Geometry and Statistics - Handbook of

Statistics, Volume 46 Frank Nielsen &

Arni S.R. Srinivasa Rao & C.R. Rao
(Editors)
Visit to download the full and correct content document:
https://ebookmass.com/product/geometry-and-statistics-handbook-of-statistics-volum
e-46-frank-nielsen-arni-s-r-srinivasa-rao-c-r-rao-editors/
Handbook of Statistics
Volume 46
Geometry and Statistics
 Handbook of Statistics

Series Editors

C.R. Rao
C.R. Rao AIMSCS, University of Hyderabad Campus,
Hyderabad, India

Arni S.R. Srinivasa Rao

Medical College of Georgia, Augusta University, United States
Handbook of Statistics
Volume 46

Geometry and
Statistics

Edited by
Frank Nielsen
Sony Computer Science Laboratories Inc.,
Tokyo, Japan

Arni S.R. Srinivasa Rao

Medical College of Georgia,
Augusta, Georgia, United States

C.R. Rao
AIMSCS, University of Hyderabad Campus,
Hyderabad, India
Academic Press is an imprint of Elsevier
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
525 B Street, Suite 1650, San Diego, CA 92101, United States
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
125 London Wall, London, EC2Y 5AS, United Kingdom

Copyright © 2022 Elsevier B.V. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording, or any information storage and
retrieval system, without permission in writing from the publisher. Details on how to seek
permission, further information about the Publisher’s permissions policies and our arrangements
with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency,
can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the
Publisher (other than as may be noted herein).

Notices
Knowledge and best practice in this field are constantly changing. As new research and experience
broaden our understanding, changes in research methods, professional
practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge
in evaluating and using any information, methods, compounds, or experiments
described herein. In using such information or methods they should be mindful of their
own safety and the safety of others, including parties for whom they have a professional
responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or
editors, assume any liability for any injury and/or damage to persons or property as a
matter of products liability, negligence or otherwise, or from any use or operation of any
methods, products, instructions, or ideas contained in the material herein.

ISBN: 978-0-323-91345-4
ISSN: 0169-7161

For information on all Academic Press publications

visit our website at https://www.elsevier.com/books-and-journals

Publisher: Zoe Kruze

Acquisitions Editor: Sam Mahfoudh
Developmental Editor: Naiza Ermin Mendoza
Production Project Manager: Abdulla Sait
Cover Designer: Victoria Pearson
Typeset by STRAIVE, India
Contents

Contributors xi
Preface xiii

Section I
Foundations in classical geometry and analysis 1
1. Geometry, information, and complex bundles 3
Steven G. Krantz and Arni S.R. Srinivasa Rao
1. Introduction 3
2. Complex planes 5
2.1 Important implications of Liouville’s theorem 9
3. Geometric analysis and Jordan curves 13
4. Summary 17
References 17

2. Geometric methods for sampling, optimization,

inference, and adaptive agents 21
Alessandro Barp, Lancelot Da Costa, Guilherme França,
Karl Friston, Mark Girolami, Michael I. Jordan, and
Grigorios A. Pavliotis
1. Introduction 22
2. Accelerated optimization 25
2.1 Principle of geometric integration 25
2.2 Conservative flows and symplectic integrators 26
2.3 Rate-matching integrators for smooth optimization 28
2.4 Manifold and constrained optimization 33
2.5 Gradient flow as a high friction limit 34
2.6 Optimization on the space of probability measures 35
3. Hamiltonian-based accelerated sampling 37
3.1 Optimizing diffusion processes for sampling 38
3.2 Hamiltonian Monte Carlo 40
4. Statistical inference with kernel-based discrepancies 46
4.1 Topological methods for MMDs 47
4.2 Smooth measures and KSDs 48
4.3 Information geometry of MMDs and natural gradient
descent 52

v
vi Contents

5. Adaptive agents through active inference 54

5.1 Modeling adaptive decision-making 54
5.2 Realizing adaptive agents 60
Acknowledgments 64
References 65

3. Equivalence relations and inference for sparse Markov

models 79
Donald E.K. Martin, Iris Bennett, Tuhin Majumder, and
Soumendra Nath Lahiri
1. Introduction 79
1.1 Improved modeling capabilities of sparse Markov models
(SMMs) 80
2. Fitting SMMs and example applications 83
2.1 Model fitting based on a collapsed Gibbs sampler 84
2.2 Fitting SMM through regularization 87
3. Equivalence relations and the computation of distributions of
pattern statistics for SMMs 90
3.1 Notation 91
3.2 Computing distributions in higher-order Markovian
sequences 91
3.3 Specializing the computation to SMM 94
3.4 Application to spaced seed coverage 96
4. Summary 100
Acknowledgments 101
References 101

Section II
Information geometry 105
4. Symplectic theory of heat and information
geometry 107
Fr
ederic Barbaresco
1. Preamble 108
2. Life and seminal work of Souriau on lie groups
thermodynamics 111
3. From information geometry to lie groups
thermodynamics 118
4. Symplectic structure of fisher metric and entropy as Casimir
function in coadjoint representation 123
4.1 Symplectic Fisher Metric structures given by Souriau
model 123
4.2 Entropy characterization as generalized Casimir invariant
function in coadjoint representation and Poisson
Cohomology 128
4.3 Koszul Poisson Cohomology and entropy
characterization 131
Contents vii

5. Covariant maximum entropy density by Souriau model 132

5.1 Gauss density on Poincare unit disk covariant with respect
to SU(1,1) Lie group 132
5.2 Gauss density on Siegel unit disk covariant with respect to
SU(N,N) Lie group 136
5.3 Gauss density on Siegel upper half plane 138
6. Conclusion 139
References 140
Further reading 143

5. A unifying framework for some directed distances in

statistics 145
Michel Broniatowski and Wolfgang Stummer
1. Divergences, statistical motivations, and connections to
geometry 147
1.1 Basic requirements on divergences (directed
distances) 147
1.2 Some statistical motivations 147
1.3 Incorporating density function zeros 152
1.4 Some motivations from probability theory 157
1.5 Divergences and geometry 159
1.6 Some incentives for extensions 161
2. The framework 163
2.1 Statistical functionals S and their dissimilarity 163
2.2 The divergences (directed distances) D 168
2.3 The reference measure λ 171
2.4 The divergence generator ϕ 171
2.5 The scaling and the aggregation functions m1, m2,
and m3 173
2.6 Auto-divergences 195
2.7 Connections with optimal transport and coupling 197
3. Aggregated/integrated divergences 201
4. Dependence expressing divergences 203
5. Bayesian contexts 205
6. Variational representations 208
7. Some further variants 211
Acknowledgments 214
References 214

6. The analytic dually flat space of the mixture family of

two prescribed distinct Cauchy distributions 225
Frank Nielsen
1. Introduction and motivation 226
2. Differential-geometric structures induced by smooth convex
functions 227
2.1 Hessian manifolds and Bregman manifolds 227
2.2 Bregman manifolds: Dually flat spaces 230
viii Contents

3. Some illustrating examples 234

3.1 Exponential family manifolds 234
3.2 Regular cone manifolds 237
3.3 Mixture family manifolds 239
4. Information geometry of the mixture family of two distinct
Cauchy distributions 241
4.1 Cauchy mixture family of order 1 241
4.2 An analytic example with closed-form dual
potentials 249
5. Conclusion 253
Acknowledgments 253
Appendix. Symbolic computing notebook in MAXIMA 253
References 255

7. Local measurements of nonlinear embeddings with

information geometry 257
Ke Sun
1. Introduction 257
2. α-Divergence and autonormalizing 260
3. α-Discrepancy of an embedding 262
4. Empirical α-discrepancy 267
5. Connections to existing methods 268
5.1 Neighborhood embeddings 268
5.2 Autoencoders 270
6. Conclusion and extensions 272
Acknowledgment 274
Appendices 274
Appendix A. Proof of Lemma 1 274
Appendix B. Proof of Proposition 1 275
Appendix C. Proof of Proposition 2 276
Appendix D. Proof of Theorem 1 277
References 279

Section III
Advanced geometrical intuition 283
8. Parallel transport, a central tool in geometric statistics
for computational anatomy: Application to cardiac
motion modeling 285
Nicolas Guigui and Xavier Pennec
1. Introduction 286
1.1 Diffeomorphometry 287
1.2 Longitudinal models 288
1.3 Parallel transport for intersubject normalization 291
1.4 Chapter organization 292
Contents ix

2. Parallel transport with ladder methods 294

2.1 Numerical accuracy of Schild’s and pole ladders 294
2.2 A short overview of the LDDMM framework 298
2.3 Ladder methods with LDDMM 300
3. Application to cardiac motion modeling 304
3.1 The right ventricle and its diseases 305
3.2 Motion normalization with parallel transport 306
3.3 An intuitive rescaling of LDDMM parallel transport 309
3.4 Changing the metric to preserve relative volume
changes 313
3.5 Analysis of the normalized deformations 316
4. Conclusion 321
Acknowledgments 322
References 322

9. Geometry and mixture models 327

Paul Marriott
1. Introduction 327
1.1 Fundamentals of modeling with mixtures 327
1.2 Mixtures and the fundamentals of geometry 329
1.3 Structure of article 330
2. Identification, singularities, and boundaries 331
2.1 Mixtures of finite distributions 334
3. Likelihood geometry 337
4. General geometric structures 341
5. Singular learning theory 343
5.1 Bayesian methods 343
5.2 Singularities and algebraic geometry 345
5.3 Singular learning and model selection 348
6. Nonstandard testing problems 350
7. Discussion 353
References 353

10. Gaussian distributions on Riemannian symmetric

spaces of nonpositive curvature 357
Salem Said, Cyrus Mostajeran, and Simon Heuveline
1. Introduction 358
2. Gaussian distributions and RMT 359
2.1 From Gauss to Shannon 360
2.2 The “right” Gaussian 361
2.3 The normalizing factor Z(σ) 363
2.4 MLE and maximum entropy 366
2.5 Barycenter and covariance 367
2.6 Z(σ) from RMT 369
2.7 The asymptotic distribution 371
2.8 Duality: The Θ distributions 372
x Contents

3. Gaussian distributions and Bayesian inference 373

3.1 MAP versus MMS 375
3.2 Bounding the distance 376
3.3 Computing the MMS 377
3.4 Proof of Proposition 13 381
Appendix A. Riemannian symmetric spaces 383
A.1 The noncompact case 385
A.2 The compact case 386
A.3 Example of Propositions A.1 and A.2 387
Appendix B. Convex optimization 388
B.1 Convex sets and functions 388
B.2 Second-order Taylor formula 390
B.3 Taylor with retractions 391
B.4 Riemannian gradient descent 393
Appendix C. Proofs for Section B 397
References 399

11. Multilevel contours on bundles of complex planes 401

Arni S.R. Srinivasa Rao
1. Introduction 401
2. Infinitely many bundles of complex planes 403
3. Multilevel contours in a random environment 415
3.1 Behavior of X (zl(t), ℂl) at (ℂl \ ℂ0) 420
3.2 Loss of spaces in bundle Bℝ(ℂ) 433
4. Islands and holes in Bℝ(ℂ) 444
4.1 Consequences of Bℝ(ℂ)\ℂl on multilevel contours 453
4.2 PDEs for the dynamics of lost space 459
5. Concluding remarks 463
Acknowledgments 463
References 463

Index 465
Contributors

ed

Fr
eric Barbaresco (107), THALES Land & Air Systems, Meudon, France
Alessandro Barp (21), Department of Engineering, University of Cambridge,
Cambridge; The Alan Turing Institute, The British Library, London,
United Kingdom
Iris Bennett (79), Department of Statistics, North Carolina State University; Corteva
Agriscience, Raleigh, NC, United States
Michel Broniatowski (145), LPSM, Sorbonne Universit
e, Paris, France
Lancelot Da Costa (21), Department of Mathematics, Imperial College London;
Wellcome Centre for Human Neuroimaging, University College London, London,
United Kingdom
Guilherme França (21), Computer Science Division, University of California,
Berkeley, CA, United States
Karl Friston (21), Wellcome Centre for Human Neuroimaging, University College
London, London, United Kingdom
Mark Girolami (21), Department of Engineering, University of Cambridge,
Cambridge; The Alan Turing Institute, The British Library, London,
United Kingdom
Nicolas Guigui (285), Universit
e C^ote d’Azur and Inria, Epione team, Sophia-
Antipolis, Biot, France
Simon Heuveline (357), Centre for Mathematical Sciences, University of Cambridge,
Cambridge, United Kingdom
Michael I. Jordan (21), Computer Science Division; Department of Statistics,
University of California, Berkeley, CA, United States
Steven G. Krantz (1), Department of Mathematics, Washington University in St.
Louis, St. Louis, MO, United States
Soumendra Nath Lahiri (79), Department of Mathematics and Statistics, Washington
University in St. Louis, St. Louis, MO, United States
Tuhin Majumder (79), Department of Statistics, North Carolina State University,
Raleigh, NC, United States
Paul Marriott (327), Department of Statistics and Actuarial Science, University of
Waterloo, Waterloo, ON, Canada
Donald E.K. Martin (79), Department of Statistics, North Carolina State University,
Raleigh, NC, United States

xi
xii Contributors

Cyrus Mostajeran (357), Department of Engineering, University of Cambridge,

Cambridge, United Kingdom
Frank Nielsen (225), Sony Computer Science Laboratories Inc., Tokyo, Japan
Grigorios A. Pavliotis (21), Department of Mathematics, Imperial College London,
London, United Kingdom
Xavier Pennec (285), Universit
e C^ote d’Azur and Inria, Epione team, Sophia-
Antipolis, Biot, France
Salem Said (357), CNRS, Laboratoire LJK, Universit
e Grenoble-Alpes, Grenoble,
France
Arni S.R. Srinivasa Rao (1, 401), Laboratory for Theory and Mathematical Modeling,
Medical College of Georgia; Department of Mathematics, Augusta University,
Augusta, GA, United States
Wolfgang Stummer (145), Department of Mathematics, University of Erlangen–
N€urnberg, Erlangen; School of Business, Economics and Society, University of
Erlangen-N€urnberg, N€urnberg, Germany
Ke Sun (257), CSIRO Data61, Sydney, NSW; The Australian National University,
Canberra, ACT, Australia
Preface

Volume 46 of the Handbook of Statistics with the theme “Geometry and

Statistics” provides state-of-the-art research topics focusing on the interface
of statistics with geometry. Geometrical intuition and clarity have always
helped statistical and mathematical analyses, and this volume presents the
fundamental concepts of the recent advancements on this interface in an enter-
taining and engaging manner. We have also included some chapters purely
dealing with the statistical aspects and others presenting new venues in
complex analysis in the hope that these chapters will foster interactions of
statistics and geometry in the near future.
The contents of the volume range from the basics of complex plane
geometry to classical geometry, dually flat surfaces, deeper geometrical
foundations that can improve our understanding of statistical inferences,
Riemannian surfaces and applications, complex bundles, information
geometry, random matrix theory, and numerical simulations.
This volume will engage both new researchers and experienced authors,
and the contents are introduced by keeping readers from statistics, mathe-
matics, data science, computer scientists, and related fields in mind. All the
authors prepared their chapters carefully and skillfully by keeping the
traditions of the Handbook of Statistics series alive.
The 11 chapters in this volume are divided into 3 sections, namely,
Section I: Foundations in Classical Geometry and Analysis
Section II: Information Geometry
Section III: Advanced Geometrical Intuition
Section I contains three chapters. The first chapter by Arni S.R. Srinivasa Rao
and Steven G. Krantz describes the foundations of complex planes and
analytic functions and introduces the development of newer applications.
A special Jordan curve theorem that passes through a complex plane bundle
and points on the boundary of a ball is proved in this chapter. The advantages
of such special Jordan curves are discussed. The kinds of complex bundles
considered are similar to the bundles introduced in Chapter 11 of this volume.
The second chapter by Alessandro Barp, Lancelot Da Costa, Guilherme
França, Karl Friston, Mark Girolami, Michael I. Jordan, and Grigorios
A. Pavliotis first provides an overview of the fundamental geometric struc-
tures that underlie sampling, optimization, inference, and adaptive behavior
problems. Then, the authors show how to design efficient algorithms to solve

xiii
xiv Preface

these problems using these geometric structures. The third chapter by Donald
E.K. Martin, Iris Bennett, Tuhin Majumder, and Soumendra Nath Lahiri is
about equivalence relations in statistical inference and geometrical analysis.
The authors demonstrate how sparse Markov modeling helps improve the
understanding of statistical inferences. The chapter touches on higher-order
Markov models and derivations of certain conditional probability distributions
and their applications.
Section II contains four chapters. The first chapter by Frederic Barbaresco
based on the foundations of “Lie Groups Thermodynamics” describes a novel
formulation of heat theory and information geometry. The chapter describes
the utility of the Gaussian distribution on the space of Symmetric Positive
Definite matrices, constructions of the Koszul–Fisher metric, and the Casimir
function, which is characterized by Koszul–Poisson cohomology. The second
chapter by Michel Broniatowski and Wolfgang Stummer introduces a general
framework for density-based and distribution function-based divergence
approaches to the relative entropy of Kullback–Leibler information distances
and distribution function-based divergences. The authors also describe in
detail and provide foundations for Cramer–von Mises test statistics and
Anderson–Darling test statistics. The third chapter by Frank Nielsen recalls
the construction of dually flat spaces from a pair of convex conjugate func-
tions related to the Legendre–Fenchel transform. This dually flat structure is
then illustrated for exponential families, mixture families, and regular homo-
geneous convex cones. For mixture families, the Shannon negentropy defines
the convex function inducing the dually flat space. It is, however, usually not
available in closed form for continuous mixtures. The chapter reports a
closed-form formula for the mixture family of two Cauchy distributions and
uses this formula to explicitly build a dually flat mixture family manifold.
The fourth chapter by Ke Sun considers the problem of embedding a
low-dimensional latent space into a high-dimensional observation space.
The author tackles the definitions of the simplicity of such embeddings based
on the framework of information geometry and discusses the relationships
between parametric and nonparametric embeddings.
Section III contains four chapters. The first chapter by Nicolas Guigui and
Xavier Pennec is richly illustrated and presents the tool of parallel transport
via affine or Riemannian connection for problems in computational anatomy.
Parallel transport defines how tangent vectors are related between tangent
planes that are infinitesimally close to each other. The authors discuss the
choice of parallel transport and its numerical accuracy in ladder method
implementations for statistical analysis of subject-specific longitudinal
changes or motions with respect to common template anatomy. The authors
then apply their novel parallel transport method to motion modeling of the
cardiac right ventricle under pressure or volume overload. To resolve this
problem, parallel transport is shown to be insufficient for normalizing
large-volume deformations, and the authors propose a novel normalization
Preface xv

procedure in which parallel transport is demonstrated to be a useful tool for

choosing the appropriate metric adapted to the data. The second chapter by
Paul Marriott presents a tutorial survey on the use of several geometries for
characterizing the inference properties of statistical mixture models. The
viewpoints of affine geometry, likelihood convex geometry, information
geometry, and algebraic geometry are explained and illustrated with numer-
ous examples that highlight the complex inferential nature of mixture models
in general. The third chapter by Salem Said, Cyrus Mostajeran, and Simon
Heuveline presents the state-of-the-art theory of Gaussian distributions on
Riemannian symmetric spaces. It is shown that when the Riemannian sym-
metric spaces have nonpositive curvature, the maximum entropy distribution
with the prescribed barycenter and dispersion defines a Riemannian Gaussian
distribution such that the maximum likelihood estimator for that family
amounts to computing a Riemannian barycenter of the observations. In the
second part of this chapter, the authors tackle the mathematical tractability
of the normalizing constants of these Riemannian Gaussian distributions and
show an interesting connection with the random matrix theory. In the fourth
chapter by Arni S.R. Srinivasa Rao, the author introduces newer principles
of constructing contours passing through complex plane bundles. The geome-
try created through this process led to new concepts such as “multilevel
contours” within the bundle, “islands,” and “holes” within complex planes.
Information is transported from one plane to another through the geometrical
shapes of the contours.
We express our sincere thanks to Mr. Sam Mahfoudh, acquisitions editor
(Elsevier and North-Holland), for his overall administrative support through-
out the preparation of this volume. His valuable involvement in the project
is highly appreciated. We thank Ms. Naiza Mendoza, developmental editor
(Elsevier), for providing excellent assistance to the editors and for engaging
authors in all kinds of technical queries throughout the preparation and until
the proofing stage and production. Our thanks also go to Md. Sait Abdulla,
the project manager of book production, RELX India Private Limited, Chen-
nai, India, for leading the production, responding to several rounds of queries
by the authors, being available at all times for printing activities, and
providing assistance to the editors. Our sincere thanks and gratitude go to
all the authors for writing brilliant chapters by keeping to our requirements
of the volume. We very much thank our referees for their timely assessment
of the chapters.
We firmly believe that this volume has come up at the right time and it
gives us great satisfaction to have been involved in its production. We are
convinced that this collection will be a useful resource for beginners and
advanced scientists working in statistics, mathematics, and geometry.
Frank Nielsen
Arni S.R. Srinivasa Rao
C.R. Rao
This page intentionally left blank
Section I

Foundations in classical
geometry and analysis
This page intentionally left blank
Chapter 1

Geometry, information,
and complex bundles
Steven G. Krantza and Arni S.R. Srinivasa Raob,c,∗
a
Department of Mathematics, Washington University in St. Louis, St. Louis, MO, United States
b
Laboratory for Theory and Mathematical Modeling, Medical College of Georgia, Augusta
University, Augusta, GA, United States
c
Department of Mathematics, Augusta University, Augusta, GA, United States
∗
Corresponding author: e-mail: arni.rao2020@gmail.com

Abstract
In this chapter, we will describe information geometric principles on complex planes.
Using the geometric constructions, we prove two theorems on special types of
constructions of Jordan curves around a ball within a complex bundle. Essentials on
complex planes and bundles required for understanding the contour constructions done
in the chapter are provided.
Keywords: Information geometry, Complex analysis, Jordan curves, Complex bundles

1 Introduction
The idea of combining Riemann surfaces with probability densities for under-
standing distances between two populations was introduced by C.R. Rao in
1945 (Rao, 1949). This led to the development of the subject of information
geometry (Amari, 2016; Amari and Nagaoka, 2000; Ay et al., 2017; van
Rijsbergen, 2004). The principles of information geometry were helpful in
statistical decisions and inferences (Amari et al., 1987; Plastino et al., 2021)
and in statistical physics (Bhattacharyya and Keerthi, 2000; Dehesa et al.,
2012; Frieden, 1992, 2021; Jaiswal et al., 2021). Most of these articles and
associated articles focused on the Cramer–Rao inequality and on Fisher
Information (Efron, 1975; Rao, 1973), and in obtaining deformation proper-
ties for the exponential family of distributions. The idea of transportation
of information from one region to another region through topological struc-
tures and through complex plane bundles was introduced in Rao (2021).

Handbook of Statistics, Vol. 46. https://doi.org/10.1016/bs.host.2022.03.002

Copyright © 2022 Elsevier B.V. All rights reserved. 3
4 SECTION I Foundations in classical geometry and analysis

Through such an analysis a new concept called multilevel contours was

developed. Such translation of information from one plane to another plane
is new in the subject of geometric structures within complex bundles and
arguably has implications in climate analysis. The connectedness properties
of complex vector bundles through homomorphism principles help in under-
standing four-dimensional manifolds; see, for example, Chern (1989) and
Chern (1977). Understanding the curvature of objects coming from a bundle
and Einstein–Hermitian conditions on such curvatures helps to understand
the stability of the system of bundles. These are different from understanding
geodesic distances on complex bundles. The idea of constructing contours
passing through the bundle helps in creating multiple paths and these are
conceptually different from analysis of vector bundles constructed earlier
(Kobayashi, 2014). Deformation of complex vector bundles can be performed
in various ways; for example, see Green and Lazarsfeld (1991), Kruglikov
(2007), and Wu and Yau (2016).
The information geometry principles were also used to obtain deformed
λ-exponential families (Tsallis, 1988; Zhang and Wong, 2021). The princi-
ples associated with obtaining distances between two probability densities
through information geometry theory have been attracting several theoreti-
cians worldwide. The idea of Riemann surfaces introduced through informa-
tion geometry caused scientists to think of the geometry of shapes formed
while building measures to obtain distances between two independent
populations.
Information geometry principles on the complex plane and their applica-
tions were introduced by Rao and Krantz (2020, 2021). The advantages of
such newer ideas are shown to be helpful in virtual tourism through the imple-
mentation of technologies of virtual reality. The distances between three-
dimensional objects of tourist locations and angles between them can be
preserved to improve existing VR technology and the ideas of contour inte-
grals of complex planes and path-connectedness properties of planes were
shown to be helpful in these newer applications. The distance measures of
regular information geometric principles and geodesic distances do not pre-
serve angular information. Hence the newer idea of geometry on complex
planes (especially conformality) and information preserved in the geometric
objectives and their potential applications enhances the utility of infor-
mation geometry. The geometric interpretations of structures formed on
complex planes and, more generally, automorphisms of Riemann surfaces
are well established—see, for example, Krantz (2004), Krantz and Parks
(1999), and Greene et al. (2011). The Bergman metric and Bergman geom-
etry are central tools in the analysis of such complex manifolds (Greene
et al., 2011; Yoo, 2017). See also the Kobayashi–Royden metric
(Choi, 2012).
 Geometry, information, and complex bundles Chapter 1 5

Our chapter is structured as follows: In the next section we present the fun-
damentals of complex analysis. Section 3 describes the ideas of geometry on
complex planes and Section 4 summarizes the newer advantages of informa-
tion geometry on complex planes.

2 Complex planes
Let  be the complex plane, and let S
 be a region. Let z be a complex
number in S, and z ¼ x + iy for x, y   (the set of real numbers). We define
a function f as follows:
f :S! (1)
such that f(z) ¼ w for w ¼ u + iv and u, v  . Such a function f is said to be
analytic(or holomorphic) in an open set S if there exists a complex derivative
at z for every z  S. If f is analytic at each point in the entire complex plane,
then f is called an entire function. Suppose f ¼ u + iv is analytic in a domain U
(an open and connected set), then the first-order partial differential equations
of u and v satisfy
∂u ∂v ∂u ∂v
¼ and ¼ : (2)
∂x ∂y ∂y ∂x
The equations in (2) are known as the Cauchy–Riemann (CR) equations.
Geometric structures can be constructed on a domain in the plane and, using
such a domain, information on geometric structures can be transported (Rao,
2021). A smooth function uðx1 , x2 , …, xn Þ that satisfies the equation
∂2 u ∂2 u ∂2 u
+ 2 +⋯+ 2¼0 (3)
∂x12 ∂x2 ∂xn
is called a harmonic function, and Eq. (3) is called Laplace’s equation. Eq. (3)
can be also written as Δu ¼ 0 for the operator
∂2 ∂2 ∂2
Δu ¼ + + ⋯ + :
∂x21 ∂x22 ∂x2n
A two-variable function u(x, y) is called harmonic if
∂2 u ∂2 u
+ 2 ¼ 0: (4)
∂x 2 ∂y
A standard result that can be proved using Laplace’s equation, and f ¼ u + iv,
where u ¼ u(x, y) and v ¼ v(x, y) is stated below. See Krantz (2004), Churchill and
Brown (1984), Ahlfors (1978), Krantz (2017), and Rudin (1987).
Theorem 1. If f ¼ u + iv is analytic in U, then u and v are harmonic in U.

Proof. Using (2) we can write

6 SECTION I Foundations in classical geometry and analysis

∂ ∂u ∂ ∂v
¼
∂x ∂x ∂x ∂y
∂ ∂v
¼
∂y ∂x

 (5)
∂ ∂u
¼ 
∂y ∂y
∂2 u
¼ :
∂y2
This implies that Δu ¼ 0. Similarly,

∂ ∂v ∂ ∂u
¼
∂x ∂x ∂x ∂y
∂ ∂u
¼
∂y ∂x
∂ ∂v (6)
¼
∂y ∂y
∂2 u
¼ 2:
∂y
This implies that Δv ¼ 0. □

The Eqs. (5) and (6) imply that u and v are harmonic. Harmonic functions
combined with CR can assist in understanding the conjugate of components
and information transportation. Analytic functions are also important in the
formation of contours which were shown to transport information from one
complex plane to another complex plane within a finite and infinite complex
plane bundle (Rao, 2021). The set γðtÞ ¼ ðxðtÞ, yðtÞÞ
 for a set of real
values t  [a, b], and for a continuous x(t) and y(t) is said to be an arc. The
arc γ(t) is called a Jordan arc if it is simple and closed, i.e., if γ(t1) 6¼ γ(t2)
for all t1 6¼ t2 except for γ(a) ¼ γ(b). A closed arc is an arc for which
γ(a) ¼ γ(b), and such arcs are also referred as Jordan curves. See Fig. 1.

FIG. 1 Arcs and Jordan arcs.

 Geometry, information, and complex bundles Chapter 1 7

Let f be a complex-valued function defined on the complex values of γ(t) for

t  [a, b]. The contour integral of f on γ(t) is defined using the Riemann–
Stieltjes integral as
I Z b
f dzðtÞ ¼ f ½γðtÞdγðtÞ: (7)
γ a

Suppose we map the value of t onto another real-valued function ϕ(ζ) for
(a1  ζ  b1), then γ(t) values within [a, b] are transformed into, say, Γ(ζ) ¼
γ[ϕ(t)]. The length of the arc γ(t), say L(γ(t)), is defined to be
Z b
LðγðtÞÞ ¼ jγ 0 ðtÞjdt: (8)
a

After the change variables described, Eq. (8) becomes

Z b1
LðγðtÞÞ ¼ jγ 0 ½ϕðtÞjϕ0 ðtÞdt: (9)
a1

Here γ 0 and ϕ0 indicate derivatives of γ and ϕ, respectively. The parametric

description described above also suggests that γ could be a homeomorphism
when the domain [a, b] is mapped into [γ(a), γ(b)]. Similarly, Γ could form
a homeomorphism when mapped from [γ(a), γ(b)] to [γ(a1), γ(b1)]. Hence,
isometryproperties can be observed depending upon whether
γ : ½a, b ! ½γðaÞ, γðbÞ
is one to one or not. A piecewise smooth arc is called a contour. For example:
suppose we form two arcs γ 1(t1) and γ 2(t2) for t1  [a, b] and t2  [b, c] and
γ 1(b) ¼ γ2(b). Let γ be the concatenation of these two curves: γ 1 followed by
γ 2. Then γ(t) for t  [a, c] represents a contour provided γ(t) is continuous and
γ 0 (t) is piecewise continuous. Let L(γ 1(t)) and Lðγ 2 ðt2 ÞÞ be the lengths of the
two arcs γ 1(t1) and γ 2(t2), respectively. The length of the contour γ(t), say
LðγðtÞÞ, can be computed as
LðγðtÞÞ ¼ Lðγ 1 ðtÞÞ + Lðγ 2 ðt2 ÞÞ
Z b1 Z
 0  c1   (10)
¼ γ ½ϕ1 ðt1 Þϕ0 ðt1 Þdt1 + γ 0 ½ϕ2 ðt2 Þϕ0 ðt2 Þdt2 :
1 1 2 2
a1 b1

In Eq. (10) the two parametric representation functions are ϕ1(t1) and ϕ2(t2)
for t1  [a, b] and t2  [b, c]. See Fig. 2. Suppose there are multiple para-
metric representation functions ϕj(tj) for tj in [aj, aj+1] corresponding to the
piecewise smooth arcs γ j for j ¼ 1, 2, …, k. The contour γ can be represented
using piecewise smooth arcs as
Z
γðtj Þdtj : (11)
8 SECTION I Foundations in classical geometry and analysis

FIG. 2 Contour formation from piecewise smooth arcs. Here γ i(ti) for i ¼ 1, 2, …, 6 are piece-
wise smooth arcs defined on the real number intervals ½a1 , a2 , ½a2 , a3 , …, ½a5 , a6 . After the para-
metric representation described in the text, one can compute the total length of the contour, say C,
using piecewise lengths of γ i(ti).

Then the length L(γ(t)) of the contour γ which is formed by concatenating

k piecewise smooth arcs can be obtained from the lengths of these piecewise
arcs, L(γ(tj)) as
Xk Z aj + 1  
 0 
LðγðtÞÞ ¼ γ j ½ϕj ðtj Þϕ0j ðtj Þdtj : (12)
j¼i aj

Analytic functions in a domain U on Jordan curves have interesting properties

due to the Cauchy integral theorem and the Cauchy integral formula stated
below.
Theorem 2. Suppose f is analytic on an open set U except for a finite number
of points within U at which f is only continuous. Then, for every piecewise
smooth closed arc that is homotopic to a point within U, we have
I
f ðzÞ dz ¼ 0: (13)
z

Definition 1 (Cauchy integral formula). Suppose that f is an analytic function

on a simply connected open set U. Let γ be a smooth piecewise closed coun-
terclockwise (positively oriented) arc in U. Then, for any point z0 interior to γ,
we have
Z
1 f ðzÞ
f ðz0 Þ ¼ dz: (14)
2πi γ z  z0
 Geometry, information, and complex bundles Chapter 1 9

Suppose we define another function g(z) on U as

8
< f ðzÞ  f ðz0 Þ if z 6¼ z
gðzÞ ¼ z  z0 0
: (15)
: 0
f ðzÞ if z ¼ z0
Then, g is also analytic. Closed smooth arcs can be utilized to transport infor-
mation from one region to another region within .

Theorem 3. Suppose that f is analytic everywhere inside and on a closed

smooth arc or closed contour γ. If z0 is interior to γ, then
Z
1 f ðzÞdz
f 0 ðzÞ ¼ , (16)
2πi γ ðz  z0 Þ2
Z
1 f ðzÞdz
f 00 ðz0 Þ ¼ , (17)
πi γ ðz  z0 Þ3
and by mathematical induction, we will have
Z
ðnÞ n! f ðzÞdz
f ðz0 Þ ¼ : (18)
2πi γ ðz  z0 Þn+1

Here the superscript (n) denotes a derivative. One of the important conse-
quences of the Cauchy integral formula (18) is
 
 ðnÞ  n! max j f ðzÞjA
 f ðz0 Þ  , ðn ¼ 1, 2, …Þ (19)
ðr A Þn
where f is analytic inside a circle A (with radius rA). The inequality (19) is
also called the Cauchy estimate and is used in proving Liouville’s theorem
on entire functions.
Theorem 4 (Liouville’s theorem). A bounded entire function is constant
throughout the complex plane.

2.1 Important implications of Liouville’s theorem

Liouville’s theorem helps prove the fundamental theorem of algebra, which
states that, for any nth degree polynomial with n at least 1, f(z), there exists
at least one point z0 such that f(z0) ¼ 0. There are several proofs of the funda-
mental theorem of algebra available; see the recent discussion in Krantz
(2020). Suppose that ρ(z) is a complex-valued polynomial given by
ρðzÞ ¼ c0 + c1 z + c2 z2 + ⋯ + cn zn :
Let
1
f ðzÞ ¼
ρðzÞ
10 SECTION I Foundations in classical geometry and analysis

and assume ρ(z) is not zero for all z  . After some algebraic constructions
and applying triangular inequality, we arrive at
1
j f ðzÞj ¼ is bounded: (20)
jρðzÞj
Eq. (20) implies f is bounded in the entire plane. But, by Liouville’s theo-
rem, f(z) is constant, which is a contradiction because ρ(z) is not constant.
Suppose a function f is analytic throughout an annular domain with radii
γ A and γ B and center z0 such that r A < jz  z0 j < r B . Let γ be a Jordan curve
around z0 within the z values for γ A < jz  z0 j < γ B. Then, at each such z, the
function f(z) can be represented as the following series:
X
∞
f ðzÞ ¼ An ðz  z0 Þn , for rA < jz  z0 j < rB (21)
n¼∞

where
Z
1 f ðzÞdz
An ¼ for n ¼ 0,  1,  2, …: (22)
2πi γ ðz  z0 Þn+1
The series in Eq. (21) is called the Laurent series. Suppose we write g(z) ¼
f(z + z0). Then, g(z) is analytic on the annulus rA < jzj < rB and we can write
g(z) with the following Laurent series expression:
X
∞ X
∞
gðzÞ ¼ Bn zn + Cn zn , (23)
n¼0 n¼0

where
Z
1 gðzÞdz
Bn ¼ for n ¼ 0, 1, 2, …, (24)
2πi γ zn+1
Z
1 gðzÞdz
Cn ¼ for n ¼ 0, 1, 2, …, (25)
2πi γ zn+1
See Fig. 3. Let us consider the disks D(z0, rA), and D(0, rA) as in Fig. 3. If
we excise the centers z0 and 0 from these disks, respectively, then the sets of
remaining points of the disks are called deleted neighborhoodsof z0 and 0,
respectively. We call z0 an isolated singularity of f if f is not analytic at z0
and f is analytic on a deleted neighborhood of z0. Similarly, an isolated singu-
larity of f at 0 can be defined.
When 0 in Fig. 3 is the isolated singular point of f, then f(z) can be
expressed as
X
∞
f ðzÞ ¼ Bn zn + C1 z1 + C2 z2 + ⋯ + Cn zn + ⋯ , (26)
n¼0
 Geometry, information, and complex bundles Chapter 1 11

FIG. 3 (A) Jordan curve within the domain r A < jz  z0 j < r B and expression of f(z) for a
z value within this region. (B) Jordan curve within the domain r A < jzj < r B and g(z) in
Eq. (23) is analytic within this domain.

where Bn and Cn are defined as in Eqs. (24) and (25). The number C1 in
Eq. (26) is called the residue of f at 0, and is denoted by
Res f ðzÞ:
z¼0

Similarly, when z0 is an isolated singular point, one can write the Laurent
series expression. The Cauchy residue theorem is a helpful tool to compute
a contour integral when there are a finite number k of isolated singular points
within a simple, closed contour γ. Suppose f is analytic within and on γ except
for a finite number of isolated singular points within γ, then
Z X
k
f ðzÞdz ¼ 2πi Res f ðzÞ: (27)
γ z¼zn
n¼1

The structure of disks and domains described earlier in the Laurent series
expression can be applied in the transportation of information within complex
planes. See Rao and Krantz (2021). In Rao and Krantz (2021) we have
described the basics and the importance of conformal mapping, and preserva-
tion of angles in 3D objects. Conformality is an important feature of analytic
functions. See Rao and Krantz (2021) and Ahlfors (1978), Churchill and
Brown (1984), Krantz (2004), and Rudin (1987) for general ideas and founda-
tions on conformal mapping of two piecewise smooth arcs, and especially
regarding angle preservations. In this chapter, we demonstrate the conformal
mapping principle on two intersecting Jordan curves.
Let us consider two Jordan curves J1 and J2 within an annulus with radii
rA and rB, respectively. Assume that the two curves J1 and J2 intersect at
z1 , z2 with
12 SECTION I Foundations in classical geometry and analysis

 
γ A < z1  < γ B , (28)
and
 
γ A < z2  < γ B : (29)
The curve J1 on the plane is created from the real values of the interval, say,
[a1, b1], and the curve J2 is created on the plane from the real-valued interval,
say, [a2, b2] (Fig. 4).
Let
c1 ¼ arg½J 02 ðt2 Þ  arg½J 01 ðt1 Þ
c2 ¼ arg½J 02 ðt4 Þ  arg½J 01 ðt3 Þ,
where t1, t3  [a1, b1], and t2, t4  [a2, b2]. Here J1(t1) ¼ J2(t2) ¼ c1, and J1(t3)
¼ J2(t4) ¼ c2. Let f1 and f2 be two analytic functions mapped from J1 and J2,
respectively. Assume f1(c1) 6¼ 0 and f2(c2) 6¼ 0. Due to the conformality the
angle from two mapped curves, say, S2 and S1, at c1 will be equal to the angle
at f1(c1) from J2 and J1, and the angle from S2 and S1 at c2 will be equal to
the angle at f2(c2) from J2 and J1,
We have discussed conformal mapping of curves within an annulus. One
can bring a similar description in any region U C in which the two curves
J1 and J2 are created.

FIG. 4 Conformal mappings of two intersecting Jordan curves J1 and J2 onto two intersecting
Jordan curves S1 and S2. Note that although we have demonstrated here a Jordan curve mapping
to another Jordan curve, a Jordan curve need not map always to a Jordan curve. The two curves
J1 and J2 are generated independently from each other. We also mention here that a Jordan curve
within an annulus need not map to a curve situated in another annulus.
Another random document with
no related content on Scribd:
Snakes. Such organs are found in the Sting-rays, the tail of which is
armed with one or more powerful barbed spines. Although they lack
a special organ secreting poison, or a canal in or on the spine by
which the venomous fluid is conducted, the symptoms caused by a
wound from the spine of a Sting-ray are such as cannot be
accounted for merely by the mechanical laceration, the pain being
intense, and the subsequent inflammation and swelling of the
wounded part terminating not rarely in gangrene. The mucus
secreted from the surface of the fish and inoculated by the jagged
spine evidently possesses venomous properties. This is also the
case in many Scorpænoids, and in the Weaver (Trachinis), in which
the dorsal and opercular spines have the same function as the
caudal spines of the Sting-rays; however, in the Weavers the spines
are deeply grooved, the groove being charged with a fluid mucus. In
Synanceia the poison-organ (Fig. 99,) is still more developed: each
dorsal spine is in its terminal half provided with a deep groove on
each side, at the lower end of which lies a pear-shaped bag
containing the milky poison; it is prolonged into a membranous duct,
lying in the groove of the spine, and open at its point. The native
fishermen, well acquainted with the dangerous nature of these
fishes, carefully avoid handling them; but it often happens that
persons wading with naked feet in the sea, step upon the fish, which
generally lies hidden in the sand. One or more of the erected spines
penetrate the skin, and the poison is injected into the wound by the
pressure of the foot on the poison-bags. Death has not rarely been
the result.
Fig. 99.—A dorsal spine, with
poison-bags, of Synanceia
verrucosa. Indian Ocean.
 Fig. 100.—Opercular part of the
Poison-apparatus of
Thalassophryne (Panama).
1. Hinder half of the head, with the
venom-sac* in situ. a, Lateral
line and its branches; b, Gill-
opening; c, Ventral fin; d, Base
of Pectoral fin; e, Base of
dorsal.
2. Operculum with the
perforated spine.

The most perfect poison-organs hitherto discovered in fishes are

those of Thalassophryne, a Batrachoid genus of fishes from the
coasts of Central America. In these fishes the operculum again and
the two dorsal spines are the weapons. The former (Fig. 100, 2) is
very narrow, vertically styliform and very mobile; it is armed behind
with a spine, eight lines long, and of the same form as the hollow
venom-fang of a snake, being perforated at its base and at its
extremity. A sac covering the base of the spine discharges its
contents through the apertures and the canal in the interior of the
spine. The structure of the dorsal spines is similar. There are no
secretory glands imbedded in the membranes of the sacs; and the
fluid must be secreted by their mucous membrane. The sacs are
without an external muscular layer, and situated immediately below
the thick loose skin which envelops the spines to their extremity; the
ejection of the poison into a living animal, therefore, can only be
effected, as in Synanceia, by the pressure to which the sac is
subjected the moment the spine enters another body.
Finally, a singular apparatus found in many Siluroids may be
mentioned in connection with the poison-organs, although its
function is still problematical. Some of these fishes are armed with
powerful pectoral spines and justly feared on account of the
dangerous wounds they inflict; not a few of them possess, in addition
to the pectoral spines, a sac with a more or less wide opening in the
axil of the pectoral fin; and it does not seem improbable that it
contains a fluid which may be introduced into a wound by means of
the pectoral spine, which would be covered with it, like the barbed
arrow-head of an Indian. However, whether this secretion is equally
poisonous in all the species provided with that axillary sac, or
whether it has poisonous qualities at all, is a question which can be
decided by experiments only made with the living fishes.
 CHAPTER XV.

DISTRIBUTION OF FISHES IN TIME.

Of what kind the fishes were which were the first to make their
appearance on the globe; whether or not they were identical with, or
similar to, any of the principal types existing at present; are
questions which probably will for ever remain hidden in mystery and
uncertainty. The supposition that the Leptocardii and Cyclostomes,
the lowest of the vertebrate series, must have preceded the other
sub-classes, is an idea which has been held by many Zoologists:
and as the horny teeth of the Cyclostomes are the only parts of their
body which under favourable circumstances might have been
preserved, Palæontologists have ever been searching for this
evidence.

Fig. 101. Right

dental plate of
Myxine affinis.
Indeed, in deposits belonging to the Lower Silurian and
Devonian, in Russia, England, and North America, minute, slender,
pointed horny bodies, bent like a hook, with sharp opposite margins,
have been found and described under the name of Conodonts. More
frequently they possess an elongated basal portion, in which there is
generally a larger tooth with rows of similar but smaller denticles on
one or both sides of the larger tooth, according as this is central or at
one end of the base. In other examples there is no prominent central
tooth, but a series of more or less similar teeth is implanted on a
straight or curved base. Modifications of these arrangements are
very numerous, and many Palæontologists entertain still doubts
whether the origin of these remains is not rather from Annelids and
Mollusks than from Fishes.
 [See G. J. Hinde, in “Quarterly Journal of the Geological Society,”
1879.]
The first undeniable evidence of a fish, or, indeed, of a vertebrate
animal, occurs in the Upper Silurian Rocks, in a bone-bed of the
Downton sandstone, near Ludlow. It consists of compressed, slightly
curved, ribbed spines, of less than two inches in length (Onchus); of
small shagreen-scales (Thelodus); the fragment of a jaw-like bar with
pluricuspid teeth (Plectrodus); the cephalic bucklers of what seems
to be a species of Pteraspis; and, finally, the coprolitic bodies of
phosphate and carbonate of lime, including recognisable remains of
the Mollusks and Crinoids inhabiting the same waters. But no
vertebra or other part of the skeleton has been found. The spines
and scales seem to have belonged to the same kind of fish, which
probably was a Plagiostome. It is quite uncertain whether or not the
jaw (if it be the jaw of a fish[16]) belonged to the buckler-bearing
Pteraspis, the position of which among Ganoids, with which it is
generally associated, is open to doubt.
No detached undoubted tooth of a Plagiostome or Ganoid scale
has been discovered in the Ludlow deposits: but so much is certain
that those earliest remains in Palæozoic rocks belonged to fishes
closely allied to forms occurring in greater abundance in the
succeeding formation, the Devonian, where they are associated with
undoubted Palæichthyes, Plagiostomes as well as Ganoids.

These fish-remains of the Devonian or Old Red Sandstone, can

be determined with greater certainty. They consist of spines or the
so-called Ichthyodorulites, which show sufficiently distinctive
characters to be referred to several genera, one of them, Onchus,
still surviving from the Silurian epoch. All these spines are believed
to be those of Chondropterygians, to which order some pluricuspid
teeth (Cladodus) from the Old Red Sandstone in the vicinity of St.
Petersburg have been referred likewise.
 The remains of the Ganoid fishes are in a much more perfect
state of preservation, so that it is even possible to obtain a tolerably
certain idea of the general appearance and habits of some of them,
especially of such as were provided with hard carapaces, solid
scales, and ordinary or bony fin-rays. A certain proportion of them,
as might have been expected, remind us, with regard to external
form, of Teleosteous fishes rather than of any of the few still existing
Ganoid types; but it is contrary to all analogy and to all
palæontological evidence to suppose that those fishes were, with
regard to their internal structure, more nearly allied to Teleosteans
than to Ganoids. If they were not true Ganoids, they may be justly
supposed to have had the essential characters of Palæichthyes.
Other forms exhibit even at that remote geological epoch so
unmistakably the characteristics of existing Ganoids, that no one can
entertain any doubt with regard to their place in the system. In none
of these fishes is there any trace of vertebral segmentation.
The Palæichthyes of the Old Red Sandstone, the systematic
position of which is still obscure, are the Cephalaspidæ from the
Lower Old Red Sandstone of Great Britain and Eastern Canada;
Pterichthys, Coccosteus, and Dinichthys: genera which have been
combined in one group—Placodermi; and Acanthodes and allied
genera, which combined numerous branchiostegals with
chondropterygian spines and a shagreen-like dermal covering.
Among the other Devonian fishes (and they formed the majority)
two types may be recognised, both of which are unmistakably
Ganoids. The first approaches the still living Polypterus, with which
some of the genera like Diplopterus singularly agree in the form and
armature of the head, the lepidosis of the body, the lobate pectoral
fins, and the termination of the vertebral column. Other genera, as
Holoptychius, have cycloid scales; many have two dorsal fins
(Holoptychius), and, instead of branchiostegals, jugular scutes;
others one long dorsal confluent with the caudal (Phaneropleuron).
In the second type the principal characters of the Dipnoi are
manifest, and some of them, for example Dipterus, Palædaphus,
Holodus, approach so closely the Dipnoi which still survive, that the
differences existing between them warrant a separation into families
only.
Devonian fishes are frequently found under peculiar
circumstances, enclosed in the so-called nodules. These bodies are
elliptical flattened pebbles, which have resisted the action of water in
consequence of their greater hardness, whilst the surrounding rock
has been reduced to detritus by that agency. Their greater density is
due to the dispersion in their substance of the fat of the animal which
decomposed in them. Frequently, on cleaving one of these nodules
with the stroke of the hammer, a fish is found embedded in the
centre. At certain localities of the Devonian, fossil fishes are so
abundant that the whole of the stratum is affected by the
decomposing remains emitting a peculiar smell when newly opened,
and acquiring a density and durability not possessed by strata
without fishes. The flagstones of Caithness are a remarkable
instance of this.

The fish-remains of the Carboniferous formation show a great

similarity to those of the preceding. They occur throughout the
series, but are very irregularly distributed, being extremely scarce in
some countries, whilst in others entire beds (the so-called bone-
beds) are composed of ichthyolites. In the ironstones they frequently
form the nuclei of nodules, as in the Devonian.
Of Chondropterygians the spines of Onchus and others still
occur, with the addition of teeth indicative of the existence of fishes
allied to the Cestracion-type (Cochliodus, Psammodus): a type which
henceforth plays an important part in the composition of the extinct
marine fish faunæ. Another extinct Selachian family, that of
Hybodontes, makes its appearance, but is known from the teeth
only.
Of the Ganoid fishes, the family Palæoniscidæ (Traquair) is
numerously represented; others are Cœlacanths (Cœlocanthus,
Rhizodus), and Saurodipteridæ (Megalichthys). None of these fishes
have an ossified vertebral column, but in some (Megalichthys) the
outer surface of the vertebræ is ossified into a ring; the termination of
their tail is heterocercal. The carboniferous Uronemus and the
Devonian Phaneropleuron are probably generically the same; and
the Devonian Dipnoi are continued as, and well represented by,
Ctenodus.

The fishes of the Permian group are very similar to those of the
Carboniferous. A type which in the latter was but very scantily
represented, namely the Platysomidæ, is much developed. They
were deep-bodied fish, covered with hard rhomboid scales
possessing a strong anterior rib, and provided with a heterocercal
caudal, long dorsal and anal, short non-lobate paired fins (when
present), and branchiostegals. The Palæoniscidæ are represented
by many species of Palæoniscus, Pygopterus and Acrolepis, and
Cestracionts by Janassa and Strophodus.

The passage from the Palæozoic into the Mesozoic era is not
indicated by any marked change as far as fishes are concerned. The
more remarkable forms of the Trias are Shark-like fishes represented
by ichthyodorulithes like Nemacanthus, Liacanthus, and Hybodus;
and Cestracionts represented by species of Acrodus and
Strophodus. Of the Ganoid genera Cœlacanthus, Amblypterus
(Palæoniscidæ), Saurichthys persist from the Carboniferous epoch.
Ceratodus appears for the first time (Muschel-Kalk of Germany).
Thanks to the researches of Agassiz, and especially Sir P.
Egerton, the ichthyological fauna of the Lias is, perhaps, the best
known of the Mesozoic era, 152 species having been described. Of
the various localities, Lyme Regis has yielded more than any other,
nearly all the Liassic genera being represented there by not less
than seventy-nine species. The Hybodonts and Cestracionts
continue in their fullest development. Holocephales (Ischyodus), true
Sharks (Palæoscyllium), Rays (Squaloraja, Arthropterus), and
Sturgeons (Chondrosteus) make their first appearance; but they are
sufficiently distinct from living types to be classed in separate
genera, or even families. The Ganoids, especially Lepidosteoids,
predominate over all the other fishes: Lepidotus, Semionotus,
Pholidophorus, Pachycormus, Eugnathus, Tetragonolepis, are
represented by numerous species; other remarkable genera are
Aspidorhynchus, Belonostomus, Saurostomus, Sauropsis,
Thrissonotus, Conodus, Ptycholepis, Endactis, Centrolepis,
Legnonotus, Oxygnathus, Heterolepidotus, Isocolum, Osteorhachis,
Mesodon. These genera offer evidence of a great change since the
preceding period, the majority not being represented in older strata,
whilst, on the other hand, many are continued into the succeeding
oolithic formations. The homocercal termination of the vertebral
column commences to supersede the heterocercal, and many of the
genera have well ossified and distinctly segmented spinal columns.
Also the cycloid form of scales becomes more common: one genus
(Leptolepis) being, with regard to the preserved hard portions of its
organisation, so similar to the Teleosteous type that some
Palæontologists refer it (with much reason) to that sub-class.
[See E. Sauvage, Essai sur la Faune Ichthyologique de la période
Liasique. In “Bibl. de l’école des hautes études,” xiii. art. 5. Paris 1875.
8o.]
As already mentioned, the Oolithic formations show a great
similarity of their fish-fauna to that of the Lias; but still more apparent
is its approach to the existing fauna. Teeth have been found which
cannot even generically be distinguished from Notidanus. The Rays
are represented by genera like Spathobatis, Belemnobatis,
Thaumas; the Holocephali are more numerous than in the Lias
(Ischyodus, Ganodus). The most common Ganoid genera are
Caturus, Pycnodus, Pholidophorus, Lepidotus, Leptolepis, all of
which had been more or less fully represented in the Lias. Also
Ceratodus is continued into it.
 The Cretaceous group offers clear evidence of the further
advance towards the existing fauna. Teeth of Sharks of existing
genera Carcharias (Corax), Scyllium, Notidanus, and Galeocerdo,
are common in some of the marine strata, whilst Hybodonts and
Cestracionts are represented by a small number of species only; of
the latter one new genus, Ptychodus, appears and disappears. A
very characteristic Ganoid genus, Macropoma, comprises
homocercal fishes with rounded ganoid scales sculptured externally
and pierced by prominent mucous tubes. Caturus becomes extinct.
Teeth and scales of Lepidotus (with Sphærodus as subgenus),
clearly a freshwater fish, are widely distributed in the Wealden, and
finally disappear in the chalk; its body was covered with large
rhomboidal ganoid scales. Gyrodus and Aspidorhynchus occur in the
beds of Voirons, Coelodus and Amiopsis (allied to Amra), in those of
Comen, in Istria. But the Palæichthyes are now in the minority;
undoubted Teleosteans have appeared, for the first time, on the
stage of life in numerous genera, many of which are identical with
still existing fishes. The majority are Acanthopterygians, but
Physostomes and Plectognaths are likewise well represented, most
of them being marine. Of Acanthopterygian families the first to
appear are the Berycidæ, represented by several very distinct
genera: Beryx; Pseudoberyx with abdominal ventral fins; Berycopsis
with cycloid scales; Homonotus, Stenostoma, Sphenocephalus,
Acanus, Hoplopteryx, Platycornus with granular scales; Podocys
with a dorsal extending to the neck; Acrogaster, Macrolepis,
Rhacolepis from the chalk of Brazil. The position of Pycnosterynx is
uncertain, it approaches certain Pharyngognaths. True Percidæ are
absent, whilst the Carangidæ, Sphyrænidæ, Cataphracti, Gobiidæ,
Cottidæ, and Sparidæ are represented by one or more genera.
Somewhat less diversified are the Physostomes, which belong
principally to the Clupeidæ and Dercetidæ, most of the genera being
extinct; Clupea is abundant in some localities. Scopelidæ
(Hemisaurida and Saurocephalus) occur in the chalk of Comen in
Istria, and of Mæstricht. Of all cretaceous deposits none surpass
those of the Lebanon with regard to the number of genera, species,
and individuals; the forms are exclusively marine, and the remains in
the most perfect condition.
 In the Tertiary epoch the Teleosteans have almost entirely
replaced the Ganoids; a few species only of the latter make their
appearance, and they belong to existing genera, or, at least, very
closely allied forms (Lepidosteus, Amia, Hypamia, Acipenser). The
Chondropterygians merge more and more into recent forms;
Holocephali continue, and still are better represented than in the
present fauna. The Teleosteans show even in the Eocene a large
proportion of existing genera, and the fauna of some localities of the
Miocene (Oeningen) is almost wholly composed of them. On the
whole, hitherto more than one-half have been found to belong to
existing genera, and there is no doubt that the number of seemingly
distinct extinct genera will be lessened as the fossils will be
examined with a better knowledge of the living forms. The
distribution of the fishes differed widely from that of our period, many
of our tropical genera occurring in localities which are now included
within our temperate zone, and being mixed with others, which
nowadays are restricted to a colder climate: a mixture which
continues throughout the Pliocene.

A few families of fishes, like the freshwater Salmonidæ, seem to

have put in their appearance in Post-pliocene times; however, not
much attention has been paid to fish-remains of these deposits; and
such as have been incidentally examined offer evidence of the fact
that the distribution of fishes has not undergone any further essential
change down to the present period.
[See E. Sauvage, Mémoire sur la Faune Ichthyologique de la
période Tertiaire. Paris 1873. 8°.]
Fig. 102.—Pycnodus rhombus, a Ganoid from the Upper
Oolite.
 CHAPTER XVI.

THE DISTRIBUTION OF EXISTING FISHES OVER THE EARTH’S SURFACE—GENERAL

REMARKS.

In an account of the geographical distribution of fishes the

Freshwater forms are to be kept separate from the Marine. However,
when we attempt to draw a line between these two kinds of fishes,
we meet with a great number of species and of facts which would
seem to render that distinction very vague. There are not only
species which can gradually accommodate themselves to a sojourn
in either salt or fresh water, but there are also such as seem to be
quite indifferent to a rapid change from one into the other: so that
individuals of one and the same species (Gastrosteus, Gobius,
Blennius, Osmerus, Retropinna, Clupea, Syngnathus, etc.), may be
found at some distance out at sea, whilst others live in rivers far
beyond the influence of the tide, or even in inland fresh waters
without outlet to the sea. The majority of these fishes belong to forms
of the fauna of the brackish water, and as they are not an
insignificant portion of the fauna of almost every coast, we shall have
to treat of them in a separate chapter.
Almost every large river offers instances of truly marine fishes
(such as Serranus, Sciænidæ, Pleuronectes, Clupeidæ, Tetrodon,
Carcharias, Trygonidæ), ascending for hundreds of miles of their
course; and not periodically, or from any apparent physiological
necessity, but sporadically throughout the year, just like the various
kinds of marine Porpoises which are found all along the lower course
of the Ganges, Yang-tseKiang, the Amazons, the Congo, etc. This is
evidently the commencement of a change in a fish’s habits, and,
indeed, not a few of such fishes have actually taken up their
permanent residence in fresh waters (as species of Ambassis,
Apogon Dules, Therapon, Sciæna, Blennius, Gobius, Atherina,
Mugil, Myxus, Hemirhamphus, Clupea, Anguilla, Tetrodon, Trygon):
all forms originally marine.
On the other hand, we find fishes belonging to freshwater genera
descending rivers and sojourning in the sea for a more or less limited
period; but these instances are much less in number than those in
which the reverse obtains. We may mention species of Salmo (the
Common Trout, the Northern Charr), and Siluroids (as Arius,
Plotosus). Coregonus, a genus so characteristic of the inland lakes
of Europe, Northern Asia, and North America, nevertheless offers
some instances of species wandering by the effluents into the sea,
and taking up their residence in salt water, apparently by preference,
as Coregonus oxyrhynchus. But of all the Freshwater families none
exhibit so great a capability of surviving the change from fresh into
salt water, as the Gastrosteidæ (Sticklebacks), of the northern
Hemisphere, and the equally diminutive Cyprinodontidæ of the
tropics; not only do they enter into, and live freely in, the sea, but
many species of the latter family inhabit inland waters, which, not
having an outlet, have become briny, or impregnated with a larger
proportion of salts than pure sea water. During the voyage of the
“Challenger” a species of Fundulus, F. nigrofasciatus, which inhabits
the fresh and brackish waters of the Atlantic States of North
America, was obtained, with Scopelids and other pelagic forms, in
the tow-net, midway between St. Thomas and Teneriffe.
Some fishes annually or periodically ascend rivers for the
purpose of spawning, passing the rest of the year in the sea, as
Sturgeons, many Salmonoids, some Clupeoids, Lampreys, etc. The
two former evidently belonged originally to the freshwater series, and
it was only in the course of their existence that they acquired the
habit of descending to the sea, perhaps because their freshwater
home did not offer a sufficient supply of food. These migrations of
freshwater fishes have been compared with the migrations of birds;
but they are much more limited in extent, and do not impart an
additional element to the fauna of the place to which they migrate, as
is the case with the distant countries to which birds migrate.
The distinction between freshwater and marine fishes is further
obscured by geological changes, in consequence of which the salt
water is gradually being changed into fresh, or vice versa. These
changes are so gradual and spread over so long a time, that many of
the fishes inhabiting such localities accommodate themselves to the
new conditions. One of the most remarkable and best studied
instances of such an alteration is the Baltic, which, during the second
half of the Glacial period, was in open and wide communication with
the Arctic Ocean, and evidently had the same marine fauna as the
White Sea. Since then, by the rising of the land of Northern
Scandinavia and Finland, this great gulf of the Arctic Ocean has
become an inland sea, with a narrow outlet into the North Sea, and
its water, in consequence of the excess of the fresh water pouring
into it over the loss by evaporation, has been so much diluted as to
be nearly fresh at its northern extremities: and yet nine species, the
origin of which from the Arctic Ocean can be proved, have survived
the changes, propagating their species, agreeing with their brethren
in the Arctic Ocean in every point, but remaining comparatively
smaller. On the other hand, fishes which we must regard as true
freshwater fishes, like the Rudd, Roach, Pike, Perch, enter freely the
brackish water of the Baltic.[17] Instances of marine fishes being
permanently retained in fresh water in consequence of geological
changes are well known: thus Cottus quadricornis in the large lakes
of Scandinavia; species of Gobius, Blennius, and Atherina in the
lakes of Northern Italy; Comephorus, of the depths of the Lake of
Baikal, which seems to be a dwarfed Gadoid. Carcharias gangeticus
in inland lakes of the Fiji Islands, is another instance of a marine fish
which has permanently established itself in fresh water.
In the miocene formation of Licata in Sicily, in which fish remains
abound, numerous Cyprinoids are mixed with littoral and pelagic
forms. Sauvage found in 450 specimens from that locality, not less
than 266, which were Leucisci, Alburni, or Rhodei. Now, although it
is quite possible that in consequence of a sudden catastrophe the
bodies of those Cyprinoids were carried by a freshwater current into,
and deposited on the bottom of, the sea, the surmise that they lived
together with the littoral fishes in the brackish water of a large
estuary, which was not rarely entered by pelagic forms, is equally
admissible. And, if confirmed by other similar observations, this
instance of a mixture of forms which are now strictly freshwater or
marine, may have an important bearing on the question to what
extent fishes have in time changed their original habitat.
Thus there is a constant exchange of species in progress
between the freshwater and marine faunæ, and in not a few cases it
would seem almost arbitrary to refer a genus or even larger group of
fishes to one or the other; yet there are certain groups of fishes
which entirely, or with but few exceptions are, and, apparently, during
the whole period of their existence have been, inhabitants either of
the sea or of fresh water; and as the agencies operating upon the
distribution of marine fishes differ greatly from those influencing the
dispersal of freshwater fishes, the two series must be treated
separately. The most obvious fact that dry land, which intervenes
between river systems, offers to the rapid spreading of a freshwater
fish an obstacle which can be surmounted only exceptionally or by a
most circuitous route, whilst marine fishes may readily and
voluntarily extend their original limits, could be illustrated by a great
number of instances. Without entering into details, it may suffice to
state as the general result, that no species or genus of freshwater
fishes has anything like the immense range of the corresponding
categories of marine fishes; and that, with the exception of the
Siluroids, no other freshwater family is so widely spread as the
families of marine fishes. Surface temperature or climate which is, if
not the most, one of the most important physical factors in the
limitation of freshwater fishes, similarly affects the distribution of
marine fishes, but in a less degree, and only those which live near to
the shore or the surface of the ocean; whilst it ceases to exercise its
influence in proportion to the depth, the true deep-sea forms being
entirely exempt from its operation. Light, which is pretty equally
distributed over the localities inhabited by freshwater fishes, cannot
be considered as an important factor in their distribution, but it
contributes towards constituting the impassable barrier between the
surface and abyssal forms of marine fishes. Altitude has stamped
the fishes of the various Alpine provinces of the globe with a certain
character, and limited their distribution; but the number of these
Alpine forms is comparatively small, ichthyic life being extinguished
at great elevations even before the mean temperature equals that of
the high latitudes of the Arctic region, in which some freshwater
fishes flourish. On the other hand, the depths of the ocean, far
exceeding the altitude of the highest mountains, still swarm with
forms specially adapted for abyssal life. That other physical
conditions of minor and local importance, under which fresh water
fishes live, and by which their dispersal is regulated, are more
complicated than similar ones of the ocean, is probable, though
perhaps less so than is generally supposed: for the fact is that the
former are more accessible to observation than the latter, and are,
therefore, more generally and more readily comprehended and
acknowledged. Thus, not only because many of the most
characteristic forms of the marine and freshwater series are found,
on taking a broader view of the subject, to be sufficiently distinct, but
also because their distribution depends on causes different in their
nature as well as the degree of their action, it will be necessary to
treat of the two series separately. Whether the oceanic areas
correspond in any way to the terrestrial will be seen in the sequel.

Fig. 103.—Ganoid scales of Tetragonolepis.