Variational Source Conditions

Quadratic Inverse Problems Sparsity

Promoting Regularization New Results
in Modern Theory of Inverse Problems
and an Application in Laser Optics Jens
Flemming
Visit to download the full and correct content document:
https://textbookfull.com/product/variational-source-conditions-quadratic-inverse-proble
ms-sparsity-promoting-regularization-new-results-in-modern-theory-of-inverse-proble
ms-and-an-application-in-laser-optics-jens-flemming/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Inverse Problems: Basics, Theory and Applications in

Geophysics 2nd Edition Mathias Richter

https://textbookfull.com/product/inverse-problems-basics-theory-
and-applications-in-geophysics-2nd-edition-mathias-richter/

An Introduction to the Mathematical Theory of Inverse

Problems 3rd Edition Andreas Kirsch

https://textbookfull.com/product/an-introduction-to-the-
mathematical-theory-of-inverse-problems-3rd-edition-andreas-
kirsch/

Geometry of the generalized geodesic flow and inverse

spectral problems Second Edition Petkov

https://textbookfull.com/product/geometry-of-the-generalized-
geodesic-flow-and-inverse-spectral-problems-second-edition-
petkov/

Stochastic Linear Quadratic Optimal Control Theory

Differential Games and Mean Field Problems Jingrui Sun

https://textbookfull.com/product/stochastic-linear-quadratic-
optimal-control-theory-differential-games-and-mean-field-
problems-jingrui-sun/
Inverse Abdominoplasty An Illustrated Guide 1st Edition
Kemal Tunc Tiryaki (Eds.)

https://textbookfull.com/product/inverse-abdominoplasty-an-
illustrated-guide-1st-edition-kemal-tunc-tiryaki-eds/

Java Coding Problems: Become an expert Java programmer

by solving over 200 brand-new, modern, real-world
problems, 2nd edition Anghel Leonard

https://textbookfull.com/product/java-coding-problems-become-an-
expert-java-programmer-by-solving-over-200-brand-new-modern-real-
world-problems-2nd-edition-anghel-leonard/

Java Coding Problems, 2nd Edition: Become an expert

Java programmer by solving over 200 brand-new, modern,
real-world problems Anghel Leonard

https://textbookfull.com/product/java-coding-problems-2nd-
edition-become-an-expert-java-programmer-by-solving-
over-200-brand-new-modern-real-world-problems-anghel-leonard/

Residual Stress, Thermomechanics & Infrared Imaging,

Hybrid Techniques and Inverse Problems, Volume 9:
Proceedings of the 2015 Annual Conference on
Experimental and Applied Mechanics 1st Edition Sven
Bossuyt
https://textbookfull.com/product/residual-stress-thermomechanics-
infrared-imaging-hybrid-techniques-and-inverse-problems-
volume-9-proceedings-of-the-2015-annual-conference-on-
experimental-and-applied-mechanics-1st-edition-sven/

Recurrent Sequences Key Results Applications and

Problems Dorin Andrica

https://textbookfull.com/product/recurrent-sequences-key-results-
applications-and-problems-dorin-andrica/
 Frontiers in Mathematics

Jens Flemming

Source
Variational
Conditions,
Quadratic
Inverse Problems,
Sparsity Promoting
Regularization
Frontiers in Mathematics

Advisory Editorial Board

Leonid Bunimovich (Georgia Institute of Technology, Atlanta)

William Y. C. Chen (Nankai University, Tianjin, China)
Benoît Perthame (Université Pierre et Marie Curie, Paris)
Laurent Saloff-Coste (Cornell University, Ithaca)
Igor Shparlinski (Macquarie University, New South Wales)
Wolfgang Sprößig (TU Bergakademie Freiberg)
Cédric Villani (Institut Henri Poincaré, Paris)

More information about this series at http://www.springer.com/series/5388

Jens Flemming

Variational Source Conditions,

Quadratic Inverse Problems,
Sparsity Promoting
Regularization
New Results in Modern Theory of Inverse
Problems and an Application in Laser Optics
Jens Flemming
Fakultät für Mathematik
Technische Universität Chemnitz
Chemnitz, Germany

ISSN 1660-8046 ISSN 1660-8054 (electronic)

Frontiers in Mathematics
ISBN 978-3-319-95263-5 ISBN 978-3-319-95264-2 (eBook)
https://doi.org/10.1007/978-3-319-95264-2

Library of Congress Control Number: 2018950663

Mathematics Subject Classification (2010): 65J22, 65J20, 47A52, 47J06

© Springer Nature Switzerland AG 2018

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.

This book is published under the imprint Birkhäuser, www.birkhauser-science.com by the registered company
Springer Nature Switzerland AG part of Springer Nature.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
This book is dedicated to Bernd Hofmann on the
occasion of his retirement.
Preface

This book grew out of the author’s habilitation thesis, which has been completed in January
2018. Parts II and III cover and slightly extend the material of the thesis. Part I, on the one
hand, provides an introduction to the other parts and, on the other hand, contains new
results on variational source conditions in the context of convergence rates theory for ill-
posed inverse problems.
The intention of writing this book was to demonstrate new and to some extent nonortho-
dox ideas for handling ill-posed inverse problems. This book is not a comprehensive
introduction to inverse problems. Instead, it focuses on few research topics and handles
them in depth.
The three topics of the book, variational source conditions, quadratic inverse problems,
and 1 -regularization, seem to be quite different. The first one is of great generality and
establishes the basis for several more concrete results in the book. The second one is
concerned with nonlinear mappings in a classical Hilbert space setting, whereas the third
deals with linear mappings in non-reflexive Banach spaces.
At the second sight, quadratic inverse problems and linear inverse problems with
sparsity context have similar structures and their handling shows several parallels.
Nevertheless, I decided to divide the book into three more or less independent parts and to
give hints on cross connections from time to time. The advantage of this decision is that
the reader may study the three parts in arbitrary order.
Finishing this book would not have been possible without constant support and advice
by Prof. Bernd Hofmann (TU Chemnitz). I thank him a lot for his efforts in several regards
during all the years I have been working in his research group. I also want to thank
my colleagues and coauthors, especially Steven Bürger and Daniel Gerth, for interesting
and fruitful discussions. Last but not least I have to express my thanks to the Faculty
of Mathematics at TU Chemnitz as a whole for the cordial and cooperative working
atmosphere.

Chemnitz, Germany Jens Flemming

May 2018

vii
Contents

Part I Variational Source Conditions

1 Inverse Problems, Ill-Posedness, Regularization .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 3
1.1 Setting . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 3
1.2 Ill-Posedness.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 4
1.2.1 Global Definitions by Hadamard and Nashed . . . . .. . . . . . . . . . . . . . . 4
1.2.2 Local Definitions by Hofmann and Ivanov .. . . . . . .. . . . . . . . . . . . . . . 5
1.2.3 Interrelations .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 8
1.2.4 Nashed’s Definition in Case of Uncomplemented Null
Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 10
1.3 Tikhonov Regularization .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 12
2 Variational Source Conditions Yield Convergence Rates . . . . . .. . . . . . . . . . . . . . . 15
2.1 Evolution of Variational Source Conditions . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 15
2.2 Convergence Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 17
3 Existence of Variational Source Conditions . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 21
3.1 Main Theorem .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 21
3.2 Special Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 25
3.2.1 Linear Equations in Hilbert Spaces .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 25
3.2.2 Bregman Distance in Banach Spaces . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 26
3.2.3 Vanishing Error Functional.. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 28

Part II Quadratic Inverse Problems

4 What Are Quadratic Inverse Problems? .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 31
4.1 Definition and Basic Properties .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 31
4.2 Examples . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 35
4.2.1 Autoconvolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 35
4.2.2 Kernel-Based Autoconvolution in Laser Optics. . .. . . . . . . . . . . . . . . 40
4.2.3 Schlieren Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 45
4.3 Local Versus Global Ill-Posedness . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 48

ix
x Contents

4.4 Geometric Properties of Quadratic Mappings’ Ranges.. . . .. . . . . . . . . . . . . . . 49

4.5 Literature on Quadratic Mappings.. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 52
5 Tikhonov Regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 55
6 Regularization by Decomposition.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 59
6.1 Quadratic Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 59
6.2 Decomposition of Quadratic Mappings . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 63
6.3 Inversion of Quadratic Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 67
6.4 A Regularization Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 75
6.5 Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 78
7 Variational Source Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 97
7.1 About Variational Source Conditions . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 97
7.2 Nonlinearity Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 98
7.3 Classical Source Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 99
7.4 Variational Source Conditions Are the Right Tool . . . . . . . . .. . . . . . . . . . . . . . . 100
7.5 Sparsity Yields Variational Source Conditions . . . . . . . . . . . . .. . . . . . . . . . . . . . . 101

Part III Sparsity Promoting Regularization

8 Aren’t All Questions Answered? .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 109
9 Sparsity and 1 -Regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 111
9.1 Sparse Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 111
9.2 1 -Regularization .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 113
9.3 Other Sparsity Promoting Regularization Methods.. . . . . . . .. . . . . . . . . . . . . . . 118
9.4 Examples . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 119
9.4.1 Denoising . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 119
9.4.2 Bidiagonal Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 119
9.4.3 Simple Integration and Haar Wavelets. . . . . . . . . . . . .. . . . . . . . . . . . . . . 120
9.4.4 Simple Integration and Fourier Basis . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 121
10 Ill-Posedness in the 1 -Setting .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 125
11 Convergence Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 129
11.1 Results in the Literature .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 129
11.2 Classical Techniques Do Not Work . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 131
11.3 Smooth Bases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 132
11.4 Non-smooth Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 138
11.5 Convergence Rates Without Source-Type Assumptions.. . .. . . . . . . . . . . . . . . 145
11.6 Convergence Rates Without Injectivity-Type Assumptions .. . . . . . . . . . . . . . 148
11.6.1 Distance to Norm Minimizing Solutions . . . . . . . . . .. . . . . . . . . . . . . . . 148
11.6.2 Sparse Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 152
11.6.3 Sparse Unique Norm Minimizing Solution . . . . . . .. . . . . . . . . . . . . . . 155
Contents xi

11.6.4 Non-sparse Solutions .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 158

11.6.5 Examples .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 160
A Topology, Functional Analysis, Convex Analysis. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 165
A.1 Topological Spaces and Nets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 165
A.2 Reflexivity, Weak and Weak* Topologies .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 166
A.3 Subdifferentials and Bregman Distances . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 168
B Verification of Assumption 11.13 for Example 11.18 . . . . . . . . . . .. . . . . . . . . . . . . . . 171
References .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 177
Index . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 181
 Part I
Variational Source Conditions
Inverse Problems, Ill-Posedness, Regularization
1

Abstract
We introduce the mathematical setting as well as basic notation used throughout the
book. Different notions of ill-posedness in the context of inverse problems are discussed
and the need for regularization leads us to Tikhonov-type methods and their behavior
in Banach spaces.

1.1 Setting

Let X and Y be Banach spaces over R or C and let F : X ⊇ D(F ) → Y be a mapping

between them with domain D(F ). We aim to solve equations

F (x) = y † , x ∈ D(F ), (1.1)

with exact and attainable data y † in Y . Solving such equations requires, in some sense,
inversion of F . Hence the term inverse problem.
The mathematical field of inverse problems is not concerned with Eq. (1.1) in general
but only with equations that are ill-posed. Loosely speaking, an equation is ill-posed if
the inversion process is very sensitive to perturbations in the right-hand side y † . Such
perturbations cannot be avoided in practice because y † represents some measured quantity
and measurements always are corrupted by noise. We provide and discuss different precise
definitions of ill-posedness in the next section.

© Springer Nature Switzerland AG 2018 3

J. Flemming, Variational Source Conditions, Quadratic Inverse Problems,
Sparsity Promoting Regularization, Frontiers in Mathematics,
https://doi.org/10.1007/978-3-319-95264-2_1
4 1 Inverse Problems, Ill-Posedness, Regularization

To analyze and overcome ill-posedness noise has to be taken into account. In other
words, the exact right-hand side y † is not available for the inversion process. Instead, we
only have some noisy measurement y δ at hand, which is assumed to belong to Y , too, and
to satisfy
 δ 
y − y †  ≤ δ (1.2)

with nonnegative noise level δ.

For later reference we list the following restrictions on our setting.

Assumption 1.1 We assume that

(i) equation (1.1) has a solution,

(ii) the domain D(F ) is weakly sequentially closed,
(iii) the mapping F is weakly sequentially continuous.

Items (ii) and (iii) are satisfied if and only if for each sequence (xn )n∈N in D(F ) and
each x in X we have

xn  x ⇒ x ∈ D(F ), F (xn )  F (x).

1.2 Ill-Posedness

1.2.1 Global Definitions by Hadamard and Nashed

The classical definition of ill-posedness was introduced by Hadamard.

Definition 1.2 The mapping F in Eq. (1.1) is well-posed in the sense of Hadamard if

(i) for each y † in Y there exists a solution,

(ii) for each fixed right-hand side y † there is at most one solution,
(iii) solutions depend continuously on the data.

Else F is ill-posed in the sense of Hadamard.

Items (i) and (ii) of the definition require that F is bijective and item (iii) says that the
inverse mapping has to be continuous with respect to the norm or some other topology. Due
to its restrictive nature Hadamard’s definition only plays a minor role in modern theory of
1.2 Ill-Posedness 5

inverse problems. Existence of solutions usually is formulated as an assumption, cf. item

(i) in Assumption 1.1, and uniqueness is not required because the developed theory will
cover the case of multiple solutions.
In [1] Nashed proposed a definition of ill-posedness for bounded linear mappings F
between Banach spaces X and Y with domain D(F ) = X.

Definition 1.3 Let F in Eq. (1.1) be linear and bounded. Then F is well-posed in the sense
of Nashed if the range of F is closed in Y and ill-posed in the sense of Nashed if the range
of F is not closed in Y .

Nashed’s definition does not consider existence and uniqueness of solutions, but
focusses on continuous (generalized) invertibility. If a generalized inverse exists, then it is
continuous if and only if F is well-posed in the sense of Nashed, see [2, Theorem 5.6(b)].
But one should be aware of the fact, that in general Banach spaces generalized inverses
are not always available, because the null space of F or the closure of the range may be
uncomplemented, see Proposition 1.10 and Sect. 1.2.4 below. An important example for
this situation is the setting used for analyzing 1 -regularization in Part III.
If F is injective, then the inverse F −1 : Y ⊇ R(F ) → X is continuous on R(F ) if and
only if R(F ) is closed. If X and Y are Hilbert spaces, then the Moore–Penrose inverse is
a generalized inverse which always exists. Thus, in Hilbert spaces well-posedness in the
sense of Nashed is equivalent to continuity of the Moore–Penrose inverse.
Nashed distinguished two types of ill-posedness in [1]. In Chap. 10 we have a closer
look at this distinction in the context of 1 -regularization.

1.2.2 Local Definitions by Hofmann and Ivanov

Hadamard’s and Nashed’s definitions of ill-posedness are of global nature. For nonlinear
mappings F properties may vary from point to point and ill-posedness has to be understood
in a local manner. Following the ideas in [3] we have to distinguish between local ill-
posedness at a point x in X and local ill-posedness at a point y in Y .
The aim of defining precisely what is meant by ill-posedness is to describe the following
situation mathematically: Given a sequence (yn )n∈N in R(F ) approximating the unknown
exact data y † in (1.1), a sequence (xn )n∈N of corresponding solutions to F (x) = yn ,
x ∈ D(F ), does not converge to a solution of (1.1). The difficulties are to choose concrete
types of approximation and convergence and to handle the case of multiple solutions.
One possibility for defining ill-posedness locally at a point of the domain D(F ) has
been suggested in [4] by Hofmann, see also [5].
6 1 Inverse Problems, Ill-Posedness, Regularization

Definition 1.4 The mapping F is locally well-posed in the sense of Hofmann at a point x0
in D(F ) if there is some positive ε such that for each sequence (xn )n∈N in Bε (x0 ) ∩ D(F )
the implication

F (xn ) → F (x0 ) ⇒ xn → x0

is true. Otherwise, F is locally ill-posed in the sense of Hofmann at the point x0 .

Local well-posedness in the sense of Hofmann implies that x0 has to be an isolated

solution to F (x) = F (x0 ), x ∈ D(F ). In this sense, local uniqueness is part of this type
of local well-posedness.
Ivanov introduced a similar concept in [6], but locally in Y . Thus, he gets around the
question of uniqueness. See also [3, Definition 1].

Definition 1.5 The mapping F is locally well-posed in the sense of Ivanov at a point y0
in R(F ) if for each sequence (yn )n∈N in R(F ) the implication

yn → y0 ⇒ sup inf x̃ − x → 0

−1
x̃∈F −1 (yn ) x∈F (y0 )

is true. Otherwise, F is locally ill-posed in the sense of Ivanov at the point y0 .

The set-to-set distance

sup inf x̃ − x

x∈M
x̃∈M̃

between two subsets M̃ and M of X used in the Definition 1.5 is not symmetric. It
expresses the maximum distance of elements in M̃ to the set M. Since we cannot control
which of possibly many approximate solutions is chosen by an inversion method, this type
of distance is the right choice.
The only drawback of Definition 1.5 is that norm convergence cannot be replaced easily
by other types of convergence to define ill-posedness with respect to the weak topology,
for example. The following proposition provides an equivalent reformulation which avoids
explicit use of norms. The proposition was already mentioned briefly in [3, Remark 1].

Proposition 1.6 The mapping F is well-posed in the sense of Ivanov at a point y0 in R(F )
if and only if for each sequence (yn )n∈N in R(F ) converging to y0 and for each sequence
(x̃n )n∈N of preimages x̃n from F −1 (yn ) there exists a sequence (xn )n∈N in F −1 (y0 ) with
x̃n − xn → 0.
1.2 Ill-Posedness 7

Proof Let F be well-posed in the sense of Ivanov at the point y0 and let (yn )n∈N be a
sequence in R(F ) converging to y0 . Given a sequence (x̃n )n∈N with x̃n ∈ F −1 (yn ) we
immediately see

inf x̃n − x → 0.
x∈F −1 (y0 )

Fixing ε, for each n we find xn in F −1 (y0 ) with

x̃n − xn  ≤ inf x̃n − x + ε.

x∈F −1 (y0 )

Thus, we obtain x̃n − xn  ≤ 2 ε for all sufficiently large n, which implies convergence
x̃n − xn → 0.
Now let y0 be in R(F ) and let (yn )n∈N be a sequence in R(F ) converging to y0 .
Further, assume that for each sequence (x̃n )n∈N of preimages x̃n from F −1 (yn ) there exists
a sequence (xn )n∈N in F −1 (y0 ) with x̃n − xn → 0. If there would be some positive fixed
ε with

sup inf x̃ − x > ε,

−1
x̃∈F −1 (yn ) x∈F (y0 )

we would find a sequence (x̃n )n∈N with

inf x̃n − x > ε

x∈F −1 (y0 )

for all n. Thus, there would be a sequence (xn )n∈N with

ε< inf x̃n − x ≤ x̃n − xn  → 0,

x∈F −1 (y0 )

which contradicts ε > 0. This shows

sup inf x̃ − x → 0.

−1
x̃∈F −1 (yn ) x∈F (y0 )

Remark 1.7 From Proposition 1.6 we easily see that the following condition is sufficient
for local well-posedness in the sense of Ivanov at y0 : Each sequence (xn )n∈N in D(F )
with F (xn ) → y0 contains a convergent subsequence and the limits of all convergent
subsequences are solutions corresponding to the right-hand side y0 .
8 1 Inverse Problems, Ill-Posedness, Regularization

Throughout this book ill-posedness is to be understood in the sense of Ivanov if not

otherwise stated.

1.2.3 Interrelations

The definitions of Hofmann and Ivanov are closely connected, but differ in two aspects.
On the one hand, Hofmann’s definition works in X and Ivanov’s definition works in Y . On
the other hand, and also as a consequence of the first difference, in Hofmann’s definition
well-posedness is restricted to isolated solutions whereas Ivanov’s definition works for
arbitrary solution sets.
Both views have their advantages. Hofmann’s definition allows for a deeper analysis
of ill-posedness phenomena. Due to its locality in X at each element of a set of isolated
solutions we can distinguish between well-posedness and ill-posedness. That is, for one
fixed data element at the same time there might exist solutions at which the mapping
is well-posed and solutions at which the mapping is ill-posed in the sense of Hofmann.
Analyzing an inverse problem with Hofmann’s definition allows to identify regions of
well-posedness and regions of ill-posedness. Thus, restricting the domain of the mapping
F with the help of Hofmann’s definition could make the inverse problem well-posed.
Ivanov’s definition does not allow for such a detailed analysis. But its advantage is that
it is closer to the issue of numerical instability. Given a data element, we want to know
whether a sequence of approximate solutions based on noisy data becomes arbitrarily close
to the set of exact solutions if the noise is reduced until it vanishes. This is exactly what
Ivanov’s definition expresses.
The interrelations between Hofmann’s definition and Ivanov’s definition are made
precise by the following two propositions. The first proposition is a slightly extended
version of [3, Proposition 2] and the second stems from oral communication with Bernd
Hofmann (Chemnitz).

Proposition 1.8 If the mapping F is locally well-posed in the sense of Ivanov at some
point y0 in R(F ), then F is locally well-posed in the sense of Hofmann at each isolated
solution corresponding to the data y0 .

Proof Let F be locally well-posed in the sense of Ivanov at y0 and let x0 be an isolated
solution to data y0 . Take a positive radius ε such that x0 is the only solution to data y0
in B2 ε (x0 ). For each sequence (x̃n )n∈N in Bε (x0 ) ∩ D(F ) and for the corresponding
sequence (yn )n∈N with yn := F (x̃n ) Proposition 1.6 yields a sequence (xn )n∈N in D(F )
with F (xn ) = y0 and x̃n −xn → 0. Since (x̃n )n∈N lies in Bε (x0 ) and x0 is the only solution
in B2 ε (x0 ), we obtain xn = x0 for all n. Consequently, x̃n → x0 , which proves local well-
posedness in the sense of Hofmann at x0 .
1.2 Ill-Posedness 9

Proposition 1.9 There exist mappings F and points x0 in D(F ) such that F is locally
well-posed in the sense of Hofmann at x0 but locally ill-posed in the sense of Ivanov at
F (x0 ).
2
Proof Choose X := R, Y := R and F (x) := 1+x x
4 with D(F ) = X. Then x0 := 0 is the
only solution to F (x) = 0, x ∈ X, and continuous invertibility of F near zero immediately
implies local well-posedness in the sense of Hofmann.
On the other hand, we may consider a sequence (yn )n∈N with elements yn := F (xn )
such that xn → ∞. Then yn → 0, but

sup inf x̃ − x ≥ inf xn − x = xn  → 0.

−1 x∈F −1 (0)
x̃∈F −1 (yn ) x∈F (0)

Thus, F is locally ill-posed in the sense of Ivanov at F (0).

Finally, we state the interrelation between Nashed’s definition and Ivanov’s definition.
The special case of Hilbert spaces, where each closed subspace is complemented, can be
found in [3, Proposition 1].

Proposition 1.10 Let F be a bounded linear operator with domain D(F ) = X between
the Banach spaces X and Y and let the null space N (F ) be complemented in X. Then F
is well-posed in the sense of Nashed if and only if F is locally well-posed in the sense of
Ivanov at every point of R(F ) and F is ill-posed in the sense of Nashed if and only if F is
locally ill-posed in the sense of Ivanov at every point of R(F ).

Proof Let N (F ) be complemented by U , that is, U is a closed linear subspace of X and

X = N (F )⊕U . One easily shows, that the restriction F |U of F to U is bijective between
U and R(F ). Thus, the inverse (F |U )−1 is a well-defined linear operator, which due to
R(F ) = R(F |U ) is bounded if and only if R(F ) is closed. We see that F is well-posed
in the sense of Nashed if and only if (F |U )−1 is bounded.
Let (F |U )−1 be bounded. To show local well-posedness in the sense of Ivanov at an
arbitrary point y0 in R(F ) we choose sequences (yn )n∈N with yn → y and (x̃n )n∈N in X
with F (x̃n ) = yn . By Proposition 1.6 we have to show that there exists a sequence (xn )n∈N
with F (xn ) = y0 and x̃n − xn → 0. Such a choice is given by
 
xn := x̃n − (F |U )−1 F (x̃n ) + (F |U )−1 (y0 ),

because

F (xn ) = F (x̃n ) − F (x̃n ) + y0 = y0

10 1 Inverse Problems, Ill-Posedness, Regularization

and
 
x̃n − xn = (F |U )−1 F (x̃n ) − (F |U )−1 (y0 ) = (F |U )−1 (yn − y0 ) → 0.

Now let (F |U )−1 be unbounded. We want to show local ill-posedness in the sense of
Ivanov at y0 = 0. Local ill-posedness for arbitrary y0 in R(F ) then follows easily by
translation.
Since (F |U )−1 is unbounded, there exists a sequence (yn )n∈N in R(F ) with yn → 0
but (F |U )−1 (yn ) → 0. Set x̃n := (F |U )−1 (yn ) for all n in N. By Proposition 1.6 we have
to show that there is no sequence (xn )n∈N in X with F (xn ) = 0 and x̃n − xn → 0. Suppose
there would exist such a sequence and denote the projection of X onto U along N (F ) by
PU : X → U . This projection exists and it is a bounded linear operator because U and
N (F ) are complementary subspaces of X. Then

x̃n = PU (x̃n − xn ) → 0,

because x̃n ∈ U and xn ∈ N (F ). But this contradicts x̃n → 0. Thus, F has to be locally
ill-posed in y0 = 0.

1.2.4 Nashed’s Definition in Case of Uncomplemented Null Spaces

Given a closed subspace U of a Banach space X we always find a second subspace V

such that X = U ⊕ V is the direct sum of U and V , see [7, Proposition 3.2.3]. The
space U is said to be algebraically complemented by V . If, in addition, V is closed,
then U is said to be topologically complemented or simply complemented by V . Banach
spaces which are not isomorphic to a Hilbert space always contain uncomplemented closed
subspaces, see [8]. An example is ∞ with the uncomplemented closed subspace c0 , cf.
[7, Theorem 3.2.20]. In Hilbert spaces each closed subspace is complemented.
In Proposition 1.10 we followed the lines of [1], [2, Section 5.2], that is, we excluded
the case that the null space of the considered linear operator might by uncomplemented.
But since this might happen in every Banach space which is not isomorphic to a Hilbert
space, we want to discuss implications of Nashed’s definition of well- and ill-posedness on
invertibility of (generalized) inverses and on Ivanov’s definition. Such a discussion seems
to be missing in the literature.
In this subsection we only consider bounded linear operators F between Banach spaces
X and Y with domain D(F ) = X. To highlight linearity we use the more common notation

A := F.

The null space N (A) of A is a closed subspace of X. Given an algebraic complement U

of N (A) we may define an inverse A+ U : R(A) → X of A with respect to U as the inverse
1.2 Ill-Posedness 11

of the restriction A|U of A to U . Obviously, A|U is a bijective bounded linear operator

from U onto R(A). Note, that if the closure R(A) of the range of A is (topologically)
complemented by V in Y , then we could extend A+ U to a generalized inverse on the dense
subspace R(A) ⊕ V of Y .
For operators with uncomplemented null space both well-posedness in the sense of
Nashed and ill-posedness in the sense of Nashed may occur. An example for well-
posedness is the mapping A : 1 → 2 defined by

A x := xn z(n) ,
n∈N

where (z(n) )n∈N is a dense subset of the open unit ball in 2 . See [9, Proof of Theo-
rem 2.3.1] for details. We now show that if N (A) is uncomplemented, then Nashed’s
definition of well-posedness does not tell anything about continuity of the inverse A+
U.

Proposition 1.11 Let N (A) be uncomplemented and let U be an algebraic complement

of N (A). Then the corresponding inverse A+
U is unbounded.

Proof Since N (A) is uncomplemented we have N (A) ∩ U = {0}. Choose some u in

(N (A) ∩ U ) \ {0} and a sequence (un )n∈N in U with un → u. If A+
U would be bounded,
then

un = A + + +
U A un → AU A u = AU 0 = 0,

which contradicts un → u = 0.

It remains to clarify the relation between Nashed’s definition and Ivanov’s definition.
We only have the partial result that a closed range implies well-posedness in the sense of
Ivanov, regardless of possible uncomplementedness of N (A).

Proposition 1.12 Let R(A) be closed. Then A is locally well-posed in the sense of Ivanov
at every point of R(A).

Proof Obviously, it suffices to show local well-posedness in the sense of Ivanov at zero.
Let (yn )n∈N be a sequence in R(A) with yn → 0 and take some positive ε. Then the set
 
Mε := x̃ ∈ X : inf x̃ − x < ε
x∈N (A)

is open and the open mapping theorem implies that the image A Mε is open, too. Since 0 ∈
A Mε , there is n with yn ∈ A Mε . Consequently, we find some x̃n in Mε with A x̃n = yn .
Letting ε → 0 we obtain a sequence (x̃n )n∈N with A x̃n = yn and infx∈N (A) x̃n −x → 0.
12 1 Inverse Problems, Ill-Posedness, Regularization

Noting

inf x̃ − x = inf x̃n − x for all x̃ ∈ A−1 yn

x∈N (A) x∈N (A)

we arrive at

sup inf x̃ − x = inf x̃n − x → 0.

x̃∈A−1 yn x∈N (A) x∈N (A)

If N (A) is uncomplemented and R(A) is not closed, then the author does not have
any result on well- or ill-posedness in the sense of Ivanov. But he conjectures that both
situations are possible.

1.3 Tikhonov Regularization

To solve Eq. (1.1) numerically we have to take into account noisy data and lacking stability
of the inversion process due to ill-posedness. Noisy data need not belong to the range of
F . Thus, there need not exist a solution. To overcome this difficulty we content ourselves
with ‘least squares’ solutions defined by

1
F (x) − y δ p → min , (1.3)
p x∈D(F )

where y δ denotes noisy data with noise level δ as introduced in (1.2). The exponent p can
be used to ease numerical minimization. We assume p > 1.
Without further assumptions on F there need not exist a solution to the minimization
problem (1.3). Even if we have a sequence (y δn )n∈N of noisy data with decreasing noise
levels δn , that is, δn → 0, for which solutions to (1.3) exist, we cannot guarantee that they
converge to the solution set of (1.1). This is a simple consequence of local ill-posedness in
the sense of Ivanov at the exact right-hand side y † , cf. Definition 1.5.
There exist several approaches to stabilize the minimization problem and to obtain
convergence of the stabilized problem’s solutions to the solution set of (1.1). We restrict
our attention to the very flexible class of Tikhonov-type regularization methods

1
Tα (x, y δ ) := F (x) − y δ p + α Ω(x) → min . (1.4)
p x∈D(F )

The penalty functional Ω : X → (−∞, ∞] has to be chosen in a way which stabilizes the
minimization problem, see Assumption 1.13 below. The positive regularization parameter
1.3 Tikhonov Regularization 13

α controls the trade-off between data fitting and stabilization. To obtain convergence of
the Tikhonov minimizers to the solution set of (1.1) if the noise level decreases, we have
to choose α depending on δ or y δ or on both. Tikhonov-type methods are well established
in the field of inverse problems and detailed information can be found in almost all
monographs and textbooks in the field, see, e.g., [10–12].

Assumption 1.13 We assume that

(i) Assumption 1.1 is true,

(ii) there exists a solution to (1.1) at which Ω is finite,
(iii) Ω is convex and proper,
(iv) Ω is weakly sequentially lower semicontinuous,
(v) each sequence in a sublevel set {x ∈ X : Ω(x) ≤ c}, c ∈ R, contains a weakly
convergent subsequence.

Under these assumptions one easily shows that there exist solutions to (1.1) which
minimize Ω in the set of all solutions. Such solutions will be referred to as Ω minimizing
solutions and are usually denoted by x † . Tikhonov minimizers only converge to this special
type of solutions.

Theorem 1.14 Let Assumption 1.13 be true. Then the following assertions on existence,
stability and convergence of Tikhonov regularized solutions are true.

(i) For each y in Y and each positive α the minimization problem (1.4) has a solution.
(ii) Let α > 0. If (yn )n∈N is a sequence in Y converging to some fixed y in Y and if
(xn )n∈N is a sequence of corresponding minimizers of Tα (·, yn ), then (xn )n∈N has a
weakly convergent subsequence and the limit of each weakly convergent subsequence
is a minimizer of Tα (·, y). Further Ω(xnk ) → Ω(x̄) for each weakly convergent
subsequence (xnk )k∈N with limit x̄.
(iii) Let (δn )n∈N be a sequence of noise levels converging to zero and let (y δn )n∈N be a
sequence of corresponding data elements. Further assume that (αn )n∈N is a sequence
p
of positive regularization parameters with αn → 0 and δαnn → 0. Then each
sequence (xn )n∈N of corresponding minimizers of Tαn (·, y δn ) has a weakly convergent
subsequence and the limit of each weakly convergent subsequence is an Ω minimizing
solution to (1.1). Further Ω(xnk ) → Ω(x †) for each weakly convergent subsequence
(xnk )k∈N with limit x † .

Proof See [12, Section 4.1] or [13, Chapter 3].

14 1 Inverse Problems, Ill-Posedness, Regularization

Convergence to Ω minimizing solutions may be arbitrarily slow and we are interested

in estimates for the convergence speed. Of course, additional assumptions are required for
such estimates. This issue will be discussed in the next chapter.
At first, we have to decide how to measure the speed of convergence. To cover all
relevant situations we introduce a general error functional E † : X → [0, ∞), where
the symbol † indicates that the error functional depends on one or more Ω minimizing
solutions. The canonical choice in Hilbert spaces is the squared norm distance

E † (x) = x − x † 2

between regularized and exact solution. If there are multiple solutions, the point-to-set
distance

E † (x) = inf x − x † 2
x † ∈S

is a suitable choice if S denotes the set of norm minimizing solutions. In Banach spaces
alternatives like the Bregman distance

E † (x) = BξΩ† (x, x † )

(see Sect. A.3 for a definition) proved to be useful. But Banach space norms can also be
used. In 1 (N) we could choose

E † (x) = x − x † 1 or E † (x) = inf x − x † 1

x † ∈S

with S denoting again the set of norm minimizing solutions. All these examples will be
discussed in Sect. 3.2 in more detail.
Denoting by xαδ the minimizers of the Tikhonov minimization problem (1.4), we aim at
asymptotic estimates

E † (xαδ ) = O(ϕ(δ)), δ → 0, (1.5)

where α may depend on δ and y δ . The function ϕ shall be an index function in the
following sense.

Definition 1.15 A function ϕ : [0, ∞) → [0, ∞) is an index function if it is continuous,

monotonically increasing, strictly increasing in a neighborhood of zero, and satisfies
ϕ(0) = 0.
Variational Source Conditions Yield Convergence
Rates 2

Abstract
We introduce variational source conditions and derive convergence rates for Tikhonov-
type regularization methods.

2.1 Evolution of Variational Source Conditions

Different techniques have been developed to prove convergence rates (1.5). The most
prominent tool are source conditions for linear ill-posed inverse problems in Hilbert
spaces. The classical concept is described in [10, Section 3.2] and general source
conditions are studied in [14]. See also the references given in [14] for the origins of
general source conditions. In both cases the norm distance between exact and regularized
solution is used as error functional E † .
For Banach spaces usage of source conditions is quite limited. But in 2007 variational
source conditions were introduced in [15] and thoroughly studied and developed during the
past 10 years, see, e.g., [11,13,16–18]. This type of condition allows to prove convergence
rates for many different settings, especially for nonlinear operators and general penalty
functionals in (1.4). In its original version Bregman distances (see Sect. A.3) were used as
error functional E † .
Variational source conditions are also known as variational inequalities, but this term
conflicts with the already existing mathematical field with the same name. A second alter-
native was introduced in the book [13]. There the term variational smoothness assumption
is used, because several kinds of smoothness (not only of the underlying exact solution
as it is the case for classical source conditions) are jointly described by one expression.
The term variational source condition rouses associations to classical source conditions.
But the new concept has no similarity to classical source conditions, most notably there

© Springer Nature Switzerland AG 2018 15

J. Flemming, Variational Source Conditions, Quadratic Inverse Problems,
Sparsity Promoting Regularization, Frontiers in Mathematics,
https://doi.org/10.1007/978-3-319-95264-2_2
16 2 Variational Source Conditions Yield Convergence Rates

is no source element. Nevertheless, in most recent literature ‘variational source condition’

seems to be used more often than ‘variational inequality’, whereas ‘variational smoothness
assumption’ is not used by other authors. Thus, we write ‘variational source condition’ to
name the technique described and applied below and, to avoid drawing too many parallels
to source conditions, we read it as ‘variational replacement for source conditions’.
The definition of variational source conditions in this book, see also [19], will be
slightly more general than other variants before, because it is not connected to one fixed
solution of (1.1). Instead, we allow multiple Ω minimizing solutions and refer to [20] and
Sect. 7.1 for a concrete example. For this purpose we denote by Ω † the value of Ω at the
Ω minimizing solutions, that is,

Ω † := min{Ω(x † ) : x † ∈ D(F ), F (x † ) = y † }.

Assumption 1.13 guarantees existence of Ω minimizing solutions.

Definition 2.1 Let β > 0 be a constant and let ϕ : [0, ∞) → [0, ∞) be an index function.
A variational source condition for fixed right-hand side y † holds on a set M ⊆ D(F ) if

β E † (x) ≤ Ω(x) − Ω † + ϕ(F (x) − y † ) for all x in M. (2.1)

If the set M is large enough to contain all minimizers of the Tikhonov functional
(1.4), then a variational source condition (2.1) implies the desired convergence rate
(1.5). Although our variant is slightly more general, the proofs of this fact given in
[13, Chapter 4] or in [21] still work with trivial modifications. Suitable choices of the
regularization parameter α are discussed there, too. For the sake of completeness we
provide the proof in the next section.
The constant β plays only a minor role. In principle we could hide it in the functional
E , but then E † would depend on the chosen index function ϕ and not solely on exact
†

and regularized solutions. The implied convergence rate does not depend on β, only the
O-constant contains the factor β1 .
Variational source conditions originally were developed to obtain rates for Tikhonov-
type regularization, but can also be used in the context of other methods. See [22] for the
residual method and [18] for iteratively regularized Newton methods.
A major drawback of variational source conditions is that the best obtainable rate may
be slower than the best possible one. This is for instance the case for rates faster than
√ 2
O( δ) in the classical linear Hilbert space setting, where the best one is O(δ 3 ). On
the other hand, in 1 -regularization rates up to the best possible one O(δ) for the error
norm can be obtained, see [23] and Sect. 11.3. An approach to overcome technical rate
limitations was undertaken in [24], but it is limited to linear equations.
2.2 Convergence Rates 17

2.2 Convergence Rates

In this section we prove that variational source conditions (2.1) imply rates (1.5). To choose
the regularization parameter we consider an a priori parameter choice α = α(δ) specified
below and an a posteriori parameter choice α = α(δ, y δ ) known as discrepancy principle.
The later consists in choosing α such that

δ ≤ F (xαδ ) − y δ  ≤ τ δ (2.2)

where τ ≥ 1 is a parameter and xαδ denotes a Tikhonov regularized solution.

In case of the a priori choice we apply techniques from [21], but slightly improve the
constants in the error estimate. For the discrepancy principle we take the proof from [13]
and specialize it to our setting. Both proofs can also be found in [25] for the special case
of 1 -regularization.
To shorten the two proofs we mention two properties of concave index functions ϕ.
Simple calculations show that t → ϕ(tt ) is decreasing. As a consequence we see that
ϕ(c t) ≤ c ϕ(t) if c ≥ 1. Both observations will be used without further notice.

Proposition 2.2 Let the variational source condition (2.1) be satisfied with a convex index
function ϕ and choose α in (1.4) such that

δp δp
c1 ≤ α ≤ c2
ϕ(δ) ϕ(δ)

for all positive δ with positive constants c1 , c2 . Then


1 1 1
E † (xαδ ) ≤ 1+ + (1 + 2 p c2 ) p−1 ϕ(δ)
β p c1

for all positive δ.

Proof Because xαδ is a minimizer of (1.4), for an arbitrary Ω minimizing solution x † to

(1.1) we have

1 1
Ω(xαδ ) − Ω † = Tα (xαδ , y δ ) − α Ω(x †) − F (xαδ ) − y δ p
α p

1 1 1
≤ F (x † ) − y δ p − F (xαδ ) − y δ p
α p p

1 1 p 1
≤ δ − F (xαδ ) − y δ p .
α p p
18 2 Variational Source Conditions Yield Convergence Rates

and thus the variational source condition (2.1) implies

1  p   
β E † (xαδ ) ≤ δ − F (xαδ ) − y δ p + ϕ F (xαδ ) − F (x † ) . (2.3)
pα

Because β E † (xαδ ) ≥ 0, we obtain

 
F (xαδ ) − y δ p ≤ δ p + p α ϕ F (xαδ ) − F (x † ) .

If F (xαδ ) − y δ  ≤ δ, then the triangle inequality, the properties of ϕ and the parameter
choice imply

F (xαδ ) − y δ p ≤ δ p + p α ϕ(2 δ) ≤ δ p + 2 p α ϕ(δ) ≤ (1 + 2 p c2 ) δ p ,

that is,
1 1
F (xαδ ) − y δ  ≤ (1 + 2 p c2 ) p δ ≤ (1 + 2 p c2 ) p−1 δ.

If, on the other hand, F (xαδ ) − y δ  > δ, then

 
F (xαδ ) − y δ p ≤ δ p + p α ϕ F (xαδ ) − y δ  + δ
 
ϕ F (xαδ ) − y δ  + δ  
= δ +pα
p
F (xαδ ) − y δ  + δ
F (xα ) − y  + δ
δ δ

ϕ(δ)  
≤ δp + p α F (xαδ ) − y δ  + δ
δ
ϕ(δ)
≤ δ p−1 F (xαδ ) − y δ  + 2 p α F (xαδ ) − y δ 
δ

and thus,

 1
ϕ(δ) p−1 1
F (xαδ ) − y δ  ≤ δ p−1 + 2 p α ≤ (1 + 2 p c2 ) p−1 δ.
δ

In both cases (2.3) can be further estimated to obtain

1  p   
β E † (xαδ ) ≤ δ − F (xαδ ) − y δ p + ϕ F (xαδ ) − y δ  + δ
pα
δp 1
≤ +ϕ 1 + (1 + 2 p c2 ) p−1 δ
pα
δp 1
≤ + 1 + (1 + 2 p c2 ) p−1 ϕ(δ)
pα
2.2 Convergence Rates 19

and the lower bound for α leads to

ϕ(δ) 1
β E † (xαδ ) ≤ + 1 + (1 + 2 p c2 ) p−1 ϕ(δ).
p c1

Note that in the proof we used arguments similar to the ones in [21], but made changes
in the details leading to a better constant in the obtained error estimate. Corresponding
estimates in [21, Theorem 1] lead to

1 1
E † (xαδ ) ≤ 1 + 2 (2 + p) p−1 ϕ(δ),
β

which has a greater constant factor than our estimate. Our estimate with the parameter
choice from [21], that is c1 = c2 = 1, reads

1 1 1
E † (xαδ ) ≤ 1 + + (1 + 2 p) p−1 ϕ(δ).
β p

Proposition 2.3 Let the variational source condition (2.1) be satisfied and choose α in
(1.4) according to the discrepancy principle (2.2). Then

1+τ
E † (xαδ ) ≤ ϕ(δ)
β

for all positive δ.

Proof Because xαδ is a minimizer of (1.4), for an arbitrary Ω minimizing solution to (1.1)
we have

1 1
Ω(xαδ ) − Ω † = Tα (xαδ , y δ ) − α Ω(x †) − F (xαδ ) − y δ p
α p

1 1 1
≤ F (x † ) − y δ p − F (xαδ ) − y δ p
α p p

1 1 p 1
≤ δ − F (xαδ ) − y δ p
α p p

and taking into account the left-hand inequality in (2.2) we obtain

Ω(xαδ ) − Ω(x †) ≤ 0.
20 2 Variational Source Conditions Yield Convergence Rates

The variational source condition (2.1) thus implies

   
β E † (xαδ ) ≤ ϕ F (xαδ ) − F (x † ) ≤ ϕ F (xαδ ) − y δ  + δ

and the right-hand side in (2.2) yields

 
β E † (xαδ ) ≤ ϕ (1 + τ ) δ ≤ (1 + τ ) ϕ(δ).
Another random document with
no related content on Scribd:
provinces. It is not quite easy to forecast where the centre of gravity
of the Central Province will ultimately fall, but if, as is possible, the
Bauchi highlands become in time a second Simla for the Central
Executive, the headquarters of the Central Province would
presumably be fixed at Zungeru, the present capital of the Northern
Protectorate. III. The Western Province, comprising all that is now
incorporated in the existing western province of Southern Nigeria,
plus—to the north—Kabba, Ilorin and Borgu, while the right bank of
the Forcados and Niger would form the eastern boundary, the
boundaries of the Province following natural lines. Its headquarters
would be Oshogbo, or its immediate neighbourhood. IV. The Eastern
Province, comprising what is now the eastern province of South
Nigeria, but with its western frontier coterminous with the left bank of
the Niger and Forcados and its northern frontiers pushed up to the
south bank of the Benue, embracing Bassa and part of Muri, Yola,
however, being left, for political reasons, in the Central Province, as
noted above. Its headquarters would be Old Calabar, the starting-
point of the future eastern railway (see map).
Each of these great provinces would be ruled by a Lieutenant-
Governor, with Residents and Assistant Residents under him, and,
wherever possible, the present political boundaries of what are now
provinces, but would become known as districts and sub-districts,
would be retained. Thus in the Northern or Sudan Province nothing
would be changed in this respect, save the separation of
Mohammedan Zaria from pagan Zaria; nothing would be changed in
the Central Province, so far as the units remaining within it were
concerned, except the division of Muri, which would offer no political
embarrassments. The enlargement of the Eastern Province as
proposed, would in some respects facilitate the work of
administration and would not cut across any ethnic divisions. In the
Western Province the principal alteration would be the re-grouping of
the different Yoruba sections in their old state form (vide Part II.)
under a Resident who would reside at Oyo; Ilorin, Kabba, and Borgu
would remain under Residents as at present. Warri (the capital of the
existing central province of Southern Nigeria) would become the seat
of a Residency for the Bini, Sobo, Ijaw and Jekri speaking peoples.
 Lagos town would continue to be what the expenditure of much
money, and the enterprise of the Yorubas, have made it, the
commercial emporium of at least the western portion of the
Protectorate, and the headquarters of the small surrounding area
known as the “Colony” (vide Part II.), administered by a “Lagos
Council,” which would replace the present “Lagos Legislative
Council,” and be composed of much the same elements as the latter
now consists of, presided over by a Resident. The functions of the
Lagos Council would be confined to the Colony.
The headquarters of the Governor-General and the central seat of
Government would be the high plateau immediately behind Lokoja,
known as Mount Patte, situated in the very centre of the
Protectorate, commanding the Niger and the Benue, within easy
steam of Baro the starting-point of the central railway, and linked up
with the western railway by a branch line to Oshogbo as indicated on
the map. The Governor-General would be assisted by an Executive
and Legislative Council. Of the former the Lieutenant-Governors and
Senior Residents would be ex officio members, together with the
Chief Justice, the Colonial Secretary, the Financial Secretary, and
the officer commanding the troops. The official members of the
Legislative Council would include the Directors of rail and river
transport, of public works, of agriculture, of forestry and of
commercial intelligence; the Director of mining; and the Principal
Medical Officer. The unofficial members would include selected
representatives of the educated native community, and, later on, one
or two distinguished Mallams, and selected representatives of the
European commercial and mining communities.
Possibly, in course of time, the work of the Council could be
carried out in conjunction with periodical Durbars attended by all the
important Emirs, but in no case would the functions of the Council be
allowed to conflict with the Native Administrations of the
Mohammedan Provinces.
SKETCH MAP OF NIGERIA, SHOWING SUGGESTED REARRANGEMENT OF
PROVINCES.

The method of handling the finances of the Protectorate would

depend to a large extent upon the capacity of the Home
Government, in conjunction with the potential Governor-General and
other advisers, to map out ahead a considered scheme of railway
construction and improvement of fluvial communications, which
would proceed from year to year and for which provision would be
made. The whole problem of communications, both rail and river,
ought to be placed under a special department, subject to periodical
inspection by an independent expert sent out from home by the
Colonial Office, and the services of consulting engineers in England
disposed of if possible. The situation financially lends itself, in a
general sense, to a certain boldness of treatment and departure from
ordinary British West African precedent. Two distinct classes of
budgets might with advantage, perhaps, be evolved, viz. a Colonial
budget and the Provincial budgets. In other words, there would be a
central budget and four local budgets, one for each Province. The
Colonial budget would be fed by the customs revenue, the whole of
which would be credited to it. (It may be estimated that two or three
years hence the total customs revenue collected in Nigeria will
amount to £2,500,000.) It would be augmented by the profits on the
railways, the mining royalties, harbour dues, and pilotage fees (there
should be a system of public pilotage on the waterways). The
Protectorate could be authorized to raise a loan on its own
recognizances of £5,000,000 redeemable in a term of years. This
loan would be expended in a succession of public works—some of
the necessary lines of rail are indicated in the map—in accordance
with the scheme of construction mapped out as previously
suggested. The Colonial budget would determine the successive
instalments of expenditure out of loans, and would provide the
interest on the new loan and on the existing loan of £5,000,000
contracted by Southern Nigeria (for public works in Southern and
Northern Nigeria). The revenues of the Colonial budget from
whatever source derived, other than from loans, would be distributed
by the Governor-General in council for the administration of the four
Provinces in accordance with their respective needs. These needs
would show marked variation for some years to come. For instance,
the hypothetical Northern and Central Provinces (i.e. the territory
which now comprises the bulk of Northern Nigeria), relying upon the
increasing regularity and juster assessment of internal direct
taxation, the nature of which may roughly be termed a graduated
property tax, might be expected to advance steadily towards the self-
supporting stage. When that stage had been reached, the surplus
would be set aside under the Provincial budget for extending the
system of fixed salaries to native officials, for expenditure on
provincial public works and economic research, improvements in
sanitation, and so on, in collaboration with the native authorities of its
various sections. A portion of my hypothetical Northern or Sudan
Province is already self-supporting, viz. Kano. Indeed, but for the
military establishment the whole of that Province would be showing
to-day a handsome surplus and, apart from the public works to be
met out of loans, would require—even if it continued to be debited
with the military establishment—very little assistance from the
Colonial budget. The hypothetical Central Province would require
more assistance for a time, but, as in the Northern Province, the
basis of an expanding land revenue is securely laid and a not
inconsiderable mineral development bringing revenue, apart from
royalties, is assured to it. On the other hand, most of the hypothetical
Western Province and almost the whole of the Eastern Province—
i.e. in combination, Southern Nigeria of to-day—produces no internal
revenue whatever except licences, the amount derived from which
will assuredly grow but will not become really large for many years.
Therefore, until and unless the delicate problem of introducing direct
taxation among peoples—the majority of whom we have been in
touch with for years without requiring of them the payment of any
form of tribute—were approached, the Colonial budget would have to
furnish these Provinces with most of their administrative revenues.
An alternative scheme would be to abandon the idea of a Central
Legislative Council for the whole Protectorate and of a new
administrative headquarters, the Governor-General spending a
certain time at the headquarters of each Province. Lagos would,
under such a scheme, become the capital of the extended Western
Province (see map), and the action of the Lagos Legislative Council
would extend to the whole of that Province. A Legislative Council
would be created for the extended Eastern Province. The
administrative machinery of the new Central and Northern Provinces
would be left as it is now. On the finance side the alternative scheme
to the one I have sketched would be to let each Province contribute
to the Colonial budget in accordance with its capacities upon a
definite proportionate basis, the sums thus accruing to the Colonial
budget, plus the loan funds, being utilized in the creation of public
works on the lines already sketched. This alternative scheme,
amalgamation on federation, would possess some advantages over
the first, and compares unfavourably with it in others.
 It will be objected that these suggestions do not take into account
the present military expenditure of the Protectorates and are dumb
with regard to the Imperial grant to Northern Nigeria. I have left a
consideration of these two questions until now because they can, I
think, be taken together. The military establishment of Southern
Nigeria costs £100,000 per annum. That of Northern Nigeria costs
£160,000 per annum. Neither is excessive in itself, although in the
latter case it amounts to no less than 33 per cent. of the total
expenditure of the Protectorate! It is not one penny too much, and to
reduce the number of troops would be folly, having regard to the
immensity of the country and the kind of political problem facing us.
And yet could anything be more topsy-turvey? Here is a financially
struggling Protectorate urgently in need of the most vital necessities;
incapable even of building decent houses for its over-worked and
short-handed staff; forced to deprive the latter of even their travelling
allowances, and to sacrifice considerations of reasonable comfort
and, therefore, of health for its personnel; in a position to pay so little
for posts of enormous responsibility that the entire political
expenditure is only some £70,000 per annum; able to devote but a
miserable £1300 a year upon economic forestry, but saddled with
this incubus of £160,000 upon a military establishment which has
already been called upon (in the case of the last Ashanti war) to
provide contingents for service outside the Protectorate, which would
infallibly happen again, in the by no means remote contingencies of
a further outbreak in Ashanti or disturbances in the Sierra Leone
hinterland. This situation needs to be examined in conjunction with
the Imperial grant about which so much fuss is made.
The nation imagines that Northern Nigeria is costing the Imperial
Treasury something like £250,000 to £300,000 per annum. Nothing
of the kind. The grants in aid from 1906 to 1909, inclusive, amounted
to £1,220,000, or an average of £305,000. But against this must be
set the direct profit to the revenues of the United Kingdom derived
from the profit which the Mint makes upon the silver coin exported, in
ever increasing quantities (and the process will go on extending), to
the two Nigerias. The average yearly cost of silver in the last nine
years has, I believe, varied between 2s. 0¾d. and 2s. 6⅞d. The coin
at par value is issued at 5s. 6d. an ounce, and I am credibly informed
that the profit to the Mint is considerably more than half the net
import by Nigeria, seeing that half the face value of the coinage is
greater than the cost of minting, plus maintenance of gold reserve
and provision for remitting. The net export of coinage, virtually the
whole of it silver, to the two Nigerias (i.e. the total exported minus the
coin returned) amounted from 1906 to 1909 to £981,582. If the profit
of the Mint is taken at only 50 per cent., it will thus be seen that the
nation is making a direct average profit of nearly £125,000 a year out
of the two Nigerias, against an average of £305,000 paid to Northern
Nigeria by way of a temporary grant in aid. To say, therefore, that
Northern Nigeria is costing the British taxpayer a quarter of a million
a year or more, is to make a statement which is not in accordance
with fact. What the nation advances directly, it recoups itself for
directly in part; without counting that these grants are in the nature of
a capital investment. Let this grant under amalgamation be
cancelled, and let the Imperial Government, on the other hand, foot
the bill for the military expenditure (which, as we have seen,
amounts to £260,000), looking upon it, say, for the next ten years as
Imperial expenditure. Nothing would so alleviate the whole situation,
while at the same time simplifying it, and, as has been shown, the
actual disbursement of the nation on this item would be considerably
less, even now, than what it would appear nominally to be, owing to
the profit made by the Mint on the silver coin sent out.
As already explained, the above proposals, illustrated in part by
the accompanying map, are put forward merely as a basis for the
discussion of a problem of some difficulty but of great urgency. I
claim for them nothing more than that, and no conceivable scheme
of amalgamation could be set down which would not lend itself to
copious criticism. But that the mush of anomalies now obtaining
cannot be perpetuated without increasing detriment to Imperial
interests in Nigeria, I am fully persuaded. The existence of two public
policies side by side in a single territorial area, where internal peace
is rapidly fusing the indigenous communities, divided by an
imaginary line which does not even correspond to natural boundaries
and exhibiting multiple differences of aim and method—in some
cases, acutely antagonistic interests—presents many obvious
inconveniences and paves the way for future embarrassments of
every kind. If these remarks can influence in any way an early and
serious examination of the problem by the Colonial Office, they will
not, however open to criticism, have been made in vain.
Amalgamation must come. All realize that. Unforeseen events might
very well, at a given moment, compel decisions of far-reaching
moment being precipitately reached without due consideration being
given to all the features of the case, such as characterized the
amalgamation of the Lagos Colony and Protectorate with old
Southern Nigeria in 1899. The advantages of clear thinking out
ahead, and of taking the inevitable step before the situation has got
tied up into more knots than it already contains, with calm
deliberation, after a full and serious study of all the facts, surely
needs no emphasizing. As to the man, a last word. The responsibility
of selecting the official to be in supreme control over the
amalgamated Nigerias is no light one. The task confronting a
Governor-General, especially in the first five years, will be replete
with difficulties. The post will need heavy calls upon tact, patience,
and a peculiarly high type of constructive statesmanship. The only
remark I would venture to make on the point is this. Any serious
administrative error perpetrated in handling affairs in the north would
be attended with consequences of exceeding gravity. That is a
proposition I think no one will be inclined to dispute. It suggests
either that the Governor-General himself should be personally
acquainted with the political conditions of what is now known as
Northern Nigeria, or, at least, that the Lieutenant-Governors of the
hypothetical Northern and Central Provinces should be chosen from
among the most experienced of the existing Senior Residents.
 PART IV
ISLAM, COTTON GROWING, AND THE LIQUOR
TRAFFIC
 CHAPTER I
CHRISTIANITY AND ISLAM IN SOUTHERN NIGERIA

I have referred to Christian missionary propaganda in

Mohammedan Northern Nigeria. There has now to be considered the
question of Christian missionary propaganda in Southern Nigeria,
and the corresponding growth therein of Mohammedanism. The
relative failure of the one and the admitted success of the other are
at present the subject of much debate and give anxious thought to
the heads of the Church. The fundamental cause appears to lie in a
disinclination to face the fact, however obvious, that a religion which
took centuries upon centuries to take root in Europe, owing, very
largely, to its ethical demands upon man, cannot hope to establish
itself in the now accessible tropical forest regions of West Africa in a
few decades, while a religion embodying a distinct advance upon
paganism but not involving the complete structural change in native
society which the Christian Church exacts, has every chance of
doing so. Then, too, there is another question which the
ecclesiastical authorities may never, it is true, find it possible frankly
to confront, but which laymen, it seems to me, are bound to do—
those, at any rate, who are persuaded that the African race is one of
the great races of mankind, not intended by the Almighty Architect to
disappear from the scene of human affairs. I refer to the
physiological requirements, in the present age, of the Nigerian forest
peoples in their struggles with the forces of primeval Nature.
All that remains of the Portuguese attempts to Christianize the
deltaic region of Nigeria in the fourteenth and fifteenth centuries are
a few names and the addition of crucifixion to native punishments of
criminals or happy despatch of sacrificial victims. The chief obstacle
to the modern efforts of the Anglo-Saxon in Southern Nigeria and the
real explanation of the successful modern efforts of the African
Muslim, are to be sought in the appeal respectively made by
Christianity and Islam to the patriarchal communities to which they
are addressed, and in the methods and character of the respective
propaganda. Christianity in West Africa either cannot be divorced, or
cannot divorce itself, from Europeanism and the twentieth century. It
remains for the people of Nigeria, and of all West Africa, an alien
religion taught by aliens who cannot assimilate themselves to the life
of the people. Islam, on the other hand, has long ceased to be an
alien religion. It is imparted by Africans. It is disseminated by
Africans. It has its roots in the soil. It has become a religion of the
people, losing much of its rigidity and fanaticism as it works down to
the coast absorbing the true negro.
Everything is against Christianity as presented to the Nigerian (I
venture to emphasize this), and everything is in favour of Islam,
although Christianity, in itself, contains more that should appeal to
the Negro character than does Mohammedanism. The conditions of
Southern Nigeria are the conditions of the Old Testament. The crying
need of the country, as of all western tropical Africa, is the need
which is proclaimed in, and stamps itself upon, every page of the
book of Genesis, the Divinely ordained requirement—population.
Vice plays only a microscopic part in the relationship of sex in
Nigeria. Race propagation is the motive force which regulates sexual
relationship. The Nigerian, incessantly striving with the destructive
agencies of Nature, responds to the instinctive and mysterious call of
racial necessity. Infant mortality is terrible. With the Nigerian the
reproduction of the species is the paramount, if unanalyzed and, no
doubt, uncomprehended obsession. It must continue to be so for a
period whose limit will be determined by the rate of his progression
in coping with these destructive agencies.
This is not the place to discuss what the attitude of the Christian
missionary should be to this paramount racial need, but it is obvious
that his insistence upon an acceptance of a sex relationship contrary
to the promptings of Nature must present a barrier—one of the
greatest, if not the greatest—to the acceptance of the Christian faith,
or, perhaps, it would be better to say, of orthodox Christianity. One
might be permitted, perhaps, to suggest that those who are disposed
to regard the condition of the Nigerian forest-dweller in these matters
as calling for hard and rigid regulation, are too prone to forget what
Lecky describes as the “appalling amount of moral evil, festering
uncontrolled, undiscussed and unalleviated under the fair surface of
a decorous society,” in civilized Europe, monogamistic social laws
notwithstanding. Sex relationship, whatever its character and
whatever the conditions of society or climate, is never, and can never
be, free from abuses. West African polygamy contains many ugly
sores, and so does the European system.
Family bonds are equally threatened by Christianity, as
propounded to the Nigerian, for it trains the child, whether
deliberately or otherwise, to look upon his parents as living a life of
sin, thus introducing a subversive element into the household. Those
who assert the absence of affections and sanctities in Nigerian
family life assert that which is untrue. Native authority is likewise
menaced, for how can the convert entertain his former respect for
rulers whom he has been taught to regard as morally and spiritually
his inferiors? These are some of the reasons why Christianity, as
propounded to the Nigerian, at the opening of the twentieth century,
presents itself to him in the light of a hurtful and disintegrating
influence. And this creed is proffered either by aliens between whom
and the inner life of the people there yawns an unbridgeable gulf, or
by denationalized Africans who have become in the eyes of the
people, strangers well-nigh as complete as the alien himself, part
and parcel of the alien’s machinery. As if these did not constitute
sufficient deterrents to the permanency of its footing, the alien race
which tenders to the Nigerian this creed—this creed claiming for all
men equality before God—is the conquering, controlling, governing
race that scorns to admit—because, being an Imperial race, it cannot
—equality of racial status with the Nigerian whom it subjugates and
controls. Between the race of the converter and that of the would-be
convert there gapes an abyss of racial and social inequality which
does not lessen, but, if anything, widens with conversion—the colour
line.
Finally, there is the lamentable intolerance displayed by Christian
proselytizers towards one another. Only the other day I read in a
West African newspaper the address of a white American Protestant
Bishop, whose sphere of work lies in Africa, to his flock. This
episcopalian interpreter of the Gospel of Christian charity to the
benighted African is concerned in his address with the downfall of
the Portuguese Monarchy and the accession of the Republic which,
he says, “opens wide every door leading to Christian work among
millions of native Africans.” He proceeds: “Of course Rome howls.
On October 13, 1910, among weeping Jesuits, speaking of the new
nation, the Pope said ‘A cursed Republic! Yes, I curse it!’ The curse
of Balaam against the people of God was turned into a blessing by
Jehovah; and so, too, will this blasphemy be turned into a blessing to
the struggling people of Portugal.”
Islam, on the other hand, despite its shortcomings, does not, from
the Nigerian point of view, demand race suicide of the Nigerian as an
accompaniment of conversion. It does not stipulate revolutionary
changes in social life, impossible at the present stage of Nigerian
development; nor does it undermine family or communal authority.
Between the converter and converted there is no abyss. Both are
equal, not in theory, but in practice, before God. Both are African;
sons of the soil. The doctrine of the brotherhood of man is carried
out in practice. Conversion does not mean for the converted a break
with his interests, his family, his social life, his respect for the
authority of his natural rulers. He is not left stranded, as the Christian
Church, having once converted, leaves him, a pitiful, rudderless
barque upon a troubled sea. He does not become, through
conversion, an alien in thought, in custom, and in outlook; a foreigner
in his own land, a citizen of none. He remains African, attached to
his country, looking for inspiration inwards, rather than towards an
alien civilization across thousands of miles of unknown seas. No one
can fail to be impressed with the carriage, the dignity of the Nigerian
—indeed, of the West African—Mohammedan; the whole bearing of
the man suggests a consciousness of citizenship, a pride of race
which seems to say: “We are different, thou and I, but we are men.”
The spread of Islam in Southern Nigeria which we are witnessing to-
day is mainly social in its action. It brings to those with whom it
comes in contact a higher status, a loftier conception of man’s place
in the universe around him, release from the thraldom of a thousand
superstitious fears. It resembles in its progress the annual overflow
of the Niger diffusing its waters over the land. The extensive
ramifications of internal trade, now greatly fostered by the
construction of additional roads and railways and rendered wholly
safe by the pax britannica, leads to the multiplying of facilities for
human intercourse among the various peoples of the Protectorate.
The Hausa pushes ever further south his commercial operations.
The Delta, and still more the Western Province, yearly attest to the
widening area of his activities. Not to be outdone, his trading rival the
Yoruba taps in additional numbers the markets of the north. Railway
construction finds the Mohammedan labouring side by side with the
pagan in the same trench. A sense of security and the increasing
circulation of a portable medium of exchange in the shape of silver
and nickel coinage attract to the great native markets of the Central
Province, such as Onitsha, for example, the tattoed pagan Ibo and
his pagan colleagues the Anams, Katundas, and Kukurukus, where
they rub shoulders with the Mohammedan Hausa, Nupe, and Igarra.
In and around Ibadan, Oyo, and Lagos you meet the Kano and
Sokoto trader with his donkeys and pack-bullocks, and even the
Tuareg with whom you parted company months before in the far
north, travelling on the roads or camping for the night near some
local village. The road is at once the club-house and public
rendezvous for Nigerian humanity. A vast commingling, a far-
reaching fusion unexampled in the history of these peoples is taking
place. The expansion of an African religion which, somehow,
succeeds in investing the convert with a spiritual and social standing
that at once raise him among his fellows, follows as a matter of
course. The Mohammedan teacher wanders over the face of the
country visiting the centres of human activity, haunting the roads and
market-places, unattended, carrying neither purse nor scrip, making
no attempt at proselytizing beyond saying his prayers in public, not in
a manner to cause obstruction, but quietly in some corner; waiting
until people come to him, literally fulfilling the command, “Take
nothing for your journey, neither staves, nor scrip, neither bread,
neither money, neither have two coats apiece.” The Mohammedan
trader or agriculturist settles in a pagan village, marries pagan
women, enters the family and social circle of the community and
imparts to it his faith, the women making even readier converts than
the men.
This is why and this is how Islam is propagating itself and taking
root in pagan Nigeria without financial outlay, without doles and
collecting boxes. One of the oldest of Christian missionaries in
Nigeria, a man of venerable appearance and saintly character, who
for twenty-five years has laboured with hands as well as with heart
and head for moral and material improvement, not of his converts
only, but of their unconverted relatives, confessed to me his fear that
nothing could stop Islam from absorbing in course of time the whole
of West Africa. He was almost disposed reluctantly to allow that in
the providence of God, Islam might prove to be intended as the
halfway house through the portals of which it was necessary the
West African negro should pass in order to lift him out of a sterilizing
paganism and make him a fitter vessel to receive in course of time
the nobler ideals of the Christian faith. Sir Harry Johnston is right, I
think, when he says that “to Negro Africa,” Islam has come “as a
great blessing, raising up savages to a state, at any rate of semi-
civilization, making them God-fearing, self-respecting, temperate,
courageous, and picturesque.” But Islam does more than this; it
preserves racial identity. In West Africa, Christianity destroys racial
identity. It should not: as taught it does.
“Picturesque,” says Sir Harry Johnston, and there speaks the
artist. But the word covers a profound truth. A great deal of the
denationalizing or Anglicizing process which is going on and which
makes bad Africans and bad Christians, is attributable to the
discarding of the national dress. Why cannot the Administration and
the missionary societies combine in some practical, positive form, to
combat this curse of alien dress? There is absolutely nothing to be
said in its favour. The West African looks better in African dress, the
robe of the Mohammedan and of many pagan Africans. It is much
healthier for him. It is preservative of his racial identity; and that is,
perhaps, the most important of all pleas which can be put forward for
its retention. With very slight modification—such as one sees among
the native staff, and personal servants in many parts of Northern
Nigeria—it can be made suitable for any form of labour, literary or
otherwise. Clad in his national dress the African has a dignity which
in most cases he loses almost entirely when he attires himself in a
costume totally unfitted for the country, and hideous at best. Nothing
to my mind is more pitiable than to visit school after school in West
Africa, filled with little boys and girls and big boys and girls in an
alien dress, to see the denationalizing process going on day after
day and nothing whatever done to stop it. In the case of the women it
is not only dignity and nationalism which are concerned, but decency
as well. The national dress of the women in West Africa is classical
and graceful, and although leaving more of the body exposed than is
usual at home (except in the ballroom) it lacks suggestiveness. It
does not accentuate the figure. It emphasizes that racial difference—
not inequality, but difference—which it is so essential to emphasize.
With the substitution of European dress, especially of the prevailing
fashion, the West African woman loses much of what she need
never lose, and acquires that which is of no profit to her. These
things cannot be altered in a day, nor would it be possible in some
cases for the present adult generation to go back to African costume.
But it would in many cases, and the reform could be at once taken in
hand so far as the children are concerned. Government could do
much. The missionary societies could do more. The anglicised native
community could do most. I believe that if some popular Government
official, known and trusted, could be led to appeal, in private
conference to the native staff and win them over, the movement
once started would spread and have enormously beneficial results.
That many members of the anglicised community would be hostile
goes without saying—that is the fault of the wretched system
everywhere at work. That a body of thoughtful men would not, I am
satisfied by the many representations on this very subject personally
made to me. I shall always recollect, in particular, the private visit
paid to me in one of the great Yoruba towns by one of the leading
merchants of the place. A magnificent specimen of an African,
dressed in African costume and speaking our language fluently, he
came with the usual touching words and gifts, and begged me very
earnestly to take up the question of dress with his compatriots.
And, in conclusion, there is another and a very serious handicap
upon Christianity in West Africa, in Southern Nigeria especially.
Under the native social system, religion and politics—the religious
organization and the political organization—go together. It is
inconceivable to the native mind that they should be separate or
antagonistic. Islam, again, preserves this ingrained conviction. But in
West Africa the political and religious organizations of the white man
are separate and distinct. The religious organization itself is split up
into countless opposing sections. And in Southern Nigeria the
section specially identified in the native mind with the white over-lord
has for some years past played a discordant note in that white over-
lord’s political organization. Its representatives are almost
everywhere, and upon many subjects persistently hostile critics of
the Administration, begetting unrest and disloyalty to Government.
The mass of native opinion concludes there is something rotten in
the system presented to it, and the Islamic wave rolls on.
 CHAPTER II
THE COTTON INDUSTRY

Is Nigeria a cotton-growing country? Is an export trade in cotton, of

any large dimensions, a possibility—early or remote? I will
endeavour to answer these questions to the best of my ability. I am
not, however, an expert on cotton-growing, and I am in general
sympathy with the work the British Cotton-growing Association is
trying to carry out, although, as will be seen, I am not entirely in
agreement with all its methods, either here or in Nigeria. To that
extent it will be possible for any one who wants to do so to discount
the views here expressed.
One of the earliest impressions one forms out there is the contrast
between the presentation of the case at home and conditions on the
spot. The view at home—somewhat modified by recent events—has
seemed to be inspired by the idea that if the number of square miles
which Nigeria covers is totted up in one column and the number of
inhabitants it supposedly contains in another and these totals
compared with conditions in the cotton belt of North America, then
you arrive at a conclusion which enables you to speak of the “huge
possibilities” of Nigeria, and even to forecast that Northern Nigeria
alone “at some future date” will be able “to supply the whole of the
requirements of Great Britain and to leave an equal quantity over for
the other cotton-consuming countries.” Four years ago a prominent
British statesman declared publicly that “once the fly belt near the
river was passed ... cotton would be grown under exactly the same
conditions as it was grown under on such a great scale in America.”
He went so far as to say that the native of Northern Nigeria was
“beginning to cease to grow cotton” because he could get British
manufactured goods in lieu of his home-grown article. Well, between
these statements and actualities there is a “huge” gulf fixed. In the
first place it can be said of Nigeria that in a part of it only is cotton
now grown, and that in a part of it only will cotton ever be grown. To
talk of Nigeria, as a whole, being a cotton-growing country par
excellence, either now or potentially, is absurd. Three-fourths of
Southern Nigeria and a third, probably more, of Northern Nigeria are
quite unsuitable for cotton-growing, and this for many reasons. To
talk of Nigeria supplying the whole requirements of Great Britain (to
say nothing of the promised surplus) is tantamount to saying that
some day “Pleasant Sunday Afternoon” excursions to the moon will
be a regular feature of the national life. Both may become possible
“at some future date,” but there is so much future about the date that
such flights of rhetoric might well be left to the compilers of gold-
mining prospectuses. These extravagances have not helped the
Association. The sincere and sober persons connected with that
body are merely hindered by them. As to cotton being produced in
Northern Nigeria under the “same conditions” as in the States, and
the natives of the country “beginning to cease” to grow cotton, one
can only remark that they are too silly to deal with.
In Southern Nigeria, the deltaic region, the Eastern Province,
virtually the whole of the Central Province, and a considerable
portion of the Western Province—i.e., four-fifths of the whole
Protectorate—may be ruled out of account as a cotton field. The
deltaic region will not produce cotton. The forest belt behind it,
passing (with occasional breaks) from dense to secondary growth
and fading away into open country, no doubt would. But only if you
cut down the forest first. To destroy the West African forest to any
extent in order to grow cotton would be economic madness. Indeed,
the Administration is working hard to preserve the forests from the
ignorance and improvidence of primitive man, and to build up for the
native communities, and in the public interest, a source of future
revenue from the methodical exploitation of its inexhaustible wealth.
With trifling exceptions the whole of this region is the home of the oil-
palm, the most beneficent tree in the world, and such activities as
the inhabitants can spare from their own requirements are given
over, in the main, to the palm oil and kernel trades. It is the home of
valuable cabinet woods, of vegetable oils, gums, and rubbers, and in
time is likely to become a great natural nursery for the cultivation of
plantation rubber and such a moisture-loving plant as the cocoa;
never, I think, of cotton.
 In the Western Province large areas of forest have been
destroyed; the population is, in a certain measure, more enterprising,
and a fair amount of cotton for export may reasonably be expected,
especially, I venture to suggest, if certain methods now prevalent are
modified. The Egba district (1869 square miles, with a population of
260,000), the capital of which is Abeokuta, a town of about 100,000
inhabitants, is the principal but not the only centre for cotton-growing
in the Western Province, and here the Association has a large and
well-equipped ginnery, as it has at Ibadan and Oshogbo. Out of
2,237,370 lbs. of lint cotton exported from Southern Nigeria in 1908,
Abeokuta and neighbourhood was responsible for 722,893 lbs. The
Egbas are good farmers and not strangers to cotton-growing for
export. The industry owes its origin there to a Manchester man, Mr.
Clegg, who introduced it at the time of the American Civil War. In
1862 the export amounted to 1810 lbs., rising in 1868 to over
200,000 lbs., and continuing, I believe, at that figure or thereabouts
for some years. Cotton then began to fall heavily in price, and the
Egba farmer, finding no profit in growing it, turned his attention to
other crops. The industry was revived on a much larger scale by the
Association in 1905. The exports of cotton lint from Southern Nigeria
from 1906 to 1910—i.e. since the Association came upon the scene
—have been as follows:—

1906 2,695,923

1907 4,089,530

1908 2,237,370

1909 4,929,646

1910 2,399,857

The total value of these five years’ output amounts to something

like £350,000. It is entirely creditable to the Association that it should
have been instrumental in reviving a decayed industry in one district
and creating one in others, and in five years to have fathered an