Professional Documents
Culture Documents
Series Editors
C.R. Rao
C.R. Rao AIMSCS, University of Hyderabad Campus,
Hyderabad, India
Geometry and
Statistics
Edited by
Frank Nielsen
Sony Computer Science Laboratories Inc.,
Tokyo, Japan
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
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
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
v
vi Contents
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
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
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 Universite, 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), Universite 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
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
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).
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.
∂ ∂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.
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
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).
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.
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.
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.
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.