Information Geometry and Population Genetics:

The Mathematical Structure of the Wright-Fisher

Model Julian Hofrichter pdf download

https://textbookfull.com/product/information-geometry-and-
population-genetics-the-mathematical-structure-of-the-wright-
fisher-model-julian-hofrichter/

Download more ebook instantly today - get yours now at textbookfull.com

We believe these products will be a great fit for you. Click
the link to download now, or visit textbookfull.com
to discover even more!

Population Biology of Plant Pathogens Genetics Ecology

and Evolution Michael G. Milgroom

https://textbookfull.com/product/population-biology-of-plant-
pathogens-genetics-ecology-and-evolution-michael-g-milgroom/

Extending The Linear Model With R Second Edition

Julian James Faraway

https://textbookfull.com/product/extending-the-linear-model-with-
r-second-edition-julian-james-faraway/

Population Health Monitoring Climbing the Information

Pyramid Marieke Verschuuren

https://textbookfull.com/product/population-health-monitoring-
climbing-the-information-pyramid-marieke-verschuuren/

Information Geometry and Its Applications Ay

https://textbookfull.com/product/information-geometry-and-its-
applications-ay/
Random Graphs Geometry and Asymptotic Structure 1st
Edition Michael Krivelevich

https://textbookfull.com/product/random-graphs-geometry-and-
asymptotic-structure-1st-edition-michael-krivelevich/

Model Perspectives Structure Architecture and Culture

Mark R. Cruvellier

https://textbookfull.com/product/model-perspectives-structure-
architecture-and-culture-mark-r-cruvellier/

Mathematical Foundation of Railroad Vehicle Systems

Geometry and Mechanics Ahmed A Shabana

https://textbookfull.com/product/mathematical-foundation-of-
railroad-vehicle-systems-geometry-and-mechanics-ahmed-a-shabana/

The Wrong Family: A Thriller Tarryn Fisher [Fisher

https://textbookfull.com/product/the-wrong-family-a-thriller-
tarryn-fisher-fisher/

Problem-solving And Selected Topics In Euclidean

Geometry: In The Spirit Of The Mathematical Olympiads
2013th Edition Sotirios E. Louridas

https://textbookfull.com/product/problem-solving-and-selected-
topics-in-euclidean-geometry-in-the-spirit-of-the-mathematical-
olympiads-2013th-edition-sotirios-e-louridas/
Julian Hofrichter • JRurgen Jost • Tat Dat Tran

Information Geometry
and Population Genetics
The Mathematical Structure
of the Wright-Fisher Model

123
Julian Hofrichter JRurgen Jost
Mathematik in den Naturwissenschaften Mathematik in den Naturwissenschaften
Max-Planck-Institut Max Planck Institut
Leipzig, Germany Leipzig, Germany

Tat Dat Tran

Mathematik in den Naturwissenschaften
Max Planck Institut
Leipzig, Germany

ISSN 1860-0832 ISSN 1860-0840 (electronic)

Understanding Complex Systems
ISBN 978-3-319-52044-5 ISBN 978-3-319-52045-2 (eBook)
DOI 10.1007/978-3-319-52045-2

Library of Congress Control Number: 2017932889

© Springer International Publishing AG 2017

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

Population genetics is concerned with the distribution of alleles, that is, variants at
a genetic locus, in a population and the dynamics of such a distribution across gen-
erations under the influences of genetic drift, mutations, selection, recombination
and other factors [57]. The Wright–Fisher model is the basic model of mathematical
population genetics. It was introduced and studied by Ronald Fisher, Sewall Wright,
Motoo Kimura and many other people. The basic idea is very simple. The alleles
in the next generation are drawn from those of the current generation by random
sampling with replacement. When this process is iterated across generations, then
by random drift, asymptotically, only a single allele will survive in the population.
Once this allele is fixed in the population, the dynamics becomes stationary. This
effect can be countered by mutations that might restore some of those alleles that
had disappeared. Or it can be enhanced by selection that might give one allele an
advantage over the others, that is, a higher chance of being drawn in the sampling
process. When the alleles are distributed over several loci, then in a sexually
recombining population, there may also exist systematic dependencies between the
allele distributions at different loci. It turns out that rescaling the model, that is,
letting the population size go to infinity and the time steps go to 0, leads to partial
differential equations, called the Kolmogorov forward (or Fokker–Planck) and the
Kolmogorov backward equation. These equations are well suited for investigating
the asymptotic dynamics of the process. This is what many people have investigated
before us and what we also study in this book.
So, what can we contribute to the subject? Well, in spite of its simplicity,
the model leads to a very rich and beautiful mathematical structure. We uncover
this structure in a systematic manner and apply it to the model. While many
mathematical tools, from stochastic analysis, combinatorics, and partial differential
equations, have been applied to the Wright–Fisher model, we bring in a geometric
perspective. More precisely, information geometry, the geometric approach to
parametric statistics pioneered by Amari and Chentsov (see, for instance, [4, 20]
and for a treatment that also addresses the mathematical problems for continuous
sample spaces [9]), studies the geometry of probability distributions. And as a
remarkable coincidence, here we meet Ronald Fisher again. The basic concept

v
vi Preface

of information geometry is the Fisher metric. That metric, formally introduced

by the statistician Rao [102], arose in the context of parametric statistics rather
than in population genetics, and in fact, it seems that Fisher himself did not see
this tight connection. Another fundamental concept of information geometry is the
Amari–Chentsov connection [3, 10]. As we shall argue in this book, this geometric
perspective yields a very natural and insightful approach to the Wright–Fisher
model, and with its help we can easily and systematically compute many quantities
of interest, like the expected times when alleles disappear from the population.
Also, information geometry is naturally linked to statistical mechanics, and this
will allow us to utilize powerful computational tools from the latter field, like the
free energy functional. Moreover, the geometric perspective is a global one, and it
allows us to connect the dynamics before and after allele loss events in a manner
that is more systematic than what has hitherto been carried out in the literature. The
decisive global quantities are the moments of the process, and with their help and
with sophisticated hierarchical schemes, we can construct global solutions of the
Kolmogorov forward and backward equations.
Let us thus summarize some of our contributions, in addition to providing a self-
contained and comprehensive analysis of the Wright–Fisher model.
• We provide a new set of computational tools for the basic quantities of interest
of the Wright–Fisher model, like fixation or coexistence probabilities of the
different alleles. These will be spelled out in detail for various cases of increasing
generality, starting from the 2-allele, 1-locus case without additional effects like
mutation or selection to cases involving more alleles, several loci and/or mutation
and selection.
• We develop a systematic geometric perspective which allows us to understand
results like the Ohta–Kimura formula or, more generally, the properties and
consequences of recombination, in conceptual terms.
• Free energy constructions will yield new insight into the asymptotic properties
of the process.
• Our hierarchical solutions will preserve overall probabilities and model the
phenomenon of allele loss during the process in more geometric and analytical
detail than previously available.
Clearly, the Wright–Fisher model is a gross simplification and idealization
of a much more complicated biological process. So, why do we consider it
then? There are, in fact, several reasons. Firstly, in spite of this idealization, it
allows us to develop some qualitative understanding of one of the fundamental
biological processes. Secondly, mathematical population genetics is a surprisingly
powerful tool both for classical genetics and modern molecular genetics. Thirdly,
as mathematicians, we are also interested in the underlying mathematical structure
for its own sake. In particular, we like to explore the connections to several other
mathematical disciplines.
As already mentioned, our book contains a self-contained mathematical analysis
of the Wright–Fisher model. It introduces mathematical concepts that are of interest
and relevance beyond this model. Our book therefore addresses mathematicians
Preface vii

and statistical physicists who want to see how concepts from geometry, partial
differential equations (Kolmogorov or Fokker–Planck equations) and statistical
mechanics (entropy, free energy) can be developed and applied to one of the most
important mathematical models in biology; bioinformaticians who want to acquire
a theoretical background in population genetics; and biologists who are not afraid
of abstract mathematical models and want to understand the formal structure of
population genetics.
Our book consists essentially of three parts. The first two chapters introduce
the basic Wright–Fisher model (random genetic drift) and its generalizations
(mutation, selection, recombination). The next few chapters introduce and explore
the geometry behind the model. We first introduce the basic concepts of information
geometry and then look at the Kolmogorov equations and their moments. The
geometric structure will provide us with a systematic perspective on recombination.
And we can utilize moment-generating and free energy functionals as powerful
computational tools. We also explore the large deviation theory of the Wright–
Fisher model. Finally, in the last part, we develop hierarchical schemes for the
construction of global solutions in Chaps. 8 and 9 and present various applications in
Chap. 10. Most of those applications are known from the literature, but our unifying
perspective lets us obtain them in a more transparent and systematic manner.
From a different perspective, the first four chapters contain general material, a
description of the Wright–Fisher model, an introduction to information geometry,
and the derivation of the Kolmogorov equations. The remaining five chapters
contain our investigation of the mathematical aspects of the Wright–Fisher model,
the geometry of recombination, the free energy functional of the model and its
properties, and hierarchical solutions of the Kolmogorov forward and backward
equations.
This book contains the results of the theses of the first [60] and the third
author [113] written at the Max Planck Institute for Mathematics in the Sciences
in Leipzig under the direction of the second author, as well as some subsequent
work. Following the established custom in the mathematical literature, the authors
are listed in the alphabetical order of their names. In the beginning, there will be
some overlap with the second author’s textbook Mathematical Methods in Biology
and Neurobiology [73]. Several of the findings presented in this book have been
published in [61–64, 114–118].
The research leading to these results has received funding from the European
Research Council under the European Union’s Seventh Framework Programme
(FP7/2007–2013)/ERC grant agreement no. 267087. The first and the third authors
have also been supported by the IMPRS “Mathematics in the Sciences”.
We would like to thank Nihat Ay for a number of inspiring and insightful
discussions.

Leipzig, Germany Julian Hofrichter

Leipzig, Germany Jürgen Jost
Leipzig, Germany Tat Dat Tran
Contents

1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 The Basic Setting .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.2 Mutation, Selection and Recombination . . . . . . . .. . . . . . . . . . . . . . . . . . . . 3
1.3 Literature on the Wright–Fisher Model . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
1.4 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 12
2 The Wright–Fisher Model .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
2.1 The Wright–Fisher Model . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
2.2 The Multinomial Distribution . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 19
2.3 The Basic Wright–Fisher Model . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 20
2.4 The Moran Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 23
2.5 Extensions of the Basic Model . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 24
2.6 The Case of Two Alleles . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 27
2.7 The Poisson Distribution .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 28
2.8 Probabilities in Population Genetics . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 29
2.8.1 The Fixation Time . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 29
2.8.2 The Fixation Probabilities .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 30
2.8.3 Probability of Having .k C 1/ Alleles (Coexistence) . . . . . 30
2.8.4 Heterozygosity .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 30
2.8.5 Loss of Heterozygosity .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 31
2.8.6 Rate of Loss of One Allele in a Population
Having .k C 1/ Alleles . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 31
2.8.7 Absorption Time of Having .k C 1/ Alleles . . . . . . . . . . . . . . 31
2.8.8 Probability Distribution at the Absorption
Time of Having .k C 1/ Alleles. . . . . . . .. . . . . . . . . . . . . . . . . . . . 31
2.8.9 Probability of a Particular Sequence of Extinction . . . . . . . 31
2.9 The Kolmogorov Equations.. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 32
2.10 Looking Forward and Backward in Time . . . . . . .. . . . . . . . . . . . . . . . . . . . 33
2.11 Notation and Preliminaries.. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 35
2.11.1 Notation for Random Variables . . . . . . . .. . . . . . . . . . . . . . . . . . . . 35
2.11.2 Moments and the Moment Generating Functions .. . . . . . . . 36

ix
x Contents

2.11.3 Notation for Simplices and Function Spaces .. . . . . . . . . . . . . 38

2.11.4 Notation for Cubes and Corresponding
Function Spaces . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
3 Geometric Structures and Information Geometry ... . . . . . . . . . . . . . . . . . . . 45
3.1 The Basic Setting .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 45
3.2 Tangent Vectors and Riemannian Metrics. . . . . . .. . . . . . . . . . . . . . . . . . . . 46
3.3 Differentials, Gradients, and the Laplace–Beltrami Operator .. . . . . 50
3.4 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51
3.5 The Fisher Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 56
3.6 Exponential Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 58
3.7 The Multinomial Distribution . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 64
3.8 The Fisher Metric as the Standard Metric on the Sphere .. . . . . . . . . . 66
3.9 The Geometry of the Probability Simplex . . . . . .. . . . . . . . . . . . . . . . . . . . 68
3.10 The Affine Laplacian .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 70
3.11 The Affine and the Beltrami Laplacian on the Sphere .. . . . . . . . . . . . . 73
3.12 The Wright–Fisher Model and Brownian Motion
on the Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 74
4 Continuous Approximations . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
4.1 The Diffusion Limit .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
4.1.1 Convergence of Discrete to Continuous
Semigroups in the Limit N ! 1 . . . . . .. . . . . . . . . . . . . . . . . . . . 77
4.2 The Diffusion Limit of the Wright–Fisher Model . . . . . . . . . . . . . . . . . . 88
4.3 Moment Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 91
4.4 Moment Duality .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 99
5 Recombination .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 103
5.1 Recombination and Linkage . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 103
5.2 Random Union of Gametes . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 105
5.3 Random Union of Zygotes .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 107
5.4 Diffusion Approximation . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 109
5.5 Compositionality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 110
5.6 The Geometry of Recombination .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 111
5.7 The Geometry of Linkage Equilibrium States . .. . . . . . . . . . . . . . . . . . . . 114
5.7.1 Linkage Equilibria in Two-Loci Multi-Allelic Models . . . 115
5.7.2 Linkage Equilibria in Three-Loci
Multi-Allelic Models . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 117
5.7.3 The General Case . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 120
6 Moment Generating and Free Energy Functionals .. . . . . . . . . . . . . . . . . . . . 123
6.1 Moment Generating Functions . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
6.1.1 Two Alleles . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 124
6.1.2 Two Alleles with Mutation .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 128
6.1.3 Two Alleles with Selection .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 130
6.1.4 n C 1 Alleles . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 132
Contents xi

6.1.5 n C 1 Alleles with Mutation . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 136

6.1.6 Exponential Families . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 138
6.2 The Free Energy Functional . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 139
6.2.1 General Definitions .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 139
6.2.2 The Free Energy of Wright–Fisher Models . . . . . . . . . . . . . . . 145
6.2.3 The Evolution of the Free Energy . . . . .. . . . . . . . . . . . . . . . . . . . 155
6.2.4 Curvature-Dimension Conditions and
Asymptotic Behavior .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 159
7 Large Deviation Theory.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 169
7.1 LDP for a Sequence of Measures on Different State Spaces .. . . . . . 169
7.2 LDP for a Sequence of Stochastic Processes . . .. . . . . . . . . . . . . . . . . . . . 171
7.2.1 Preliminaries .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 171
7.2.2 Basic Properties.. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 173
7.3 LDP for a Sequence of -Scaled Wright–Fisher Processes . . . . . . . . 175
7.3.1 -Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 175
7.3.2 Wentzell Theory for -Processes . . . . . .. . . . . . . . . . . . . . . . . . . . 177
7.3.3 Minimum of the Action Functional Sp;q ./ . . . . . . . . . . . . . . . . 180
8 The Forward Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 195
8.1 Eigenvalues and Eigenfunctions .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 196
8.2 A Local Solution for the Kolmogorov Forward Equation .. . . . . . . . . 202
8.3 Moments and the Weak Formulation
of the Kolmogorov Forward Equation.. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 203
8.4 The Hierarchical Solution.. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 205
8.5 The Boundary Flux and a Hierarchical Extension of Solutions .. . . 210
8.6 An Application of the Hierarchical Scheme . . . .. . . . . . . . . . . . . . . . . . . . 213
9 The Backward Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 219
9.1 Solution Schemes for the Kolmogorov Backward Equation . . . . . . . 220
9.2 Inclusion of the Boundary and the Extended
Kolmogorov Backward Equation .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 221
9.3 An Extension Scheme for Solutions of the Kolmogorov
Backward Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 222
9.4 Probabilistic Interpretation of the Extension Scheme . . . . . . . . . . . . . . 230
9.5 Iterated Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 231
9.6 Construction of General Solutions via the Extension Scheme . . . . . 236
9.7 A Regularising Blow-Up Scheme for Solutions
of the Extended Backward Equation . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 238
9.7.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 239
9.7.2 The Blow-Up Transformation and Its Iteration . . . . . . . . . . . 240
9.8 The Stationary Kolmogorov Backward Equation
and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 257
9.9 The Backward Equation and Exit Times . . . . . . . .. . . . . . . . . . . . . . . . . . . . 263
xii Contents

10 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 269
10.1 The Case of Two Alleles . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 269
10.1.1 The Absorption Time .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 269
10.1.2 Fixation Probabilities and Probability
of Coexistence of Two Alleles . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 272
10.1.3 The ˛th Moments .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 274
10.1.4 The Probability of Heterozygosity .. . . .. . . . . . . . . . . . . . . . . . . . 274
10.2 The Case of n C 1 Alleles . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 275
10.2.1 The Absorption Time for Having k C 1 Alleles. . . . . . . . . . . 275
10.2.2 The Probability Distribution of the Absorption
Time for Having k C 1 Alleles . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 282
10.2.3 The Probability of Having Exactly k C 1 Alleles . . . . . . . . . 283
10.2.4 The ˛th Moments.. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 284
10.2.5 The Probability of Heterozygosity .. . . .. . . . . . . . . . . . . . . . . . . . 284
10.2.6 The Rate of Loss of One Allele in a Population
Having k C 1 Alleles . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 285
10.3 Applications of the Hierarchical Solution .. . . . . .. . . . . . . . . . . . . . . . . . . . 285
10.3.1 The Rate of Loss of One Allele in a Population
Having Three Alleles . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 285

A Hypergeometric Functions and Their Generalizations.. . . . . . . . . . . . . . . . 289

A.1 Gegenbauer Polynomials . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 289
A.2 Jacobi Polynomials .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 290
A.3 Hypergeometric Functions .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 291
A.4 Appell’s Generalized Hypergeometric Functions .. . . . . . . . . . . . . . . . . . 292
A.5 Lauricella’s Generalized Hypergeometric Functions .. . . . . . . . . . . . . . 294
A.6 Biorthogonal Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 295

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 307

Index of Notation .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 313

Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 317
Chapter 1
Introduction

1.1 The Basic Setting

Population genetics is concerned with the stochastic dynamics of allele frequencies

in a population. In mathematical models, alleles are represented as alternative values
at genetic loci.
The notions of allele and locus are employed here in a rather abstract manner.
They thus cover several biological realizations. A locus may stand for a single
position in a genome, and the different possible alleles then are simply the four
nucleotides A; C; G; T. Or a locus can stand for the site of a gene—whatever that
is—in the DNA, and since such a gene is a string of nucleotides, say of length L,
there then are 4L different nucleotide combinations. Of course, not all of them will
be realized in a population, and typically there is a so-called wildtype or default
gene, together with some mutants in the population. The wildtype gene and its
mutants then represent the possible alleles.
It makes a difference whether we admit finitely many or infinitely many such
possible values. Of course, from the preceding discussion it is clear that in biological
situations, there are only finitely many, but in a mathematical model, we may also
consider the case of infinitely many possibilities. In the finite case, they are drawn
from a fixed reservoir, and hence, there is no possibility of genetic novelty in such
models when one assumes that all those alleles are already present in the initial
population. In the infinite case, or when there are more alleles than members of the
population, not all alleles can be simultaneously present in a finite population, and
therefore, through mutations, there may arise new values in some generation that
had not been present in the parental generation.
We consider here the finite case. The finitely many possible values then are
denoted by 0; : : : ; n. The simplest nontrivial case, n D 1, on one hand, already
shows most of the features of interest. On the other hand, the general structure of
the model becomes clearer when one considers arbitrary values of n.

© Springer International Publishing AG 2017 1

J. Hofrichter et al., Information Geometry and Population Genetics,
Understanding Complex Systems, DOI 10.1007/978-3-319-52045-2_1
2 1 Introduction

We consider a population of N diploid individuals, although for the most basic

model, the case of a population of 2N haploid individuals would lead to a formally
equivalent structure. (Here, “diploid” means that at each genetic locus, there are two
alleles, whereas in the “haploid” case, there is only one.)
We start with a single genetic locus. Thus, each individual in the population car-
ries two alleles at this locus, with values taken from 0; : : : ; n. Different individuals
in the population may have different values, and the relative frequency of the value
i in the population (at some given time) is denoted by pi . We shall also consider p as
a probability measure on SnC1 WD f0; : : : ; ng, that is,

X
n
pi D 1: (1.1.1)
iD0

The relationship between the deterministic concept of a frequency and the stochastic
concept of a probability of course requires some clarification, and this will be
addressed below, through the passage to a continuum limit.1
The population is evolving in time, and members pass on genes to their offspring,
and the allele frequencies pi then change in time through the mechanisms of
selection, mutation and recombination. In the simplest case, one has a population
with nonoverlapping generations. That means that we have a discrete time index t,
and for the transition from t to t C 1, the population Vt produces a new population
VtC1 . More precisely, members of Vt can give birth to offspring that inherit their
alleles. This process involves potential sources of randomness. Most basically, the
parents for each offspring are randomly chosen, and therefore, the transition from
the allele pool of one generation to that of the next defines a random process. In
particular, we shall see the effects of random genetic drift. Mutation means that
an allele may change to another value in the transition from parent to offspring.
Selection means that the chances of producing offspring vary depending on the value
of the allele in question, as some alleles may be fitter than others. Recombination
takes place in sexual reproduction, that is, when each member of the population has
two parents. It is then determined by chance which allele value she inherits when
the two parents possess different alleles at the locus in question. Depending on how
loci from the two parents are combined, this may introduce correlations between the
allele values at different loci.
Here is a remark which is perhaps obvious, but which illuminates how the
biological process is translated into a mathematical one. As already indicated, in
the simplest case we have a single genetic locus. In the diploid case, each individual
carries two alleles at this locus. These alleles could be different or identical, but
for the basic process of creating offspring, this is irrelevant. In the diploid case,
for each individual of the next generation, two parents are chosen from the current
generation, and the individual inherits one allele from each parent. That allele then is

1
In a certain sense, we shall sidestep the real issue, and in this text, we do not enter into the issue
of objective and subjective probabilities.
1.2 Mutation, Selection and Recombination 3

randomly chosen from the two that parent carries. The parents are chosen randomly
from the population, and we sample with replacement. That means that when a
parent has produced an offspring it is put back into the population so that it has
the chance to be chosen for the production of further offspring. To be precise,
we also allow for the possibility that one and the same parent is chosen twice for
the production of an individual offspring. In such a case, that offspring would not
have two different parents, but would get both its alleles from a single parent, and
according to the procedure, then even the same allele of that parent could be chosen
twice. (Of course, when the population size N becomes large—and eventually, we
shall let it tend to infinity—, the probability that this happens becomes exceedingly
small.) But then, formally, we can look at the population of 2N alleles instead of
that of N individuals. The rule for the process then simply says that the next allele
generation is produced by sampling with replacement from the current one. In other
words, instead of considering a diploid population with N members, we can look
at a haploid one with 2N participants. That is, for producing an allele in the next
generation, we randomly choose one parent in the current population of 2N alleles,
and that then will be the offspring allele. Thus, we have the process of sampling
with replacement in a population of size 2N. The situation changes, however, when
the individuals possess several loci, and the transmission of the alleles at different
loci may be correlated through restrictions on the possible recombinations. In that
case, we need to distinguish between gametes and zygotes, and the details of the
process will depend on whether we recombine gametes or zygotes, that is, whether
we perform recombination after or before sampling. This will be explained and
addressed in Chap. 5.
Since we want to adopt a stochastic model, in line with the conceptual structure
of evolutionary biology, the future frequencies become probabilities, that is, instead
of saying that a fraction of pi of the 2N alleles in the population has the value i, we
shall rather say that the probability of finding the allele i at the locus in question is
pi . While these probabilities express stochastic effects, they will then change in time
according to deterministic rules.
Although we start with a finite population with a discrete time dynamics,
subsequently, we shall pass to the limit of an infinite population. In order to
compensate for the growing size, we shall make the time steps shorter and pass to
continuous time. Obviously, we shall choose the scaling between population size
and time carefully, and we shall obtain a parabolic differential equation for the
deterministic dynamics of the probabilities in the continuum limit.

1.2 Mutation, Selection and Recombination

The formal models of population genetics make a number of assumptions. Many of

these assumptions are not biologically plausible, and for essentially any assumption
that we shall make, there exist biological counterexamples. However, the resulting
4 1 Introduction

gain of abstraction makes a mathematical analysis possible which in the end will
yield insights of biological value.
We consider a population Vt that is changing in discrete time t with nonover-
lapping generations, that is, the population VtC1 consists of the offspring of
the members of Vt . There is no spatial component here, that is, everything is
independent of the location of the members of the population. In particular, the
issue of migration does not arise in this model.
Moreover, we shall keep the population size constant from generation to
generation.
While we consider sexual reproduction, we only consider monoecious or, in a
different terminology, hermaphrodite populations, that is, they do not have separate
sexes, and so, any individual can pair with any other to produce offspring. We
also assume random mating, that is, individuals get paired at random to produce
offspring.
The reproduction process is formally described as follows. For each individual
in generation t C 1, we sample the generation t to choose its one or two parents. The
simplest case is to take sampling with replacement. This means that the number of
offspring an individual can foster is only limited by the size of the next generation.
If we took sampling without replacement, each individual could only produce one
offspring. This would not lead to a satisfactory model. Of course, one could limit
the maximal number of offspring of individuals, but we shall not pursue this option.
Each individual in the population is represented by its genotype . We assume
that the genetic loci of the different members of the population are in one-to-one
correspondence with each other. Thus, we have loci ˛ D 1; : : : ; k. In the haploid
case, at each locus, there can be one of n˛ C 1 possible alleles. Thus, a genotype is
of the form  D . 1 ; : : :  k /, where  ˛ 2 f0; 1; : : : ; n˛ g. In the diploid case, at each
locus, there are two alleles, which could be the same or different. We are interested
in the distribution of genotypes  in the population and how that distribution changes
over time through the effects of mutation, selection, and recombination.
The trivial case is that each member of Vt by itself, that is, without recombination,
produces one offspring that is identical to itself. In that case, nothing changes in
time. This baseline situation can then be varied in three respects:
1. The offspring is not necessarily identical to the parent (mutation).
2. The number of offspring an individual produces or may be expected to produce
varies with that individual’s genotype (selection).
3. Each individual has two parents, and its genotype is assembled from the
genotypes of its parents (sexual recombination).
Item 2 leads to a naive concept of fitness as the realized or the expected number
of offspring. Fitness is a difficult concept; in particular, it is not clear what the unit
of fitness is, whether it is the allele or the genotype or the ancestor of a lineage, or in
groups of interacting individuals even some higher order unit (see for instance the
analysis and discussion in [70]). Item 3 has two aspects:
1.2 Mutation, Selection and Recombination 5

(a) Each allele is taken from one of the parents in the haploid case. In the diploid
case, each parent produces gametes, which means that she chooses one of her
two alleles at the locus in question and gives it to the offspring. Of course,
this choice is made for each offspring, so that different descendents can carry
different alleles.
(b) Since each individual has many loci that are linearly arranged on chromosomes,
alleles at neighboring loci are in general not passed on independently.
The purpose of the model is to understand how the three mechanisms of mutation,
selection and recombination change the distribution of genotypes in the population
over time. In the present treatise, item 3, that is, recombination, will be discussed in
more detail than the other two.
These three mechanisms are assumed to be independent of each other. For
instance, the mutation rates do not favour fitter alleles.
For the purpose of the model, a population is considered as a distribution
of genotypes. Probability distributions then describe the composition of future
populations. More precisely, pt ./ is the probability that an individual in generation
t carries the genotype . The model should then express the dynamics of the
probability distribution pt in time t.
For mutations, we consider a matrix M D .m / where ;  range over the
possible genotypes and m is the probability that genotype  mutates to genotype .
In the most basic version, the mutation probability m depends only on the number
d.; / (d standing for distance, of course) of loci at which  and  carry different
alleles. Thus, in this basic version, we assume that a mutation occurs at each locus
with a uniform rate m, independently of the particular allele found at that locus.
Thus, when the allele i at the locus ˛ mutates, it can turn into any of the n˛ other
alleles that could occur at that locus. Again, we assume that the probabilities are
equal, and so, it then mutates with probability nm˛ into the allele j ¤ i. In the simplest
case, there are only n C 1 D 2 alleles possible at each locus. In this case,

m D md.;/ .1  m/kd.;/ : (1.2.1)

When the number n C 1 of alleles is arbitrary, but still the same at each site, we
have instead
 m d.;/  m kd.;/
m D 1 : (1.2.2)
n n
In contrast to mutation, recombination is a binary operation, that is, an operation
that takes two parent genotypes ;  as arguments to produce one offspring genotype
. Here, a genotype consists of a linear sequence of k sites occupied by particular
alleles. We consider the case of monoecious individuals with haploid genotypes for
the moment. An offspring is then formed through recombination by choosing at
each locus the allele that one of the parents carries there. When the two parents
carry different alleles at the locus in question, we have to decide by a selection rule
which one to choose. This selection rule is represented by a mask , a binary string
6 1 Introduction

of length k. An entry 1 at position ˛ means that the allele is taken from the first
parent, say , and a 0 signifies that the allele is taken from the second parent, say .
Each genotype is simply described by a string of length k, and for k D 6, the mask
100100 produces from the parents  D 1 : : : 6 and  D 1 : : : 6 the offspring
 D 1 2 3 4 5 6 . The recombination operator
X
R D pr ./C ./ (1.2.3)


is then expressed in terms of the recombination schemes C ./ for the masks 
and the probabilities pr ./ for those masks. In the simplest case, all the possible 2k
masks are equally probable, and consequently, at each locus, the offspring obtains
an allele from either parent with probability 1=2, independently of the choices at the
other loci. Thus, this case reduces to the consideration of k independent loci.
Dependencies between sites arise in the so-called cross-over models (see for
example [11]). Here, the linear arrangement of the sites is important. Only masks
of the form c D 11 : : : 100 : : : 0 are permitted. For such a mask, at the first a.c /
sites, the allele from the first parent is chosen, and at the remaining k  a.c / sites,
the one from the second parent. As a can range from 0 to k, we then have k C 1
possible such masks c , and we may wish to assume again that each of those is
equally probable.
In the diploid case, each individual carries two alleles at each locus, one from
each parent. We think of this as two strings of alleles. It is then randomly decided
which of the two strings of each parent is given to any particular offspring.
Therefore, formally, the scheme can be reduced to the haploid case with suitable
masks, but as we shall discuss in Chap. 5, there will arise a further distinction, that
between gametes and zygotes.
With recombination alone, some alleles may disappear from the populations,
and in fact, as we shall study in detail below, with probability 1, in the long
term, only one allele will survive at each site. This is due to random genetic drift,
that is, because the parents that produce offspring are randomly selected from the
population. Thus, it may happen that no carrier of a particular allele is chosen at
a given time or that none of the chosen recombination masks preserves that allele
when the mating partner carries a different allele at the locus under consideration.
That would then lead to the ultimate extinction of that allele. However, when
mutations may occur, an allele that is not present in the population at time t may
reappear at some later time. Of course, mutation might also produce new alleles that
have not been present in the population before, and this is a main driver of biological
evolution.
For these introductory purposes, we do not discuss the order in which the
mutation and recombination operators should be applied. In fact, in most models
this is irrelevant.
Finally, we include selection. This means that we shall modify the assumptions
that individuals in generation t are randomly selected with equal probabilities as
parents of individuals in generation t C 1. Formally, this means that we need to
1.2 Mutation, Selection and Recombination 7

change the sampling rule for the parents of the next generation. The sampling
probability for an individual to become a parent for the next generation should
now depend on its fitness, that is, on its genotype, according to the naive fitness
notion employed here. Thus, there is a probability distribution ps ./ on the space of
genotypes . Again, the simplest assumption is that in the haploid case, each allele
at each locus has a fitness value, independently of which other alleles are present
at other loci. In the diploid case, each pair of alleles at a locus would have a fitness
value, again independently of the situation at other loci. Of course, in general one
should consider fitness functions depending in a less trivial manner on the genotype.
Also, in general, the fitness of an individual will depend on the composition of the
population, but we shall not address this important aspect here.
The preceding was needed to the set the stage. However, everything said so far
is fairly standard and can be found in the introduction of any book on mathematical
population genetics. We shall now turn to the mathematical structures underlying the
processes of allele dynamics. Here, we shall develop a more abstract mathematical
framework than utilized before in population genetics.
Let us first outline our strategy. Since we want to study dynamics of probability
distributions, we shall first study the geometry of the space of probability distribu-
tions, in order to gain a geometric description and interpretation of our dynamics.
For the dynamics itself, it will be expedient to turn to a continuum limit by suitably
rescaling population size 2N and generation time ıt in such a way that 2N ! 1,
but 2Nıt D 1. This will lead to Kolmogorov type backward and forward partial
differential equations for the probability distributions. This means that in the limit,
the probability density f . p; s; x; t/ WD @x1@@xn P.X.t/  xjX.s/ D p/ with s < t will
n

satisfy the Kolmogorov forward or Fokker–Planck equation

1 X @2  i i  X @  i 
n n
@
f . p; s; x; t/ D x .ı  x j
/f . p; s; x; t/  b .x; t/f . p; s; x; t/ ;
@t 2 i;jD1 @x @x
i j j
iD1
@x i

(1.2.4)

and the Kolmogorov backward equation

1X i i X
n n
@ @2 @
 f . p; s; x; t/ D p .ıj  p j / i j f . p; s; x; t/ C bi . p; s/ i f . p; s; x; t/
@s 2 i;jD1 @p @p iD1
@p
(1.2.5)

where the second order terms arise from random genetic drift, which therefore is
seen as the most important mechanism, whereas the first order terms with their
coefficients bi incorporate the effects of the other evolutionary forces.
Again, this is standard in the population genetics literature since its original
introduction by Wright and its systematic investigation by Kimura. We shall develop
a geometric framework that will interpret the coefficients of the second order terms
as the inverse of the Fisher metric of mathematical statistics. Among other things,
8 1 Introduction

this will enable us to find explicit solutions of these equations which, importantly,
are valid across loss of allele events. In particular, we can then determine all
quantities of interest, like the expected extinction times of alleles in the population,
in a more general and systematic manner than so far known in the literature.

1.3 Literature on the Wright–Fisher Model

In this section, we discuss some of the literature on the Wright–Fisher model. Our
treatment here is selective, for several reasons. First, there are simply too many
papers in order to list them all and discuss and compare their relevant contributions.
Second, we may have overlooked some papers. Third, our intention is to develop a
new and systematic approach for the Wright–Fisher model, based on the geometric
as opposed to the stochastic or analytical structure of the model. This approach
can unify many previous results and develop them from a general perspective, and
therefore, we did not delve so deeply into some of the different methods that have
been applied to the Wright–Fisher model since its inception.
Actually, there exist some monographs on population genetics with a systematic
mathematical treatment of the Wright–Fisher model that also contain extensive
bibliographies, in particular [15, 33, 39], and the reader will find there much useful
information that we do not repeat here.
But let us first recall the history of the Wright–Fisher model (as opposed to
other population genetics models, cf. for example [17, 18] for a branching process
model). The Wright–Fisher model was initially presented implicitly by Ronald
Fisher in [46] and explicitly by Sewall Wright in [125]—hence the name. A third
person with decisive contributions to the model was Motoo Kimura. In 1945,
Wright approximated the discrete process by a diffusion process that is continuous
in space and time (continuous process, for short) and that can be described by a
Fokker–Planck equation. By solving this Fokker–Planck equation derived from the
Wright–Fisher model, Kimura then obtained an exact solution for the Wright–Fisher
model in the case of two alleles in 1955 (see [79]). Shortly afterwards, Kimura [78]
produced an approximation for the solution of the Wright–Fisher model in the multi-
allele case, and in [80], he obtained an exact solution of this model for three alleles
and concluded that this can be generalized to arbitrarily many alleles. This yields
more information about the Wright–Fisher model as well as the corresponding
continuous process. We also mention the monograph [24] where Kimura’s theory
is systematically developed. Kimura’s solution, however, is not entirely satisfactory.
For one thing, it depends on very clever algebraic manipulations so that the general
mathematical structure is not very transparent, and this makes generalizations very
difficult. Also, Kimura’s approach is local in the sense that it does not naturally
incorporate the transitions resulting from the (irreversible) loss of one or more
alleles in the population. Therefore, for instance the integral of his probability
density function on its domain need not be equal to 1. Baxter et al. [14] developed
1.3 Literature on the Wright–Fisher Model 9

a scheme that is different from Kimura’s; it uses separation of variables and works
for an arbitrary number of alleles.
While the original model of Wright and Fisher works with a finite population in
discrete time, many mathematical insights into its behavior are derived from its dif-
fusion approximation that passes to the limit of an infinite population in continuous
time. As indicated, the potential of the diffusion approximation had been realized
already by Wright and, in particular, by Kimura. The diffusion approximation
also makes an application of the general theory of strongly-continuous semigroups
and Markov processes possible, and this then lead to a more systematic approach
(cf. [43, 119]). In this framework, the diffusion approximation for the multi-allele
Wright–Fisher model was derived by Ethier and Nagylaki [36–38], and a proof of
convergence of the Markov chain to the diffusion process can be found in [34, 56].
Mathematicians then derived existence and uniqueness results for solutions of the
diffusion equations from the theory of strongly continuous semigroups [34, 36, 77]
or martingale theory (see, for example [109, 110]). Here, however, we shall not
appeal to the general theory of stochastic processes in order to derive the diffusion
approximation, but rather proceed directly within our geometric framework.
As the diffusion operator of the diffusion approximation becomes degenerate
at the boundary, the analysis at the boundary becomes difficult, and this issue
is not addressed by the aforementioned results, but was dealt with by more
specialized approaches. An alternative to those methods and results some of which
we shall discuss shortly is the recent approach of Epstein and Mazzeo [29–31] that
systematically treats singular boundary behavior of the type arising in the Wright–
Fisher model with tools from the regularity theory of partial differential equations.
We shall also return to their work in a moment, but we first want to identify
the source of the difficulties. This is the possibility that alleles get lost from the
population by random drift, and as it turns out, this is ultimately inevitable, and as
time goes to infinity, in the basic model, in the absence of mutations or particular
balancing selective effects, this will happen almost surely. This is the key issue,
and the full structure of the Wright–Fisher model and its diffusion approximation
is only revealed when one can connect the dynamics before and after the loss of an
allele, or in analytic terms, if one can extend the process from the interior of the
probability simplex to all its boundary strata. In particular, this is needed to preserve
the normalization of the probability distribution. In geometric terms, we have an
evolution process on a probability simplex. The boundary strata of that simplex
correspond to the vanishing of some of the probabilities. In biological terms, when a
probability vanishes, the corresponding allele has disappeared from the population.
As long as there is more than one allele left, the probabilities continue to evolve.
Thus, we get not only a flow in the interior of the simplex, but also flows within all
the boundary strata. The key issue then is to connect these flows in an analytical,
geometric, or stochastic manner.
Before going into further details, however, we should point out that the diffusion
approximation leads to two different partial differential equations, the Kolmogorov
forward or Fokker–Planck equation on one hand and the Kolmogorov backward
equation on the other hand. While these two equations are connected by a duality
10 1 Introduction

relation, their analytical behavior is different, in particular at the boundary. The

Kolmogorov forward equation yields the future distribution of the alleles in a
population evolving from a current one. In contrast, the Kolmogorov backward
equation produces the probability distribution of ancestral states giving rise to a
current distribution. See for instance [94]; a geometric explanation of the analogous
situation in the discrete case is developed in Sect. 4.2 of [73].
The distribution produced by the Kolmogorov backward equation may involve
states with different numbers of alleles present. Their ancestral distributions,
however, do not interfere, regardless of the numbers of alleles they involve. Thus,
some superposition principle holds, and the Kolmogorov backward equation nicely
extends to the boundary. For the Kolmogorov forward equation, the situation is more
subtle. Here, the probability of some boundary state does not only depend on the
flow within the corresponding boundary stratum, but also on the distribution in the
interior, because at any time, there is some probability that an interior state loses
some allele and turns into a boundary state. Thus, there is a continuous flux into
the boundary strata from the interior. Therefore, the extension of the flow from the
interior to the boundary strata is different from the intrinsic flows in those strata,
and no superposition principle holds.
As we have already said, there are several solution schemes for the Kolmogorov
forward equation in the literature. For the Kolmogorov backward equation, the
situation is even better. The starting point of much of the literature was the
observation of Wright [126] that when one includes mutation, the degeneracy at
the boundary is removed. And when the probability of a mutation of allele i into
allele j depends only on the target j, then the backward process possesses a unique
stationary distribution, at least as long as those mutation rates are positive. This then
lead to explicit representation formulas for even more general diffusion processes,
in [25, 27, 35, 53, 54, 86, 105, 106, 112]; these, however, were rather of a local
nature, as they did not connect solutions in the interior and in boundary strata
of the domain. Finally, much useful information can be drawn from the moment
duality [68] between the Wright–Fisher model and the Kingman coalescent [81],
see for instance [26] and the literature cited there. The duality method transforms
the original stochastic process into another, simpler stochastic process. In particular,
one can thus connect the Wright–Fisher processes and its extension with ancestral
processes such as Kingman’s coalescent [81], the method of tracing lines of descent
back into the past and analyzing their merging patterns (for a brief introduction,
see also [73]; for an application to Wright–Fisher models cf. [88]). Some of these
formulas, in particular those of [35, 106] also pertain to the limit of vanishing
mutation rates. In [106], a superposition of the contributions from the various strata
was achieved whereas [35] could write down an explicit formula in terms of a
Dirichlet distribution. However, this Dirichlet distribution and the measure involved
both become singular when one approaches the boundary. In fact, Shimakura’s
formula is simply a decomposition into the various modes of the solutions of a
linear PDE, summed over all faces of the simplex; this illustrates the rather local
character of the solution scheme.
1.3 Literature on the Wright–Fisher Model 11

Some ideas from statistical mechanics are already contained in the free fitness
function introduced by Iwasa [67] as a consequence of H-theorems. Such ideas will
be developed here within the modern theory of free energy functionals. A different
approach from statistical mechanics which can also produce explicit formulae
involves master equations for probability distributions; they have been applied to
the Moran model [89] of population genetics in [65]. That model will be briefly
described in Sect. 2.4.
Large deviation theory has been systematically applied to the Wright–Fisher
model by Papangelou [96–100], although this is usually not mentioned in the
literature. In Chap. 7, we can build upon his work.
As already mentioned, the Kolmogorov equations of the Wright–Fisher model
are not accessible to standard stochastic theory, because of their boundary behavior.
In technical terms, the square root of the coefficients of the second order terms of
the operators is not Lipschitz continuous up to the boundary. As a consequence, in
particular the uniqueness of solutions to the above Kolmogorov backward equations
may not be derived from standard results.
In this situation, Epstein and Mazzeo [29–31] have developed PDE techniques to
tackle the issue of solving PDEs on a manifold with corners that degenerate at the
boundary with the same leading terms as the Kolmogorov backward equation (1.2.5)
for the Wright–Fisher model in the closure of the probability simplex in .n /1 D
n  .1; 0/. Such an analysis had been started by Feller [43] (and essentially also
[42]), who had considered equations of the form

@ @2 @
f .x; t/ D x 2 f .x; t/ C b f .x; t/ for x  0 (1.3.1)
@t @x @x
with b  0, that is, equations that have the same singularity at the boundary
x D 0 as the Fokker–Planck or Kolmogorov forward equation of the simplest
type of the Wright–Fisher model. Feller could compute the fundamental solution
for this problem and thereby analyze the local behavior near the boundary. In
particular, the case where b ! 0 is subtle; in biological terms, this corresponds
to the transition from a setting with mutation to one without, and without mutation,
the boundary becomes absorbing. For more recent work in this direction, see for
instance [21]. In any case, this approach which focusses on the precise local analysis
at the boundary and which only requires a particular type of asymptotics near the
boundary and can therefore apply general tools from analysis, should be contrasted
with Kimura’s who looked for global solutions in terms of expansions in terms of
eigenfunctions and which needs the precise algebraic structure of the equations.
Epstein and Mazzeo [29, 30] then take up the local approach and develop it much
further. A main achievement of their analysis is the identification of the appropriate
function spaces. These are anisotropic Schauder spaces. In [31], they develop a
different PDE approach and derive and apply a Moser type Harnack inequality,
that is, the probably most powerful general tool of PDE theory for studying the
regularity of solutions of partial differential equations. According to general results
in PDE theory, such a Harnack inequality follows when the underlying metric and
12 1 Introduction

measure structure satisfy a Poincaré inequality and a measure doubling property,

that is, the volume of a ball of radius 2r is controlled by a fixed constant times
the volume of the ball of radius r with the same center, for all (sufficiently small)
r > 0. Since in the case that we are interested in, that of the Wright–Fisher model,
we identify the underlying metric as the standard metric on the unit sphere, such
properties are natural in our case. Also, in our context, their anisotropic Schauder
k;
spaces CWF .n / would consist of k times continuously differentiable functions
whose kth derivatives are Hölder continuous with exponent  w.r.t. the Fisher metric
(a geometric concept to be explained below which is basic for our approach). In
terms of the Euclidean metric on the simplex, this means that a weaker Hölder
exponent (essentially 2 ) is required in the normal than in the tangential directions
at the boundary. Using this framework, they subsequently show that if the initial
k;
values are of class CWF .n /, then there exists a unique solution in that class. This
result is very satisfactory from the perspective of PDE theory (see e.g. [72]). Our
setting, however, is different, because the biological model forces us to consider
discontinuous boundary transitions. The same also applies to other works which
treat uniqueness issues in the context of degenerate PDEs, but are not adapted to the
very specific class of solutions at hand. This includes the extensive work by Feehan
[41] where—amongst other issues—the uniqueness of solutions of elliptic PDEs
whose differential operator degenerates along a certain portion of the boundary @0
of the domain is established: For a problem with a partial Dirichlet boundary
condition, i.e. boundary data are only given on @ n @0 , a so-called second-
order boundary condition is applied for the degenerate boundary area; this is that
a solution needs to be such that the leading terms of the differential operator
continuously vanishes towards @0 , while the solution itself is also of class C1
up to @0 . Within this framework, Feehan then shows that—under certain natural
conditions—degenerate operators satisfy a corresponding maximum principle for
the partial boundary condition, which assures the uniqueness of a solution. Again,
our situation is subtly different, as the degeneracy behaviour at the boundary is
stepwise, corresponding to the stratified boundary structure of the domain n , and
hence does not satisfy the requirements for Feehan’s scenario. Furthermore, in the
language of [41], the intersection of the regular and the degenerate boundary part
@@0 , would encompass a hierarchically iterated boundary-degeneracy structure,
which is beyond the scope of that work.
Finally, we should mention that the differential geometric approach to the
Wright–Fisher model was started by Antonelli–Strobeck [5]. This was further
developed by Akin [2].

1.4 Synopsis

We now briefly describe, in somewhat informal terms, our approach and results.
Again, we begin with the case of a single locus. As already indicated, we consider
the relative frequencies or probabilities p0 ; : : : ; pn on the set f0; 1; : : : ; ng of possible
1.4 Synopsis 13

alleles at our locus. This leads to the simplex

( )
n X
n
0 1
† WD . p ; p ; : : : ; p / W p  0 for all i;
n i
p D1
i

iD0

of probability distributions on a set of n C 1 elements. This means that when

n
p 2 † and we draw an allele according to the probability distribution p, we obtain
n
i with probability pi . The various faces of † then correspond to configurations
where some alleles have probability 0. Again, when we take the probabilities
as relative frequencies, this means that the corresponding alleles are not present
in the population. Concerning the oscillation between relative frequencies and
probabilities, the situation is simply that the relative frequencies of the alleles in
one generation determine the probabilities with which they are represented in the
next generation according to our sampling procedure. And in the most basic model,
we sample according to the multinomial distribution with replacement.
A fundamental observation is that there exists a natural Riemannian metric
n
on the probability simplex † . This metric is not the Euclidean metric of the
simplex, but rather the Fisher metric. Fisher here stands for the same person as
the originator of the Wright–Fisher model, but this metric did not emerge from his
work on population genetics, but rather from his work on parametric statistics, and
apparently, he himself did not realize that this metric is useful for the model. In
fact, the Fisher metric was developed not really by Fisher himself, but rather by the
statistician Rao [102]. The Fisher metric is a basic subject of the field of information
geometry that was created by Amari, Chentsov, and others. Information geometry,
that is, the theory of the geometry of probability distributions, deals with a geometric
structure that not only involves a Riemannian metric, but also two dually affine
structures which are generated by potential functions that generalize the entropy
and the free energy of statistical mechanics. We refer to the monographs [3, 10].
It will appear that the Fisher metric becomes singular on the boundary of
n
the probability simplex † . These singularities, however, are only apparent, and
they only indicate that from a geometric perspective, we have chosen the wrong
parametrization for the family of probability distributions on nC1 possible types. In
n
fact, as we shall see in Chap. 3, a better parametrization uses the positive sector SC of
the n-dimensional unit sphere. (This parametrization is obtained by p 7! q D . pi /2
i i

for a probability distribution . p0 ; p1 ; : : : ; pn / on the types 0; 1; : : : ; n.) With that

n
parametrization, the Fisher metric of † is nothing but the Euclidean metric on
S ,! R , which, of course, is regular on the boundary of SC
n nC1 n
.
More generally, the Fisher metric on a parametrized family of probability
distributions measures how sensitively the family depends on the parameter when
sampling from the underlying probability space. The higher that sensitivity, the
easier is the task of estimating that parameter. That is why the Fisher metric is
important for parametric statistics. For multinomial distributions, the Fisher metric
is simply the inverse of the covariance matrix. This indicates on one hand that
the Fisher metric is easy to determine, and on the other hand that it is naturally
14 1 Introduction

associated to our iterated sampling from the multinomial distribution. In fact, the
Kolmogorov equations can naturally be interpreted as diffusion equations w.r.t. the
Fisher metric. One should note, however, that the Kolmogorov equations are not
in divergence form, and therefore, they do not constitute the natural heat equation
for the Fisher metric, or in other words, they do not model Brownian motion
for the Fisher metric. They rather have to be interpreted in terms of the dually
affine connections of Amari and Chentsov that we mentioned earlier. From that
perspective, entropy functions emerge as potentials. In particular, this will provide
us with a beautiful geometric approach to the exit times of the process, that is,
the expected times of allele losses from the population. When considering so-
called exponential families (called Gibbs distributions in statistical mechanics),
information geometry also naturally connects with the basic quantities of statistical
mechanics. These are entropy and free energy. As is well known in statistical
mechanics, the free energy functional and its derivatives encode all the moments
of a process. We shall make systematic use of this powerful scheme, and also
indicate some connections to recent research in stochastic analysis. In Chap. 7, we
shall explore large deviation principles in the context of the Wright–Fisher model.
Moreover, the geometric structure behind the Kolmogorov equations will also guide
our analysis of the transitions between the different boundary strata of the simplex.
This will constitute our main technical achievement.
As discussed, the key is the degeneracy at the boundary of the Kolmogorov
equations. While from an analytical perspective, this presents a profound difficulty
for obtaining boundary regularity of the solutions of the equations, from a biological
or geometric perspective, this is very natural because it corresponds to the loss
of some alleles from the population in finite time by random drift. And from
a stochastic perspective, this has to happen almost surely. For the Kolmogorov
forward equation, in Chap. 8, we gain a global solution concept from the equations
for the moments of the process, which incorporate the dynamics on the entire
simplex, including all its boundary strata. This also involves the duality between
the Kolmogorov forward equation and the Kolmogorov backward equation. In
Chap. 9, we then develop a careful notion of hierarchically extended solutions of
the Kolmogorov backward equation, and we show their uniqueness both in the time
dependent and in the stationary case. The stationary case is described by an elliptic
equation whose solutions arise from the time dependent equation as time goes to
infinity.2 The stationary equation is important because, for instance, the expected
times of allele loss are solutions of an inhomogeneous stationary equation. From
our information geometric perspective, as already mentioned, we can interpret these
solutions most naturally in terms of entropies.

2
In fact, one might be inclined to say that time goes to minus infinity in the backward case, because
this corresponds to the infinite past. With this time convention, however, the Kolmogorov backward
equation is not parabolic. When we change the direction of time, it becomes parabolic, and we can
then speak of time going to infinity. This mathematically natural, although not compatible with the
biological interpretation.
1.4 Synopsis 15

In Chap. 10, we shall explore how the schemes developed in this book, namely
the moment equations and free energy schemes, information geometry, the expan-
sions of solutions of the Kolmogorov equations in terms of Gegenbauer polyno-
mials, will provide us with computational tools for deriving formulas for basic
quantities of interest in population genetics.
We mainly focus on the basic Wright–Fisher model in the absence of additional
effects like selection or mutation. Nevertheless, we shall describe, in line with
the standard literature, how this will modify the equations. Also, in Sect. 6.1, we
shall systematically apply the moment generating function and energy functional
method to those issues. The issue of recombination will be treated in more detail
in Chap. 5 because here our geometric approach on one hand leads to an important
simplification of Kimura’s original treatment and on the other hand also provides
general insight into the geometry of linkage equilibria.
Chapter 2
The Wright–Fisher Model

2.1 The Wright–Fisher Model

The Wright–Fisher model considers the effects of sampling for the distribution of
alleles across discrete generations. Although the model is usually formulated for
diploid populations, and some of the interesting effects occurring in generalizations
depend on that diploidy, the formal scheme emerges already for haploid populations.
In the basic version, with which we start here, there is a single genetic locus that
can be occupied by different alleles, that is, alternative variants of a gene.1 In the
haploid case, it is occupied by a single allele, whereas in the diploid case, there are
two alleles at the locus. Biologically, diploidy expresses the fact that one allele is
inherited from the mother and the other from the father. However, the distinction
between female and male individuals is irrelevant for the basic model. In biological
terminology, we thus consider monoecious (hermaphrodite) individuals. Inheritance
is then symmetric between the parents, without a distinction between fathers and
mothers. Consequently, it does not matter from which parent an allele is inherited,
and there will be no effective difference between the two alleles at a site, that is, their
order is not relevant. Even in the case of dioecious individuals, one might still make
the simplifying assumption that it does not matter whether an allele is inherited
from the mother or the father. While there do exist biological counterexamples, one
might argue that for mathematical population genetics, this could be considered as
a secondary or minor effect only. Nevertheless, it would not be overly difficult to
extend the theory presented here to also include such effects.
Generalizations will be discussed subsequently, and we start with the simplest
case. In particular, for the moment, we assume that there are no selective differences
between these alleles and no mutations. These assumptions will be relaxed later,
after we have understood the basic model.

1
Obviously, the term “gene” is used here in a way that abstracts from most biological details.

© Springer International Publishing AG 2017 17

J. Hofrichter et al., Information Geometry and Population Genetics,
Understanding Complex Systems, DOI 10.1007/978-3-319-52045-2_2
18 2 The Wright–Fisher Model

In order to have our conventions best adapted to the diploid case, we consider
a population of 2N alleles. In the haploid case, we are thus dealing with 2N
individuals, each carrying a single allele, whereas in the diploid case, we have N
individuals carrying two alleles each.
For each of these alleles, there are n C 1 possibilities. We begin with the simplest
case, n D 1, where we have two types of alleles A0 ; A1 . In the diploid case, an
individual can be a homozygote of type A0 A0 or A1 A1 or a heterozygote of type
A0 A1 or A1 A0 —but we do not care about the order of the alleles and therefore
identify the latter two types. The population reproduces in discrete time steps. In the
haploid case, each allele in generation m C 1 is randomly and independently chosen
from the allele population of generation m. In the diploid case, each individual in
generation m C 1 inherits one allele from each of its parents. When a parent is a
heterozygote, each allele is chosen with probability 1=2. Here, for each individual
in generation m C 1, randomly two parents in generation m are chosen. All the
choices are independent of each other. Thus, the alleles in generation m C 1 are
chosen by random sampling with replacement from the ones in generation m. In this
model, the two parents of any particular individual might be identical (that is, in
biological terminology, selfing is possible), but of course, the probability for that to
occur goes to zero like N1 when the population size increases. Also, each individual
in generation m may foster any number of offspring between 0 and N in generation
m C 1 and thereby contribute between 0 and 2N alleles.
In any case, the model is not concerned with the lineage of any particular
individual, but rather with the relative frequencies of the two alleles in each
generation. Even though the diploid case appears more complicated than the haploid
one, at this stage, the two are formally equivalent, because in either case the 2N
alleles present in generation m C 1 are randomly and independently sampled from
those in generation m. In fact, from a mathematical point of view, the individuals
play no role, and we are simply dealing with multinomial sampling in a population
of 2N alleles belonging to n C 1 different classes. The only reason at this stage to
talk about the diploid case is that that case will offer more interesting perspectives
for generalization below.
The quantity of interest therefore is the number2 Ym of alleles A0 in the population
at time m. This number then varies between 0 and 2N. The distribution of allele
numbers thus follows the binomial distribution. When n > 1, the principle remains
the same, but we need to work more generally with the multinomial distribution. We
shall now discuss the basic properties of that distribution.

2
The random variable Y will carry two different indices in the course of our text. Sometimes, the
index m is chosen to indicate the generation time, but at other occasions, we rather use the index
2N for the number of alleles in the population, that is, more shortly, (twice) the population size.
2.2 The Multinomial Distribution 19

2.2 The Multinomial Distribution

We consider the basic situation of probabilities p0 ; : : : ; pn on the set f0; 1; : : : ; ng.

That is, we consider the simplex
( )
X
n
0 1
†n WD . p ; p ; : : : ; p / W p  0 for all i;
n i
p D1
i

iD0

of probability distributions on a set of n C1 elements. When we consider an element

p 2 †n and draw one of those elements according to the probability distribution p,
we obtain the element i with probability pi .
For each time step of the Wright–Fisher model, we draw 2N times independently
from such a distribution p, to create the next generation of alleles from the current
i
one. Call the corresponding random variables Y2N , standing for the number of alleles
Ai drawn that way. We utilize the index 2N for the total number of alleles here as
subsequently we wish to consider the limit N ! 1. For simplicity, we shall write
i in place of Ai . When we draw once, we obtain a single element i, that is, Y1i D 1
j
and Y1 D 0 for all j ¤ i. Since that element had probability pi , we have

E.Y1i / D E..Y1i /2 / D pi for all i; and E.Y1i Y1 / D 0 for j ¤ i

j
(2.2.1)

and hence
j
Var.Y1i / D pi .1  pi /; Cov.Y1i Y1 / D pi p j for i ¤ j: (2.2.2)

When we draw 2N times independently from the same probability distribution p,

i
we consequently get for the corresponding random variables Y2N

j
i
E.Y2N / D 2Npi ; i
Var.Y2N / D 2Npi .1  pi /; i
Cov.Y2N Y2N / D 2Npi p j for i ¤ j:
(2.2.3)
By the same kind of reasoning, we also get
i ˛
E..Y2N / / D O.2N/ (2.2.4)

for all other moments (where ˛ is a multi-index with j˛j  3 whose convention will
be explained below in Sect. 2.11).
We also point out the following obvious lumping lemma.
Lemma 2.2.1 Consider a map

`W †n ! †m
0
.p ;:::;p / n
7 .q0 ; : : : ; qm /
!
P
with q j D iDij1 C1;:::;ij pi where i0 D 1; im D n; (2.2.5)
20 2 The Wright–Fisher Model

that is, we lump the alleles Aij1 C1 ; : : : ; Aij into the single super-allele Bj . Then the
j
random variable Z2N that records multinomial sampling from †m is given by

j
X
Z2N D i
Y2N : (2.2.6)
iDij1 C1;:::;ij
t
u

2.3 The Basic Wright–Fisher Model

For the Wright–Fisher model, we simply iterate this process across several genera-
tions. Thus, we introduce a discrete time m and let this time m now be the subscript
for Y instead of the 2N that we had employed so far to indicate the total number of
alleles present in the population. Instead of the absolute probabilities of multinomial
sampling, we now need to consider the transition probabilities.
That is, when we know what the allele distribution at time m is and when we
multinomially sample from that distribution, we want to know the probabilities
for the resulting distribution at time m C 1. We also not only want to know the
expectation values for the numbers of alleles—which remain constant in time—and
the variances and covariances—which grow in time in the sense that if we start at
time 0 and want to know the distribution at time m, the formulas in (2.2.3) acquire a
factor m—, but we are now interested in the entire distribution of allele frequencies.
We recall that we have n C 1 possible alleles A0 ; : : : ; An at a given locus, still in a
diploid population of fixed size N. There are therefore 2N alleles in the population
in any generation, so it is sufficient to focus on the number Ym D .Ym1 ; : : : ; Ymn / of
alleles A1 ; : : : ; An at generation time m. Assume that Y0 D 0 D .f10 ; : : : ; n0 g/
and that, as before, the alleles in generation m C 1 are derived by sampling with
replacement from the alleles of generation m. Thus, the transition probability is
given by the multinomial formula

Yn  i yi
.2N/Š 
P.YmC1 D yjYm D / D 0 ; (2.3.1)
.y /Š.y1 /Š : : : .yn /Š iD0 2N

where
( )
X
n
; y 2 Sn.2N/ 1
D  D . ; : : : ;  / W  2 f0; 1; : : : ; 2Ng;
n i
  2N
i

iD1

and

0 D 2N  jj D 2N  1  : : :  n I y0 D 2N  jyj D 2N  y1  : : :  yn :
2.3 The Basic Wright–Fisher Model 21

In particular, if  j D 0 for some j, then also

P.YmC1 D yjYm D / D 0 whenever y j ¤ 0; (2.3.2)

and

Y  i y
i
.2N/Š
P.YmC1 D yjYm D / D 0 ;
.y /Š.y1 /Š : : : .y j1 /Š.y jC1 /Š : : : .yn /Š 2N
i¤j
(2.3.3)
for y D 0. Thus, whenever allele j disappears from the population, we simply get
j

the same process with one fewer allele. Iteratively, we can let n alleles disappear so
that only one allele remains which will then live on forever.
Returning to the general case, we then also have the probability
X
P.YmC1 D yjY0 D / D P.YmC1 D yjYm D m / P.Ym D m jYm1 D m1 /
1 ;:::;m

: : : P.Y1 D 1 jY0 D /: (2.3.4)

This is the probability for finding the allele distribution y at generation m C 1

when the process started with the distribution  at generation 0. In order to go from
time 0 to time m C 1, we sum over all possibilities at intermediate times. This is also
called the Chapman–Kolmogorov equation.
In terms of this probability distribution, we can express moments as
X
E..YmC1 /˛ jY0 D / D y˛ P.YmC1 D yjY0 D /; (2.3.5)
y

assuming that the process started with the allele distribution  at time 0.
From (2.2.3), we have

E.YmC1 jYm D m / D m ; (2.3.6)

and iterating (2.3.6), we get

E.YmC1 jY0 D / D : (2.3.7)

In particular, by (2.3.6), the expected allele distribution at generation m C 1 equals

the allele distribution at generation m, and the iteration (2.3.7) then tells us that it
also equals the allele distribution at generation 0. Thus, the expected value does not
change from step to step. This, or more precisely (2.3.6), is also called the martingale
property.
Exploring the Variety of Random
Documents with Different Content
The text on this page is estimated to be only 29.10%
accurate

POPULATION of over one-fourth to 19.7 in 1931. Because

of the growth of population, years of increasing births have shown
smaller increases in crude birth rates, while years of decreasing
births have shown larger declines in crude birth rates. Thus, the
decline of about one-fourth in the birth rate from 1921 to 1931
resulted from a decrease of less than one-sixth in the number of
births. Crude birth rates are often misleading because they depend
upon the age and sex composition of the population. Most births
occur to women from 15-44 years of age, so that if one population
differs from another only in having a higher proportion of women in
these ages, it will have a correspondingly higher birth rate. Such
difficulties may be avoided by classifying births according to the age
of the mother, for example by dividing the number of births to
women aged 20-24 by the number of women of that age. The
results give births by age of women and are known as specific rates.
They may be calculated only for years near the census date, when
the number of women of each age is known, and for states which
register births by age of mother. To ascertain the recent trend, the
period 1918-1921, with the census of January 1, 1920, in the center,
can be taken as a starting point. It includes 1919, a year of low birth
rates due to the mobilization of 1918 and earlier, and also 1921, a
year of high birth rates following demobilization, so the average
should be fair. The most recent period that can be used is 1928-
1929, as the tabulation of 1930 births by age of mother has not yet
been published by the Bureau of the Census (September, 1932).
These births are divided by the number of women in each age
period on January 1, 1929, estimated by interpolation between the
1920 and 1930 censuses. Specific Birth Rates by Race and Nativity.
— Comparing the specific birth rates for 1928-1929 with those for
1918-1921, a marked downward trend is found, as is shown in
Figure 20. Native white, foreign born white, and Negro women
showed large declines in specific birth rates and in each group the
drop was greatest in the latter part of the childbearing period.
Among native white women the birth rate in the 15-19 age group
was almost unchanged, but at greater ages the decreases varied
from 11 percent at age 20-24 to 22 percent at ages 40-44. If the
native white birth rates at each age are weighted according to the
total number of women of that age in the 1930 census, and then
averaged, the standardized birth rate is obtained. This rate fell 13
percent during the period. Among foreign born white women, the
specific birth rate at age 15-19 fell over one-fourth, while at greater
ages the drop was about one-third. These declines average more
than twice as large as those of native whites. Negro women
maintained their standardized birth rate at a level nearer that of
1918-1921 than foreign born white women and native white women.
There was an increase in the Negro rate at age 15-19, the only F 41
1
The text on this page is estimated to be only 27.47%
accurate

RECENT SOCIAL TRENDS increase shown in Figure 20. At

age 20-24 the decline was smaller than that of native white women
but at greater ages it was larger. That declines in specific birth rates
were so much larger among foreign born white women than among
native whites or Negroes is probably due chiefly to the Americanizing
of the foreign born, a process that had little counterpart among
natives. Due to the smaller additions to the foreign group in 1920-
1930 because of the immigration restrictions, the immigrant women
in the 1930 population had spent more years in the United States
than those of the 1920 population. There had thus been Births per
NATIVE WHITE ,000 Women WOMEN01 ISO 1918-1921 1920-1929
60 Births per 1.000 women FOREIGN BORN WHITE NEGRO 200 r-
WOMEN « WOMEN 200 50 15-19 20-Z4 25-39 3O-34 35-39 4O-44
15-19 2O-Z4 25-29 3O-34 35-39 4O-44 Age of Women Age of
Women FIG. 20.— Birth rates by age, race, and nativity of women,
1918-1921 and 1928-1929.*
The text on this page is estimated to be only 29.11%
accurate

POPULATION The results indicate that from 1890 to 1910

the proportion of childless and one child families increased from 28.0
percent to 39.4 percent in the professional group, from 23.6 percent
to 39.4 percent in the business group, from 22.1 percent to 34.4
percent in the skilled labor group, from 16.8 percent to 31.2 percent
in the unskilled labor group, and from 17.8 percent to 20.7 percent
in the farm owner group.33 It is probable that the proportion of
small families has increased still more since 1910 and that the
tendency for small families to gain most rapidly in groups where they
were least numerous earlier has finally reached the farm owner
group.34 The omission from the last two census enumerations of the
questions bearing on this matter makes it impossible to bring the
Milbank study up to date. It is the opinion of the authors that the
increasing practice of contraception is the outstanding factor in the
decline in birth rates. The larger decreases in the rates in the older
groups are just what one would expect if the decline is due to
voluntary control. In the great masses of the laboring population
older married couples who already have all the children they can
care for will almost certainly be at greater pains to prevent additional
conceptions than younger couples who are still childless or have only
one or two children. But it would certainly be a mistake for us to
ignore other factors in reducing the birth rate which are considered
of importance by those who are well informed about them. Thus
there are many competent physicians who believe that abortion is
responsible for much of the decline of the birth rate. Another factor
is the increasing failure of the reproductive system to function
normally either because of disease or because of modern modes of
life. Much sterility, both complete and partial, is thought to arise
from disease (particularly venereal disease), from the nervous strain
of city life, from the sedentary habits of many city dwellers, or from
faults in diet attributable to the increasing distance between the
producer and consumer of food and more refined modes of
preparing it; in a word, from the general derangement of bodily
functions arising out of the changes incident to passing from an
agricultural to an industrial economy. Unfortunately, practically
nothing is known of the relative importance of these various factors;
hence it is inevitable that the social scientist, the physician and the
biologist, approaching the problem from different angles, should
hold opinions which are widely at variance. Regional Variations in
Birth Rates. — So far, the trends in specific birth rates have been
considered for the entire registration area of 1919. The degree of
change has varied considerably among the different states, as may
be seen from Figure 21, which presents the standardized birth 33
Notestein, Frank W., "The Decrease in Size of Families from 1890 to
1910," Quarterly Bulletin of the Milbank Memorial Fund, October,
1931, vol. IX, no. 4, pp. 181-188. 34 See data on size of family,
Chap. XIII. F 43 1
The text on this page is estimated to be only 27.60%
accurate

RECENT SOCIAL TRENDS rate (the weighted average of the

specific birth rates) for groups of states having similar specific rates.
In each group there was a decline in the standardized rate for whites
during the last decade and in two of the four groups the Negro rate
decreased. The native white standardized rate declined least in New
Hampshire and Vermont (4.7 percent), and most in Utah (19.3
percent) and five southern states (19.5 percent). There was
£221918-1921 1 1920-1929 NATIVE WHITE WOMEN FOREIGN
BORN WHITE WOMEN01 NEGRO WOMEN b BIRTH RATE BIRTH
RATE BIRTH RATE 150 50 NEW HAMPSHIRE, VERMONT
CONNECTICUT. MASSACHUSETTS NEW YORK PENNSYLVANIA OHIO,
INDIANA, MICHIGAN. WISCONSIN. MINNESOTA. KANSAS
MARYLAND, VIRGINIA. NORTH CAROLINA .SOUTH CAROLINA.
KENTUCKY FIG. 21. — Standardized birth rates by race and nativity
for groups of states, 1918-1921 and 1928-1929." 0 Includes
Mexicans, their births having been registered as white. 6 Negro rates
not shown for states having small Negro population. e Calculated for
the 1919 birth registration area (excluding Maine) according to
method discussed in text. a tendency for the decreases to be larger
in states having higher rates in 1918-1921 and smaller in states
having lower rates. The Pacific states were the outstanding
exception to this tendency, for their birth rate was barely half that of
Utah or five southern states, yet it decreased nearly as rapidly.
Declines in the standardized birth rates of foreign born white women
were much larger than those of native white women in every group
of states. New Hampshire and Vermont again showed the smallest
decline, [44]
The text on this page is estimated to be only 29.20%
accurate

POPULATION while the largest occurred in the Pacific states

where the 1918-1921 rates were lowest. With foreign born white
birth rates there was little if any tendency for the states with higher
standardized rates in 1918-1921 to show large decreases. The
largest decline in the standardized birth rate of Negro women took
place in the five southern states where the rate was highest in
19181921. (Figure 21.) This is typical of the bulk of the Negro
population, since southern Negroes still outnumber northern
Negroes by nearly three to one. Not too much weight should be
given to the increase in the two northern areas because the makeup
of their Negro population in 1918-1921 probably was unusual
because of the large northward migration. The relation of rural life to
the trend of specific birth rates should be noted. On the whole, the
agricultural states had higher rates in 1918-1921 and larger declines
since. This is what might be expected from the study of the Milbank
Memorial Fund, previously cited. It showed that the four groups
primarily urban (professional, business, and skilled and unskilled
labor) had somewhat lower birth rates than the rural group (farm
owners) in 1890 and suffered declines nearly twice as large from
1890 to 1910. The rural rate was thus considerably above the urban
in 1910, consequently it would be expected to have the largest
subsequent decrease. Assuming that contraception is the chief
means by which the decline of the birth rate has been effected, the
inference is inevitable that it was practiced first among the
professional and business classes, spread to the skilled labor and
unskilled labor groups and reached farmers and country dwellers
last. But after reaching them, the drop in rural birth rates was the
largest, rates of other groups having fallen previously. The more
rapid downward movement of native white specific birth rates than
of Negro rates during the past decade has cut heavily into the
differential in favor of native whites which formerly existed in most
parts of the United States. In 1918-1921 the standardized birth rate
of native white women was 8 percent higher than that of Negroes in
the five southern states (Figure 21), nearly 30 percent higher in the
six north central states, 21 percent higher in Pennsylvania and 4
percent lower in Massachusetts, Connecticut and New York. By
1928-1929, however, the white rate was 1 percent lower than the
Negro rate in Pennsylvania and the five southern states, 10 percent
lower in Massachusetts, Connecticut, and New York, and 11 percent
higher in the six north central states.35 36 This discussion is based
on the specific rates in Figure 21, which include registered births
only. Since it is possible that non-registered births amount to as
much as 5 to 10 percent of white births and 10 to 20 percent of
Negro births, the Negro birth rates in 19181921 may have been
nearer the native white rates than Figure 21 indicates, and even
above them in 1928-1929. [ 45 ]
The text on this page is estimated to be only 28.89%
accurate

RECENT SOCIAL TRENDS Here again, the more rapid

downward trends of native white than of Negro specific birth rates is
what would be anticipated if the practice of contraception is the
main immediate cause of fewer births. Regulation of the size of
families would be expected to start in the more educated urban
groups and to reach the less educated rural groups last of all. In
1930, 40 percent of the Negroes lived on farms compared with 25
percent of the native whites and this relatively larger group of
Negroes had much poorer educational facilities. For these reasons
they would be slower in learning about birth control and in practicing
it. But during some future period, perhaps not far distant, the drop
in the birth rate of rural Negroes should be greater than that of
native whites. Estimating Future Population Growth. — At various
places in this chapter references have been made to the probable
size and makeup of the population in future years. Figures on annual
births, deaths and immigration furnish a base for estimating the
immediate future, but if the probable trend over a longer interval is
desired a more complicated method must be followed; though of
course no mathematical formula can forecast population growth with
absolute accuracy, no matter how well it may describe growth in
past years.36 But even if the course of population growth cannot be
foretold exactly, it will be worth while to know what the population
will be according to certain assumptions as to immigration and
specific birth and death rates, assumptions that may seem
reasonable judging from trends during recent years. Estimates on
several different assumptions have been worked out by the Scripps
Foundation for Research in Population Problems, two of which will be
discussed here. These indicate up to 1980 the limits between which
the actual population probably will lie. Minimum and Maximum
Assumptions. — For the probable minimum it is assumed that
immigration of whites (excluding Mexicans) and Negroes will be
offset by emigration during 1930-1934 but that net immigration will
average 50,000 per year for whites and 800 for Negroes during
19351939 and 100,000 per year for whites and 800 for Negroes
thereafter. For the probable maximum an average annual net
immigration of 20,000 for whites and 800 for Negroes during 1930-
1934, 100,000 for whites and 1,600 for Negroes during 1935-1939,
and 200,000 for whites and 1,600 for Negroes thereafter is
assumed. For the probable minimum it is assumed that white births
(excluding Mexicans) will average 2,100,000 and Negro births
300,000 annually 36 An /-shaped curve has been fitted to the past
population of the United States by Pearl and Reed, and prolonged to
indicate the future population. To the authors, it seems probable that
these estimates will prove too high due to restricted immigration and
lowered birth rates. In fact, Pearl and Reed have shown that it may
be necessary to join together two /-shaped curves to fit the past
where conditions of growth have changed sufficiently in a country.
There is no way now of telling whether or how soon a second curve
may need to be joined to the present curve for the United States in
order to fit future growth. r 46 1
The text on this page is estimated to be only 27.38%
accurate

POPULATION during 1930-1934, which continues the

decline of recent years. (Figure 19.) This will mean that the average
specific birth rates in 1930-1934 will have declined from the 1925-
1929 level by 9.6 percent for native whites, 16.2 percent for foreign
born whites and 8.0 percent for Negroes. During subsequent years
the decrease in birth rates is assumed to continue at a declining rate
until a stationary condition is reached in 1970 at 67 percent of the
1930 level for native whites, 65 percent for foreign born whites, and
64 percent for Negroes. (See Figure 20 for 1930 rates.) KATIO OF
CHILDREN O-4 I.OOOI 900 800 700 CHILDREN TO WOMEN PER
I.OOO WOMEN 15-44 DEATH RATE 600 Wh tes'v" Negroes Death
RateWhites ^Death Rate-Negroes Logarithmic Sca/a 30 I65O I860
1670 I88O I89O I9OO I9IO I9ZO I93O I94O I95O I960 I97O I98O
FIG. 22. — Past and possible future trends, by race, of ratio of
children to women and of death rates, 1850-1980.° 0 Based on
expectation of life shown by selected Life Tables. Estimated for 1940
to 1980 according to probable minimum assumptions. The probable
maximum assumes 2,150,000 white births and 305,000 Negro births
annually during 1930-1934, with specific birth rates decreasing less
rapidly than minimum rates. In 1945 these rates would become
stationary at 88 percent of the 1930 level for native whites, 82
percent for foreign born whites, and 86 percent for Negroes. An idea
of how the trend of the probable minimum compares with that of
the last several decades may be had by examining Figure 22, which
is based on the ratio of children 0-4 to women 15-44. The decline of
this ratio has been less than that of the birth rate, because of the
lowering of infant [ 47 ]
The text on this page is estimated to be only 29.20%
accurate

RECENT SOCIAL TRENDS mortality. Nevertheless, it is the

only measure available for decades before 1910, since few states
then had birth registration. The infant mortality rate for whites
declined from 82.1 in 1920 to 60.2 in 1930 and for Negroes from
135.6 to 95.1. The probable minimum assumes a further decline for
whites (excluding Mexicans) to 55 in 1940 and to 52 in 1950 and for
Negroes to 85 and 80, with small decreases subsequently. The
probable maximum assumes a decline for whites to 50 in 1940 and
45 in 1950, and for Negroes to 75 and 65, with small decreases
subsequently. The expectation of life of whites was 56.4 years in
1919 according to the Foudray Life Tables and appears to have been
lengthened to almost 60 years in 1930. The probable minimum
assumes the expectation of life for whites (excluding Mexicans) will
gradually rise to 66 years in 1970 and then remain at that figure. For
Negroes, a weighted average of the expectation of life in southern
and northern states, as given in the Foudray Life Tables for 1919-
1920, was 45.3 years. An expectation of life of 47.6 years in 1930 is
assumed, rising to 54 years in 1970 and then remaining stationary.
The probable maximum assumes the expectation of life for whites
will gradually rise to 73 years in 1980 and for Negroes to 62 years.
How the probable minimum assumption continues past trends is
approximately indicated by Figure 22. The probable minimum
assumes that "other colored" (including Mexicans) will continue to
amount to 1.67 percent of the white and Negro races as they did in
1930. The probable maximum assumes that "other colored" will rise
to 2 percent of the white and Negro races in 1940, 2.5 percent in
1950, and finally to 4 percent in 1980. Taking as a starting point the
1930 census population by age periods, the deaths by age may be
calculated by means of the estimated specific death rates and the
births to women at each age by means of the estimated specific
birth rates. Subtracting the estimates of deaths and adding those of
births and net immigration gives the population by age periods one
year later. By repeating the process it is possible to calculate by age
periods the population which will result in any future year if the
trends assumed for birth rates, death rates and immigration are
actually realized. (See Figure 16 for age composition in 1950.)
Probable Population after 1930. — The maximum and minimum
assumptions above described indicate a population between
132,500,000 and 134,500,000 in 1940, between 140,500,000 and
148,500,000 in 1950 and between 145,000,000 and 190,000,000 in
1980. According to the minimum estimate, the population will reach
its greatest size (approximately 146,000,000) between 1965 and
1970 and will subsequently decline, while the maximum estimate
indicates increases up to the end of the century. It is believed by the
authors, however, that the actual population will be considerably
nearer the minimum than the maximum [ 48 ]
The text on this page is estimated to be only 29.19%
accurate

POPULATION figure, especially by 1980. The birth rate has

been declining in the United States since 1810, hence it seems more
likely that it will continue to decline until 1970 rather than become
stationary in 1945, as the maximum assumes. Even according to the
minimum assumption for birth rates in 1980, there will be about 195
births per hundred women who marry as compared with 280 in
1930. This would make families average nearly two children, which
is far different from having all families childless, the absolute
extreme to which the birth rate can decline. For this reason,
references to future population in preceding sections37 are based on
weighted average of the maximum and minimum. Equal weights are
used in 1940 but the minimum is given increasing weight up to 75
percent in 1980. Considering the probability that the 1950
population will be between 140,500,000 and 148,500,000, it should
be remembered that there is little chance of error in saying that
there will be about 96,000,000 survivors from the 1930 population in
1950. This number is obtained by applying death rates to the 1930
population and allowing for emigration. Death rates at ages over one
year have changed but little in recent years,38 while emigration has
averaged about 100,000 a year since 1920. The remainder of the
1950 population will be made up of persons born here or
immigrating after 1930. These two movements cannot be foretold
with as much accuracy as the number of deaths, but together they
will account for only about one-third of the total. Consequences of
Slower Population Growth. — The consequences upon our social and
economic life of the slower population growth which seems assured
for the future are likely to be many. In the past there has been a
widespread belief that a rapidly growing population was one of the
essential conditions of general progress. While rapid growth of
population undoubtedly has contributed to past progress, the
slowing up of growth in the future need not be accompanied by
gradual stagnation. As a slower growth in population affects a larger
and larger part of the nation, one of the most important
consequences is likely to be a revaluation of the importance of
growth. Changes should come to be appraised in other than
quantitative terms. It is impossible to foretell the direction or the
extent of the changes in mental outlook which will ensue; but it may
be hazarded that purely quantitative measurement will bulk less
large in a judgment of what constitutes progress and that the quality
of living will secure greater attention. An immediate and practical
influence of slower population growth will probably manifest itself in
efforts to adjust economic activity to such growth. In all likelihood
this adjustment will not be particularly 37 See Figures 1, 3, 16 and
accompanying discussion. 38 See Chap. XII. [ 49 ]
The text on this page is estimated to be only 29.35%
accurate

RECENT SOCIAL TRENDS difficult in most lines, once

business men are fully convinced that population growth will slacken
and are able to estimate with fair accuracy the population for five or
ten years in advance. That this change in attitude may not be easily
effected is indicated by the fact that a population of from
200,000,000 to 300,000,000 by the year 2000 is frequently assumed
by sales managers and executives. Just because the population in
1860 was eight times as large as in 1790, and in 1930 was four
times as large as in 1860, one is not justified in saying that in 2000
it will be twice as large as in 1930. Certain industries will face
difficult and extensive problems in adjusting to a slower population
growth; these will be the ones most affected by the probable future
trends in population. They include industries in which technical
improvements are rapidly increasing human efficiency, those in
which consumption per capita is relatively inelastic, those in which
productive capacity is already largely in excess of effective demand
and those in which capital (including land) is relatively durable,
nontransferable and has a high value per unit of product. Some
industries, of which agriculture is an example, will be handicapped
by a combination of several of these unfavorable factors. Farm
production has been over-expanded since the World War, efficiency
has increased rapidly, foods in general face an inelastic demand, and
the proportion of capital in land is high, as is also land value per unit
of product. Any policies for the utilization of farm land in the future
must give careful consideration to the probable growth of population
if they are to prevent the farm population from sinking to a low
economic level. There are other industries which seem directly
dependent upon population increase for their growth. These
industries will feel the effects of an approaching stationary
population in proportion to the degree they have a stable product or
have already reached the saturation point. The present radio may be
replaced by an improved model at any time, but the kitchen stove is
usually kept until worn out. The point is that some industries can
expect to expand only as population grows, even if purchasing
power grows considerably. On the other hand, there are many
industries, probably producing the majority of all industrial goods,
whose growth is largely independent of population increase. They
could sell their product in much greater quantity if the public had the
money to buy it. To such industries raising the per capita purchasing
power of the public will be a vastly greater concern as population
growth is retarded. Making better customers of the population at
large may require raising wages and salaries, which may temporarily
reduce profits to some extent. But there will be less need to use
profits for increasing plant capacity until the increased purchasing
power of the bulk of the consumers has offset slower population
growth. In [50]
The text on this page is estimated to be only 29.15%
accurate

POPULATION the future plant expansion should be based

upon probable increase in the purchasing power of the population
rather than upon the belief that population growth will soon overtake
any expansion which available capital makes possible. It may be
argued that even though the population of the United States is
growing slowly and may soon become practically stationary, industry
can continue to expand by increasing foreign trade. In the long run
it seems debatable if much relief can be found in this direction. In
the first place, population growth is slowing up in practically all of
the countries with which the United States trades on a large scale,
and will soon be stationary in many of them. Secondly, all other
industrial nations are competing more and more strenuously for such
trade. Finally, the tendency to raise tariff barriers, which still shows
no sign of abatement, handicaps international trade. The slower
growth of population is not the sole or even the chief factor in
rendering more serious the economic difficulties into which the
country has drifted. But it does seem to merit careful consideration
in future planning for the rationalization of social and economic life.
VI. POPULATION POLICY Early Encouragement of Growth. — It is
not difficult to show that, consciously or unconsciously, the United
States has had a population policy from a relatively early date. From
the time when settlement first took place most communities wanted
people, partly to increase the safety of life and property and partly
because of the effect on land values. Most of the individuals and
companies who received large grants of land, or were able to
purchase it cheaply from public authorities, made efforts to have
their lands occupied. It is well known, for example, that William
Penn made strenuous efforts to get settlers on his grant and that at
times his success was sufficient to incur the dislike of men in other
communities who felt that he was using unfair means to attract
people to his domains. Because of such interest the settlement of
the land was encouraged in a variety of ways. The land policy
provided free or cheap land in farm units to foreigners as well as
natives if they would settle and work it. Immigration policies
permitted easy entry, offered political asylum and allowed the
importation of slaves for a time. Political leaders spread the idea that
here the common man had opportunities never before open to him.
Immigrants came in great numbers and the surplus youth in the east
moved westward in a steady stream. By about 1890 the actual
settlement of the land was almost completed; but since the
industrial development of the country was also well under way by
that time, there was still need for immigrants. The steamship
companies and other [ 51 1
The text on this page is estimated to be only 29.02%
accurate

RECENT SOCIAL TRENDS interests which profited by

immigration saw to it that the advantages of coming to the United
States were well advertised. The policy of the "open door'* was a
huge success in peopling the land. Gradual Restriction of
Immigration. — Although public encouragement of immigration was
generally accepted, there have always been those who felt that the
"new" immigrants were inferior and that something should be done
to preserve the economic advantages of the country for the
descendants of early arrivals. Nevertheless, it was not until about
fifty years ago that active steps were taken to close the door to
"undesirable" groups other than criminals or those afflicted with
certain diseases. In 1882, partly as a consequence of racial troubles
in the west, the first Chinese Exclusion Act was passed. In 1885,
under pressure of organized labor, the Alien Contract Labor Law was
passed. This forbade the entrance of foreigners under contract to
individuals or firms and was intended to prevent employers from
breaking strikes and undercutting wages by using cheap labor
recruited by agents in foreign countries. These early acts clearly
indicate that, under certain circumstances, an increase of numbers
was not considered the highest good by all groups. Step by step
federal policy has thus gone from the one extreme of stimulating
immigration to the opposite extreme. The exhaustion of desirable
free land put an end to the public encouragement of agricultural
immigrants. More drastic is the present policy, which assigns to each
country in the Eastern Hemisphere an annual immigration quota.39
The various quota laws in effect since June 3, 1921, also represent
an effort to influence the makeup of the population to preserve the
composition attained before the arrival of the millions of eastern and
southern Europeans who came during the present century. The
quotas in effect since July 1, 1929, are based upon the proportions
of the population springing from the different national stocks, and
definitely favor northern and western Europeans. Quotas have not
yet been applied to Canada and Latin America but may be applied in
the future. The open door policy of the past is completely
abandoned; not only are numbers restricted but there is definite
selection as to kind. Supplementing the quota laws are various
administrative regulations. Those providing for the deportation of
aliens who have entered illegally have been vigorously enforced in
recent years, resulting in thousands of expulsions plus a large but
unknown number of departures caused by fear of deportation. A
more drastic influence in restricting immigration has been the
regulation in force since the latter part of 1930 under which a visa is
denied to a prospective immigrant if it is believed that he may
become a public charge. This practically excludes aliens without jobs
(unless wealthy), while the Alien Contract Labor Law excludes aliens
?9 NO immigration of Orientals ineligible for citizenship is allowed. [
52 ]
The text on this page is estimated to be only 29.10%
accurate

POPULATION with jobs. As a result, the stream of

immigration has been reversed, and since November, 1930, more
aliens have been leaving the United States than have been entering.
Changing Attitude Toward Large Families. — The same attitudes of
mind which counted unrestricted immigration as a good also
encouraged the raising of large families. Furthermore, large families
were of direct advantage to much of the population. Farmers with
several sons were assured of a steady labor supply with little or no
wage payment, while other workingmen had augmented family
incomes if several minor children were at work. Besides, having
numerous children was probably the most certain form of insuring
old age security in a pioneering community. The general attitude
toward birth control and large families which prevailed prior to 1870
may be illustrated by an incident which occurred in 1832. At that
time Charles Knowlton wrote a little pamphlet (published in New
York City) entitled Fruits of Philosophy; or the Private Companion of
Young Married People in which he advocated contraception and
described some of the methods by which it might be accomplished.
This was considered an offense against public morality, and
Knowlton was punished by fine and imprisonment. There is little to
record regarding the birth control movement between Knowlton's
time and 1873 when Congress passed the so-called Comstock Laws
"for the suppression of trade in, and circulation of obscene literature
and articles of immoral use," which in effect outlawed information
about practices and devices for preventing conceptions. The passage
of these laws would seem to be evidence that birth control was
becoming sufficiently common to attract the attention of those who
were opposed to it, on whatever grounds. In more recent years,
particularly since Margaret Sanger attempted to open a birth control
clinic in 1916, there has been considerable legal conflict between
those who believe that man has the right to control his family
numbers and those who believe that such control is harmful from a
moral, national or personal point of view. However, there is
increasing opposition to measures which interfere with the individual
control of the size of the family. Between the mild enforcement of
restrictive laws by public authorities and the general disregard of
them by individuals, these laws are of little consequence at present.
In this connection it should be noted that birth control legislation
undoubtedly has had some effect upon the sources of growth in the
population. Had no restrictions been placed on the spread of birth
control information and had clinics been permitted to function freely,
it is probable that birth control would have spread more evenly
through social classes and that the decline in the birth rate among
poorer and less educated people would have been more closely
comparable with the decline among the better educated and the well
to do. (See pages 42 and 43.) f 53 1
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

textbookfull.com