ω),
ω∈Ω

and – here is the ergodicity – this is exactly the limit of the ”time average”

1
n
x(T k−1 ω).
n
1

In the Markov Example 5:4, if p and q are not 0, there is no possibility

for the process to visit every state the correct number of times, since either it
starts in the invariant set A1 and then it always takes the values 1, 2, or it
starts in A2 and then it stays there forever and takes only values 3, 4. This is
the key to the definition of ergodicity.
140 Ergodic theory and mixing Chapter 5

Definition 5:3 A measure preserving transformation T on (Ω, F, P ) is called

ergodic if every invariant set A ∈ J has either P (A) = 0 or P (A) = 1, that
is, all invariant sets are trivial. The term ”metrically transitive” is sometimes
used instead of ”ergodic”.

Here is a nice example of a transformation that can be ergodic.

Example 5:5 The modulo game in Example 5:1, considered as a transfor-

mation on the unit interval, ([0, 1], B, ), ( is the Lebesgue measure, i.e. the
uniform distribution),
T x = x + θ mod 1,
is ergodic when θ is irrational, and non-ergodic for rational θ .
For θ = m/n one has T n x = x and every set of the form A ∪ {A + 1/n} ∪
{A + 2/n} . . . ∪ {A + (n − 1)/n} is invariant but can have probability ∈ (0, 1).
To see what happens when θ is irrational we need some further results for
ergodic transformations.

Theorem 5:2 Let T be a measure preserving transformation of (Ω, F, P ).

Then T is ergodic if and only if every invariant random variable x is a.s.
a constant. It is sufficient that every bounded invariant random variable is a.s.
constant.

Proof: First assume that every bounded invariant variable is constant, and
take an invariant set A. Its indicator function χA (ω) = 1 if ω ∈ A is then
an invariant random variable, and by assumption it is an a.s. constant. This
means that the sets where it is 0 and 1, respectively, have probability either 0
or 1, i.e. P (A) = 0 or 1. Hence T is ergodic.
Conversely, take an ergodic T , and consider an arbitrary invariant random
variable x. We shall show that x is a.s. constant. Define, for real x0 ,

Ax0 = {ω; x(ω) ≤ x0 }.

Then T −1 Ax0 = {ω; T ω ∈ Ax0 } = {ω; x(T ω) ≤ x0 } = Ax0 , since x is invariant.

But then, by ergodicity, P (Ax0 ) = 0 or 1, depending on x0 , and it is easy to
0 such that P (x = x
see that then there is an x 0 ) = 1, hence x is constant.
2

Example 5:6 (Irrational modulo game) We can now show that T x = x + λ

mod 1 is ergodic if λ is irrational. Take any Borel-measurable function f (x)
1
on [0, 1) with 0 f 2 (x) dx < ∞. It can be expanded in a Fourier series
∞

f (x) = ck e2πikx ,
k=−∞
Section 5.3 The Ergodic theorem, transformation view 141

with convergence in quadratic mean for almost all x and |ck |2 < ∞. Then
y = f (ω) is a random variable. We assume it is invariant, and prove that it is
a.s. constant. But f invariant means that
∞

f (x) = f (T x) = ck e2πikx · e2πikλ ,
k=−∞

which implies ck (1 − e2πikλ ) = 0 for all k . But e2πikλ = 1 for k = 0 when

λ is irrational, and hence ck = 0 for all k = 0, which means that f (x) = c0 ,
constant. By Theorem 5:2 we conclude that T is ergodic.
The study of ergodic transformations is an important part of the dynamic
systems and chaos theory; see [21].

5.3.3 The Birkhoff Ergodic theorem

In the introductory remarks, Section 5.3.1, we noted that any limit of the time
average
1
n
x(T k−1 ω)
n
1
is invariant. The limit could be a constant, and it could be a random variable,
but it needs to be invariant. If it is constant, we need to find the value of the
constant, and if it is a random variable, we want to find out as much as possible
about its distribution.
In fact, for measure preserving transformations, the limit always exists, and
is equal to the conditional expectation of x given the invariant sets. Since we
have not dealt with this concept previously in this course, we state the basic
properties of conditional expectations by giving its definition.

Definition 5:4 If y is a random variable on (Ω, F, P ) with E(|y|) < ∞,

and A a sub-σ -field of F , then by the conditional expectation of y given A,
E(y | A), is meant any A-measurable random variable u that satisfies
 
y(ω) dP (ω) = u(ω) dP (ω),
ω∈A ω∈A
for all A ∈ A. Note that the value of E(y | A) = u is defined only almost
surely, and that any A-measurable variable which has the same integral as y
when integrated over A-sets, works equally well as the conditional expectation.

In particular, if A only contains sets which have probability 0 or 1, E(y | A)

is a.s. constant and equal to E(y).
In particular, we consider the conditional expectation E(x | J ), given the
σ -field J of sets which are invariant under a measure preserving transforma-
tion.
142 Ergodic theory and mixing Chapter 5

5.3.3.1 Time averages always converge

Theorem 5:3 (Birkhoff ergodicity theorem) Let T be a measure preserving

transformation on (Ω, F, P ). Then for any random variable x with E(|x|) <
∞,
1
n−1
lim x(T k ω) = E(x | J ), a.s.
n→∞ n
0

The proof is based on the following lemma, shown by Adriano M. Garsia in

1965.

Lemma 5.1 Let T be a measure preserving transformation and x a random

variable with E(|x|) < ∞. Define

Sk (ω) = x(ω) + . . . + x(T k−1 ω), and Mn (ω) = max(0, S1 , S2 , . . . , Sn ).

Then 
x(ω) dP (ω) ≥ 0.
ω;Mn >0

Proof of lemma: Consider Sk = x(T ω) + . . . + x(T k ω) = Sk − x(ω) + x(T k ω),

and Mn (T ω) = Mn = max(0, S1 , . . . , Sn ). For k = 1, . . . , n, Mn ≥ Sk , so

x + Mn ≥ x + Sk = Sk+1 ,

and (for k = 0) x + Mn ≥ S1 (= x). Hence

x ≥ Sk − Mn , for k = 1, 2, . . . , n + 1,

which implies
x ≥ max(S1 , . . . , Sn ) − Mn .
Thus (with Mn = Mn (T ω)),
 
x(ω) dP (ω) ≥ {max(S1 (ω), . . . , Sn (ω)) − Mn (T ω)} dP (ω).
Mn >0 Mn >0

But on the set {ω; Mn (ω) > 0}, one has that Mn = max(S1 , . . . , Sn ), and
thus
 
x(ω) dP (ω) ≥ {Mn (ω) − Mn (T ω)} dP (ω)
Mn >0 Mn >0

≥ {Mn (ω) − Mn (T ω)} dP (ω) = 0,

since increasing the integration area does not change the integral of Mn (ω)
while it can only make the integral of Mn (T ω) larger. Further, T is measure
Section 5.3 The Ergodic theorem, transformation view 143

preserving, i.e. shifting the variables one step does not change the distribution,
nor the expectation. 2
Proof of theorem: We first assume that E(x | J ) = 0 and prove that the
average converge to 0, a.s. For the general case consider x − E(x | J ) and use
that
E(x | J )(T ω) = E(x | J )(ω),
since E(x | J ) is invariant by Theorem 5:1(b), page 139
We show that x = lim sup Sn /n ≤ 0 and, similarly, x = lim inf Sn /n ≥ 0,
giving lim Sn /n = 0. Take an  > 0 and denote D = {ω; x > }: we shall show
that P (D) = 0. Since, from Example 5:3, x is an invariant random variable,
also the event D is invariant. Define a new random variable,

x(ω) −  if ω ∈ D,
x∗ (ω) =
0 otherwise,

and set Sn∗ (ω) = n1 x∗ (T k−1 ω), with Mn∗ defined from Sk∗ . From Lemma 5.1
we know 
x∗ (ω) dP (ω) ≥ 0. (5.2)
Mn∗ >0
We now only have to replace this inequality by an inequality for a similar
integral over the set D to be finished. The sets
 
Fn = {Mn∗ > 0} = max Sk∗ > 0
1≤k≤n

increase towards the set

     
Sk∗ Sk
F = sup Sk∗ > 0 = sup > 0 = sup >  ∩ D.
k≥1 k≥1 k k≥1 k

But since supk≥1 Sk /k ≥ lim sup Sk /k = x, we have that F = D . In order to

take the limit of (5.2) we must be sure the expectations are finite, i.e. E(|x∗ |) ≤
E(|x|) + , and bounded convergence gives
 
∗
0≤ x (ω) dP (ω) → x∗ (ω) dP (ω). (5.3)
Mn∗ >0 D

Here the right hand side is

 
∗
x (ω) dP (ω) = x(ω) dP (ω) − P (D)
D D

= E(x | J ) dP − P (D) = −P (D),
D

since D E(x | J ) dP = 0, by assumption. Together with (5.3) this implies
P (D) = 0, and hence x ≤ . But  > 0 was arbitrary so x ≤ 0, a.s. The same
chain of arguments leads to x ≥ 0, and hence lim sup Sn /n ≤ 0 ≤ lim inf Sn /n.
The limit therefore exists and is 0, which was to be shown. 2
144 Ergodic theory and mixing Chapter 5

5.3.3.2 Time averages of ergodic transformations go to the mean

It is now a simple corollary that time averages for ergodic transformations

converge to the expected value.

Corollary 5.1 If T is a measure preserving ergodic transformation on the

probability space (Ω, F, P ), then for any random variable x with E(|x|) < ∞,

1
n−1
lim x(T k ω) = E(x), a.s.
n→∞ n
0

Proof: When T is ergodic every invariant set has probability 0 or 1 and

therefore the conditional expectation is constant, E(x | J ) = E(x), a.s. 2

Remark 5:2 If x is non-negative with E(x) = ∞ then Sn /n → ∞ if T is

ergodic. Show that as an exercise; Exercise 5:10.

5.3.3.3 Ergodic non-random walks

One can regard a measure preserving transformation T as a non-random walk
ω , T ω , T 2 ω , . . . , over the sample space. In the beginning of this section we
interpreted the ergodic statements as convergence results for the number of
visits to any neighborhood of a fixed outcome. This can now be made precise.
Take a set A ∈ F and consider its indicator function χA (ω). The ergodic
theorem says, that if T is ergodic,

1
n−1
a.s.
χA (T k ω) → P (A).
n
0

Example 5:7 In the irrational modulo game, T x = x + θ mod 1, the relative

number of points falling in an interval [a, b) converges to the length of the
interval if θ is irrational. Thus the number of points becomes asymptotically
equidistributed over [0, 1). This is the weak Weyl’s equidistribution theorem.

5.4 The Ergodic theorem, process view

It is easy to formulate the convergence theorem for transformations in terms of
time averages of stationary sequences. First we need to define invariant events
and ergodicity in (Ω, F, P ).

Definition 5:5 (Reformulation of Definition 5:2) Let {xn } be a stationary

sequence. An event A ∈ F is called invariant for {xn }, if there exists a B ∈ B∞
such that for any n ≥ 1,

A = {(xn , xn+1 , . . .) ∈ B}.

Section 5.4 The Ergodic theorem, process view 145

The sequence is called ergodic if every invariant set has probability 0 or 1.

A random variable z is called invariant for {xn }, if it is a function of
x1 , x2 , . . . and remains unchanged under the shift transformation, i.e. if there
exists a random variable φ on (R∞ , B∞ ) such that z = φ(xn , xn+1 , . . .) for all
n ≥ 1.
From the correspondence between transformations of the sample space and
stationary sequences, it is easy to formulate an ergodicity theorem for a sta-
tionary sequence {xn , n ∈ Z}.

Theorem 5:4 (a) If {xn , n ∈ Z} is a stationary sequence with E(|x1 |) < ∞,

and J denotes the σ -field of invariant sets,

1  a.s.
n
xk → E(x1 | J ).
n
1

(b) If {xn , n ∈ Z} is stationary and ergodic, then

1  a.s.
n
xk → E(x1 ).
n
1

(c) If {xn , n ∈ Z} is stationary and ergodic, and φ(x1 , x2 , . . .) is measurable

on (R∞ , B∞ ), then the process

yn = φ(xn , xn+1 , . . .)

is stationary and ergodic. The special case that φ(xn , xn+1 , . . .) = φ(xn ) only
depends on xn should be noted in particular.
Proof: It is easy to reformulate the results of Theorem 5:3 to yield the the-
orem. However, we shall give the direct proof once more, but use the process
formulation, in order to get a slightly better understanding in probabilistic
terms. We give a parallel proof of part (a), and leave the rest to the reader.

nthe counterpart of Lemma 5.1: From the sequence {xn , n ∈ Z}, define
(a) First
Sk = j=1 xj and Mn = max(0, S1 , S2 , . . . , Sn ). We prove that

E(x1 | Mn > 0) ≥ 0. (5.4)


To this end, write Sk = k+1j=2 xj = Sk − x1 + xn+1 , and define the cor-
responding maximum, Mn = max(0, S1 , S2 , . . . , Sn ), and note that, for k =
1, . . . , n,
x1 + Mn ≥ x1 + Sk = Sk+1 .
Since Mn ≥ 0 we also have x1 + Mn ≥ S1 (= x1 ), so

x1 ≥ max(S1 , . . . , Sn ) − Mn .
146 Ergodic theory and mixing Chapter 5

Now, when Mn > 0 we can replace max(S1 , . . . , Sn ) by Mn , so taking the

conditional expectation, given Mn > 0, we get

E(x1 | Mn > 0) ≥ E(max(S1 , . . . , Sn ) − Mn | Mn > 0)

= E(Mn − Mn | Mn > 0).

Further, since Mn ≥ 0, one easily argues that

E(Mn − Mn | Mn > 0) ≥ E(Mn − Mn )/P (Mn > 0),

which is 0, since Mn and Mn have the same expectation. This proves (5.4).
We continue with the rest of the proof of part (a). Suppose E(x1 | J ) = 0
and consider the invariant random variables x = lim sup Sn /n. Take an  > 0
and introduce the invariant event D = {x > }. Then, by the assumption,

E(x1 | x > ) = 0, (5.5)

a fact which is basic in the proof. We intend to prove that P (D) = 0 for every
 > 0, thereby showing that x ≤ 0.
Similarly, x = lim inf Sn /n can be shown to be non-negative, and hence
x = x = 0.
However, before we prove that P (D) = P (x > ) = 0, we need to discuss
the meaning of (5.5). A conditional expectation is defined as a random variable,
measurable with respect to the conditioning σ -field, in this case J . In (5.5)
we conditioned on one of the events D ∈ J and that is fine if P (D) > 0,
but if P (D) = 0, the claim (5.5) makes no sense. The conditional expectation
given an event of probability 0 can be given any value we like, since the only
requirement on the conditional expectation is that it should give a correct value
when integrated over a J -event. If that event has probability 0 the integral is
0 regardless of how the expectation is defined. We return to this at the end of
the proof.
From (x1 , x2 , . . .), define a new sequence of variables

xk −  if x > ,
x∗k =
0 otherwise,

and define Sk∗ = kj=1 x∗k and Mn∗ = max(0, S1∗ , S2∗ , . . . , Sn∗ ), in analogy with
Sk and Mn . The sequence {x∗k } is stationary, since x is an invariant random
variable, so we can apply (5.4) to get

E(x∗1 | Mn∗ > 0) ≥ 0. (5.6)

On the other hand, from the definition of x∗k we have

E(x∗1 | x > ) = E(x1 | x > ) −  = − < 0, (5.7)

Section 5.4 The Ergodic theorem, process view 147

since E(x1 | x > ) = 0 by the assumption. These two inequalities go in

opposite directions, and in fact they will turn out to be in conflict, unless
P (D) = 0, proving the assertion.
So we would like to have a relation between the events D = {x > } and
Fn = {Mn∗ > 0} = {max1≤k≤n Sk∗ > 0}, and we see that as n increases, then
Fn increases to
     
Sk∗ Sk 3
∗
F = sup Sk > 0 = sup > 0 = sup > {x > } .
k≥1 k≥1 k k≥1 k

But x = lim sup Sk /k ≤ supk≥1 Sk /k , so the right hand side is just D = {x >
}; Fn ↑ D . This implies (here E(|x∗1 |) ≤ E(|x1 |) +  < ∞ is needed),

0 ≤ lim E(x∗1 | Fn ) = E(x∗1 | D) = − < 0,

n→∞

which obviously is impossible.

Where did it go wrong, and where is the contradiction? By definition, for
the outcomes where x > , the variables x∗k ARE equal to xk − , and Sk∗ /k =
Sk /k − . But Fn ↑ D does not imply limn→∞ E(x∗1 | Fn ) = limn→∞ E(x∗1 | D),
since these expressions are not well defined. If P (D) > 0 our reasoning makes
sense and lead to a contradiction, if P (D) = 0 we have argued with undefined
quantities in (5.6) and (5.7). The reader who wants to be on safest possible
grounds should return to the formulation and proof of Theorem 5:3. 2

5.4.0.4 Ergodic stationary processes

For continuous time processes {x(t), t ∈ R}, one defines the shift transforma-
tion Uτ by
(Uτ x)(t) = x(t + τ ).
If x(t) is stationary, Uτ is measure preserving. For a set of functions B , the
shifted set Uτ B is the set of functions Uτ x for x ∈ B . A Borel set B ∈ BR
is called a.s. invariant, if B and Uτ B differ by, at most, sets of P -measure 0.
Let J denote the σ -field of invariant sets. The process {x(t), t ∈ R} is called
ergodic if all invariant sets have probability 0 or 1.

Theorem 5:5 (a) For any stationary process {x(t), t ∈ R} with E(|x(t)|) <
∞ and integrable sample paths, as T → ∞,

1 T a.s.
x(t) dt → E(x(0) | J ).
T 0

(b) If further {x(t), t ∈ R} is ergodic, then


1 T a.s.
x(t) dt → E(x(0)).
T 0
148 Ergodic theory and mixing Chapter 5
n
Proof: One first has to show that xn = n−1 x(t) dt is a stationary sequence,
and use the ergodic theorem to get convergence for integer n,
 n  1
1 a.s.
x(t) dt → E( x(t) dt | J ),
n 0 0

as n → ∞. By invariance and stationarity the last expectation is equal to

E(x(0) | J ). Finally, with n = [T ],
 T  n  T
1 n 1 1
x(t) dt = x(t) dt + x(t) dt.
T 0 T n 0 T n
n
The first term has the same limit as n1 0 x(t) dt, while the second term is
bounded by

1 n+1
|x(t)| dt.
T n
But also |x(t)| is a stationary process, to which we can apply Theorem 5:4(a),
getting convergence, and hence can conclude that the last term tends to 0.
Thus we obtain the desired limit. 2

5.5 Ergodic Gaussian sequences and processes

We shall now give simple conditions for ergodicity for Gaussian stationary pro-
cesses, characterized by their covariance function r(t) in continuous or discrete
time.

Theorem 5:6 Let {x(t), t ∈ R} be stationary and Gaussian with E(x(t)) = 0

and V (x(t)) = 1, and let its covariance function be r(t). Then x(t) is ergodic
if and only if its spectral distribution function F (ω) is continuous everywhere.
ω
If the spectral distribution has a density f (ω), F (ω) = −∞ f (l) dl , the F (ω) is
obviously continuous, but that is by no means necessary. It suffices that F (ω)
is a continuous function.

Proof of ”only if” part: If x(t) is ergodic, so is x2 (t), and therefore the
time average of x2 (t) tends to E(x2 (0)) = 1,
 T
ST 1
x2 (t) dt → 1,
a.s.
=
T T 0

as T → ∞. It is a property of the Gaussian distribution that E((ST /T )4 ) ≤ K

is bounded for large T , and therefore the almost sure convergent ST /T also
converges in quadratic mean, i.e. E((ST /T − 1)2 ) → 0. But this expectation
can be calculated. Since, for a standard Gaussian process, E(x(s)2 x(t)2 ) =
Section 5.5 Ergodic Gaussian sequences and processes 149

1 + 2r(t − s)2 , one gets,

 T  T

2 1 2 2
E((ST /T − 1) ) = 2 E x(s) x(t) ds dt − 1
T 0 0
 T T  T   t 
2 2 4 1 2
= 2 r (t − s) ds dt = 2 t· r (s) ds dt. (5.8)
T 0 0 T 0 t 0
t
But according to Theorem 4:7, page 94, relation (4.20), 1t 0 r 2 (s) ds tends to
the
 sum 2of squares of all jumps of the spectral distribution function F (ω),
(ΔFk ) . Hence, if this sum is strictly positive, the right hand side in (5.8)
has a positive limit, which contradicts what we proved above, and we have
concluded the ”only if” part of the theorem. 2

The ”if” part is more difficult, and we can prove it here only under the
additional condition that
 ω the process has a spectral density, i.e. the spectral
distribution is F (ω) = −∞ f (x) dx, because then

r(t) = eiωt f (ω) dω → 0,

as t → ∞, by Riemann-Lebesgue’s Lemma. (The full statement and proof can

be found in [24, 15].) So what we have to show is that if r(t) → 0, then x(t) is
ergodic. Since this is worth to remember, we formulate it as a lemma.

Lemma 5.2 A Gaussian stationary process is ergodic if its covariance function

r(t) → 0 as t → ∞.

Proof of lemma, and the ”if” part of theorem: We show that if r(t) → 0,
then every invariant set has probability 0 or 1. Let S be an a.s. invariant set
for the x(t)-process, i.e. the translated event Sτ differs from S by an event of
probability zero. But every event in F can be approximated arbitrarily well
by a finite-dimensional event, B , depending only on x(t) for a finite number of
time points tk , k = 1, . . . , n; cf. Section 1.3.3. From stationarity, also Sτ can
be approximated by the translated event Bτ = Uτ B , with the same error, and
combining S with Sτ can at most double the error. Thus, we have

|P (S) − P (B)| < ,

|P (S ∩ Sτ ) − P (B ∩ Bτ )| < 2.

Here P (S ∩ Sτ ) = P (S) since S is invariant, so P (S) can be approximated

arbitrarily well by both P (B) and by P (B ∩ Bτ ).
But B depends on x(ti ), i = 1 . . . , n, while Bτ is defined from x(τ + tj ),
j = 1, . . . , n, and they are multivariate normal with covariances

Cov(x(ti ), x(tj + τ )) = r(τ + tj − ti ) → 0

150 Ergodic theory and mixing Chapter 5

as τ → ∞. Thus these two groups of random variables become asymptotically

independent,
P (B ∩ Bτ ) − P (B) · P (Bτ ) → 0,
and by stationarity, P (B ∩ Bτ ) → P (B)2 . Thus both P (B) and P (B)2 ap-
proximate P (S), and we conclude that P (S) = P (S)2 . This is possible only if
P (S) is either 0 or 1, i.e. x(t) is ergodic. 2

5.6 Mixing and asymptotic independence

How much of the future development of a stochastic process is pre-determined
from what has already happened, and how much information about the future is
there in a piece of observation of a stochastic process and how closely dependent
are disjoint segments of the process?
In this section we will briefly mention some criteria on a stationary process
that guarantee perfect predictability and un-predictability, respectively. A dif-
ferentiable process can be predicted locally by means of a Taylor expansion. The
linear prediction theory gives the Cramér-Wold decomposition (Theorem 5:7)
into a singular component that can be predicted linearly without error and one
regular component for which the predictable part tends to zero with increas-
ing prediction horizon. The spectral representation in Chapter 4 relates the
spectrum to the number of harmonic components which are needed to build
a stationary process. The ergodic theorem in Chapter 5 touched upon the
problem of asymptotic independence; for a Gaussian process with (absolutely)
continuous spectrum and asymptotically vanishing covariance function, values
far apart are asymptotically independent and a ”law of large numbers” holds.
In this section we shall try to relate these scattered results to each other. Some
comments are for Gaussian processes only, while others are of general nature.
A new concept, mixing, will also be introduced.

5.6.1 Singularity and regularity

When predicting from x(s), s ≤ t, a question of both practical and theo-
retical (or perhaps philosophical) interest is from where does the informa-
tion about x(t + h) originate, and how much new information is added to
H(x, t) = S(x(s); s ≤ t), with increasing t.
When t → −∞, obviously

H(x, t) ↓ H(x, −∞) = ∩t≤t0 H(x, t),

and
H(x, −∞) ⊆ H(x, t) ⊆ H(x).
The subspace H(x, t) is the space of random variables that can be obtained as
limits of linear combinations of variables x(tk ) with tk ≤ t, and H(x, −∞) is
Section 5.6 Mixing and asymptotic independence 151

what can be obtained from old variables, regardless of how old they may be. It
can be called the infinitely remote past, or the the primordial randomness.4
Two extremes may occur, depending on the size of H(x, −∞):
• if H(x, −∞) = H(x), then {x(t), t ∈ R} is purely deterministic, or sin-
gular,

• if H(x, −∞) = 0, then {x(t), t ∈ R} is purely non-deterministic, or

regular.

A process is deterministic if all information about the future that can be

obtained from the past at time t, x(s), s ≤ t, can be obtained already from
x(s), s ≤ τ < t, arbitrarily far back. An example of this is the band-limited
white noise Gaussian process, which we studied in Chapter 4. Such a process is
infinitely differentiable – this follows from the general rules for differentiability
– and the entire sample functions can be reconstructed from the values in an
arbitrarily small interval located anywhere on the time axis. A summary of
facts pertaining reconstruction and prediction is given later in these notes; see
Section 5.6 on general results on asymptotic independence and its opposite,
complete dependence.
The following theorem was proved by Cramér (1962) in the general param-
eter case; the discrete stationary case was given by Wold (1954).

Theorem 5:7 (The Cramér-Wold decomposition) Every stochastic process

{x(t), t ∈ R}

with E(|x(t)|2 ) < ∞, is the sum of two uncorrelated processes,

x(t) = y(t) + z(t),

where {y(t), t ∈ R} is regular (purely non-deterministic) and {z(t), t ∈ R} is

singular (deterministic).
Proof: Construct H(x, t) and H(x, −∞) = limt↓−∞ H(x, t), and define

z(t) = P−∞ (x(t)) = the projection of x(t) on H(x, −∞),

y(t) = x(t) − z(t).

To prove the theorem, we have to show that

1. H(z, −∞) = H(z, t), and z(t) is deterministic,

2. H(y, −∞) = 0, and y(t) is non-deterministic,

3. H(y) ⊥ H(z), and y(s) and z(t) are uncorrelated.

4
Nowadays called ”the primordial soup”.
152 Ergodic theory and mixing Chapter 5

Number (3) follows from the projection properties, since the residual y(s) =
x(s) − P−∞ (x(s)) is uncorrelated with every element in H(x, −∞), i.e. y(s) ⊥
H(x, −∞). Since z(t) ∈ H(x, −∞), number (3) follows.
Further, H(y, t) ⊆ H(x, t) and H(y, t) ⊥ H(x, −∞). Therefore H(y, −∞)
is equal to 0, because if y is an element of H(y, −∞) then both y ∈ H(y, t) ⊂
H(x, t) for all t, i.e. y ∈ H(x, −∞), and at the same time y ⊥ H(x, −∞). The
only element that is both in H(x, −∞) and is orthogonal to H(x, −∞) is the
zero element, showing (2).
Finally, H(z, t) = H(x, −∞) for every t. To see this, note that

H(x, t) ⊆ H(y, t) ⊕ H(z, t)

for all t, and therefore also

H(x, −∞) ⊆ H(y, −∞) ⊕ H(z, t).

Since H(y, −∞) = 0,

H(x, −∞) ⊆ H(z, t) ⊆ H(x, −∞).

Thus H(x, −∞) = H(z, t), and (1) is proved. 2

As the band-limited white noise example shows, there are natural deter-
ministic processes. Other common process models are regular. Examples of
processes combining the two properties seem to be rather artificial.

Example 5:8 An AR(1)-process with an added component,

x(t) = ax(t − 1) + e + e(t),

with uncorrelated e and e(t)-variables, can be decomposed into

∞

y(t) = ak e(t − k),
k=0

which is regular, and

1
e,
z(t) =
1−a
which is singular. The common ARMA-process is regular.

5.6.2 Asymptotic independence, regularity and singularity

As we saw in the proof of Lemma 5.2, if r(t) → 0 as t → ∞, then in a Gaussian
process, finitely many variables taken sufficiently far apart are almost indepen-
dent. But this does definitely not mean that the process observed in an entire
interval, x(s), s ∈ I is asymptotically independent of x(s + t), s ∈ I . These two
Section 5.6 Mixing and asymptotic independence 153

segments can be completely dependent of each other in a deterministic way.

For example, the realizations can be infinitely differentiable and part of an an-
alytic function that can be reconstructed from its derivatives in an arbitrarily
small interval. It was shown by Belyaev [3] that if the covariance function of a
separable process is an entire function, i.e. analytic in the entire complex plane,
then the sample functions are a.s. also entire functions, which can be expressed
as a convergent power series
∞
 tk
x(t) = x(k) (0) .
k!
k=0
2
Examples of such covariance functions are r(t) = e−t /2 and r(t) = sint t . Ya-
glom [38, Ch. 8] contains a readable account of prediction in this case; [18, 11]
give more mathematical details.

5.6.2.1 Regularity, singularity and the spectrum

The Cramér-Wold decomposition deals with the problem of predicting future
values by means of linear combinations of past observations. A singular (or
purely deterministic) process can be perfectly predicted linearly from old values.
In a regular process the predictable part tends to zero with increasing prediction
horizon. Simple conditions for singularity/regularity can be formulated in terms
of the spectral distribution function F (ω).
d
Let f (ω) = dω F (ω) be the derivative of the bounded and non-decreasing
function F (ω). For almost
 ∞all ω this derivative exists and is non-negative, and
its integral is bounded, −∞ f (ω) dω ≤ F (∞) − F (−∞). Write
 ω
(ac)
F (ω) = f (x) dx ≤ F (ω).
−∞

The spectrum is called absolutely continuous with spectral density f (ω) if

F (ω) = F (ac) (ω).
In general, F (ac) (ω) need not be equal to F (ω). In particular, this is of course
the case when the spectrum has jumps ΔFk at frequencies ωk . Write

F (d) (ω) = ΔFk ,
ωk ≤ω

so the spectrum is discrete if F (ω) = F (d) (ω). The part of the spectrum which
is neither absolutely continuous nor discrete is called the singular part:
F (s) (ω) = F (ω) − F (ac) (ω) − F (d) (ω).
Note that both F (d) (ω) and F (s) (ω) are bounded non-decreasing functions,
differentiable almost everywhere, with zero derivative. The question of singu-
larity or regularity of the process {x(t), t ∈ R} depends on the behavior of the
spectrum for large |ω|.
154 Ergodic theory and mixing Chapter 5

5.6.2.2 Conditions for stationary sequences

π
Since −π f (ω) dω < ∞ and −∞ ≤ log f (ω) ≤ f (ω) = d
dω F (ω), the integral
 π
1
P = log f (ω) dω (5.9)
2π −π

is either finite or equal to −∞.

Theorem 5:8 For a stationary sequence xt , t ∈ Z the following cases can oc-
cur.
a) If P = −∞, then xt is singular.

b) If P > −∞ and the spectrum is absolutely continuous with f (ω) > 0 for
almost all ω , then xt is regular.

c) If P > −∞, but F (ω) is either discontinuous, or is continuous with non-

vanishing singular part, F (s) (ω) = 0, then xt is neither singular nor regular.

This theorem is quite satisfying, and it is worth making some comments

on its implications. First, if the spectrum is discrete with a finite number of
jumps, then f (ω) = 0 for almost all ω and P = −∞, so the process is singular.
As seen from the spectral representation (4.18), the process then depends only
on a finite number of random quantities which can be recovered from a finite
number of observed values.
If the spectrum is absolutely continuous with density f (ω), singularity and
regularity depends on whether f (ω) comes close to 0 or not. For example, if
f (ω) vanishes in an interval, then P = −∞ and x(t) is singular.5 If f (ω) ≥
c > 0 for −π < ω ≤ π , then the integral is finite and the process regular.
A stationary sequence x(t) is regular, if and only if it can be represented
as a one-sided moving average


t
xt = ht−k yk ,
k=−∞

with uncorrelated yk ; cf. Theorem 4:8 which also implies that it must have a
spectral density.
A sequence that is neither singular nor regular can be represented as sum
of two uncorrelated sequences,

(s) (r) (s)


t
xt = xt + xt = xt + ht−k yk .
k=−∞

5
Singularity also occurs if f (ω) = 0 at a single point ω0 and is very close to 0 nearby,
1
such as when f (ω) ∼ exp(− (ω−ω 0)
2 ) when ω → ω0 .
Section 5.6 Mixing and asymptotic independence 155

(r)
The regular part xt has absolutely continuous spectrum,
 ω
(ac)
F (ω) = f (x) dx,
−π

(s)
while the singular part xt has spectral distribution F (d) (ω) + F (s) (ω).
It is also possible to express the prediction error in terms of the integral
 π
1
P = log f (ω) dω.
2π −π

In fact, the one step ahead prediction error

⎛ 2 ⎞
 ∞ 
 
σ02 = inf E ⎝xt+1 − hk xt−k  ⎠ = 2π exp(P ).
h0 ,h1 ,...  
k=0

5.6.2.3 Conditions for stationary processes

Conditions for regularity and singularity for stationary processes x(t) with
continuous parameter, can be expressed in terms of the integral
 ∞
log f (ω)
Q= 2
dω,
−∞ 1 + ω

d
where as before, f (ω) = dω F (ω) is the a.s. existing derivative of the spectral
distribution function.

Theorem 5:9 For a stationary process {x(t), t ∈ R}, one has that
a) if Q = −∞, then x(t) is singular,
b) if Q > −∞ and the spectrum is absolutely continuous then x(t) is regular.
The decomposition of x(t) into one singular component which can be pre-
dicted, and one regular component which is a moving average, is analogous to
the discrete time case,
 t
x(t) = x(s) (t) + h(t − u) dζ(u),
u=−∞

where {ζ(t), t ∈ R} is process with uncorrelated increments.

2
We apply the theorem to processes with covariance functions r(t) = e−t /2
and r(t) = sint t . They have absolutely continuous spectra with spectral densi-
ties,
1 2
f (ω) = √ e−ω /2 and f (ω) = 1/2 for |ω| < 1,
2π
respectively. Obviously Q = −∞ is divergent in both cases, and we have
verified the statement that these processes are deterministic, although their
156 Ergodic theory and mixing Chapter 5

covariance functions tend to 0 as t → ∞. The Ornstein-Uhlenbeck process

with covariance function r(t) = e−α|t| and spectral density f (ω) = π(α2α+ω2 ) is
an example of a regular process with Q > −∞.
Note that in all three examples the covariance function r(t) tends to 0 as
t → ∞, but with quite different rates. For the regular Ornstein-Uhlenbeck
process it tends to 0 exponentially fast, while for the two singular (and hence
2
predictable) processes, the covariance falls off either much faster, as e−t /2 , or
much slower, as 1/t. Hence, we learn that stochastic determinism and non-
determinism are complicated matters, even if they are entirely defined in terms
of correlations and distribution functions.

5.6.3 Uniform, strong, and weak mixing

Predictability, regularity, and singularity are probabilistic concepts, defined and
studied in terms of prediction error moments and correlations. When it comes
to ergodicity, we have seen examples of totally deterministic sequences which
exhibit ergodic behavior, in the sense that the obey the law of large numbers.
To complicate matters, we have seen that a Gaussian process, with covariance
function tending to 0 at infinity, implying asymptotic independence, is always
ergodic, even if the remote past is in a deterministic sense determined by the
arbitrarily remote.
Ergodicity is a law of large numbers, time averages converge to a limit. In
statistics, one would also like to have some idea of the asymptotic distribution;
in particular, one would like a central limit theorem for normalized sums or
integrals, N
N
k=1 x(k) − AN t=0 x(t) dt − AN
, ,
BN BN
as N → ∞. This asks for general concepts, that guarantee the asymptotic
independence of functionals of a stochastic process.
For a general stationary process, only the ergodic theorem can be expected
−1 N
to hold. For example, if x(t) ≡ x, then BN k=1 x(k) = x with AN =
0 and BN = N , and this has the same distribution as x, regardless of N .
Here, there is a strong dependence between the x-variables. But even a very
weak dependence would not help us much to obtain any interesting limiting
distributions. For example, if y(k) is a sequence of independent, identically
distributed random variables, and

x(t) = y(t + 1) − y(t),

then
1 
N
y(N + 1) − y(1)
x(k) = .
BN BN
k=1

This has a non-trivial asymptotic distribution only if BN has a non-zero limit.

Central limit theorems for normalized sums typically require BN → ∞ and
Section 5.6 Mixing and asymptotic independence 157

then some mixing condition has to be imposed. (Martingale arguments are

another class of ”almost independence” conditions for asymptotic normality.;
see [36, 37].)
For any stochastic process {x(t), t ∈ R} define the σ -field Mba as the σ -
field generated by x(t); a ≤ t ≤ b. Taking a and b as ∓ infinity, we get
Mb−∞ and M∞ a as the σ -fields generated by {x(t); t ≤ b} and by {x(t); t ≥
a}, respectively. The following mixing conditions represent successively milder
conditions on the asymptotic independence.

uniform mixing: x(t) is called uniformly mixing (or φ-mixing) if there is a

non-negative function φ(n) such that φ(n) → 0 as n → ∞, and for all t
and events A ∈ Mt−∞ and B ∈ M∞ t+n ,

|P (A ∩ B) − P (A)P (B)| ≤ φ(n) P (A).

strong mixing: x(t) is called strongly mixing (or α-mixing) if there is a non-
negative function α(n) such that α(n) → 0 as n → ∞, and for all t and
events A ∈ Mt−∞ and B ∈ M∞ t+n ,

|P (A ∩ B) − P (A)P (B)| ≤ α(n).

weak mixing: x(t) is called weakly mixing if, for all events A ∈ Mt−∞ and
B ∈ M∞t ,

1
n
lim |P (A ∩ U −k B) − P (A)P (B)| = 0.
n→∞ n
k=1

Here, U −k B = {x(·); x(k + ·) ∈ B} ∈ M∞

k .

Of these, uniform mixing is the most demanding and weak mixing the least.

Theorem 5:10 Let {x(t); t ∈ Z} be a stationary Gaussian sequence. Then

a) x(t) is uniformly mixing if and only if it is m-dependent, i.e. there is an m

such that the covariance function r(t) = 0 for |t| > m.

b) x(t) is strong mixing if it has a continuous spectral density f (ω) ≥ c > 0 on

−π < ω ≤ π .

Proof: a) That m-dependence implies uniform mixing is obvious. To prove

the necessity, assume that r(t) is not identically 0 for large t. That x(t)
is uniformly mixing implies that for all A ∈ M0−∞ , with P (A) > 0 and all
B ∈ M∞ n ,
|P (B | A) − P (B)| < φ(n) → 0 (5.10)
158 Ergodic theory and mixing Chapter 5

as n → ∞. This directly implies that r(t) → 0 as t → ∞, since otherwise

there would be an infinite number of t = tk for which r(tk ) ≥ r0 > 0, say. Take
A = {x(0) > 0} ∈ M00 and B = {x(tk ) > 0} ∈ Mttkk . It is easy to see that

P (B | A) − P (B) ≥ c0 > 0

is bounded away from 0 by a positive constant c0 , and hence can not tend to
0.
For simplicity, assume E(x(t) = 0, E(x(t)2 ) = 1 and assume there is an
infinite number of t = tk for which ρk = r(tk ) > 0, but still r(tk ) → 0. Define
A = {x(0) > 1/ρk }, (obviously P (A) > 0) and let B = {x(tk ) > 1}. Since
x(0), x(tk ) have a bivariate normal distribution, the conditional distribution of
x(tk ) given x(0) = x is normal with mean ρk x and variance 1−ρ2k . As ρk → 0,
the conditional distribution of x(0) given x(0) > 1/ρk will be concentrated near
1/ρk and then x(tk ) will be approximately N (1, 1). Therefore,
∞ as ρk → 0,
P (B | A) → 1/2. On the other hand, P (B) = (2π)−1/2 1 exp(−y 2 /2) dy <
0.2. Hence (5.10) does not hold for φ(tk ) → 0.
b) For a proof of this, see [18] or [17]. 2

Theorem 5:11 A stationary process x(t) that is strong mixing is ergodic.

Proof: The proof is almost a complete repetition of the proof of Lemma 5.2.
Take an invariant event S and approximate it by a finite-dimensional event B
with |P (S) − P (B)| < . Suppose B ∈ Mba ⊂ Mb−∞ . Then the translated
event
Bn = Un B ∈ Ma+n
b+n
⊂ M∞ a+n ,

and hence
P (B ∩ Bn ) − P (B) · P (Bn ) → 0.
As in the proof of Lemma 5.2, it follows that P (B)2 = P (B) and hence P (S)2 =
P (S). 2

The reader could take as a challenge to prove that also weak mixing implies
that x(t) is ergodic; see also Exercise 14.
Section 5.6 Mixing and asymptotic independence 159

Exercises
5:1. Show that the following transformation of Ω = [0, 1), F = B , P =
Lebesgue measure, is measurable and measure preserving,

T x = 2x mod 1.

5:2. (Continued.) Define the random variable x(ω) = 0 if 0 ≤ ω < 1/2,

x(ω) = 1 if 1/2 ≤ ω < 1. Show that the sequence xn (ω) = x(T n−1 ω)
consists of independent zeros and ones, with probability 1/2 each.

5:3. Show that if T is measure preserving on (Ω, F, P ) and x is a random

variable, then E(x(ω)) = E(x(T ω)).
5:4. Show that every one-sided stationary sequence {xn , n ≥ 0} can be ex-
tended to a two-sided sequence {xn , n ∈ Z} with the same finite-dimen-
sional distributions.

5:5. Find a distribution for x0 which makes xk+1 = 4xk (1 − xk ) a stationary

sequence.

5:6. Prove that if E(|xn | < ∞, then P (xn /n → 0, as n → ∞) = 1; this was

used on page 138.
5:7. For any sequence of random variables, xn , and event B ∈ B , show that
the event {xn ∈ B, infinitely often} is invariant under the shift transfor-
mation.

5:8. Give an example of an ergodic transformation T on (Ω, F, P ) such that

T 2 is not ergodic.
5:9. Show that xn is ergodic if and only if for every A ∈ Bk , k = 1, 2, . . . ,

1
n
χA (xj , . . . , xj+k ) → P ((x1 , . . . , xk+1 ) ∈ A).
n
j=1

5:10. Showthat if x is non-negative with E(x) = ∞ and xn (ω) = x(T n−1 ω),
Sn = n1 xn , then Sn /n → ∞ if T is ergodic.
5:11. Prove Theorem 5:4.

5:12. Take two stationary and ergodic sequences xn and yn . Take one of the
two sequences at random with equal probability, zn = xn , n = 1, 2, . . . or
zn = yn , n = 1, 2, . . . . Show that zn is not ergodic.
5:13. Let {xn } and {yn } be two ergodic sequences, both defined on (Ω, F, P ),
and consider the bivariate sequence zn = (xn , yn ). Construct an exam-
ple that shows that zn need not be ergodic, even if {xn } and {yn } are
independent.
160 Ergodic theory and mixing Chapter 5

5:14. Prove that a sufficient condition for z(n) = (x(n), y(n)) to be ergodic,
if {x(n)} and {y(n)} are independent ergodic sequences, is that one of
{x(n)} and {y(n)} is weakly mixing.
Chapter 6

Vector processes and random

fields

6.1 Cross-spectrum and spectral representation

The internal correlation structure of a stationary process is defined by the
covariance function; the spectrum distributes the correlation over different fre-
quencies and in the spectral representation the process is actually built by
individual components, with independent amplitudes and phases, like in the
discrete spectrum case (4.18), on page 94. The phases are all independent and
uniformly distributed over (0, 2π). The spectrum does not contain any phase
information!
When we have two stationary processes {x(t), t ∈ R} and {t(t), t ∈ R}
which are correlated, their individual amplitudes and phases are still, of course,
independent, but for every frequence the amplitudes in the two processes can be
dependent, and there can exist a complicated dependence between the phases.
This cross dependence is describes by the cross-covariance function and the
cross-spectrum.
A stationary vector-valued process is a vector of p stationary processes,
x(t) = (x1 (t), . . . , xp (t)),
with stationary cross-covariances, for E(xj (t)) = 0,
rjk (t) = E(xj (s + t) · xk (s)) = rkj (−t).
If the processes are real, which we usually assume,
rjk (t) = E(xj (s + t) · xk (s)) = rkj (−t).
The covariance function
R(t) = (rjk (t))
is a matrix function of covariances, where each auto-covariance
 function rkk (t)
has its marginal spectral representation rkk (t) = eiωt dFkk (ω).

161
162 Vector processes and random fields Chapter 6

6.1.1 Spectral distribution

Theorem 6:1 (a) To every continuous covariance matrix function R(t) there
exists a spectral distribution F(ω) such that
 ∞
R(t) = eiωt dF(ω),
−∞

where F(ω) is a function of positive type, i.e. for every pair j, k , complex
z = (z1 , . . . , zn ), and frequency interval ω1 < ω2 ,

(Fjk (ω2 ) − Fjk (ω1 ))zj zk ≥ 0.
j,k

This says that ΔF(ω) = (Fjk (ω2 )− Fjk (ω1 )) is a non-negative definite Hermite
matrix.
(b) If Fjj (ω), Fkk (ω) are absolutely continuous with spectral densities fjj (ω),
fkk (ω), then Fjk (ω) is absolutely continuous, with

|fjk (ω)|2 ≤ fjj (ω) fkk (ω).

Proof: (a) Take the z and define the stationary process


y(t) = zj xj (t)
j

with covariance function

 
r z (t) = rjk (t)zj zk = eiωt dGz (ω),
jk

where Gz (ω) is a real, bounded and non-decreasing spectral distribution func-

tion. We then take, in order, zj = zk = 1, and zj = i, zk = 1, the rest being
0. This gives two spectral distribution G1 (ω), and G2 (ω), say, and we have,

rjj (t) + rkk (t) + rjk (t) + rkj (t) = eiωt dG1 (ω),

rjj (t) + rkk (t) + irjk (t) − irkj (t) = eiωt dG2 (ω).

 
Together with rjj (t) = eiωt dFjj (ω), and rkk (t) = eiωt dFkk (ω), we get

rjk (t) + rkj (t) = eiωt (dG1 (ω) − dFjj (ω) − dFkk (ω)) ,

irjk (t) − irkj (t) = eiωt (dG2 (ω) − dFjj (ω) − dFkk (ω)) ,
Section 6.1 Cross-spectrum and spectral representation 163

which implies

1
rjk (t) = rkj (t) = eiωt · (dG1 (ω) − idG2 (ω) − (1 − i)(dFjj (ω) + dFkk (ω)))
2

= eiωt dFjk (ω),

say, which is the spectral representation of rjk (t).

It is easy to see that ΔF(ω) has the stated properties; in particular that

ΔFjk (ω)zj zk ≥ 0. (6.1)
jk

(b) From (6.1), by taking only zj and zk to be non-zero, it follows that

ΔFjj |zj |2 + ΔFkk |zk |2 + 2 Re(ΔFjk zj zk ) ≥ 0,

which in turn implies that for any ω -interval, |ΔFjk |2 ≤ ΔFjj · ΔFkk .Thus, if
Fjj and Fkk have spectral densities, so does Fjk and

|fjk (ω)|2 ≤ fjj (ω) fkk (ω).

For real-valued vector processes, the spectral distributions may be put in

real form as for one-dimensional processes. In particular, the cross-covariance
function can be written
 ∞
rjk (t) = {cos ωt dGjk (ω) + sin ωt dHjk (t)} , (6.2)
0

where Gjk (ω) and Hjk (ω) are functions of bounded variation.

6.1.2 Spectral representation of x(t)

6.1.2.1 The spectral components
Each component xj (t) in a stationary vector process has its spectral representa-
tion xj (t) = eiωt dZj (ω) in terms of a spectral process Zj (ω) with orthogonal
increments. Further, for j = k , the increments of Zj (ω) and Zk (ω) over dis-
joint ω -intervals are orthogonal, while for equal ω -intervals,

E(dZj (ω) · dZk (ω)) = dFjk (ω),

E(dZj (ω) · dZk (μ)) = 0, for ω = μ.

The cross-correlation between the components of x(t) are determined by the

correlations between the spectral components. To see how it works, consider
164 Vector processes and random fields Chapter 6

processes with discrete spectrum, for which the spectral representation are sums
of the form (4.29),

xj (t) = σj (n) (Uj (n) cos ω(n)t + Vj (n) sin ω(n)t) .
n

Here Uj (n) and Vj (n) are real random variables with mean 0, variance 1,
uncorrelated for different n-values. The correlation between the xj - and the
xk -process is caused by a correlation between the U :s and V :s in the two
representations:
E(Uj (n)Uk (n)) = E(Vj (n)Vk (n)) = ρjk (n), j = k,
E(Uj (n)Vk (n)) = −E(Vj (n)Uk (n)) = −
ρjk (n), j = k,
E(Uj (n)Vj (n)) = 0,
for some ρjk (n) and ρjk (n) such that 0 ≤ ρjk (n)2 + ρjk (n)2 ≤ 1. Direct
calculation of auto- and cross-covariances gives

rjk (t) = σj (n)σk (n) (ρjk (n) cos ω(n)t + ρjk (n) sin ωn t) ,
n

= Ajk (n) cos (ω(n)t − Φjk (n)) , j = k, (6.3)
n

rkj (t) = σj (n)σk (n) (ρjk (n) cos ω(n)t − ρjk (n) sin ωn t) ,
n

= Akj (n) cos (ω(n)t − Φkj (n)) , j = k, (6.4)
n

rj (t) = σj (n)2 cos ω(n)t, j = 1, . . . , p. (6.5)
n

Here, Ajk (n) = Akj (n) = σj (n)σk (n) ρjk (n)2 + ρjk (n)2 , represent the corre-
lation between the amplitudes, while Φjk (n) = −Φkj (n), with
ρjk (n) ρjk (n)
cos Φjk (n) =  , sin Φjk (n) =  ,
ρjk (n)2 + ρjk (n)2 ρjk (n)2 + ρjk (n)2
represent the phase relations.
The corresponding spectral distributions Fkk (ω) are symmetric with mass
ΔFk = σk (n)2 /2
at the frequencies ±ω(n), while for j = k , the cross spectrum Fjk (ω) is skewed
if ρjk (n) = 0, with mass
⎧
⎪ 1 ρ (n) + i ρjk (n)
⎪
⎪ A (n) e−iΦjk (n) = 12 Ajk (n)  jk , at ω = ωn ,
⎨ 2 jk ρjk (n)2 + ρjk (n)2
ΔFjk =
⎪
⎪ ρ (n) − i ρjk (n)
⎪
⎩ 12 Ajk (n) eiΦjk (n) = 12 Ajk (n)  jk , at ω = −ωn .
ρjk (n) + ρjk (n)2
2
Section 6.2 Some random field theory 165

6.1.2.2 Phase, amplitude, and coherence spectrum

The frequency dependent function Φjk (n) is called the phase spectrum and it
defines the time delay between components in xk (s) and xj (s + t); the correla-
tion between the components Uj (n) cos(ω(n)(s+t))+Vj (n) sin(ω(n)(s+t)) and
Uk (n) cos(ω(n)s) + Vk (n) sin(ω(n))s have their maxima at t = Φjk (n)/ω(n).
Further, Ajk (n)/2 is called the amplitude spectrum and the squared coher-
ence spectrum is defined as

|ΔFjk (n)|2
= ρjk (n)2 + ρjk (n)2 .
ΔFjj (n)ΔFkk (n)

For processes with continuous spectra and cross spectral density

1
fjk (ω) = Ajk (ω) eiΦjk (ω) ,
2
the amplitude spectrum, phase spectrum, and squared coherence spectrum are
defined as Ajk (ω)/2, Φjk (ω), and |fjk (ω)|2 /(fjj (ω)fkk (ω)), respectively.

6.1.2.3 Cross-correlation in linear filters

In a linear filter, the cross-covariance and cross spectrum describe the relation
 iωt
between
 the input process
 x(t) = e dZx (ω) and the output process y(t) =
h(t − u)x(u) du = g(ω)e dZx (ω). One calculates easily,
iωt

rxy (t) = E(x(s + t) · y(s))

 

 s 
=E eiω(s+t)
dZx (ω) · g(ω )eiω dZx (ω ) = eiωt g(ω) dFx (ω),

so the cross spectral distribution is

dFxy (ω) = g(ω)dFx (ω).

By estimating the cross-covariance or the cross spectrum between input and

output in an unknown filter, one can estimate the frequency function as g∗ (ω) =
∗ (ω)/f ∗ (ω) = f ∗ (−ω)/f ∗ (ω), when f ∗ (ω) > 0.
fxy x xy x x

6.2 Some random field theory

A random field is a stochastic process x(t) with multi-dimensional parameter

t = (t1 , . . . , tp ).

For example, if t = (t1 , t2 ) is two-dimensional we can think of (t1 , t2 , x(t)) as a

random surface. A time-dependent random surface is a field (s1 , s2 , x(t, s1 , s2 ))
with t = (t, s1 , s2 ), where t is time and (s1 , s2 ) ∈ R2 is location. In the general
166 Vector processes and random fields Chapter 6

theory we use t as generic notation for the parameter; in special applications

to random time dependent surfaces we use (t, (s1 , s2 )) as parameter.
From being mainly used in geoscience, like geology under the name of geo-
statistics, and in marine science as models for a random sea, cf. Section 1.6.3,
random fields are now widely used in all sorts of applications with spatial or
spatial-temporal variability.

6.2.1 Homogeneous fields

Define the mean value and covariance functions for random fields in the natural
way as m(t) = E(x(t)) and r(t, u) = C(x(t), x(u)).
The analogue of a stationary process is a homogeneous field. The field is
called homogeneous if m(t) is constant m and r(t, u) depends only on the
difference t − u, i.e. assuming a real field with m = 0,

r(t) = r(u + t, u) = E(x(u + t) · x(u)).

The covariance of the process values at two parameter points depends on dis-
tance as well as on direction of the vector between the two points.
In spatial applications it is popular to use the variogram defined by
 
2γ(u, v) = E |x(u) − x(v)|2

or the semi-variogram γ(t).

The variogram plays the same role as the incremental variance does in a
Wiener process. There E(|w(s + t) − w(s)|2 = t · σ 2 is independent of s. A field
for which the variogram only depends on the vector u − v is called intrinsically
stationary. The variogram for a homogeneous field is γ(t) = r(0) − r(t). A
homogeneous field is also intrinsically stationary, but as seen from the Wiener
process the converse is not sure.

Theorem 6:2 (a) The covariance function r(t) of a homogeneous random field
has a spectral distribution

r(t) = eiω·t dF (ω),

where ω·t = ω1 t1 +. . .+ωp tp , and F (ω) is a p-dimensional spectral distribution

function depending on the frequency parameter ω = (ω1 , . . . , ωp ).1
(b) There exists a stochastic spectral process Z(ω) with orthogonal increments
ΔZ(ω) over rectangles Δω = [ω1 , ω1 + Δ1 ] × . . . × [ωp , ωp + Δp ], such that
E(ΔZ(ω)) = 0 and

E(|ΔZ(ω)|2 ) = ΔF (ω), and E(ΔZ(ω 1 ) · ΔZ(ω2 )) = 0,

1
F (ω) is equal to a regular p -dimensional probability distribution function multiplied by
a positive constant, equal to the variance of the process.
Section 6.2 Some random field theory 167

for disjoint rectangles Δω1 and Δω2 , and


x(t) = eiω·t dZ(ω). (6.6)

6.2.2 Isotropic fields

In an homogeneous isotropic field, the correlation properties are the same in
all directions,
) i.e. the covariance function r(t) depends only on the distance
t = t21 + . . . + t2p . This type of process model is natural when one can
not identify any special directional dependent stochastic properties in the field,
but rather it keeps its distributions after rotation (and translation). Many
phenomena in the natural world share this property while other naturally do
not. Ocean waves are non-isotropic, chemical concentration field in the bottom
sediment in the ocean or the disturbance field in mobile phone communication
might well be isotropic.
As we have seen, a spectral distribution for a honogeneous field need only
satisfy the requirement that it is non-negative, integrable and symmetric. The
spectral distribution for an isotropic field needs to satisfy a special invariance
condition, giving it a particularly simple structure.

Theorem 6:3 The covariance function r(t) to a homogeneous isotropic field

a mixture of Bessel functions,
 ∞
J(p−2)/2 (ω · t)
r(t) = r(t) = dG(ω),
0 (ω · t)(p−2)/2
where G(ω) is a bounded, non-decreasing function, and Jm (ω) is a Bessel func-
tion of the first kind of order m,
∞
 (ω/2)2k+m
Jm (ω) = (−1)k .
k! Γ(k + m + 1)
k=0

Proof: We have r(t) = eiω·t  dF (ω). Introduce the spherical coordinates,
ω = ω · (1 , . . . , p ), ω = ω, 2k = 1, and let 1 = cos θp−1 . For every
θp−1 , (2 , . . . , p ), defines a point on the sphere with radius 1 − cos2 θp−1 =
sin θp−1 .
Since r(t) depends only on t, we can calculate the integral for the special
point t = (t, 0, . . . , 0), to get
 
r(t) = e iω1 t
dF (ω) = eiω t cos θp−1 dF (ω).

With G(ω) = ω≤ω dF (ω) we find that
 ∞  
iωt cos θp−1
r(t) = e dσ(θ) dG(ω),
ω=0 θ
168 Vector processes and random fields Chapter 6

where dσ is the rotation invariant measure on the (p − 1)-dimensional unit

sphere. For fixed θp−1 we integrate (θ1 , . . . , θp−2 ) over the sphere with radius
sin θp−1 , with area Cp−1 sinp−2 θp−1 . We find
 ∞  π 
iωt cos θ p−2
r(t) = Cp−2 e sin θ dθ dG(ω),
ω=0 θ=0

which is the stated form, if we incorporate the constant Cp−2 into G. 2

6.2.2.1 Isotropic fields with special structure

Of course, a homogeneous field x(t1 , t2 ) with two-dimensional parameter and
spectral density fx (ω1 , ω2 ) and covariance function rx (t1 , t2 ) gives rise to a
stationary process when observed along a single straight line, for example along
t2 = 0. Then x(t  1 , 0) has covariance function r(t1 ) = rx (t1 , 0) and spectral
density f (ω1 ) = ω2 fx (ω1 , ω2 ) dω2 .
It is important to realise that even if any non-negative definite function can
act as covariance function for a stationary process, not all stationary processes,
and corresponding covariance functions, can occur as an observation of a section
in an isotropic field with two-dimensional parameter. Only those functions that
can be expressed as  ∞
r(t) = J0 (ωt) dG(ω),
0
with bounded non-decreasing G(ω) are possible. A simple sufficient condition
for a one-dimensional spectral density f (ω) to be obtained from a section of
an isotropic field is the following:

• Any function f (ω), ω ∈ R that is non-increasing for ω ≥ 0 and integrable

over R can act as the spectral density for a section, e.g. x(t1 , 0), of an
isotropic field in two dimensions.

One should also note the particularly simple form of the covariance function
for the case p = 3,  ∞
sin(ωt)
r(t) = dG(ω).
0 ωt
Another special case that needs special treatment is when the parameter t
is composed of both a time and a space parameter, t = (t, s1 , . . . , sp ), one could
hope for isotropy in (s1 , . . . , sp ) only, in which case the spectral form becomes
 ∞  ∞
r(t, (s1 , . . . , sp )) = eiνt · Hp (ω(s1 , . . . , sp )) dG(ν, ω),
ν=−∞ ω=0

where
Hp (x) = ((2/x))(p−2)/2 Γ(p/2) J(p−2)/2 (x).
The form of the covariance function for an isotropic field as a mixture of
Bessel functions is useful for non-parametric estimation from data. It is only
Section 6.2 Some random field theory 169

the weight function G(ω) that needs to be estimated, for example as a discrete
distribution.
The theorem gives all possible isotropic covariance functions valid in the
special dimension p. Some functions can be used as covariance function in any
dimension, for example

r(t) = σ 2 exp(−φtα ),

for any α ∈ (0, 2).

Another popular class of covariance functions valid in any dimension is the
Whittle-Matérn family, which is

σ2 √ √
r(t) = ν−1
(2 ν tφ)ν Kν (2 ν tφ),
2 Γ(ν)

where Kν is a modified Bessel function of order ν . The value of ν determines

the smoothness of the field. When ν → ∞, the Whittle-Matérn covariance
functions tends to a Gaussian shape, in which case the field is infinitely differ-
entiable; cf. Chapter 2.
In all these forms, φ is a spatial scale parameter. Often one adds an extra
variance term σ02 at the center, for t = 0, to account for the so called nugget
effect, a variance component with a covariance range too short to be observed.
A modern account of spatial data modeling is given in [2], also by means of
spectral and correlation models.

6.2.3 Randomly moving surfaces

The spectral representation makes it easy to imagine the structure of a ran-
domly moving surface, homogeneous in time as well as in space. The spectral
representation (6.6) is a representation of real field x(t) as a packet of directed
waves, At cos(ω · t + φω ), with random amplitude and phase, and constant on
each plane parallel to ω · t = 0. For example, with t = (s1 , s2 , t) and t is time
and (s1 , s2 ) is space, and ω = (κ1 , κ2 , ω), the elementary waves are

Aω cos(κ1 s1 + κ2 s2 + ωt + φω ).

For fixed t this is a cosine-function in the plane, which is zero along lines
κ1 s1 +κ2 s2 +ωt+φω = π/2+kπ , k integer. For fixed (s1 , s2 ) it is a cosine wave
with frequency ω . The parameters κ1 and κ2 are called the wave numbers.
In general there is no particular relation between the time frequency ω and
the space frequencies ν , except for water waves, which we shall deal with later.
However, one important application of space-time random fields is the model-
ing of environmental variables, like the concentration of a hazardous pollutant.
Over a reasonably short period of time the concentration variation may be
regarded as statistically stationary in time, at least averaged over a 24 hour
170 Vector processes and random fields Chapter 6

period. But it is often unlikely that the correlation structure in space is inde-
pendent of the absolute location. Topography, location of cities and pollutant
sources, makes the process inhomogeneous in space.
One way to overcome the inhomogeneity is to make a transformation of the
space map and move each observation point (s1 , s2 ) to a new location (ŝ1 , ŝ2 )
so that the field x̂(t, ŝ1 , ŝ2 ) = x(t, s1 , s2 ) is homogeneous. This may not be
exactly attainable, but the technique is often used in environmental statistics
for planning of measurements.

6.2.4 Stochastic water waves

Stochastic water waves are special cases of homogeneous random fields, for
which there is a special relation between time and space frequencies (wave
numbers). For a one-dimensional time dependent Gaussian wave x(t, s), where
s is distance along an axis, the elementary waves have the form

Aω cos(ωt − κs + φω ).

By physical considerations one can derive an explicit relation, the dispersion

relation, between wave number κ and frequency ω . If h is the water depth,

ω 2 = κg tanh(hκ),

which for infinite depth2 reduces to ω 2 = κg. Here g is the constant of gravity.
A Gaussian random wave is a mixture of elementary waves of this form, in
spectral language, with κ > 0 solving the dispersion relation,
 ∞  ∞
i(ωt−κs)
x(t, s) = e dZ+ (ω) + ei(ωt+κs) dZ− (ω)
ω=−∞ ω=−∞
= x+ (t, s) + x− (t, s).

Here is a case when it is important to use both positive and negative frequencies;
cf. the comments in Section 4.3.3.4. Waves described by x+ (t, s) move to the
right and waves in x− (t, s) move to the left with increasing t.
Keeping t = t0 or s = s0 fixed, one obtains a space wave, x(t0 , s), and a time
wave, x(t, s0 ) respectively. The spectral density of the time wave, x(t, s0 ) =
x+ (t, s) + x− (t, s0 ) is called the wave frequency spectrum,

fxf req (ω) = f+ (ω) + f− (ω),

and we see again that it is not possible to distinguish between the two wave
directions by just observing the time wave.
2
tanh x = (ex − e−x )/(ex + e−x ) .
Section 6.2 Some random field theory 171

The space wave has a wave number spectrum given by the equation, for
infinite water depth, with ω 2 = gκ > 0,
2ω space 2
fxtime (ω) = f (ω /g),
g x
%
space 1 g time √
fx (κ) = f ( gκ).
2 κ x
One obvious effect of these relations is that the space process seems to have
more short waves than can be inferred from the time observations. Physically
this is due to the fact that short waves travel with lower speed than long waves,
and they are therefore not observed as easily in the time process. Both the
time wave observations and the space registrations are in a sense “biased” as
representatives for the full time-space wave field. 
In Chapter 3, Remark 3:3, we introduced the mean period 2π ω0 /ω2 of
a stationary time process. The corresponding quantity for the space wave is
the mean wave length, i.e. the average distance between two upcrossings of the
mean level by the space process x(t0 , s). It is expressed in terms of the spectral
moments of the wave number spectrum, in particular
 
κ0 = fxspace(κ) dκ = fxtime dω = ω0 ,
κ ω
  %
2 space ω4 g2 time 2ω
κ2 = κ fx (κ) dκ = 2
f
2 x
(ω) dω = ω4 /g2 .
κ ω 2g ω g

The average wave length is therefore 2πg ω0 /ω4 . We see that the average wave
length is more sensitive to the tail of the spectral density than is the average
wave period. Considering the difficulties in estimating the high frequency part
of the wave spectrum, all statements that rely on high spectral moments are
unreliable.
The case of a two-dimensional time dependent Gaussian wave x(t, s1 , s2 ),
the elementary waves with frequency ω and direction θ become
Aω cos(ωt − κ(s1 cos θ + s2 sin θ) + φω ), (6.7)
where ω > 0 and κ > 0 is given by the dispersion relation. With this choice of
sign, θ determines the direction in which the waves move.
The spectral density for the time-space wave field specifies the contribution
to x(t, (s1 , s2 )) from elementary waves of the form (6.7). Summed (or rather
integrated) over all directions 0 ≤ θ < 2π , they give the time wave x(t, s0 ), in
which one cannot identify the different directions. The spectral distribution,
called the directional spectrum, is therefore often written in polar form, based
on the spectral density fxtime (ω) for the time wave, as
f (ω, θ) = fxtime (ω)g(ω, θ).
 2π
The spreading function g(ω, θ), with 0 g(ω, θ) dθ = 1, specifies the relative
contribution of waves from different directions. It may we frequency dependent.
172 Vector processes and random fields Chapter 6

Directional Spectrum; level curves at 2, 4, 6, 8, 10 300

90
0.8
120 60

0.6 250

150 0.4 30
200

0.2

150
180 0

100

210 330

50

240 300

270 0
0 50 100 150 200 250 300

Figure 6.1: Left: Level curves for directional spectrum with frequency depen-
dent spreading. Right: Level curves for simulated Gaussian space sea.

Example 6:1 Wave spectra for the ocean under different weather conditions
are important to characterize the input to (linear or non-linear) ship models.
Much effort has been spent on design and estimation of typical wave spectra.
One of the most popular is the Pierson-Moskowitz spectrum,
α −1.25(ωm /ω)4
fPtime
M (ω) = e .
ω5
or the variant, the Jonswap spectrum, in which an extra factor γ > 1 is
introduced to enhance the peak of the spectrum,
α −1.25(ωm /ω)4 exp(−(1−ω/ωm )2 /2σm
2 )
fJtime (ω) = 5
e γ .
ω
In both spectra, α is a main parameter for the total variance, and ωm defines
the “peak frequency”. The parameters γ and σm determine the peakedness
of the spectrum. The spectrum and a realization of a Gaussian process with
Jonswap spectrum was shown in Example 1:4.
Figure 6.1 shows to the right, the level curves for a simulated Gaussian
wave surface with the directional spectrum with frequency dependent spreading,
shown on the left. Frequency spectrum is of Jonswap type.
As mentioned in the historical Section 1.6.3, Gaussian waves have been used
since the early 1950, with great success. However, since Gaussian processes are
symmetric, x(t) has the same distribution as −x(t) and as x(−t), they are not
very realistic for actual water waves except in special situations; deep water, no
strong wind. Much research is presently devoted to development of “non-linear”
stochastic wave models, where elementary waves with different frequencies can
interact, in contrast to the “linear” Gaussian model, where they just add up.
Section 6.2 Some random field theory 173

Exercises
6:1. To be written.
174 Vector processes and random fields Chapter 6
Appendix A

The axioms of probability

Here is a collection of the basic probability axioms, together with proofs of the
extension theorem for probabilities on a field (page 5), and of Kolmogorov’s
extension theorem, from finite-dimensional probabilities to infinite-dimensional
ones (page 10).

A.1 The axioms

A probability P is a countably additive measure on a probability space, as
defined here.

Definition A:1 (a) A family of subsets F0 to an arbitrary space Ω is called

a field (or algebra) if it contains the whole set Ω and is closed under the set
operations complement, A∗ , union, A ∪ B , and intersection, A ∩ B , i.e. if A
and B are sets in the family F0 , then also the complement A∗ and the union
A ∪ B belong to F0 , etc. It then also contains all unions of finitely many sets
A1 , . . . , An in F0 .
(b) A field F of subsets is called a σ -field (or σ -algebra) if it contains all
countable unions and intersections of its sets, i.e. if it is a field and furthermore,

A1 , A2 , . . . ∈ F implies ∪∞
n=1 An ∈ F.

(c) A probability measure P on a sample space Ω with a σ -field F of events,

is a function defined for every A ∈ F , with the properties

(1) 0 ≤ P (A) ≤ 1 for all A ∈ F .

(2) P (Ω) = 1.
∞
(3) For disjoint sets Ak ∈ F , k = 1, 2, . . . one has P (∪∞
1 Ak ) = 1 P (Ak ).

(d) Equivalent with (c) is

175
176 Appendix A.

(1) 0 ≤ P (A) ≤ 1 for all A ∈ F .

(2) P (Ω) = 1.
(3’) If A1 , A2 ∈ F are disjoint, then P (A1 ∪ A2 ) = P (A1 ) + P (A2 ).
(3”) If A1 ⊇ A2 ⊇ . . . ∈ F with ∩∞
1 An = ∅, then limn→∞ P (An ) = 0.

A typical field F0 in R is the family of sets which are unions of a finite

number of intervals. The smallest σ -field that contains all sets in F0 is the
family B of Borel sets.

A.2 Extension of a probability from field to σ -field

How do we define probabilities? For real events, of course, via a statistical
distribution function F (x). Given a distribution function F , we can define a
probability P for finite unions of real half open disjoint intervals,

n
P (∪n1 (ak , bk ]) = (F (bk ) − F (ak )) . (A.1)
1

If we define the probability of a single point as P ([a]) = F (a) − limx↑a F (x),

we can define a probability to every finite union of real intervals.
Sets which are unions of a finite number real intervals, (a, b], (a, b), [a, b),
[a, b] with −∞ ≤ a < b ≤ ∞, form a field on R, and they can be given
probabilities via a distribution function. The question is, does this also give
probabilities to the more complicated events (Borel sets) in the σ -field F gen-
erated by the intervals. The answer is yes, as stated in the following extension
theorem, Carathéodory’s extension theorem, which is valid not only for intervals
and Borel sets, but for any field F0 and generated σ -field F . For a proof, the
reader is referred to any text book in probability or measure theory, e.g. [36].

Theorem A:1 Suppose P is a function which is defined for all sets in a field
F0 , there satisfying the probability axioms, i.e. (with three equivalent formula-
tions of Condition 4),

(1) 0 ≤ P (A) ≤ 1 for all A ∈ F0 .

(2) P (Ω) = 1.
(3) If A1 , A2 ∈ F0 are disjoint, then P (A1 ∪ A2 ) = P (A1 ) + P (A2 ).
(4a) If A1 ⊇ A2 ⊇ . . . ∈ F0 with ∩∞
1 An = ∅, then limn→∞ P (An ) = 0.

(4b) If A1 , A2 , . . . ∈ F0 are disjoint and ∪∞

k=1 Ak ∈ F0 , then
∞

P (∪∞
k=1 Ak ) = P (Ak ).
k=1
Section A.2 Extension of a probability from field to σ -field 177
∞
(4c) If A1 , A2 , . . . ∈ F0 are disjoint and ∪∞
k=1 Ak = Ω then k=1 P (Ak ) = 1.
Then one can extend P to be defined, in one and only one way, for all sets in
the σ -field F generated by F0 , so that it still satisfies the probability axioms.
We can now state and prove the existence of probability measures on the
real line with a given distribution function.

Theorem A:2 Let F (x) be a statistical distribution function on R, i.e. a non-

decreasing, right-continuous function with F (−∞) = 0, F (∞) = 1. Then
there exists exactly one probability measure P on (R, B), such that P ((a, b]) =
F (b) − F (a).
Proof: We shall use the extension Theorem A:1. The Borel sets B equal the
σ -field generated by the field F0 of unions of finitely many intervals. Equation
(A.1) extended by singletons, defines P for each set in F0 , and it is easily
checked that properties (1), (2), and (3) hold. The only difficult part is (4a),
which we prove by contradiction.
The idea is to use Cantor’s theorem that every decreasing sequence of com-
pact, non-empty sets, has a non-empty intersection. Assume, for a decreasing
sequence of sets An ∈ F0 , that P (An ) ↓ h > 0. We show that then the in-
tersection of the An -sets is not empty. Each An consists of finitely many half
open intervals. It is then possible to remove from An a short piece from the
left end, to make it closed and bounded, i.e. there exists a compact, nonempty,
Kn ⊂ An , such
P (An − Kn ) ≤ ε/2n .
(Convince yourself that P (An − Kn ) is defined.) Then
Lm = ∩ m
1 Kn ⊆ Km ⊆ Am ,

form a decreasing sequence,

L1 ⊇ L2 ⊇ . . . .
If we can prove that the Lm can be taken nonempty, we can use Cantor’s
theorem, and conclude that they have a nonempty intersection, i.e. there exists
at least one point x ∈ ∩∞ ∞
1 Lm , which also implies x ∈ ∩1 Am , so the Ak do not
decrease to the empty set. The proof would be finished.
Hence, it remains to prove that we can choose each Lm nonempty. Take
ε < h. Then
P (Am − Lm ) = P (Am − ∩m
1 Kn ) = P (∪1 (Am − Kn ))
m


m 
m 
m
≤ P (Am − Kn ) ≤ P (An − Kn ) ≤ ε/2n ≤ ε,
1 1 1

P (Lm ) = P (Am ) − P (Am − Lm ) ≥ h − ε > 0,

and Lm is non-empty. 2
178 Appendix A.

A.3 Kolmogorov’s extension to R∞

Kolmogorov’s existence, or extension, theorem from 1933, allows us to define
a stochastic process through its family of finite-dimensional distribution func-
tions. Kolmogorov’s book [20] appeared after a period of about 30 years at-
tempts to give probability theory a solid mathematical foundation; in fact,
Hilbert’s sixth problem (1900) asked for a logical investigation of the axioms of
probability.

Theorem A:3 Extension formulation: Every consistent family {Pn } of prob-

ability measures on (Rn , Bn ), n = 1, 2, . . . , can be uniquely extended to a prob-
ability measure P on (R∞ , B∞ ), i.e. in such a way that

P ((a1 , b1 ] × (a2 , b2 ] × . . . × (an , bn ] × R∞ )

= Pn ((a1 , b1 ] × (a2 , b2 ] × . . . × (an , bn ]) .

Existence formulation: To every consistent family of finite-dimensional dis-

tribution functions, F = {Ftn }∞
n=1 , there exists one and only one probability
∞
measure P on (R , B∞ ) with

P (x1 ≤ b1 , . . . , xn ≤ bn ) = Fn (b1 , . . . , bn ).

Proof: We prove first the Extension formulation. We are given one probability
measure Pn on each one of (Rn , Bn ), n = 1, 2, . . . . Consider the intervals in
R∞ , i.e. the sets of the form

I = {x = (x1 , x2 , . . .); xi ∈ (ai , bi ], i = 1, . . . , n},

for some n, and unions of a finite number of intervals. Define P for each
interval by  n 
4
P (I) = Pn (ai , bi ] .
1

Let I1 and I2 be two disjoint intervals. They may have different dimensions
(n1 = n2 ), but setting suitable ai or bi equal to ±∞, we may assume that
they have the same dimension. The consistency of the family {Pn } guarantees
that this does not change their probabilities, and that the additivity property
holds, that is, if also I1 ∪ I2 is an interval, then P (I1 ∪ I2 ) = P (I1 ) + P (I2 ). It
is easy to extend P with additivity to all finite unions of intervals. By this we
have defined P on the field F0 of finite unions of intervals, and checked that
properties (1), (2), and (3) of Theorem A:1 hold.
Now check property (4a) in the same way as for Theorem A:2, for a decreas-
ing sequence of non-empty intervals with empty intersection,

I1 ⊇ I2 ⊇ . . . , with ∩∞
1 In = ∅,
Section A.3 Kolmogorov’s extension to R∞ 179

and suppose P (In ) ↓ h > 0.1 We can always assume In to have dimension n,
(n) (n)
In = {x ∈ R∞ ; aj < xj ≤ bj , j = 1, . . . , n},

and we can always assume the aj and bj to be bounded. As in the proof of

Theorem A:2, remove a small piece of the lover side of each interval In to get a
compact Kn and define Lm = ∩m 1 Kn . By removing a small enough piece one
can obtain that P (Lm ) ≥ h/2 > 0 so Lm is non-empty.
If we write
(1) (1)
L1 : α1 ≤ x1 ≤ β1
(2) (2) (2) (2)
L2 : α1 ≤ x1 ≤ β1 , α2 ≤ x2 ≤ β2
(3) (3) (3) (3) (3) (3)
L3 : α1 ≤ x1 ≤ β1 , α2 ≤ x2 ≤ β2 , α3 ≤ x3 ≤ β3
.. .. .. ..
. . . .

(n) (n)
For each j , [αj , βj ], n = j, j + 1, . . . , is a decreasing sequence of non-empty,
closed and bounded intervals, and by Cantor’s theorem they have at least one
(n) (n)
common point, xj ∈ ∩∞ n=j [αj , βj ]. Then, x = (x1 , x2 , . . .) ∈ Ln for all n.
Hence x ∈ Ln ⊆ In for all n and the intersection ∩∞ 1 In is not empty. This
contradiction shows that P (In ) ↓ 0, and (3”) is shown to hold.
The conditions (1), (2), (3), and (4a) of Theorem A:1 are all satisfied, and
hence P can be extended to the σ -field F generated by the intervals.
To get the Existence formulation, just observe that the family of finite-dimen-
sional distributions uniquely defines Pn on (Rn , Bn ), and use the extension.
2

Exercises
A:1. Let Z be the integers, and A the family of subsets A, such that either
A or its complement Ac is finite. Let P (A) = 0 in the first case and
P (A) = 1 in the second case. Show that P can not be extended to a
probability to σ(A), the smallest σ -field that contains A.

1
Property (4a) deals with a decreasing sequence of finite unions of intervals. It is easy to
convince oneself that it suffices to show that (4a) holds for a decreasing sequence of intervals.
180 Appendix A.
Appendix B

Stochastic convergence

Here we summarize the basic types of stochastic convergence and the ways we
have to check the convergence of a random sequence with specified distributions.

Definition B:1 Let {xn }∞ n=1 be a sequence of random variables x1 (ω), x2 (ω),
. . . defined on the same probability space, and let x = x(ω) be a random variable,
defined on the same probability space. Then, the convergence xn → x as n → ∞
can be defined in three ways:
a.s.
• almost surely, with probability one (xn → x): P ({ω; xn → x}) = 1;


• in quadratic mean (xn → x): E |xn − x|2 → 0;
q.m.

P
• in probability (xn → x): for every  > 0, P (|xn − x| > ) → 0.
L
Furthermore, xn tends in distribution to x, (in symbols xn → x) if

P (xn ≤ a) → P (x ≤ a)

for all a such that P (x ≤ u) is a continuous function of u at u = a.

B.1 Criteria for convergence almost surely

In order for a random sequence xn to converge almost surely (i.e. with proba-
bility one), to the random variable x, it is necessary and sufficient that

lim P (|xn − x| > δ for at least one n ≥ m) = 0 (B.1)

m→∞

for every δ > 0.

181
182 Appendix B.

To prove this, note that if ω is an outcome such that the real sequence
xn (ω) does not converge to x(ω), then
⎧ ⎫
⎨ ∞ 3∞  ∞ ⎬
ω∈ |xn (ω) − x(ω)| > 1/q .
⎩ n=m
⎭
q=1 m=1

Here, the innermost event has probability

P (∪∞
n=m |xn (ω) − x(ω)| > 1/q) = P (|xn − x| > 1/q for at least one n ≥ m) ,

and this is 0 for all q if and only if (B.1) holds for all δ > 0. The reader should
complete the argument, using that P (∪k Ak ) = 0 if and only if P (Ak ) = 0 for
all k , and the fact that if B1 ⊇ B2 ⊇ . . . is a non-increasing sequence of events,
then P (∩k Bk ) = limk→∞ P (Bk ). Now,
∞

P (|xn − x| > δ for at least one n ≥ m ) ≤ P (|xn − x| > δ),
n=m

and hence a simple sufficient condition for (B.1) and a sufficient condition for
almost sure convergence is that for all δ > 0,
∞

P (|xn − x| > δ) < ∞. (B.2)
n=1

(In fact, the first Borel-Cantelli lemma directly shows that (B.2) is sufficient
for almost sure convergence.)
A simple moment condition is obtained from the inequality P (|xn − x| >
δ) ≤ E(|xn − x|h )/δh , giving that a sufficient condition for almost sure conver-
gence is
∞

E(|xn − x|h ) < ∞, (B.3)
n=1
for some h > 0.
A Cauchy convergence type condition is the following: sufficient condition
for almost sure convergence:
 if there exist
∞ two sequences of positive numbers
δn and n such that ∞ δ
n=1 n < ∞ and n=1 n < ∞, and such that

P (|xn+1 − xn | > δn ) < n , (B.4)

a.s.
then there exists a random variable x such that xn → x.
To see this, use the Borel-Cantelli lemma to conclude that
P (|xn+1 − xn | > δn for infinitely many n) = 0.
Thus, for almost all ω , there is a number N , depending on the outcome ω ,
such that
|xn+1 − xn | < δn for all n ≥ N .

Since δn < ∞, the sequence xn (ω) converges to a limit x(ω) for these
outcomes. For ω where the limit does not exist, set x(ω) = 0, for example.
a.s.
Then xn → x, as was to be proved.
Section B.2 Criteria for convergence in quadratic mean 183

B.1.0.1 Uniform convergence of random functions

A sequence of random variables can converge almost surely, and we have just
given sufficient conditions for this. But we shall also need convergence of a
sequence of random functions {xn (t); t ∈ T }, where T = [a, b] is a closed
bounded interval.

Definition B:2 A sequence of functions {xn (t); a ≤ t ≤ b} converges uni-

formly to the function {x(t); a ≤ t ≤ b} if

max |xn (t) − x(t)| → 0, as n → ∞,

a≤t≤b

that is, if xn lies close to the limiting function x in the entire interval [a, b]
for all sufficiently large n.

It is a basic result in real analysis that if a sequence of continuous functions

converges uniformly in a closed and bounded interval, then the limiting function
is also continuous. This fact will be useful when we show the almost sure sample
function continuity of a random function.
Condition (B.4) can be restated to deal with almost sure uniform conver-
gence of random functions:
 if there exist
∞two sequences of positive numbers δn
and n such that ∞ n=1 nδ < ∞ and n=1 n < ∞, and such that

P ( max |xn+1 (t) − xn (t)| > δn ) < n , (B.5)

a≤t≤b

a.s.
then there exists a random function x(t); a ≤ t ≤ b, such that xn (t) → x(t)
uniformly for t ∈ [a, b].

B.2 Criteria for convergence in quadratic mean

Some of the representation theorems for stationary processes express a process
as a complex stochastic integral, defined as a limit in quadratic mean of ap-
proximating sums of complex-valued random variables. To define a quadratic
mean integral, or other limit of that kind, one needs simple convergence criteria
q.m.
for when xn → x for a sequence of random variables with E(|xn |2 ) < ∞.
The Cauchy convergence criterion for convergence in quadratic mean states
that a necessary and sufficient condition for that there exists a (possibly com-
q.m.
plex) random variable x such that xn → x is that

E(|xm − xn |2 ) → 0, (B.6)

as n and m tend to infinity, independently of each other. (In mathematical

language, this is the completeness of the space L2 .)
184 Appendix B.

The limit x has E(|x|2 ) = lim E(|xn |2 ) < ∞, and E(xn ) → E(x). If there
q.m. q.m.
are two convergent sequences, xn → x and yn → y , then

E(xn ȳn ) → E(xȳ). (B.7)

To show quadratic mean convergence of stochastic integrals, the following

criterion is useful:
the Loève criterion: the sequence xn converges in quadratic mean if and
only if
E(xm x̄n ) has a finite limit c, (B.8)

when m and n tend to infinity independently of each other.

The if part follows from E(|xm −xn |2 ) = E(xm x̄m )−E(xm x̄n )−E(xn x̄m )+
E(xn x̄n ) → c−c−c+c = 0. The only if part follows from E(xm x̄n ) → E(xx̄) =
E(|x|2 ).

B.3 Criteria for convergence in probability

Both almost sure convergence and convergence n quadratic mean imply con-
P
vergence in probability. Further, if xn → x then there exists a subsequence
a.s.
nk → ∞ as k → ∞, such that xnk → x.
To prove this, we use criterion (B.2). Take any sequence k > 0 such that
∞

k < ∞.
k=1

P
If xn → x, take any δ > 0 and consider P (|xn − x| > δ) → 0 as n → ∞. The
meaning of the convergence is that for each k there is an N k such that

P (|xn − x| > δ) < k ,

for all n ≥ N k . In particular, with nk = N k , one has

∞
 ∞

P (|xnk − x| > δ) < k ,
k=1 k=1

which is finite by construction. The sufficient criterion (B.2) gives the desired
almost sure convergence of the subsequence xnk .

Exercises
B:1. Prove the Borel-Cantelli lemma:
Section B.3 Criteria for convergence in probability 185

a) If Ak are events in a probability space (Ω, F, P ), then


P (Ak ) < ∞,
k

implies P (Ak infinitely often) = 0.

b) If the events Ak are independent, then

P (Ak ) = ∞,
k

implies P (Ak infinitely often) = 1.

B:2. Let x1 , x2 , . . . be independent identically distributed random variables.

Show that

E(|xk | < ∞ if and only if P (|xk | > k infinitely often) = 0.

B:3. Suppose the random sequences xn and xn have the same distribution.
a.s.
Prove that if xn → x then there exists a random variable x such that
a.s.
xn → x .
186 Appendix B.
Appendix C

Hilbert space and random

variables

C.1 Hilbert space and scalar products

A Hilbert space is a set of elements which can be added and multiplied by
complex numbers, and for which there is defined an inner product. The inner
product in a Hilbert space has the same mathematical properties as the covari-
ance between two random variables with mean zero, and therefore it is natural
to think of random variables as elements in a Hilbert space. We summarize
here the basic properties of a Hilbert space, for use in Chapters 3 and 4. For
further reading on Hilbert spaces and on metric spaces, see e.g. the classical
book by Royden [29].

Definition C:1 A general Hilbert space H over the complex numbers C is a

set of elements, usually called points or vectors, with the following properties:

1. The operations addition and subtraction are defined, and there exists a
unique ”zero” element 0 ∈ H and to each x ∈ H there is a unique
inverse −x:
x + y = y + x ∈ H,
x + 0 = x,
x + (−x) = 0.

2. Multiplication with complex scalar is defined (usually written cx = c · x):

c · x ∈ H,
0 · x = 0,
1 · x = x.

187
188 Appendix C.

3. A scalar (inner) product (x, y) is defined such that:

(x, y) = (y, x) ∈ C,
(ax + by, z) = a(x, z) + b(y, z),
(x, x) ≥ 0,
(x, x) = 0 if and only if x = 0.

4. A norm x and a distance d(x, y) = x−y are defined, and convergence
has the standard meaning: if x ∈ H then x = (x, x)1/2 , and if xn , x ∈
H then limn→∞ xn = x if and only if xn − x → 0.

5. The space is complete in the sense that if xn , x ∈ H and xm − xn  → 0

as m, n → ∞ then there is a point x ∈ H such that limn→∞ xn = x.

Remark C:1 If H is a space that satisfies(1-3) in the definition, then it can

be completed and made a Hilbert space that satisfies also (5).

We list some further properties of Hilbert spaces and scalar products, which
will be seen to have parallels as concepts for random variables:

Schwarz inequality: |(x, y)| ≤ x · y with equality if and only if (y, x)x =
(x, x)y ,

Triangle inequality: x + y ≤ x + y,

Continuity: if xn → x and yn → y then (xn , yn ) → (x, y),

Pythagorean theorem: if x and y are orthogonal, i.e. (x, y) = 0, then

x + y2 = x2 + y2 .

C.1.0.2 Linear subspaces:

Let L = {xj ∈ H; j = 1, 2, . . .} be a set of elements in a Hilbert space H , and
let
M0 = {a1 x1 + . . . + ak xk ; k = 1, 2, . . . ; aj ∈ C}
be the family of all finite linear combinations of elements in L. Then
 
M = M0 = S(L) = x ∈ H; x = lim xn for some xn ∈ M0
n→∞

is called the subspace of H spanned by L. It consists of all elements in H

which are linear combinations of elements in L or are limits of such linear
combinations. It is a subspace in the sense that it is closed under addition,
multiplication by scalar, and passage to a limit.
Section C.2 Projections in Hilbert space 189

C.2 Projections in Hilbert space

Two elements in a Hilbert space are called orthogonal, written x ⊥ y , if (x, y) =
0. Two subsets L1 and L2 are said to be orthogonal, L1 ⊥ L2 , if all elements
x ∈ L1 are orthogonal to all elements y ∈ L2 . Similarly, two subspaces M1
and M2 are orthogonal, M1 ⊥ M2 , if all elements in M1 are orthogonal to
all elements in M2 . The reader should check that if L1 ⊥ L2 , then S(L1 ) ⊥
S(L2 ).
For a sequence of subspaces, M1 , . . . , Mk of H , write

V = M1 ⊕ . . . ⊕ Mk

for the vector sum of M1 , . . . , Mk , which is the set of all vectors x1 + . . . + xk ,

where xj ∈ Mj , for j = 1, . . . , k .

C.2.0.3 The projection theorem

Let M be a subspace of a Hilbert space H , and let x be a point in H not in
M. Then x can be written in exactly one way as a sum

x=y+z

with y ∈ M and z = (x − y) ⊥ M. Furthermore, y is the point in M which

is closest to x,
d(x, y) = min d(x, w).
w∈M

and equality holds if and only if w = y .

The most common use of the projection theorem is to approximate a point
x in a general Hilbert space by a linear combination, or a limit thereof, of a
finite or infinite number of certain elements in H .

C.2.0.4 Separable spaces and orthogonal bases

A Hilbert space H is called separable if it contains a countable set of elements
x1 , x2 , . . . such that the subspace spanned by all the xj is equal to H . If the x-
variables are linearly independent, i.e. there is no non-trivial linear combination
equal to 0, a1 x1 + . . . + an xn = 0, it is possible to find orthogonal elements
y1 , y2 , . . . , such that

y1 = c11 x1 ,
y2 = c21 x1 + c22 x2 ,
···
yn = cn1 x1 + cn2 x2 + . . . + cnn xn ,
···
190 Appendix C.

This is the Gram-Schmidt orthogonalization process, and orthogonal means that


1, j = k
(yj , yk ) = δjk =
0, j = k.

The sequence y1 , y2 , . . . is called a complete orthogonal basis for the Hilbert

space H . It is a basis, i.e. every element in H can be written as a linear com-
bination of yk -elements or as a limit of such combinations, and it is orthogonal
by construction. It is furthermore complete, i.e. there is no element z ∈ H such
that
z > 0, (z, yj ) = 0, for all j .

C.3 Stochastic processes and Hilbert spaces

A Hilbert space is a set of elements which can be added and multiplied by
complex numbers, and for which there is defined an inner product. The inner
product in a Hilbert space has the same mathematical properties as the covari-
ance between two random variables with mean zero, and therefore it is natural
to think of random variables as elements in a Hilbert space; see Appendix C
for a summary of elementary properties of Hilbert spaces.
We shall consider a very special Hilbert space, namely the space of all
random variables X on a probability space (Ω, F, P ), which have zero mean
and finite variance.

Theorem C:1 If (Ω, F, P ) is a probability space, then

H = {random variables x on (Ω, F, P ) such that E(x) = 0, E(|x|2 ) < ∞}

with the scalar product

(x, y) = E(xy)

is a Hilbert space; it will be denoted H(Ω).

First, it is clear that (x, y) = E(xy) has the properties of a scalar product;
check that. It is also clear that we can add random variables with mean zero
and finite variance
 to obtain new random variables with the same properties.
Also, x = E(|x|2 ), which means that if x = 0, then P (x = 0) = 1,
so random variables which are zero with probability one, are, in this context,
defined to be equal to the zero element 0. Convergence in the norm  ·  is
equal to convergence in quadratic mean of random variables, and if a sequence
of random variables xn is a Cauchy sequence, i.e. xm −xn  → 0 as m, n → ∞,
then we know that it converges to a random variable x with finite mean, which
means that H(Ω) is complete. Therefore it has all the properties of a Hilbert
space.
Section C.3 Stochastic processes and Hilbert spaces 191

C.3.0.5 A stochastic process as a curve in H(Ω)

A random variable with E(x) = 0 and finite variance is a point in the Hilbert
space H(Ω). Two equivalent random variables x and y are represented by the
same point in H(Ω), since P (x = y) = 1 and hence x − y2 = E(|x − y|2 ) = 0.

A stochastic process is a family of random variables, and thus a stochastic

process {x(t), t ∈ R} with one-dimensional parameter
 t is a curve in H(Ω).
Further, from the definition of the norm x = E(|x|2 ), we see that conver-
gence in this norm is equivalent to convergence in quadratic mean. In other
words, if a stochastic process is continuous in quadratic mean, then the corre-
sponding curve in H(Ω) is continuous.

C.3.0.6 The generated subspace

A set of points in a Hilbert space generates a subspace, which consists of all
finite linear combinations and their limits. If {x(t); t ∈ T } is a stochastic
process, write
H(x) = S(x)
for the subspace spanned by x(·). Also, for a process {x(t), t ∈ R}, define

H(x, t) = S(x(s); s ≤ t)

as the subspace spanned by all variables observed up till time t. At time t

it contains all variables which can be constructed by linear operations on the
available observations. Examples of random variables in H(x, t) are
 t
x(t) + x(t − 1) + . . . + x(t − n + 1)
, e−(t−u) x(u) du, x− (t) + 3x− (t),
n −∞

where x− (t), x− (t) denote left derivatives.

Example C:1 Take an MA(1)-process, i.e. from uncorrelated variables

e(t), t = . . . , −1, 0, 1, 2, . . . ,

with E(e(t)) = 0, V (e(t)) = 1, we construct

x(t) = e(t) + b1 e(t − 1).

If |b1 | < 1, the process can be inverted and e(t) simply retrieved from x(s), s ≤
t:

e(t) = x(t) − b1 e(t − 1) = x(t) − b1 (x(t − 1) − b1 e(t − 2))

n
= (−b1 )k x(t − k) + (−b1 )n+1 e(t − n − 1) = yn (t) + zn (t), say.
k=0
192 Appendix C.

Here,
yn (t) ∈ S(x(s); s = t − n, . . . , t) ⊆ S(x(s); s ≤ t) = H(x, t),
while
zn (t) = |b1 |n+1 → 0
as n → ∞. Thus e(t) − yn (t) → 0 and we have that
∞
 
n
e(t) = (−b1 ) x(t − k) = lim
k
(−b1 )k x(t − k) ∈ H(x, t)
n→∞
k=0 k=0

if |b1 | < 1. The representation of e(t) as a limit of finite linear combinations

of x(t − k)-values is explicit and obvious.
For |b1 | = 1 it is less obvious that e(t) ∈ H(x, t), but it is still possible to
represent e(t) as a limit. For example, if b1 = −1, x(t) = e(t) − e(t − 1), and
zn (t) = e(t − n − 1) does not converge to anything. But in any case,


n
e(t) = x(t − k) + e(t − n − 1),
k=0

and so, since the left hand side does not depend on n,

1  1  1 
N N n N
e(t) = e(t) = x(t − k) + e(t − n − 1)
N N N
n=1 n=1 k=0 n=1
N


1 
N
k
= 1− x(t − k) + e(t − n − 1) = yN (t) + zN (t).
N N
k=0 n=1
N
Now, zN (t) = N1 n=1 e(t − n − 1) = e(t) − yN (t) → 0 by the law of large
numbers, since all e(t) are uncorrelated with E(e(t)) = 0 and V (e(t)) = 1.
We have shown that e(t) is in fact the limit of a finite linear combination of
x(s)-variables, i.e. e(t) ∈ H(x, t).
Appendix D

Spectral simulation of random

processes

D.1 The Fast Fourier Transform, FFT

A stationary process {x(t), t ∈ R} with continuous spectrum f (ω) can be
efficiently simulated by Fourier methods from its spectral representation. One
then has to discretize the continuous spectrum and use the approximation (4.29)
from Section 4.3.3.
Fourier-simulation is most effectively performed with the help of the Fast
Fourier Transform (FFT), or rather the inverse transform. This algorithm
transforms a sequence of real or complex numbers Z(0), Z(1), . . . , Z(N − 1)
into its (inverse) discrete Fourier transform


N −1
z(n) = Z(k) exp(i2πkn/N ), (D.1)
k=0

for n = 0, 1, . . . , N − 1, where the integer N is a power of 2, N = 2m . In

the literature, there are as many ways to write the Fourier sum as there are
combinatorial possibilities, with or without a factor N in the denominator and
with or without a minus-sign in the exponential function. Almost every mathe-
matical computer software toolbox contains efficient algorithms to perform the
FFT according to (D.1).
The basis for the use of (D.1) to generate a sample sequence lies in the
representation of a stationary process as an approximating sum of harmonic
functions with random phase and amplitude; see (4.29) and the alternative form
(4.30). The Z(k) will then be chosen as complex random variables with absolute
value and argument equal to the desired amplitude and phase. When using the
formula for simulation purposes, there are however a number of questions which
have to be resolved, concerning the relation between the sampling interval and
the frequency resolution, as well as the aliasing problem.

193
194 Appendix D.

Before we describe the steps in the simulation we repeat the basic facts
about processes with discrete spectrum, and the special problems that arise
when sampling a continuous time process.

D.2 Random phase and amplitude

To see how (D.1) can be used to generate a sample function we consider first
the special stationary process (4.18) with discrete spectrum in Section 4.3.3, or
the normalized form (4.30). Including the spectral jump at zero frequency it
has the form,
∞
x(t) = ρ0 + ρk cos(ωk t + φk ). (D.2)
k=1

Here ρ0 is a random level shift, while {ρk } are the amplitudes and {φk } the
phases of the different harmonic components of x(t). The frequencies ωk > 0
can be any set of fixed positive frequencies.
If we define

Z(0) = ρ0 ,
Z(k) = ρk exp(iφk ), for k = 1, 2, . . ..

it is easy to see that x(t) in (D.2) is the real part of a complex sum, so if we
write y(t) for the imaginary part, then
∞

x(t) + iy(t) = Z(k) exp(iωk t). (D.3)
k=0

We repeat the fundamental properties of this representation.

If amplitudes and phases in (D.2) are independent and the phases

φk are uniformly distributed over [0, 2π), then {x(t), t ∈ R} is
stationary and has a discrete spectral distribution with mass σk2 =
1 2 2 2
2 E(ρk ) and σ0 = E(ρ0 ) at the frequencies ωk > 0, and ω0 = 0,
respectively. Further, Z(k) = ρk exp(iφk ) = σk (Uk + iVk ) have the
desired properties if the real and imaginary parts are independent
standardized Gaussian random variables, with E(Uk ) = E(Vk ) = 0
and variance V (Uk ) = V (Vk ) = 1.

It is possible to approximate every spectral distribution by a discrete spectrum.

The corresponding process is then an approximation of the original process.

D.3 Aliasing
If a stationary process {x(t), t ∈ R} with continuous twosided spectral den-
sity fx (ω), is sampled with a sampling interval d, the sequence {x(nd), n =
Section D.4 Simulation scheme 195

(d)
0, ±1, . . .} has a spectral density fx (ω) that can be restricted to any interval
of length π/d, for example the interval (−π/d, π/d]. There it can be written
as a folding of the original spectral density,
∞


2πj
fx(d) (ω) = fx ω+ , for −π/d < ω ≤ π/d.
d
j=−∞

(d)
The corresponding one-sided spectral density gx (ω) can then be defined on
[0, π/d] as
fx(d) (ω) + fx(d) (−ω).

For reasons that will become clear later (Section D.6) we prefer to define it
instead on [0, 2π/d) by
∞


2πj
gx(d) (ω) = fx ω+ , for 0 ≤ ω < 2π/d. (D.4)
d
j=−∞

D.4 Simulation scheme

In view of (D.1) and (D.3) we would like to generate a finite part of the sum
in (D.3) to get z(n) and then take the real part to get x(t) for t = nd,
n = 0, 1, . . . , N − 1. To see the analogy clearly we repeat the expressions:


N −1
z(n) = Z(k) exp(i2πkn/N ) (D.5)
k=0
∞

x(t) = Re Z(k) exp(iωk t) (D.6)
k=0

Here is the scheme to follow.

We have: A real spectral density gx (ω) for ω ≥ 0 for a stationary process
{x(t), t ∈ R}.
We want: A discrete time sample x(nd), n = 0, 1, . . . , N − 1 of {x(t), t ∈ R}
of size N = 2m with sampling interval d, equally spaced over the time interval
[0, T ) with T = N d;
Means: Generate random variables Z(k) = σk (Uk + iVk ), k = 0, 1, . . . , N − 1,
−1
with distribution described below, and take z(n) = Nk=0 Z(k) exp(i2πkn/N )
according to (D.1). Then set

x(nd) = Re z(n), n = 0, 1, . . . , N − 1.

This will give the desired realization.

196 Appendix D.

D.5 Difficulties and details

The Fourier simulation scheme rises a number of questions which have to be
dealt with before it can be implemented. Here we shall comment on the impor-
tant issues.
Frequency spacing: We have requested N time points regularly spaced in
[0, T ) in steps of d, and we want to use the special sum (D.1). This will impose
a restriction both on the frequency spacing and on the maximum frequency
accounted for. Comparing (D.5) and (D.6), bearing in mind that t = nd, we
find that only frequencies that are of the form
2πk 2πk
ωk = = for k = 0, 1, . . . , N − 1,
Nd T
appear in the simulation, and further that the highest frequency in the sum is
2π(N −1)
dN , just barely below
2π
ωmax = .
d
Discretization of spectrum: The continuous spectrum with density gx (ω)
has to be replaced by a discrete spectrum with mass only at the frequencies
ωk = 2πk
dN which enter into the sum (D.5). The mass at ωk should be equal to
2π (d)
σk2 = g (ωk ), k = 0, 1, . . . , N − 1. (D.7)
dN x
Generation of the Z(n): Generate independent random variables
Z(k) = σk (Uk + iVk )
with Uk and Vk from a normal distribution with mean zero and variance 1, for
instance by the Box-Müller technique,

Uk = cos(2πR1 ) −2 ln R2 ,

Vk = sin(2πR1 ) −2 ln R2 ,
where R1 and R2 are independent random numbers uniformly distributed in
(0, 1].
Aliasing: The restricted 
frequency range in (D.5), implies that the generated
−1 2 2
x(nd) will have variance N k=0 σk , where each σk is an infinite sum:
∞
2π  2πj
σk2 = fx (ωk + ).
dN d
j=−∞

In practice one has to truncate the infinite series and use

2π 
J
2πj
σk2 = fx (ωk + ), k = 0, 1, . . . , N − 1, (D.8)
dN d
j=−J

where J is taken large enough. If fx (ω) ≈ 0 for ω ≥ ωmax one can take J = 0.
Section D.6 Simulation of the envelope 197

D.6 Simulation of the envelope

The Fourier simulation will not only yield a realization of x(nd) = Re z(n)
but also of its Hilbert transform y(nd) = Im z(n). Therefore we can get the
envelope as a byproduct, 
x(nd)2 + y(nd)2 .
Thus generation of 2N Gaussian random numbers Uk , Vk , for k = 0, 1, . . . , N −
1 will result in 2N useful data points. If the aim is to generate only the x(nd)-
series, one could restrict the sum (D.5) to only n = 0, 1, . . . , N/2 − 1 and thus
generate only N Gaussian variates.

D.7 Summary
In order to simulate a sample sequence of a stationary process {x(t), t ∈ R}
with spectral density fx (ω) over a finite time interval one should do the follow-
ing:

1. Choose the desired time interval [0, T ).

2. Choose a sampling interval d or the number of sample points N = 2m .

This will give a sequence of N process values x(nd), k = 0, 1, . . . , N − 1.

3. Calculate and truncate the real discretized spectrum

2π 
J
2πj
σk2 = fx (ωk + ), k = 0, 1, . . . , N − 1,
dN d
j=−J

and take J so large that fx (ω) ≈ 0 for ω > 2π(J + 1)/d.

4. Generate independent standard normal variables

Uk , Vk , for k = 0, 1, . . . , N − 1

with mean zero and variance 1.

5. Set Z(k) = σk (Uk + iVk ) and calculate the (inverse) Fourier transform


N −1
z(n) = Z(k) exp(i2πkn/N ).
k=0

6. Take the real part,

x(nd) = Re z(n), n = 0, 1, . . . , N − 1;

this is the desired sequence.

198 Appendix D.

7. To generate the envelope, take the imaginary part

y(nd) = Im z(n), n = 0, 1, . . . , N − 1;

the envelope is then 

x(nd)2 + y(nd)2 .
Literature
[1] Adler, R. (1990): An introduction to continuity, extrema, and related top-
ical for general Gaussian processes. IMS Lecture Notes-Monograph Series,
Vol. 12.

[2] Banerjee, S., Carlin, B.P. and Gelfand, A.E. (2004): Hierarchical mod-
eling and analysis for spatial data. Chapman & Hall/CRC, Boca Raton.
indexCarlin, B.P.

[3] Belyaev, Yu.K. (1959): Analytic random processes. Theory Probab. and its
Applications, English edition, 4, 402.

[4] Belyaev, Yu.K. (1961): Continuity and Hölder’s conditions for sample func-
tions of stationary Gaussian processes. Proc. Fourth Berk. Symp. on Math.
Stat. and Probability, 2, 22-33.

[5] Breiman, L. (1968): Probability. Addison-Wesley, Reading. Reprinted 199?

in SIAM series.

[6] Cramér, H. (1942): On harmonic analysis in certain function spaces. Arkiv

Mat. Astron. Fysik, 28B, no. 12.

[7] Cramér, H. (1945): On the theory of stochastic processes. Proc. Tenth

Scand. Congr. of Math., Copenhagen, pp. 28-39.

[8] Cramér, H. (1945): Mathematical Methods of Statistics. pronceton Univer-

sity Press, 1945..

[9] Cramér, H. and Leadbetter, M.R. (1967): Stationary and related stochastic
processes. Wiley, New York. Reprinted by Dover Publications, 2004.

[10] Dobrushin, R.L. (1960): Properties of the sample functions of a stationary

Gaussian process. Teoriya Veroyatnostei i ee Primeneniya, 5, 132-134.

[11] Doob, J.L. (1953): Stochastic processes. Wiley, New York.

[12] Durrett, R. (1996): Probability: Theory and examples. Doxbury Press.

[13] Einstein, A. (1905): Investigations on the theory of Brownian novement.

Reprinted by Dover Publications, 1956.

199
200 Literature

[14] Jordan, D.W. and Smith, P. (1999): Nonlinear ordinary differential equa-
tions. 3rd Ed. Oxford University Press.

[15] Grenander, U. (1950): Stochastic processes and statistical inference. Arkiv

Mat. 1, 195-277.

[16] Gut, A. (1995): An intermediate course in probability. Springer-Verlag.

[17] Ibragimov, I.A. and Linnik, Yu.V. (1971): Independent and stationary
sequences of random variables. Wolters-Noordhoff, Groningen.

[18] Ibragimov, I.A. and Rozanov, Y.A. (1978): Gaussian random processes.
Springer-Verlag, New York.

[19] Kac, M. and Slepian, D. (1959): Large excursions of Gaussian processes.

Ann. Math. Statist., 30, 1215–1228.

[20] Kolmogorov, A. (1933): Grundbegriffe der Wahrscheinlichkeitsrechnung.

Springer-Verlag, Berlin.

[21] Lasota, A. and Mackey, M.C. (1994): Chaos, fractals, and noise; stochastic
aspects of dynamics. Springer-Verlag, New York.

[22] Leadbetter, M.R., Lindgren, G. and Rootzén, H. (1983): Extremes and

related properties of random sequences and processes. Springer-Verlag, New
York.

[23] Lindgren, G. (1975): Prediction from a random time point. Annals of

Probability, 3, 412–423.

[24] Maruyama, G. (1949): The harmonic analysis of stationary stochastic pro-

cesses. Mem. Fac. Sci. Kyusyu Univ. A4, 45-106.

[25] Petersen, K. (1983): Ergodic Theory, Cambridge University Press, Cam-

bridge.

[26] von Plato, J. (1994): Creating modern probability. Cambridge University

Press, Cambridge.

[27] Rice, S.O. (1944, 1945): Mathematical analysis of random noise. Bell Sys-
tems technical Journal, 23, 282-332, and 24, pp 46-156. Reprinted in:
Wax, N. (1954): Selected papers on noise and stochastic processes. Dover
Publications, New York.

[28] Rice, S.O. (1963): Noise in FM-receivers. In: Time Series Analysis, Ed:
M. Rosenblatt, Chapter 25, pp. 395-422. Wiley, New York.

[29] Royden, H.L. (1988): Real Analysis, 3rd Ed. Prentice Hall.
Literature 201

[30] Rychlik, I. (2000): On some reliability applications of Rice’s formula for

the intensity of level crossings. Extremes, 3, 331–348.

[31] Slepian, D. (1963): On the zeros of Gaussian noise. In: Time Series Anal-
ysis, Ed: M. Rosenblatt, pp. 104-115. Wiley, New York.

[32] St. Denis, M. and Pierson, W.J. (1954): On the motion of ships in confused
seas. Transactions, Soc. Naval Architects and Marine Engineers, Vol. 61,
(1953) pp. 280-357.

[33] van Trees, H. (1968): Detection, Estimation, and Modulation Theory. Part
I”, John Wiley & Sons.

[34] WAFO, Wave Analysis for Fatigue and Oceanography. Available at

http://www.maths.lth.se/matstat/wafo/.

[35] Wong, E. and Hajek, B. (1985): Stochastic processes in engineering sys-

tems. Springer-Verlag, New York.

[36] Williams, D. (1991): Probability with Martingales, Cambridge University

Press, Cambridge.

[37] Williams, D. (2001): Weighting the Odds. Cambridge University Press,

Cambridge.

[38] Yaglom, A.M. (1962): An introduction to the theory of stationary random

functions. Prentice Hall, Englewood Cliffs.

[39] B. Øksendal: Stochastic differential equations, an introduction with appli-

cations. Springer-Verlag, 5th Ed 2000.
Index

α Boltzmann, L., 133

spectral width parameter, 63, 77 Boole’s inequality, 32
 Borel
spectral width parameter, 77 field in C, 37
ω field in R∞ , 9
angular frequency, 2, 14 field in RT , 11
elementary outcome, 2 field in R, 3
σ -algebra, 3, 175 measurable function, 6
σ -field, 3, 175 set, 3
generated by events, 3 Borel-Cantelli lemma, 25, 32, 53, 184
generated by random variables, Box-Müller’s method, 196
6 Breiman, L., 199
of invariant sets, 139 Brown, R., ix, 19
Øksendal, B., 201 Brownian motion, ix, 19, 112
Bulinskaya, E.V., 43
additive Bulinskaya,E.V., 56
countably, 4, 175
finitely, 4 Cantor’s theorem, 177, 179
Adler, R., 199 Carathéodory’s extension theorem,
alarm prediction, 64 176
algebra, see σ -algebra Cauchy principal value, 86
aliasing, 88, 194, 196 characteristic equation, 108
amplitude, 89, 94 characteristic function, 83
for narrow band process, 118 Chebysjev’s inequality, 34
ARMA-process, 152 complete
auto-covariance function, 161 basis, 123, 190
auto-regressive process, 152 space, 92, 188
completion
Banach’s theorem, 59 of probability measure, 5
band-limited white noise, 151 of vector space, 188
Banerjee, S., 199 conditional covariance
Belyaev, Yu.K., 36, 153, 199 normal distribution, 17, 69
Bennet, W.R., 97 conditional expectation, 7, 64, 141
Bessel function, 167 horizontal window, 66
modified, 169 normal distribution, 17, 69
Birkhoff ergodic theorem, 142 vertical window, 66
Bochner’s theorem, 82 consistent family of distributions, 8

202
Index 203

continuity linear, 108

absolute, 13 directional spectrum, 171
conditions for, 33 dispersion relation, 23, 96, 170
in L2 , 27, 45, 90 Dobrushin, R.L., 36, 199
in quadratic mean, 27, 45, 90 Doob, J.L, 199
modulus, 41 Durbin’s formula, 74
of Gaussian process, 35 Durbin, J., 74
of sample function, 30 Durrett, R., 199
of stationary process, 34, 35
probabilities on C, 37 eigenfunction, 122
summary of conditions, 49 Einstein, A., ix, 20, 97, 199
convergence ensemble average, 52, 139
almost surely, 27, 41, 181 envelope, 63, 117
of narrow band process, 118
Cauchy condition, 182
simulation, 197
in distribution, 181
equivalent processes, 29
in probability, 27, 181
ergodicity, 65, 133
in quadratic mean, 27, 181, 190
almost surely, 52
Loève criterion, 184
and mixing, 156
uniform, of random functions,
definition, 140
183
in continuous time, 147
countable
in quadratic mean, 52
additivity, 10
Markov chain, 139
basis, 11
of Gaussian process, 148
covariance function, 13
of random sequence, 145
Cramér, H., 21, 24, 151, 199
upcrossings, 67
Cramér-Wold decomposition, 151, 190
excursion
cross spectrum, 161
above a level, 73
cross-covariance function, 161
exponential smoothing, 106
cross-spectrum, 162 extension
crossings of probability measure, 5, 176
expected number, 21, 57
finite number, 44 fast Fourier transform, 193
FFT, 193
derivative field, see σ -field
as linear filter, 103 homogeneous field, 166
at upcrossing, 56 isotropic random field, 15, 167
deterministic process, 151, 153 random, 165
differentiability filter, 101, 106
conditions for, 37 amplification, 103
in quadratic mean, 28, 46 causal, 104
of Gaussian process, 39 cross-correlation, 165
of sample function, 37, 46 linear, 100
summary of conditions, 49 phase shift, 103
differential equation time-invariant, 101
204 Index

finite-dimensional inner product, 188, 190

approximation by event, 12 integrability
distribution, 8 conditions for, 49
FM-noise, 21 in quadratic mean, 49
folding, 88 interval
Fourier inR∞ , 9
integral, 13 invariant
inversion, 86 measure, 137
transform, 21, 193 random variable, 138
frequency, 89 set, 138
mean frequency, 63 inversion of spectrum
frequency function for filter, 101 continuous time, 86
Fubini’s theorem, 60 discrete time, 94
isometry, 90
Garsia, A.M., 142
Gaussian process, 16, 96 JONSWAP spectrum, 14, 74, 77, 172
continuity, 35 Jordan, D.W., 200
differentiability, 39 jump discontinuity, 40
ergodicity, 148
Slepian model, 68 Kac and Slepian, horizontal window,
unboundedness, 36 66
Gelfand, A.E., 199 Kac, M., 57, 58
generating function, 108 Kac. M., 200
geostatistics, 166 Karhunen-Loève expansion, 122
Gram-Schmidt orthogonalization, 190 Kolmogorov
Grenander, U., ix, 24, 127, 200 existence theorem, 10, 80, 178
Gut, A., 200 extension theorem, 178
Grundbegriffe, 9, 178
Hölder condition, 40–42 probability axiom, 3, 175
Hajek, B., 201 Kolmogorov, A.N., 3, 200
Hamming, R.W., 22
harmonic oscillator, 106 Langevin’s equation, 20, 98
Herglotz’ lemma, 88 Lasota, A., 200
Hermitian function, 80 law of large numbers, 192
Hilbert space, 187, 190 Leadbetter, M.R., 199, 200
generated by a process, 191 level crossing, 57
Hilbert transform, 115, 197 likelihood ratio test, 128
Hilbert’s sixth problem, 178 Lindgren, G., 200
homogeneous field linear
spectral distribution, 15, 166 filter, 100, 101
spectral representation, 15, 166 interpolation, 100
horizontal window conditioning, 66 oscillator, 14, 106
prediction, 18, 100
Ibragimov, I.A., 200 process, 104
impulse response, 101, 104 regression, 101
Index 205

subspace, 188 orthogonal

time-invariant filter, 102 basis, 189
Linnik, Yu.V., 200 increments, 89
Lipschitz condition, 41 oscillator
Loève criterion, 45, 50, 184 harmonic, 106
Lord Rayleigh, 22, 97 linear, 106

m-dependence, 157 Palm distribution, 66

Mackey, M.C., 200 Perrin, J.B., 20
Markov’s inequality, 34 Petersen, K., viii, 200
Maruyama, G., 200 phase, 89, 94
Matérn, B., 169 Pierson, W.J., ix, 22, 201
Mercer’s theorem, 127 Pierson-Moskowitz spectrum, 172
metrically transitive, 140 point process, 66
mixing Poisson process, 40
for Gaussian processes, 157 prediction, 100
relation to ergodicity, 158 after upcrossing, 68
strong, 157 linear, 18
uniform, 157 quadratic mean, 7
weak, 157 primordial
modulo game, 135 randomness, 151
as ergodic sequence, 140 soup, 151
Monotone convergence, 61 principal components, 121
Monte Carlo simulation, 135, 193 probability measure, 175
moving average projection
discrete time, 191 in Hilbert space, 100, 189
infinite, 105 uncorrelated residuals, 101, 152
multivariate normal distribution, 15 uniqueness, 101
pseudo randomness, 134
narrow banded spectrum, 14, 63 Pythagorean theorem, 188
non-deterministic process, 151
non-negative function, 80 quadratic mean
non-random walk continuity, 45
as ergodic sequence, 144 convergence, 45
norm, 188 Loève criterion, 45
normal distribution differentiability, 46
bivariate, 56 integrability, 49
conditional, 16 optimal prediction, 7
inequalities, 18
multivariate, 15 random
nugget effect, 169 amplitude, 89, 94, 194
moving surface, 23, 169
observable, 9, 11, 124 phase, 89, 94, 194
Ornstein-Uhlenbeck process, 39, 98, waves, 23, 170
112, 156 Rayleigh distribution, 117
206 Index

for slope at upcrossing, 70 distribution, 13

RC-filter, 106 for process, 82
reconstruction, 18 for random field, 166
rectangle for sequence, 88
n-dimensional, 4, 25 for vector process, 162
generalized, 4, 9 inversion, 86
regression approximation, 74 moment, 13, 49
regular process, 150, 151 representation
as moving average, 154 of process, 21, 89
Rice’s formula, 21, 57 of random field, 166
alternative proof, 60 of sequence, 100
for Gaussian process, 62 of vector process, 163
for non-stationary process, 61 width, 63, 77
Rice, S., ix spectrum
Rice, S.O., 21, 22, 57, 58, 79, 97, absolutely continuous, 94, 153
200 amplitude spectrum, 165
Riemann-Lebesgue’s lemma, 149 coherence spectrum, 165
Rootzén, H., 200 directional, 171
Royden, H.L., viii, 200 discrete, 93, 153
Rozanov, Y.A., 200 encountered, 23
Rychlik, I., 201 JONSWAP, 14, 74, 77, 172
one-sided, 87, 95
sample function average, 52 phase spectrum, 165
sampling theorem, 119 Pierson-Moskowitz, 172
scalar product, 188 singular, 153
Schwarz’ inequality, 188 Whittle-Matérn, 169
second order stationary process, 79 spreading function, 23, 96, 171
separable stationary
process, 29 intrinsically, 166
space, 189 stationary process, 13
shift transformation, 147 as curve in Hilbert space, 191
signal detection, 127 complex, 82
simulation second order, 13
by Fourier method, 193 strictly, 79
of envelope, 197 weakly, 13, 79
singular StDenis, M., ix, 22, 201
process, 150, 151 stochastic calculus, 113
spectral conditions, 154 stochastic differential equation, 98,
spectrum, 153 108
Slepian model, 57, 68 Stone-Weierstrass theorem, 92
Slepian, D., 57, 200, 201 strong Markov property, 64
Smith, P., 200 subspace
spectral generated, 191
density, 13, 86 spanned, 191
Index 207

tangents, non-existence of, 43 Hölder condition, 42

time average, 52, 138 infinite variation, 55
time-invariant filter, 101 integral, 51
total variation, 59 Karhunen-Loève expansion, 124
transfer function, 101 Wiener, N., 22
transformation Williams, D., viii, 201
generated by stationary process, Wold, H., 151
137 Wong, E., 201
generating stochastic process, 136
measurable, 136 Yaglom, A.M., 153, 201
measure preserving, 136
quadratic, 135
shift, 137
triangle inequality, 188
Tukey, J.W., 22

upcrossings
counting, 60
expected number, 57
intensity, 58
slope at, 56

variogram, 166
vertical window conditioning, 66
von Plato, J., 133, 200

WAFO, 72, 74, 201

wave
average length, 171
average period, 171
characteristics, 75
frequency, 23, 169, 170
number, 23, 169, 170
Pierson-Moskowitz spectrum, 119
wave direction, 171
wave frequency spectrum, 170
wave number spectrum, 171
Weyl’s equidistribution theorem, 144
white noise, 14, 94, 98, 112
Whittle, P., 169
Whittle-Matérn spectrum, 169
Wiener process, 19, 112
as spectral process, 97
continuity, 34
derivative of, 20, 99, 112
208 Index
 October 2006
Mathematical Statistics
Centre for Mathematical Sciences
Lund University
Box 118, SE-221 00 Lund, Sweden
http://www.maths.lth.se/