Generalized Multiresolution Analyses

Kathy D. Merrill
Visit to download the full and correct content document:
https://textbookfull.com/product/generalized-multiresolution-analyses-kathy-d-merrill/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Control Performance Assessment: Theoretical Analyses

and Industrial Practice Pawe■ D. Doma■ski

https://textbookfull.com/product/control-performance-assessment-
theoretical-analyses-and-industrial-practice-pawel-d-domanski/

Information Technology Project Management Kathy

Schwalbe

https://textbookfull.com/product/information-technology-project-
management-kathy-schwalbe/

Generalized linear models and extensions Fourth Edition

Hardin

https://textbookfull.com/product/generalized-linear-models-and-
extensions-fourth-edition-hardin/

Color Atlas of Pediatric Anatomy Laparoscopy and

Thoracoscopy 1st Edition Merrill Mchoney

https://textbookfull.com/product/color-atlas-of-pediatric-
anatomy-laparoscopy-and-thoracoscopy-1st-edition-merrill-mchoney/
Property Principles and Policies Third Edition Thomas
W. Merrill & Henry E. Smith

https://textbookfull.com/product/property-principles-and-
policies-third-edition-thomas-w-merrill-henry-e-smith/

Chemistry Structure and Properties Solutions manual

Kathy Thrush Shaginaw

https://textbookfull.com/product/chemistry-structure-and-
properties-solutions-manual-kathy-thrush-shaginaw/

Existence Theory for Generalized Newtonian Fluids 1st

Edition Breit

https://textbookfull.com/product/existence-theory-for-
generalized-newtonian-fluids-1st-edition-breit/

An Introduction to Generalized Linear Models Annette J.

Dobson

https://textbookfull.com/product/an-introduction-to-generalized-
linear-models-annette-j-dobson/

Higher Gradient Materials and Related Generalized

Continua Holm Altenbach

https://textbookfull.com/product/higher-gradient-materials-and-
related-generalized-continua-holm-altenbach/
 Lecture Notes in Applied and Numerical Harmonic Analysis

Kathy D. Merrill

Generalized
Multiresolution
Analyses
Lecture Notes in Applied and Numerical
Harmonic Analysis

Series Editor
John J. Benedetto
University of Maryland
College Park, MD, USA

Editorial Advisory Board

Emmanuel Candes
Stanford University
Stanford, CA, USA

Peter Casazza
University of Missouri
Columbia, MO, USA

Gitta Kutyniok
Technische Universität Berlin
Berlin, Germany

Ursula Molter
Universidad de Buenos Aires
Buenos Aires, Argentina

Michael Unser
Ecole Polytechnique Federale de Lausanne
Lausanne, Switzerland

More information about this series at http://www.springer.com/series/4968

Kathy D. Merrill

Generalized Multiresolution
Analyses
Kathy D. Merrill
Department of Mathematics
The Colorado College
Colorado Springs, CO, USA

ISSN 2296-5009 ISSN 2296-5017 (electronic)

Applied and Numerical Harmonic Analysis
ISSN 2512-6482 ISSN 2512-7209 (electronic)
Lecture Notes in Applied and Numerical Harmonic Analysis
ISBN 978-3-319-99174-0 ISBN 978-3-319-99175-7 (eBook)
https://doi.org/10.1007/978-3-319-99175-7

Library of Congress Control Number: 2018954118

Mathematics Subject Classification: 42-02, 42C40, 42C15, 43A40, 28A80, 20H15, 65T60

© 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
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
LN-ANHA Series Preface

The Lecture Notes in Applied and Numerical Harmonic Analysis (LN-ANHA) book
series is a subseries of the widely known Applied and Numerical Harmonic Analysis
(ANHA) series. The Lecture Notes series publishes paperback volumes, ranging
from 80 to 200 pages in harmonic analysis as well as in engineering and scientific
subjects having a significant harmonic analysis component. LN-ANHA provides a
means of distributing brief-yet-rigorous works on similar subjects as the ANHA
series in a timely fashion, reflecting the most current research in this rapidly
evolving field.
The ANHA book series aims to provide the engineering, mathematical, and
scientific communities with significant developments in harmonic analysis, ranging
from abstract harmonic analysis to basic applications. The title of the series
reflects the importance of applications and numerical implementation, but richness
and relevance of applications and implementation depend fundamentally on the
structure and depth of theoretical underpinnings. Thus, from our point of view,
the interleaving of theory and applications and their creative symbiotic evolution
is axiomatic.
Harmonic analysis is a wellspring of ideas and applicability that has flourished,
developed, and deepened over time within many disciplines and by means of
creative cross-fertilization with diverse areas. The intricate and fundamental rela-
tionship between harmonic analysis and fields such as signal processing, partial
differential equations (PDEs), and image processing is reflected in our state-of-the-
art ANHA series.
Our vision of modem harmonic analysis includes mathematical areas such as
wavelet theory, Banach algebras, classical Fourier analysis, time-frequency analysis,
and fractal geometry, as well as the diverse topics that impinge on them.
For example, wavelet theory can be considered an appropriate tool to deal with
some basic problems in digital signal processing, speech and image processing,
geophysics, pattern recognition, bio-medical engineering, and turbulence. These
areas implement the latest technology from sampling methods on surfaces to fast
algorithms and computer vision methods. The underlying mathematics of wavelet

v
vi LN-ANHA Series Preface

theory depends not only on classical Fourier analysis but also on ideas from abstract
harmonic analysis, including von Neumann algebras and the affine group. This leads
to a study of the Heisenberg group and its relationship to Gabor systems and of the
metaplectic group for a meaningful interaction of signal decomposition methods.
The unifying influence of wavelet theory in the aforementioned topics illustrates
the justification for providing a means for centralizing and disseminating informa-
tion from the broader, but still focused, area of harmonic analysis. This will be a key
role of ANHA. We intend to publish with the scope and interaction that such a host
of issues demands.
Along with our commitment to publish mathematically significant works at the
frontiers of harmonic analysis, we have a comparably strong commitment to publish
major advances in applicable topics such as the following, where harmonic analysis
plays a substantial role:

Bio-mathematics, bio-engineering, Image processing and super-resolution;

and bio-medical signal processing; Machine learning;
Communications and RADAR; Phaseless reconstruction;
Compressive sensing (sampling) Quantum informatics;
and sparse representations; Remote sensing;
Data science, data mining Sampling theory;
and dimension reduction; Spectral estimation;
Fast algorithms; Time-frequency and time-scale analysis
Frame theory and noise reduction; – Gabor theory and wavelet theory.
The above point of view for the ANHA book series is inspired by the history of
Fourier analysis itself, whose tentacles reach into so many fields.
In the last two centuries Fourier analysis has had a major impact on the
development of mathematics, on the understanding of many engineering and
scientific phenomena, and on the solution of some of the most important problems
in mathematics and the sciences. Historically, Fourier series were developed in the
analysis of some of the classical PDEs of mathematical physics; these series were
used to solve such equations. In order to understand Fourier series and the kinds
of solutions they could represent, some of the most basic notions of analysis were
defined, for example, the concept of “function.” Since the coefficients of Fourier
series are integrals, it is no surprise that Riemann integrals were conceived to deal
with uniqueness properties of trigonometric series. Cantor’s set theory was also
developed because of such uniqueness questions.
A basic problem in Fourier analysis is to show how complicated phenomena,
such as sound waves, can be described in terms of elementary harmonics. There are
two aspects of this problem: first, to find, or even define properly, the harmonics or
spectrum of a given phenomenon, e.g., the spectroscopy problem in optics; second,
to determine which phenomena can be constructed from given classes of harmonics,
as done, for example, by the mechanical synthesizers in tidal analysis.
Fourier analysis is also the natural setting for many other problems in engineer-
ing, mathematics, and the sciences. For example, Wiener’s Tauberian theorem in
LN-ANHA Series Preface vii

Fourier analysis not only characterizes the behavior of the prime numbers but also
provides the proper notion of spectrum for phenomena such as white light; this
latter process leads to the Fourier analysis associated with correlation functions in
filtering and prediction problems, and these problems, in turn, deal naturally with
Hardy spaces in the theory of complex variables.
Nowadays, some of the theory of PDEs has given way to the study of Fourier
integral operators. Problems in antenna theory are studied in terms of unimodular
trigonometric polynomials. Applications of Fourier analysis abound in signal
processing, whether with the fast Fourier transform (FFT), or filter design, or the
adaptive modeling inherent in time-frequency-scale methods such as wavelet theory.
The coherent states of mathematical physics are translated and modulated Fourier
transforms, and these are used, in conjunction with the uncertainty principle, for
dealing with signal reconstruction in communications theory. We are back to the
raison d’être of the ANHA series!

University of Maryland John J. Benedetto

College Park, MD, USA Series Editor
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Wavelets and Multiresolution Structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Classes of Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 The Invariance of the Core Subspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1 Arbitrary Hilbert Spaces and Affine Structures. . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Affine Structures with Abelian Γ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 The Classical Setting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 The Multiplicity Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1 Examples of Multiplicity Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 The Consistency Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 The Dimension Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Characterizing Multiplicity Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4 Wavelet Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2 Wavelet Sets from Multiplicity Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3 Wavelet Sets from the Consistency Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4 Wavelet Sets from Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5 Generalized Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1 Classical Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.2 Extended Uses of Classical Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.3 Generalized Filters in GMRAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.4 Building GMRAs and Parseval Wavelets from Filters in L2 (RN ). . . . 68
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

ix
x Contents

6 Fractal Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.1 GMRAs and Wavelets on Enlarged Fractal Spaces . . . . . . . . . . . . . . . . . . . 75
6.2 Generalized Filters in Fractal Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.3 Fourier Transforms on Enlarged Fractal Spaces . . . . . . . . . . . . . . . . . . . . . . . 81
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7 Composite Dilations and Crystallographic Groups . . . . . . . . . . . . . . . . . . . . . . 85
7.1 The Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2 Haar Type Wavelets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.3 Wavelet Set Wavelets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.4 Filters and Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8 Abstract Constructions of GMRAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
8.1 Super-Wavelets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
8.2 Direct Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
8.3 A Classifying Space for GMRAs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Chapter 1
Introduction

The history of wavelets is a story that demonstrates the power of collaboration

between different specialties within mathematics, physics, engineering, and com-
puter science. Many parallel beginnings of wavelets emerged during the twentieth
century, as researchers in all of these fields applied and modified Fourier’s 1807
decomposition of functions into canonical pieces. (See [21] or [42] for details.)
Applications in signal and image processing drove innovation, while theory from
harmonic analysis and theoretical physics broadened and deepened the ideas. One
early example of the fruitfulness of this interaction between application and theory
coincided with the naming of wavelets in the early 1980s. Geophysical engineer
J. Morlet and theoretical physicist A. Grossman collaborated to adapt windowed
Fourier transform functions into a tool they called a wavelet [27]. Their paper
already incorporated abstract harmonic analysis, a theme of this book, in the
form of representations of the ax + b group. A second important example of
collaboration between application and theory happened in the mid-1980s. Stéphane
Mallat brought his knowledge of pyramidal algorithms in image processing to a
collaboration with Yves Meyer, who brought insights from harmonic analysis dating
back to the work of Littlewood-Paley and Calderon. The result was a formulation
of the concept of multiresolution analyses, the ancestor of this book’s topic, and the
basis of a construction method for wavelets [34, 41].
Since these beginnings, tools from classical harmonic analysis involving the
standard Fourier transform have been an essential part of the theoretical side of the
interactions that built and refined wavelets. Abstract harmonic analysis, while not
quite so constant a presence in wavelet theory, has reappeared regularly, after being
introduced into the beginnings of the subject by Grossman and Morlet. For example,
in 1988 Feichtinger and Gröchenig [24] used group representations to unify Gabor
and wavelet transforms into a single theory [28], and later Hartmut Führ extended
their work to more general settings [25].
In 1995, Baggett, Carey, Moran, and Ohring [3] introduced a representation
theoretic approach to the study of multiresolution structures. This volume will

© Springer Nature Switzerland AG 2018 1

K. D. Merrill, Generalized Multiresolution Analyses, Applied and Numerical
Harmonic Analysis, https://doi.org/10.1007/978-3-319-99175-7_1
2 1 Introduction

describe the resulting strand of research, which is centered around the concept of
a generalized multiresolution analysis. The use of abstract harmonic analysis to
further develop this concept has been carried out by several groups during the late
1990s and into the first two decades of 2000. This work has led to more powerful and
more general methods to build new wavelets, and also to an ability to use wavelets
in broader contexts.

1.1 Wavelets and Multiresolution Structures

Historically, wavelets have used the operations of dilation and translation to

decompose functions into canonical pieces, so that each piece has a prescribed
location and level of zoom. Ideally, the pieces decay rapidly outside of a narrow
focus, and thus overcome the Fourier system’s weakness of being non-localized.
Multiresolutions have been used to break the space up according to level of zoom,
with a subspace Vj containing elements zoomed in up to level j .
The development of multiresolution structures has roots in subband filtering in
signal processing and pyramid algorithms in image reconstruction [15], and such
structures occur naturally in these and other applied fields. For example, pattern
recognition requires the ability to examine an image at different resolutions, since
patterns of varying sizes are often being sought [35]. The use of multiresolutions
in computer vision applications is encouraged by evidence showing that the
human visual system itself uses multiresolution signal processing [1]. Further,
the decomposition of a signal into a coarser component together with a detail
component at each level enables a recursive method perfectly adapted to computer
reconstruction. Conversely, the recursive structure behind dynamical systems such
as iterated function systems can be naturally adapted to build multiresolution
structures in that context. (See, for example, Chapter 6, and for a different approach,
[36].)
Wavelets and multiresolution structures were originally developed in L2 (R) with
translations by integers and √ dilation by 2, using the unitary operators2 τn f (x) =
f (x − n) and δf (x) = 2f (2x). Later, they were extended √ to L (R N ) with

translations by the integer lattice and dilation δA f (x) = | det A|f (Ax) by an
expansive (all eigenvalues have absolute value greater than 1) integer matrix A.
The work of Baggett, Carey, Moran, and Ohring [3] generalized these operations
to interrelated operators on an abstract Hilbert space. In particular, they replaced
standard translations by a countable, discrete, not necessarily abelian group Γ of
unitary operators on a separable Hilbert space H , and replaced dilation by another
unitary operator δ on H such that δ −1 Γ δ ⊂ Γ. They called such a collection of
operators an Affine Structure. In this context, they defined a wavelet:
Definition 1.1 A (orthonormal) wavelet in a Hilbert space H is a finite set
{ψ1 , ψ2 , · · · , ψL } ⊂ H such that {δ j (γ (ψi ))} forms an orthonormal basis of H ,
with −∞ < j < ∞, γ ∈ Γ, and 1 ≤ i ≤ L.
1.1 Wavelets and Multiresolution Structures 3

The authors in [3] then generalized the Mallat/Meyer definition of a multireso-

lution analysis not only to allow operators on an abstract Hilbert space, but also to
replace the requirement of an orthonormal basis of translates in the core subspace
by that of the core subspace being invariant under the action of Γ . They called
this structure simply a multiresolution; it was later given the name generalized
multiresolution analysis in [4], which is the name we will use in this book. We
will follow the convention of reserving the name multiresolution analysis for such a
structure that also has a scaling vector as defined below.
Definition 1.2 A generalized multiresolution analysis (GMRA) of a Hilbert space
H , relative to Γ and δ, is a collection {Vj }∞
j =−∞ of closed subspaces of H that
satisfy:
1. Vj ⊆ Vj +1 for all j .
2. δ(Vj ) = Vj +1 for all j .
3. ∪Vj is dense in H and ∩Vj = {0}.
4. V0 (called the core subspace) is invariant under the action of Γ .
A multiresolution analysis (MRA) is a collection {Vj }∞
j =−∞ of closed subspaces of
H that satisfy conditions 1–3 above, together with
4 . There exists a scaling vector φ ∈ V0 such that {γ φ : γ ∈ Γ } gives an
orthonormal basis for V0 .
Condition (4) allows the application of the theory of group representations to
prove theorems about the inter-relationship between wavelets and multiresolution
structures. In [3], the authors showed that all (orthonormal) wavelets have associated
GMRAs. This makes GMRAs a natural tool to pair with wavelets. While they found
an example of a GMRA that can have no associated orthonormal wavelet, they also
used an analysis of the representation of the group Γ on V0 to show that all abstract
MRAs for groups Γ where δ −1 Γ δ is of finite index d in Γ do have associated
wavelets, and established that the number L of wavelet functions must equal d − 1.
A parallel line of research was begun in 1994 by de Boor, DeVore, and Ron [22],
with the study of shift invariant spaces, spaces invariant under integer translations.
They exploited the concept of a range function, introduced by Helson [29], to
determine conditions under which a finitely generated shift invariant space has a
generating set that is orthogonal or stable. Ron and Shen [44] later introduced
techniques involving Gramian fibers, functions on RN whose values are nonnegative
Hermitian matrices with coefficients that are inner products that sum over translates
of L2 functions. They also used what they called quasi-affine systems to overcome
the fact that sets of positive dilates of wavelets are not shift invariant [45]. We will
point out connections to this parallel approach throughout the book.
In the late 1990s and early 2000s, other researchers proposed different gener-
alizations of Mallat and Meyer’s multiresolution analysis that were based on the
concept of a frame.

 {v1 , v2 , · · · }
Definition 1.3 A frame for a Hilbert space H is a collection of vectors
for which there exist positive frame bounds A and B with A v 2 ≤ i | v, vi |2 ≤
B v 2 for all v ∈ H . A Parseval frame (or normalized tight frame) is a frame with
bounds A = B = 1.
4 1 Introduction

Benedetto and Li [10] defined a frame multiresolution analysis (FMRA) in

L2 (R), in which condition (4) of Definition 1.2 is replaced by the requirement
that the subspace V0 have a frame consisting of translates of a single function.
M. Papadakis [43] later defined a more general concept of a generalized frame
multiresolution analysis (GFMRA) for an abelian group of operators acting on an
abstract Hilbert space, in which the frame for V0 is allowed to consist of the group
orbit of an arbitrary collection of functions in V0 . These authors also considered
wavelets that provide frames rather than orthonormal bases for H .
Definition 1.4 A Parseval wavelet in a Hilbert space H is a set {ψ1 , ψ2 , · · · } ⊂ H
such that {ψi,γ ,j = δ j (γ (ψi ))} forms a Parseval frame for H , with −∞ < j <
∞, γ ∈ Γ, and i ≥ 1. If, in addition, ψi,γ ,j , ψi ,γ ,j = 0 whenever j = j , then
{ψ1 , ψ2 , · · · } is called a semiorthogoonal Parseval wavelet.

Parseval wavelets allow reconstruction of elements of H by v = v, ψi,γ ,j ψi,γ ,j ,
while offering more robustness and mitigation of noise. Baggett, Courter, and
Merrill showed in [5] that a quite general class of GMRAs are guaranteed to have
Parseval wavelets. Chapter 2 will pursue these ideas by developing the properties of
GMRAs that follow from the invariance of V0 under the action of the group. Using
these properties, it will describe the relationship between GMRAs and both types of
wavelets, as well as how GMRAs relate to the other multiresolution structures.
The paper by Baggett, Medina and Merrill [4] that named GMRAs also analyzed
more fully the case where the group Γ is abelian. Here they showed the existence
of a multiplicity function and measure class that characterize the GMRA. In the
particular case of L2 (RN ), the measure class was shown to be Haar. In this case,
by comparing the representation of Γ on V0 with its representation on V1 , they
determined that the multiplicity function has to obey a consistency equation, a result
that is key to many of the applications of GMRA theory to wavelets in L2 (RN ).
In the classical setting of dilations and translations in L2 (R), the multiplicity
function has been shown to equal the wavelet dimension function [13, 46, 47],
thus tying the results obtained by this abstract approach to classical results using
the Fourier transform. Work by Bownik, Rzeszotnik, and Speegle [14] on the
dimension function, and later by Baggett and Merrill [2] on the multiplicity function,
established exactly which functions on the N −dimensional torus can be multiplicity
functions for GMRAs in L2 (RN ), thus paving the way for constructions described
below that start with a multiplicity function. The theory of the multiplicity function
is developed in Chapter 3.

1.2 Classes of Examples

As with most mathematical concepts, much of the value of the GMRA definition
lies in interesting and useful examples. All MRAs are also GMRAs, but in the
classical setting, the most interesting examples are GMRAs that are not also MRAs.
The earliest of these examples were tied to wavelet sets, sets whose characteristic
1.2 Classes of Examples 5

functions are the Fourier transforms of wavelets. The first was discovered by Journé
(see [20], p. 136) before a GMRA was defined. He found a wavelet set for dilation
by 2 in L2 (R), which he could show has no associated MRA, but which necessarily
(by Theorem 2.1) has an associated GMRA.
By the L = d − 1 result from [3] mentioned above, single wavelets in L2 (RN )
for dilations of determinant greater than 2 must be non-MRA wavelets, and the
earliest examples of these also were wavelet set wavelets. Dai, Larson and Speegle
surprised the wavelet community in the late 1990s by showing that single wavelet set
wavelets exist for all expansive dilations in all dimensions [16]. In general, wavelet
sets for expansive matrix dilations in L2 (RN ) are directly linked to the multiplicity
function of their GMRAs, so that examples of wavelet sets and GMRAs in L2 (RN )
are naturally built together. In [4], the multiplicity function was used to develop
a construction procedure for all wavelet sets in RN . Later, this procedure led to
the discovery of wavelet sets that are finite unions of convex sets for any scalar
dilation in L2 (RN ) (see [37, 40]), as well as to a partial determination of which
matrix dilations have wavelet sets with this property [39]. Chapter 4 develops the
relationship between GMRAs and wavelet sets in L2 (RN ) and shows how it leads
to these and other wavelet set results.
Although wavelet sets are useful in providing key examples that settle questions
in wavelet theory, their associated wavelets are not well-localized, and thus not
the most useful for applications. In general, it has been shown that non-MRA
(orthonormal) wavelets cannot be both well-localized and smooth. For example,
any orthonormal wavelet ψ ∈ L2 (R) such that |ψ̂| is continuous and |ψ̂(ω)| =
1
O(|ω|− 2 − ) for some positive , must have an associated MRA [30]. Thus, to
associate smooth and well-localized wavelets in L2 (RN ) with GMRAs that are not
MRAs, we must turn to Parseval wavelets.
The technique for building well-localized and smooth Parseval wavelets for
GMRAs uses generalized filters. Classical filters, periodic functions which are
multiplied by Fourier transforms, were used by Mallat and Meyer [34, 41] to build
wavelets from MRAs in L2 (R), and later by Daubechies [19] to build smooth
compactly supported wavelets in the same setting. For dilation by 2 in L2 (RN ),
A. Ron and Z. Shen [45] used filters with the Fourier transform in what they called
the Unitary Extension Principle to build Parseval wavelets from given frames of
translates for V0 . Benedetto/Li [10] and Papadakis [43] also used filters coming
from given frames in their FMRA and GFMRA settings to build Parseval wavelets.
Generalized filters in GMRAs are functions defined on the support of the
multiplicity function, and in the case of L2 (RN ), can be viewed in terms of the
Fourier transform in a way similar to classical filters. Baggett, Courter, and Merrill
[5], and later Baggett, Jorgensen, Merrill, and Packer [6], developed a theory of
generalized filters defined in terms of the multiplicity function, and used them
to build both the GMRA itself and Parseval wavelets. One result [7] was the
construction of a non-MRA Parseval wavelet with the same multiplicity function
as the Journé wavelet, yet which has a C ∞ Fourier transform and can be made to
be C r for an arbitrary integer r > 0. Smooth well-localized Parseval wavelets were
6 1 Introduction

also built with the same multiplicity functions as well-known wavelet sets in R2
[38]. Chapter 5 describes generalized filters, and shows how they are used to build
GMRAs and wavelets.
Other noteworthy examples of GMRAs have been MRAs built-in different
settings than ordinary translations and dilations in L2 (RN ). In 2006, Dutkay and
Jorgensen [23] constructed a fractal Hilbert space from the Hausdorff measure on R
associated with the Cantor set, and then built an MRA and (orthonormal) wavelet on
this space. Their results were later extended to other fractals, such as the Sierpinski
gasket [18], Sierpinski carpet [17], and fractals from non-linear iterated function
systems [12]. In [17], Parseval wavelets were also constructed for fractal spaces.
All of these constructions use generalized filters in this non-traditional setting. The
application of GMRA theory to fractal spaces is developed in Chapter 6.
At about the same time, wavelets with composite dilations were introduced by
Guo, Labate, Lim, Weiss, and Wilson [26]. These wavelets live in GMRAs on
L2 (RN ), with Γ a (not necessarily discrete) subgroup of the affine group of RN
and δ a dilation by a compatible expansive matrix. Note that unlike the previous
examples, this group Γ is not abelian. Shearlets are examples of this type of
construction that are useful in applications. Another important example of the
theory of composite dilations is the subclass of the two-dimensional crystallographic
groups consisting of those groups that split as semidirect products of translations and
finite point groups. Recently, MacArthur and Taylor [33] described how composite
dilations fit into GMRA theory, and also worked out how the action of the non-
semidirect product crystallographic groups on RN can be combined with compatible
dilations to give GMRAs and build wavelets. A related theory by Larson and
Massopust [32] replaced translations entirely by an affine Weyl group of reflections
in a family of hyperplanes, and found wavelet sets for that affine structure. Chapter 7
of this volume presents the theory of composite dilations, as well as the actions of
crystallographic groups that are not semidirect products, as examples of GMRA
theory.
The final chapter of this book, Chapter 8, looks at abstract constructions of
GMRAs using direct limits and direct sums. Both techniques use multiplicity
functions to build filters and then GMRAs, and both give new ways of looking
at the examples of GMRAs described above. The direct limit construction was first
developed by Larsen and Raeburn [31], and further explored in [8]. These authors
showed that there are fewer restrictions on which functions can be multiplicity
functions in contexts other than L2 (RN ), and used direct limits to build from
the multiplicity function both new examples and familiar ones such as the Journé
wavelet and the Cantor fractal space. Bildea, Dutkay, and Picioroaga in [11] used the
ideas of abstract GMRAs to develop what they call super-wavelets in a Hilbert space
composed of direct sums of L2 (R). Finally, Baggett, Furst, Merrill, and Packer [9]
described a direct sum method of building an abstract GMRA from any multiplicity
function and filter for which a GMRA exists. They used this construction to form a
classification space for GMRAs.
References 7

References

1. Akansu, A., Haddad, R.: Multiresolution Signal Decomposition: Transforms, Subbands and
Wavelets. Academic Press, San Diego (2001)
2. Baggett, L., Merrill, K.: Abstract harmonic analysis and wavelets in Rn . Contemp. Math. 247,
17–27 (1999)
3. Baggett, L., Carey, A., Moran, W., Ohring, P.: General existence theorems for orthonormal
wavelets, an abstract approach. Publ. Res. Inst. Math. Sci. 31, 95–111 (1995)
4. Baggett, L., Medina, H., Merrill, K.: Generalized multiresolution analyses and a construction
procedure for all wavelet sets in Rn . J. Fourier Anal. Appl. 5, 563–573 (1999)
5. Baggett, L., Courter, J., Merrill, K.: The construction of wavelets from generalized conjugate
mirror filters in L2 (Rn ). Appl. Comput. Harmon. Anal. 13, 201–223 (2002)
6. Baggett, L., Jorgensen, P., Merrill, K., Packer, J.: Construction of Parseval wavelets from
redundant filter systems. J. Math. Phys. 46, 1–28 (2005)
7. Baggett, L., Jorgensen, P., Merrill, K., Packer, J.: A non-MRA C r frame wavelet with rapid
decay. Acta Appl. Math. 89, 251–270 (2006)
8. Baggett, L., Larsen, N., Merrill, K., Packer, J, Raeburn, I.: Generalized multiresolution analyses
with given multiplicity functions. J. Fourier Anal. Appl. 15, 616–633 (2009)
9. Baggett, L., Furst, V., Merrill, K., Packer, J.: Classification of generalized multiresolution
analyses. J. Funct. Anal. 258, 4210–4228 (2010)
10. Benedetto, J., Li, S.: The theory of multiresolution analysis frames and applications to filter
banks. Appl. Comput. Harmon. Anal. 5, 389–427 (1998)
11. Bildea, S., Dutkay, D., Picioroaga, G.: MRA super-wavelets. N. Y. J. Math. 11, 1–19 (2005)
12. Bohnstengel, J., Kesseböhmer, M.: Wavelets for iterated function systems. J. Funct. Anal. 259,
583–601 (2010)
13. Bownik, M., Rzeszotnik, Z.: The spectral function of shift-invariant spaces. Mich. Math. J. 51,
387–414 (2003)
14. Bownik, M., Rzeszotnik, Z., Speegle, D.: A characterization of dimension functions of
wavelets. Appl. Comput. Harmon. Anal. 10, 71–92 (2001)
15. Bratteli, O., Jorgensen, P.: Wavelets Through a Looking Glass. Birkhäuser, Boston (2002)
16. Dai, X., Larson, D., Speegle, D.: Wavelet sets in Rn . J. Fourier Anal. Appl. 3, 451–456 (1997)
17. D’Andrea, J.: Constructing fractal wavelet frames. Numer. Funct. Anal. Optim. 33, 906–927
(2012)
18. D’Andrea, J., Merrill, K., Packer, J.: Fractal wavelets of Dutkay-Jorgensen type for the
Sierpinski gasket space. Contemp. Math. 451, 69–88 (2008)
19. Daubechies, I.: Orthonormal bases of compactly supported wavelets. Commun. Pure Appl.
Math. 41, 909–996 (1988)
20. Daubechies, I.: Ten Lectures on Wavelets. Society for Industrial and Applied Mathematics,
Philadelphia (1992)
21. Daubechies, I.: Where do wavelets come from?–A personal point of view. Proc. IEEE 84, 510–
513 (1996)
22. de Boor, C., DeVore, R., Ron, A.: The structure of finitely generated shift-invariant subspaces
of L2 (Rd ). J. Funct. Anal. 119, 37–78 (1994)
23. Dutkay, D., Jorgensen, P.: Wavelets on fractals. Rev. Math. Iberoamericana 22, 131–180 (2006)
24. Feichtinger, H., Gröchenig, K.: A unified approach to atomic decompositions through inte-
grable group representations. In: Cwikel, M., et al. (eds.) Function Spaces and Applications,
pp. 52–73. Springer, Berlin (1988)
25. Führ, Hartmut: Abstract Harmonic Analysis of Continuous Wavelet Transforms. Springer-
Verlag, Berlin, Heidelberg (2005)
26. Guo, K., Labate, D., Lim, W., Weiss, G., Wilson, E.: The theory of wavelets with composite
dilations. In: Heil, C. (ed.) Harmonic Analysis and Applications, pp. 231–250. Birkhäuser,
Boston (2006)
8 1 Introduction

27. Grossmann, A., Morlet, J.: Decomposition of Hardy functions into square integrable wavelets
of constant shape. SIAM J. Math. 15, 723–736 (1984)
28. Heil, C., Walnut, D.: Continuous and discrete wavelet transforms. SIAM Rev. 31, 628–666
(1989)
29. Helson, H.: Lectures on Invariant Subspaces. Academic Press, New York (1964)
30. Hernandez, E., Weiss, G.: A First Course on Wavelets. CRC Press, Boca Raton (1996)
31. Larsen, N., Raeburn, I.: From filters to wavelets via direct limits. Contemp. Math. 414, 35–40
(2006)
32. Larson, D., Massopust, P.: Coxeter groups and wavelet sets. Contemp. Math. 451, 187–218
(2008)
33. MacArthur, J., Taylor, K.: Wavelets with crystal symmetry shifts. J. Fourier Anal. Appl. 17,
1109–1118 (2011)
34. Mallat, S.: Multiresolution approximations and wavelet orthonormal bases of L2 (R). Trans.
Am. Math. Soc. 315, 69–87 (1989)
35. Mallat, S.: A theory for multiresolution signal decomposition: the wavelet representation. IEEE
Trans. Pattern Anal. Mach. Intell. 11, 674–693 (1989)
36. Massopust, P.: Fractal Functions, Fractal Surfaces, and Wavelets. Academic Press, Orlando
(1995)
37. Merrill, K.: Simple wavelet sets for scalar dilations in L2 (R2 ). In: Jorgensen, P., Merrill, K.,
Packer, J. (eds.) Wavelets and Frames: A Celebration of the Mathematical Work of Lawrence
Baggett, pp. 177–192. Birkhäuser, Boston (2008)
38. Merrill, K.: Smooth well-localized Parseval wavelets based on wavelet sets in R2 . Contemp.
Math. 464, 161–175 (2008)
39. Merrill, K.: Simple wavelet sets for matrix dilations in R2 . Numer. Funct. Anal. Optim. 33,
1112–1125 (2012)
40. Merrill, K.: Simple wavelet sets in Rn . J. Geom. Anal. 25, 1295–1305 (2015)
41. Meyer, Y.: Ondelettes, fonctions splines et analyses graduées. Rapport Ceremade 8703 (1987)
42. Meyer, Y.: Wavelets: Algorithms and Applications. Society for Industrial and Applied Mathe-
matics, Philadelphia (1993)
43. Papadakis, M.: Generalized frame multiresolution analysis of abstract Hilbert spaces. In:
Benedetto, J., Zayed, A. (eds.) Sampling, Wavelets and Tomography, pp. 179–223. Birkhäuser,
Boston (2004)
44. Ron, A., Shen, Z.: Frames and stable bases for shift-invariant subspaces of L2 (Rd ). Can. J.
Math. 47, 1051–1094 (1995)
45. Ron, A., Shen, Z.: Affine systems in L2 (Rd ): the analysis of the analysis operator. J. Fourier
Anal. Appl. 3, 408–447 (1997)
46. Ron, A., Shen, Z.: The wavelet dimension function is the trace function of a shift-invariant
system. Proc. Am. Math. Soc. 131, 1385–1398 (2002)
47. Weber, E.: Applications of the wavelet multiplicity function. Contemp. Math. 247, 297–306
(1999)
Chapter 2
The Invariance of the Core Subspace

The invariance of the core subspace V0 of a GMRA under the group Γ of unitary
operators implies the existence of a unitary representation of Γ on V0 , thus enabling
the use of tools from abstract harmonic analysis. Because of the required condition
δ −1 Γ δ ⊂ Γ , the invariance of V0 also gives invariance of all Vj for j ≥ 0, and
thus representations of Γ there as well. The invariance of V1 in turn implies the
invariance of W0 = V1  V0 , where the representations of Γ are useful in proving
the existence of orthonormal or Parseval wavelets. In this chapter, we will explore
results that follow from analyzing these representations of Γ . We start with the most
general setting of an arbitrary affine structure on a separable abstract Hilbert space,
and later consider the case of abelian Γ , and in particular, the classical setting of
ordinary translation and expansive matrix dilations in L2 (RN ).

2.1 Arbitrary Hilbert Spaces and Affine Structures

Let H be a separable Hilbert space with an affine structure given by a countable,

discrete group Γ of unitary operators on H , together with another unitary operator
δ, such that δ −1 Γ δ ⊂ Γ . Note that the relationship between Γ and δ implies that to
be a part of an affine structure, Γ must have a chain of subgroups {Γi } = {δ −i Γ δ i },
for which the index of Γi in Γi−1 is constant. This gives an implicit restriction on
which groups can be part of an affine structure associated with a GMRA. The map
α(γ ) = δ −1 γ δ is an isomorphism of Γ onto Γ1 , and it induces a map α ∗ on the dual
group Γ, defined by α ∗ (ω) = ω ◦ α.
We begin with two theorems from Baggett, Carey, Moran, and Ohring [3] that
establish a relationship between wavelets and GMRAs in the general case.
Theorem 2.1 Let {ψ1 , · · · ψL } be an orthonormal wavelet in a Hilbert space H
relative to the affine structure given by Γ and δ, and define Vj = span{δ k (γ (ψi )) :

© Springer Nature Switzerland AG 2018 9

K. D. Merrill, Generalized Multiresolution Analyses, Applied and Numerical
Harmonic Analysis, https://doi.org/10.1007/978-3-319-99175-7_2
10 2 The Invariance of the Core Subspace

1 ≤ i ≤ L, γ ∈ Γ, −∞ < k < j }. Then {Vj }∞

j =−∞ is a GMRA of H relative to Γ
and δ.
Proof Properties (1), (2), and the first half of (3) of Definition 1.2 follow immedi-
ately from the choice of Vj . The intersection of the Vj is trivial since any element
of Vj must be orthogonal to δ k (γ (ψi )) for k ≥ j. To see that the core subspace V0
is invariant under Γ , it will suffice to show that for k ≥ 0, each Wk ≡ Vk+1  Vk is
invariant, since V0 is the orthogonal complement of ⊕∞ k=0 Wk . The invariance of Wk
follows from

γ (δ k (γ (ψi ))) = δ k ((δ −k γ δ k )(γ (ψi ))). (2.1)



Remark 2.1 The situation for Parseval wavelets is not quite so straightforward. If
H has a semiorthogonal Parseval wavelet, then the same proof shows that H has
a GMRA. For other Parseval wavelets, it is not known whether ∩Vj = {0} always
holds, where Vj is defined as in Theorem 2.1. The answer is negative for general
frame wavelets, i.e.,{ψ1 , · · · , ψL } such that {δ j γ (ψi )} forms a frame for H . In fact,
Bownik and Rzeszotnik [8] found an example of a frame wavelet in L2 (R) with
frame bounds 1 and 1 + δ that has ∩Vj = L2 (R).
A partial converse to Theorem 2.1 follows from a comparison of the representa-
tions of Γ on V0 and V1 with the left regular representation Λ of Γ, which acts on
ξ ∈ l 2 (Γ ) by Λγ (ξ(γ )) = ξ(γ −1 γ ).
Theorem 2.2 Suppose the affine structure on a Hilbert space H is given by a group
Γ and operator δ with the property that δ −1 Γ δ is of finite index d in Γ . If there is
an MRA of H , then there exists an orthonormal wavelet {ψ1 , · · · , ψd−1 } for H .
Proof The existence of an MRA means that the action of the group Γ on V0 is
equivalent to the left regular representation Λ of Γ , with the unitary operator U :
V0 → l 2 (Γ ) by U (γ (φ)) = 1γ establishing the equivalence.
On the other hand, we claim that the action of Γ on V1 is equivalent to the
direct sum of d copies of Λ. To see this, first note that {δ(γ (φ)) : γ ∈ Γ } forms
an orthonormal basis of V1 . Let Γ = Γ1 ∪ · · · ∪ Γ d be a division of Γ into its
d distinct right cosets of δ −1 Γ δ, and define V1,i = { γ ∈Γi cγ δγ (φ)}. Write V1 =
d 
i=1 V1,i . The action of Γ on each V1,i is equivalent to Λ, since γ ∈Γi cγ δγ (φ) =
 −1 −1 −1
γ ∈Γi cγ δγ δ δ(φ), and as γ ranges over a right coset of δ Γ δ in Γ , δγ δ
ranges over a right coset of Γ in δΓ δ .−1

As before, we write Wj = Vj +1  Vj . If we let σ be the representation of Γ on

W0 , we can also write the representation of Γ on V1 = V0 ⊕ W0 as Λ ⊕ σ , obtaining


d
Λ=Λ⊕σ (2.2)
i=1
2.1 Arbitrary Hilbert Spaces and Affine Structures 11

Since Γ is discrete, the commutant of Λ, and hence also that of the direct sum of
d copies of Λ, is a finite von Neumann algebra. Thus, the cancellation property for
representations of groups
d−1 with finite von Neumann algebras applies (see [13] 3.2.3
Prop. 6), to yield σ = i=1 Λ. Since V0 is a proper subspace of V1 (else H would
be trivial), we have W
d−1 0 =
 0, so that d − 1 > 0.
Write W0 = i=1 W0,i , where the action of Γ is equivalent to Λ on each
W0,i , and denote by Ui : W0,i → l 2 (Γ ) the unitary operator that establishes this
equivalence. Let ψi = Ui−1 (1e ), where e is the identity of Γ . Then {γ (ψi ) : γ ∈
Γ, 0 ≤ i ≤ d − 1} forms an orthonormal basis for the subspace W0 , so that
{δ j (γ (ψi
)) : γ ∈ Γ, 0 ≤ i ≤ d − 1, −∞ < j < ∞} forms an orthonormal
basis for Wj = H . 

Corollary 2.1 If {ψ1 , . . . ψl } is an orthonormal wavelet associated with an MRA
in a Hilbert space H , then l = d − 1, where d is the index of δ −1 Γ δ in Γ .
Proof By the cancellation property for representations of groups with finite von
Neumann algebras referenced above,


d−1 
l
Λ= Λ (2.3)
i=1 i=1

implies that l = d − 1. 

Note that the proofs of both Theorem 2.2 and Corollary 2.1 use the assumption that
Γ is a discrete group.
Remark 2.2 An example is given in [3] to show that the hypothesis of the existence
of a scaling vector cannot be entirely removed. However, MacArthur and Taylor in
[22] were able to relax this hypothesis in the following natural way. They define a
finite scaling ensemble (FSE) to be {φ1 , · · · , φr } ⊂ V0 such that {γ φi : γ ∈ Γ, 1 ≤
i ≤ r} forms an orthonormal basis of V0 . They show that the same proof as above
establishes the existence of a wavelet {ψ1 , · · · ψr(d−1) } when the requirement of an
MRA is replaced by that of a GMRA that has an FSE.
In addition to Theorem 2.2, the authors in [3] also developed abstract general
theorems about building wavelets that belong to desirable dense subspaces of the
Hilbert space H , including one that generalizes classical results about the existence
of smooth wavelets with compact support. Both the strengths and the limitations of
Theorem 2.2 and these variations lie in their abstract generality. While the theorems
are not constructive, they can be used to build wavelets in situations where finding
an orthonormal basis for W0 from the scaling function(s) is straightforward. This
occurs, for example, when the scaling functions are characteristic functions, and is
exploited in [16] and [22] to build Haar wavelets for composite dilations and the
crystallographic groups. (See Chapter 7.) Here is a simple example of Theorem 2.2
with non-abelian Γ that fits within both the composite and crystallographic
classifications. This MRA was mentioned in [7].
12 2 The Invariance of the Core Subspace

Example 2.1 Let H = L2 (R2 ), and take Γ to be the crystallographic group pm,
which can be written as the semidirect product of a translation group and a two
element point group, Γ = Z2  {I, σy }, where σy is a reflection in the y axis. Γ is
a subset of the affine group, and acts on R2 by [n, S] · x = S(x + n), and hence on
L2 (R2 ) by
1
[n, S] · f (x) = | det S|− 2 f (S −1 x − n) = f (Sx − n) (2.4)

for n ∈ Z2 , S ∈ {I, σy }. We take δ to be dilation by 2,

δf (x) = 2f (2x), (2.5)

and observe that δ −1 [n, S]δ = [2n, S], so that δ −1 Γ δ is a subgroup of order 4 in Γ.
Let R = [0, 12 ] × [0, 1], which is a fundamental domain for the action of Γ on R2 ,
√
and define φ = 21R . If we let V0 be the closed linear span of {γ φ : γ ∈ Γ }, and
Vj = δ j V0 , we obtain an MRA for L2 (R2 ). Thus Theorem 2.2 applies to guarantee
the existence of a 3-wavelet for L2 (R2 ) relative to the action of the group Γ = pm
and dilation by 2. For this simple φ, we can find the wavelet explicitly by completing
the basis for V0 into a basis for V1 . We note that V1 consists of functions that are
constant on rectangles of width 14 and height 12 . Thus it will suffice to include, along
with {γ φ}, the Γ orbits of the three functions
√
ψ1 = 2(1[0, 1 ]×[0, 1 ] − 1[0, 1 ]×[ 1 ,1] )
2 2 2 2
√
ψ2 = 2(1[0, 1 ]×[0, 1 ] + 1[ 1 , 1 ]×[ 1 ,1] − 1[ 1 , 1 ]×[0, 1 ] − 1[0, 1 ]×[ 1 ,1] )
4 2 4 2 2 4 2 2 4 2
√
ψ3 = 2(1[0, 1 ]×[0,1] − 1[ 1 , 1 ]×[0,1] ).
4 4 2

In general, explicitly finding the wavelet from the scaling function is not so
simple. Wavelet sets and filters have often been used to carry out this construction,
particularly in the classical setting of ordinary dilations and translations on L2 (RN ).
We will discuss the generalizations of these techniques in Chapters 4 and 5.
In addition to the use of non-abelian groups in composite and crystallographic
examples such as Example 2.1, a currently active strand of research builds multires-
olution structures with Γ a non-abelian Lie group. Early work by P. G. Lemarié
[21] constructed a variant of a multiresolution analysis and spline wavelets using
a stratified nilpotent Lie group. More recently, A. Mayeli built a wavelet frame
multiresolution analysis with translation on the Heisenberg group [24]. See also
recent papers by B. Currey, A. Mayeli, and V. Oussa, e.g., [11].
Theorem 2.2 and its extension in Remark 2.2 both depend on the existence of
scaling functions, as does the traditional use of filters to build wavelets. In order to
make GMRAs more useful in building wavelets, the concept of a scaling function
2.2 Affine Structures with Abelian Γ 13

was generalized in [5, 6, 10] and [25] to include scaling functions that provide
frames instead of orthonormal bases for the core subspace.
Definition 2.1 A set of generalized scaling vectors for a GMRA on a Hilbert space
H is a countable (possibly finite) collection {φ1 , φ2 , · · · } ⊂ V0 such that {γ φi : γ ∈
Γ } form a Parseval frame for V0 .
We will show below that if we restrict our attention to abelian Γ , we have such
scaling vectors in a large class of GMRAs that includes all in the traditional setting
of L2 (RN ). By applying a similar idea to the representation of Γ on W0 , we also
will prove the existence of semiorthogonal Parseval wavelets for this large class of
GMRAs.

2.2 Affine Structures with Abelian Γ

We now use decomposition into irreducibles to look more closely at the repre-
sentations of Γ on V0 and W0 in the case of an abelian Γ . Let π be a unitary
representation of Γ on an invariant subspace V ⊂ H . When Γ is abelian, Stone’s
theorem generalized to locally compact abelian groups, or equivalently, the spectral
theorem for a commuting family of operators (see, e.g., [14] Th. 4.44), shows that
there exists a unique projection valued measure (spectral measure) p on the dual
group Γ such that for every γ ∈ Γ ,

πγ = γ (ω)dp(ω). (2.6)
Γ

To make this characterization of π more useful in our setting, we apply spectral

multiplicity theory. The idea is to describe p completely in terms of canonical
projection valued measures q, which act in L2 (ν) spaces with finite measures ν
on Γ by qE f = 1E f . Note that if K is a cyclic subspace of V for p, with cyclic
vector x of norm 1, we can define a measure ν on Γ by ν(E) = pE x, x , and
the unitary map U (1E ) = pE (x) extended to L2 (ν) will intertwine the canonical
projection valued measure on L2 (ν) with p on K.
To use this for our V , first find a maximal subspace K1 ⊂ V that is cyclic for
p. To do this, start with a countable dense subset {zi } of vectors in V , and use the
Hausdorff Maximality Principle to find a maximal linearly ordered set of cyclic
subspaces for p containing z1 . Let K1 be the closure of the union of this set of
cyclic subspaces. Then K1 has a cyclic vector y1 and associated measure μ1 defined
by μ1 (E) = pE y1 , y1 . By maximality, we have μ1 ≡ p. Let U1 : L2 (μ1 ) → K1
be the intertwining operator. Next use the same process to find a maximal cyclic
subspace K2 containing the projection onto K1⊥ of the first vector in {zi } that is not
in K1 . The associated measure for K2 will satisfy μ2 ≡ p|K ⊥ , and we will have a
1
map U2 : L2 (μ2 ) → K2 .
14 2 The Invariance of the Core Subspace

Repeating this processbreaks the separable Hilbert space V into a direct sum of
∞
cyclic subspaces V = i=1 K i , with associated measures μi and maps Ui , such
that the measures satisfy μi+1  μi . (The process may terminate in a finite number
of steps,
 so that the Ki andμi are all trivial after i = n.) The unitary operator
J = ∞ U
i=1 i
−1
maps V to ∞ 2
i=1 L (μi ). Since J intertwines the projection valued
measure p with the direct sum of canonical projection valued measures on L2 (μi ),
it intertwines the direct integral πγ with multiplication by γ (as a function on Γ) in
each component. As a result, we have the following, where μ = μ1 and Si is the
support of μi :
Theorem 2.3 Given a unitary representation π of Γ on V , there exists a unique
Borel measure class [μ] on Γ, Borel subsets 
S1 2⊇ S2 ⊇ · · · of Γ , which are unique
(a.e.μ), and a unitary operator J : V → L (Si ) (not necessarily unique) such
that for γ ∈ Γ, f ∈ V ,

[J (πγ (f ))](ω) = γ (ω)[J (f )](ω). (2.7)

Proof See [17], [19] (Sec. 1.7), or [23] (Theorem 1.21) for general proofs. See also
[9] (Chapter 3) for more details of the specific outline given above. 

Remark 2.3 Spectral multiplicity theory is often described as showing that the
projection valued measure p is uniquely determined by the measure class [μ] on Γ,
and a multiplicity function m : Γ → {0, 1, 2, · · · } ∪ {∞} defined a.e. μ. Chapter 3
will further explore the multiplicity function, which is derived from Theorem 2.3 by
∞

m= 1 Si , (2.8)
i=1

or equivalently, Si = {ω ∈ Γ : m(ω) ≥ i}.

We will apply Theorem 2.3 to the invariant subspaces V = V0 and V = W0 . We will
follow the convention of using J , μ, Si , and m for the components of Theorem 2.3
in the case of V = V0 , and J,  μ,   for these components in the case of
Si , and m
V = W0 . Also, for V = V0 , we will refer to the measure class [μ] of Theorem 2.3
as the core measure class, and the multiplicity function m of the Remark as the
core multiplicity function, or just the multiplicity function. We refer to m as the
complementary multiplicity function.
In the case where the core measure class [μ] is absolutely continuous with respect
to Haar measure, following [2, 5, 10] and [15], we can use the map J to define
generalized scaling vectors for the GMRA.
Theorem 2.4 Let {Vj } be a GMRA on a Hilbert space H with respect to an affine
structure with abelian Γ . Suppose that the core measure class [μ] is absolutely
continuous with respect to Haar measure on Γ. Using the map J , and setsSj , from
Theorem 2.3, define φj = J −1 (1Sj ), where 1Sj denotes the function in L2 (Si )
2.2 Affine Structures with Abelian Γ 15

whose j th component is 1Sj and other components are 0. Then {φ1 , φ2 , · · · } are
generalized scaling vectors for {Vj }.
Proof Write λ for normalized Haar measure on Γ. Since μ  λ, we can take
μ = λ|S1 , and in general, μ|Sj = λ|Sj . We have then that {γ 1Sj : γ ∈ Γ } is a
Parseval frame for L2 (Sj ) (see, e.g.,[18]), and thus that {γ 1Sj : γ ∈ Γ, j ≥ 1}
 2
is a Parseval frame for L (Sj ). The result then follows from the fact that J is a
unitary map that intertwines multiplication by γ with the action of πγ . 

As mentioned in Chapter 1, M. Papadakis in [25] defined an alternative mul-
tiresolution structure called a generalized frame multiresolution analysis (GFMRA).
A GFMRA is defined in the same context as GMRAs except that the group Γ is
not required to be countable or discrete. A GFMRA replaces condition (4) in the
definition of GMRA by the requirement that V0 has a frame consisting of the Γ
orbit of a countable set of vectors. Theorem 2.4 shows that in the case of countable
discrete Γ , when the core measure μ is absolutely continuous with respect to Haar
measure, the definitions of GMRA and GFMRA are equivalent.
When the measure μ of Theorem 2.3 is not absolutely continuous with respect to
Haar measure, it is possible that there exist GMRAs that are not also GFMRAs. Han
and Larson in [18] (Theorem 3.11) and Weber in [27] showed that a countable group
Γ acting on a finite set {φ1 , · · · φn } of vectors cannot give a frame for a subspace of
a Hilbert space unless the measure μ is absolutely continuous with respect to Haar
measure. Thus, a GMRA or GFMRA with core measure not absolutely continuous
with respect to Haar measure could only have generalized scaling vectors involving
an infinite collection of φi . GMRAs (for countable discrete groups) that do have
such a collection are also GFMRAs; GMRAs that do not are exactly those GMRAs
that are not also GFMRAs.
Examples of multiresolution structures in which the core measure class is not
absolutely continuous with respect to Haar are rare in the literature. In particular,
we will see in the next section that such examples cannot occur for translations by
integers and dilation by expansive matrices in L2 (RN ). Here is a sketch of a 2010
example with singular measure by Baggett [1]:
Example 2.2 Let S be the countable set
∞
1 2n − 1
S= , , (2.9)
2n 2n
n=1

and define an atomic probability measure ν on S by ν( 12 ) = 12 , and ν( 21n ) =

ν( 2 2−1
n
n ) = for n ≥ 2. Set K0 = L2 (ν), and for n ≥ 1, Kn = C2 . Define
1
∞ 2n+1
H = n=0 Kn , so that f ∈ H is of the form

f = {f0 , f1 , f2 , · · · }, fn ∈ Kn (2.10)
16 2 The Invariance of the Core Subspace

Let Γ = {T n : n ∈ Z}, where T is the operator on H given by

⎧ 2π ix
⎨e f0 (x) n=0
[T (f )]n (x) = (f1 1 , −f1 2 ) n=1 . (2.11)
⎩
fn n≥2

Define the operator D by

[D(f )]n = fn+1 for n ≥ 1, (2.12)

and
⎧
⎪
⎪ f1 1 x= 1
⎪
⎨ √2(f (2x) + f )
2
⎪
0 12 x= 1
[D(f )]0 (x) = √ 4
, (2.13)
⎪
⎪ 2(f0 (2x) − f1 2 ) x= 3
⎪
⎪
⎩√
4
2f0 (2x) else

and take δ = D −1 . Then δ −1 T δ = T 2 , so that Γ and δ form an affine structure for

H.
We define a GMRA on H relative to Γ and δ as follows: For j ≥ 0, set Vj =
j
n=0 Kn . For j < 0, we define Vj in terms of a Ruelle operator Sh ,√a class of
operators we will discuss more in Chapter 5. Specifically, we let h = 2 1S\{ 1 } ,
2
and define the isometry Sh on L2 (ν) by [Sh (f )](x) = h(x)f (2x). Then, for j < 0,
−j
set Vj = Sh (V0 ).
The map J of Theorem 2.3 can be taken to be the identity map on V0 = L2 (ν),
so that the core measure class [μ] is that of the given measure ν. See [1] for more
details.
We now apply the ideas of this section to the representation of Γ on W0 =
V1  V0 in order to prove the existence of Parseval wavelets for a large class of
GMRAs. Let [ μ] be the measure class and J : W0 → L2 (
Si ) be the map given by
Theorem 2.3 for the representation σ of Γ on W0 . The following theorem, a version
of which first appeared in [5], is the analog of Theorem 2.4 for this representation.
Theorem 2.5 Let {Vj } be a GMRA for the Hilbert space H . Suppose the affine
structure on H has an abelian group Γ whose representation σ on W0 has
associated measure absolutely continuous with respect to Haar measure on Γ. Then
H has a semiorthogonal Parseval wavelet {ψ1 , ψ2 , · · · }, where ψi = J−1 (1
Si ).
Proof The same argument as used in the proof of Theorem 2.4 shows that {γ 1 Si :
 2
γ ∈ Γ, i ≥ 1} is a Parseval frame for   −1
L (Si ). Hence, applying J , we have
that {γ ψi : γ ∈ Γ, i ≥ 1} is a Parseval frame for W0 , so that {δ j γ ψi : γ ∈ Γ, i ≥
1, j ∈ Z} is a Parseval frame for H = ∪Wj . Thus {ψ1 , ψ2 , · · · , ψL } is a Parseval
wavelet, which is semiorthogonal because Wj ⊥ Wk . 

2.2 Affine Structures with Abelian Γ 17

Here is an example from [15] of a frequency space GMRA, in which we can

explicitly construct both generalized scaling vectors and Parseval wavelets from the
maps J and J.
Example 2.3 Let E = [− 14 . 12 ) ∪ [ 34 , 1), and define Vj = L2 (2j E). It is easy to
check that {Vj } forms a GMRA for L2 (R) with Γ = Z acting by γ · f (ω) =
e2π iγ ω f (ω), and δf (ω) = √1 f ( ω2 ). We have Γ = T, which we parametrize as
2
[− 12 , 12 ). Since Γ is already acting on copies of L2 (Γ) in L2 (R) via multiplication
by characters, we can take J to be an operator that translates copies of Γ into E in
the argument of the function. Since we have two copies of parts of Γ in E, J will
map to a direct sum of two L2 (Si ) spaces with Si ⊂ Γ. The set S1 will be the set of
points in Γ that have at least one translate in E, and S2 the set of points that have
two translates in E. That is, define

1 1 1
S1 = [− , ) and S2 = [− , 0),
4 2 4
and let J (f )(ω) = f (ω)1S1 (ω) + f (ω + 1)1S2 (ω). (Recall we write 1Si for the
element of L2 (S1 ) ⊕ L2 (S2 ) whose i th component is 1Si and other component is 0.)
The core measure class is absolutely continuous with respect to Lebesgue measure
restricted to S1 , and {φ1 , φ2 } are generalized scaling vectors, where

φ1 = 1[− 1 , 1 ) , φ2 = 1[ 3 ,1) .
4 2 4

 
We have W0 = L2 (2E \ E) = L2 [− 12 , − 14 ) ∪ [ 12 , 34 ) ∪ [ 32 , 2) . Following the
same procedure as with V0 , define

 1  1 1  1 1
S1 = [− , 0), S2 = [− , − ), S3 = [− , − ),
2 2 4 2 4

and J(f )(ω) = f (ω + 2)1

S1 (ω) + f (ω)1
S2 (ω) + f (ω + 1)1
S3 (ω). The semiorthog-
onal Parseval wavelet of Theorem 2.5 is thus

ψ1 = 1[ 3 ,2) , ψ2 = 1[− 1 ,− 1 ) , ψ3 = 1[ 1 , 3 ) .
2 2 4 2 4

Example 2.3 can be used to illustrate the non-uniqueness of the operator J . We

could, for example, define
 
Jˇ(f )(ω) = f (ω + 1)1[− 1 ,0) (ω) + f (ω)1[0, 1 ) (ω) 1S1 (ω) + f (ω)1S2 (ω),
4 2

and note that Jˇ also satisfies the intertwining requirement of Theorem 2.3 for V =
V0 . The scaling vectors φi = Jˇ−1 (1Si ) would become φ̌1 = 1[0, 1 )∪[ 3 ,1) and φ̌2 =
2 4
1 1 . Similar variations are possible for J.
[− 4 ,0)
18 2 The Invariance of the Core Subspace

We will see in the next section that in the classical case, the measure class of the
representation of Zn on W0 is always absolutely continuous with respect to Haar
measure, so that semiorthogonal Parseval wavelets exist for every GMRA.

2.3 The Classical Setting

Now consider the classical wavelet setting in which H = L2 (RN ), and Γ =

ZN acts on H by translation. Here, we have Γ = TN , which we parametrize
by ω ∈ [− 12 , 12 )N , with ω(n) = e2π i n,ω for n ∈ ZN . In this case, the
operator J of Theorem 2.3 is closely related to the Fourier transform, since both
operators intertwine translation with multiplication by exponentials. This leads to
the following theorem from [4]:
Theorem 2.6 Let {Vj } be a GMRA on H √ = L2 (RN ) relative to Γ = ZN and δA ,
where γ ·f (x) = f (x −γ ), and δA f (x) = | det A|f (Ax) for an expansive integer
matrix A. Then the measure classes of the representations of Γ on V0 and W0 are
absolutely continuous with respect to Haar measure.
Proof By taking Fourier transforms, the representation of ZN on all of RN is
equivalent to multiplication by exponentials ei n,ω . On the other hand, when
we parametrize Γ as above, the regular representation Λ of Γ is equivalent to
multiplication by exponentials on L2 ([− 12 , 12 )N ). Thus, by writing RN as the
disjoint union of translates of [− 12 , 12 )N by elements of Γ , we have that the
representation of Γ on all of H is equivalent to an infinite multiple of the regular
representation of Γ . Therefore, the projection valued measure associated with the
representation of Γ on H is equivalent to Haar measure on Γ, so that the projection
valued measure associated with any subrepresentation will be absolutely continuous
with respect to Haar measure. 

Theorems 2.4, 2.5, and 2.6 together imply that GMRAs in the classical setting
always have generalized scaling vectors, and always have semiorthogonal Parseval
wavelets. Recall from Chapter 1 that a frame multiresolution analysis (FMRA),
defined by Benedetto and Li in [6] for dilation by 2 in L2 (R), adds to the
requirements of a GMRA that the core subspace V0 has a frame consisting of the
group Γ acting on a single vector φ. We see then that an FMRA is the same as
a GMRA in which the operator J of Theorem 2.3 maps V0 to a single L2 space,
L2 (μ1 ). Equivalently, we can characterize an FMRA as a GMRA in L2 (R) for
which the multiplicity function takes on only the values 0 and 1. Benedetto and Li
use filters to build wavelets from FMRAs, a process we will describe for GMRAs in
Chapter 5. They show that filters for FMRAs can be narrow band, resulting in better
signal reconstruction than with MRAs.
The FMRAs of Benedetto and Li, and the GFMRAs of Papadakis, have frames
for the core subspace specified in their definition. The multiresolution structure can
thus explicitly include particular frames that are useful for the purpose at hand. This
2.3 The Classical Setting 19

also avoids the sometimes difficult task of finding a frame for the core subspace.
While Theorem 2.4 gives a way of finding a frame using the operator J , that operator
is often elusive.
In L2 (RN ) however, the fact that the operator J and the Fourier transform both
intertwine translation with multiplication by exponentials, make the frame given by
Theorem 2.4 useful even when an explicit formula for J is hard to find. In particular,
it was shown in [2] that the Fourier transforms of the generalized scaling vectors
defined by Theorem 2.4 satisfy the following orthogonality condition, which gives a
concrete connection between these generalized scaling vectors and the multiplicity
function. The quantities evaluated in the Lemma are instances of what are called
Gramian fibers and bracket products in [12] and [26]. We will make use of this
condition to study the multiplicity function in Chapter 3, and also to build filters in
Chapter 5.
Lemma 2.1 Let {φi } = {J −1 1Si } be the generalized scaling vectors produced by
Theorem 2.4 for L2 (RN ). Then

i (ω + γ )φ
φ j (ω + γ ) = δi,j 1Si (ω). (2.14)
γ ∈ZN

for almost all ω ∈ Γ = [− 12 , 12 )N .

Proof Since both sides are in L2 ([− 12 , 12 )N ), it will suffice to show their Fourier
coefficients, cγ̃ agree. We have
  
cγ̃ ( i (ω+γ )φ
φ j (ω+γ )) = e−2π i γ̃ ,ω i (ω+γ )φ
φ j (ω + γ )dω
γ ∈Zn [− 12 , 12 )N
γ ∈ZN
 
= e−2π i γ̃ ,ω i (ω)φ
φ j (ω)dω
[− 12 , 12 )N +γ
γ ∈ZN

i , e2π i
= φ γ̃ ,· j
φ
= φi , πγ̃ φj

= 1Si , e2π i γ̃ ,·
1Sj
= cγ̃ (δi,j 1Si (ω))



The close relationship between the Fourier transform and the operator J of
Theorem 2.3 was also used by V. Furst [15] to generalize the Fourier equations that
characterize classical wavelets in L2 (R) to an abstract Hilbert space. The classical
result, attributed to Y. Meyer, states that a function ψ ∈ L2 (R), with ψ = 1 is an
orthonormal wavelet if and only if the equations
Another random document with
no related content on Scribd:
The Baharia (sailors) on board said, ‘You Washenzi (pagans, bush
people) from the interior, you will vomit yourselves to death.’ But we
came safe to Lindi after all, and said (to the sailors): ‘You mocked at
God (by saying that we should die), but we came safe to land.’”
This love of singing is characteristic of the Wanyamwezi. In the
course of my enforced detention here, I have taken many a
photographic stroll, in which my men are always eager to accompany
me. On these occasions I have to divide the small amount of
apparatus necessary to be taken with me among as many of them as
possible, so that everyone may have something to carry. It is never
very long before Pesa mbili the Mnyampara or caravan headman,
lifts up his voice—a very good one too—whereupon the chorus
promptly falls in in excellent time. I may here give a specimen of
these little marching songs:—
Kabowe kabowe ku meso; Namuki kabowe ku meso. (1)
Wambunga kabowe ku meso; Namuki kabowe ku meso.
Ki! kabowe ku meso; Wamwera kabowe ku meso.
Ki! kabowe ku meso; Wakumbwa kabowe ku meso.

(1) We shoot with our eyes—we shoot the Namuki with our eyes,
The Wambunga, we shoot them with our eyes—the Namuki, we shoot them
with our eyes;
Bang! we shoot with our eyes—the Wamwera, we shoot them with our
eyes;
Bang! we shoot with our eyes—the Wakumbwa—we shoot them with our
eyes.

To judge by the words of this song, the Wanyamwezi must be

exceedingly loyal to the German Government, for they march against
all the rebellious southern tribes in turn and annihilate them. The
Namuki are identical with the Majimaji, the insurgents of 1905–6.
The time is a frantic recitative which makes a reproduction in our
notation impossible. The exclamation “ki” conveys, according to the
unanimous testimony of Pesa mbili and the most intelligent among
his friends, the expression of the force with which the Rugaruga (the
auxiliaries) smash the skulls of the wounded enemy, even though it
should have to be done with a stamp of the heel. At every repetition
of the ki the singers stamp on the ground so that it quivers—so
completely can these peaceable Northerners throw themselves into
all the horrors of the late rising; one can almost hear the skulls crash
at every ki. This song of defiance is certainly not an original
composition of my people’s, but has been borrowed by them from
some of their tribesmen who served in the last campaign as
Rugaruga and are now lounging about Lindi out of work. I have been
obliged to engage some of these men as carriers for the march to
Masasi; they are in their whole behaviour much more decided and
defiant than my gentle grown-up children from Dar es Salam, so that
I shall be glad to get rid of them when my destination is reached. I
think the above song must belong to them.
Now that I am on the subject I will reproduce a march of the
Sudanese soldiers which in its meaning closely resembles the one
just given. This was sung into the phonograph for me by Sol
(Sergeant-Major) Achmed Bar Shemba and a couple of divisions
from the third company of the Field Force by order of that excellent
African veteran, Captain Seyfried. The little non-com. stood like a
bronze statue in front of the machine, and the gaunt brown warriors
from Darfur and Kordofan closed up behind him, as if they had been
on the drill-ground, in two ranks, each man accurately behind the
one in front. We had no little trouble in making them take up the
wedge formation necessary to produce the desired effect. The song
runs thus:—
 YAO WOMEN AT MTUA

The singers, who are principally Nubians, state that this song is in
their mother tongue, the Darfur dialect. I have not yet succeeded in
obtaining a literal translation. The general meaning of the words,
which are sung with enviable lung-power and indefatigable energy, is
somewhat as follows:
“We are always strong. The Jumbe (headman) has been hanged by
the command of Allah. Hongo (one of the insurgent leaders) has
been hanged by the command of Allah.”
Thus much as to the results of my musical inquiries so far as they
concern the foreign elements (foreign, that is to say, here at Lindi) of
the Wanyamwezi and Nubians. I have obtained some records of
ngoma songs from Yaos and other members of inland tribes, but I
cannot tell for the present whether they are a success, as I find to my
consternation that my cylinders are softening under the influence of
the damp heat, so that I can take records, but cannot risk
reproducing them for fear of endangering the whole surface. A
cheerful prospect for the future!
Very interesting from a psychological point of view is the
behaviour of the natives in presence of my various apparatus. The
camera is, at any rate on the coast, no longer a novelty, so that its use
presents comparatively few difficulties, and the natives are not
particularly surprised at the results of the process. The only
drawback is that the women—as we found even at Dar es Salam—
usually escape being photographed by running away as fast as their
legs will carry them. The cinematograph is a thing utterly outside
their comprehension. It is an enchini, a machine, like any other
which the mzungu, the white man, has brought into the country—
and when the said white turns a handle on the little black box,
counting at the same time, in a monotonous rhythm, “Twenty-one,
twenty-two, twenty-one, twenty-two,” the native may be pleasantly
reminded of the droning measures which he is accustomed to chant
at his work; but what is to be the result of the whole process he
neither knows nor cares.
 GIRLS FROM LINDI

The phonograph, on the contrary, is an enchini after the very

heart, not only of the black man, but even of the black woman. If I
should live to the age of Methuselah the scene in Mr. Devers’s
compound at Dar es Salam will always remain one of the most
delightful recollections connected with my stay in Africa. After
spending some time in the native quarter, watching the dances of
various tribes—here a Manyema ngoma, there one of the Wazaramo,
or yonder again that of some coast people’s club, and observing the
costumes of the performers, sometimes hideous but always
picturesque, I returned to my own quarters, at the head of a
procession numbering some hundreds of the dancers, male and
female, in order to take down the audible part of the proceedings.
Everything had gone off in the most satisfactory way; but every time
I changed the diaphragms, took out the recorder and put in the
reproducer, when the full-voiced melody poured forth from the
mysterious funnel in exactly the same time and with the precise
timbre which had been sung into it—what measureless and at the
same time joyful astonishment was painted on the brown faces, all
moist and shining with their exertions in singing and dancing!
Whenever this happened, all the more unsophisticated souls joined
in the chorus, to be speedily enlightened by the derisive laughter of
the more “educated” element.
But the most delightful instance of naiveté came at the close of the
proceedings, after I had used up my small stock of Swahili idioms in
expressing my pleasure at a successful afternoon. Two women, who
had previously attracted my notice by their tremendous vocal power,
as well as by the elegance of their attire, came forward again; and, as
the crowd fell back, leaving a clear space in front of the phonograph,
first one and then the other approached the apparatus, dropped a
curtsy in the finest Court style, and waving her hand towards the
mouthpiece said, “Kwa heri, sauti yangu!”—“Good-bye, my voice!”
This incident illustrates the way in which the native mind cannot get
away from what is most immediately obvious to the senses. In the
very act of uttering their farewell, these two women could hear for
themselves that they had not lost their voices in the least, and yet
because they had a moment ago, heard them distinctly coming out of
the phonograph, they regarded themselves as deprived of them from
that instant, and solemnly took leave of them.
As to my inquiries into the artistic aptitudes of the natives, I prefer
to give the results in a connected form later on, when I shall have
brought together a larger amount of material on which to form a
judgment. So much, however, I can say even now: c’est le premier
pas qui coûte is true, not only for the executant artist but also for the
investigator. At Dar es Salam, the matter was simpler. My “boy”
Kibwana (literally, “the Little Master”), a youth of the Wazegeju tribe
from Pangani, though, like Omari the cook (a Bondei from the north
of the colony), he had never had a pencil or a piece of paper in his
hand before, had been too long in the service of Europeans to
venture any objections when desired to draw something for me—say
the palm in front of my window, or my piece of India-rubber. He set
to work, and cheerfully drew away, with no anxieties as to the artistic
value to be expected from the result.
In the case of my Wanyamwezi, with whom I have made a
beginning here, in order to give them something to do, a mere order
is of little use. If I put a sketch-book and pencil into the hands of one
of my followers with the invitation to draw something, the inevitable
answer is a perplexed smile and an embarrassed “Sijui, bwani”—“I
don’t know how, sir.” Then one has to treat the man according to his
individuality—with an energetic order, or a gentle request; but in
every case I found that the best plan was to approach him on the side
of his ambition. “Why, you’re a clever fellow, you know—a mwenyi
akili—just look at your friend Juma over there—he is not nearly as
clever as you—and yet, see how he can draw! Just sit down here and
begin drawing Juma himself!” This subtle flattery proved irresistible
to all but a few, who, despite everything I could say by way of
encouragement, stuck to it that they could not do what was wanted.
The rest are like the lion who has once tasted blood: they are
insatiable, and if I had brought two dozen sketch-books with me,
they would all be continuously in use. I found that, instead of leaving
the beginner to choose his own subject, it was a better plan (as it is
also educationally a sounder one) to suggest in the first instance
something quite familiar—a Nyamwezi hut, a fowl, a snake, or the
like. Then one finds that they set to work with some confidence in
themselves, and that they are inordinately proud of their
masterpieces, if their mzungu gives them the smallest word of praise.
It is obvious that I should never dream of finding fault—my object
being, not criticism with a view to improvement, but merely the
study of the racial aptitudes and the psychological processes involved
in artistic production.

RUINED TOWER, LINDI (BUILT BY THE PORTUGUESE)

My way of getting at the latter is to stipulate that each of my

draughtsmen, as soon as he feels that his degree of proficiency
entitles him to a reward, is to show me his work. Then comes a
shauri[5] usually of long duration, but extremely amusing for both
parties. “What is this?” I ask, pointing with my pencil to what looks a
perplexing complication of lines. “Mamba—a crocodile,” comes the
answer, either with a slight undertone of indignant astonishment at
the European who does not even know a crocodile when he sees it, or
somewhat dejectedly on finding the work to be so unsuccessful that
even the omniscient mzungu cannot tell what it is meant for. “Oh! a
crocodile—very good!” I reply, and write the word beside the
drawing. “Yes,” the artist never fails to add, “but it is a mamba of
Unyamwezi,” or “of Usagara,” or “in the Ngerengere,” as the case
may be. One is brought up short by this information, and asks,
“Why? How so?” and then comes a long story in explanation. This is
a crocodile which the artist and his friends (here follow their names
in full), saw on the march from Tabora to the coast with such and
such a European, and which came very near being the death of him
at the crossing of such and such a swamp, or of the Ngerengere river.
When writing down the first few of these commentaries, I did not
pay any special attention to the fact of their always being connected
with a particular incident; but now, after having acquired a large
collection of drawings representing either single objects (animals,
plants, implements, etc.), or scenes from native life, it has become
clear to me that the African is incapable of drawing any object in the
abstract, so to speak, and apart from its natural surroundings—or
indeed from some particular surroundings in which he has met with
it on some particular occasion. If he is told to draw a Mnyamwezi
woman he draws his own wife, or at any rate some relative or
personal acquaintance, and if he is to draw a hut, he proceeds in
exactly the same way, and depicts his own or his neighbour’s. Just so
with the genre pictures, which are not such in our sense of the word,
but might almost be termed a species of historical painting. I have
already a whole series of sketches representing a lion springing on a
cow, or a hyæna attacking a man, or some similar scene from the life-
struggle of the higher organisms, and the explanation is always
something like this:—“This is a lion, and this is a cow, but the cow
belonged to my uncle and the lion carried it off about four years ago.
And this is a hyæna, and this man is my friend—say, Kasona—who
was taken ill on the march from Tabora to Mwanza and had to stay
behind, and the hyæna came and was going to bite him, but we drove
it away and saved Kasona.”
These are only one or two specimens of my methods and results. I
am convinced that I am on the right tack, though no doubt I shall
make many mistakes and need much additional experience.
 My dynamometer, which did such excellent service on board the
Red Sea steamer in promoting friendly international relations, has
not lost its virtue here. When I am at the end of my resources for
amusing my men and the friends whom they have gathered round
them since our arrival in Lindi, I put the steel oval into the hand of
honest Pesa mbili, who, of course, must have the precedence in
everything. He presses it, and then, with the whole troop of his black
friends crowding round, gazes with the greatest excitement at the
dial, as if he could read the mysterious signs engraved on the brass
arc. When I have glanced at the scale and announced the result—of
course the numbers only, as the kilogrammes would merely serve to
perplex them—it is received with a certain quite comprehensible
feeling of doubt; they do not yet know if the number means much or
little, having no standard of comparison. The second man begins to
excite interest; if, instead of his predecessor’s 35 kilogrammes, he
can only reach 30, he is greeted by mildly derisive laughter, but if he
excels his rival, he is a mwenyi nguvu—a strong man, worthy of the
tribute of admiration which he receives with smiling dignity.
So each man takes his turn, and they will go on for hours without
tiring. One thing only is felt by the more intelligent to be wanting—it
interests them to know which among themselves is the strongest or
weakest, but in order to get a higher and absolute standard of
comparison, they are all eagerness to know what their lord and
master can do. Of course I am willing to oblige them, at the close of
the meeting, and press the instrument, first in my right hand and
then in my left. When they hear the result (which, to my great
satisfaction, requires no cooking), a unanimous “A-ah! bwana
mkubwa!” bursts from the admiring circle—literally, “Ah! Great
master!”—but about equivalent to, “What a giant you are for
strength!”
 UNDER THE PALMS

In fact we Europeans, as far as the spontaneous putting forth of

strength goes, are as giants compared with the African. I made fairly
careful records of the figures for each man, not once only, but in
several successive trials, so that no allowance need be made for
novelty or want of practice, but how inferior they are to us! None of
them could compass a greater pressure than 35 kilos with the right
hand and 26 with the left, with the exception of one man who
attained to something over 40 kilos; while I, even here in the damp
heat of the coast region can still manage over 60 with the right and
over 50 with the left. And yet nearly all my men are professional
carriers, sturdy fellows with tremendous chest-measurement, broad
shoulders and splendidly developed upper arm muscles. What they
lack, as has so often been pointed out, is the power of concentrating
the strength of the whole body at a given moment of time. These very
Wanyamwezi are famous for their almost incredible powers of
endurance.
The natives thus, as a whole, indisputably present a picture not
without attractions from a psychological point of view; but in the six
weeks or so which I have by this time spent on the coast, the
Europeans have appeared to me almost more interesting still. Dar es
Salam is so large and contains so many of our race that the new-
comer does not have the contrasts between black and white forced on
his notice, while the contrasts to be found among the white
population are less observable on the wider field of a large
settlement. Lindi, being very much smaller, leaves no room for either
possibility; in the narrowness of its environment and the monotony
of its life, there is nothing to modify the shock of contrasted and
clashing individualities, and in such a place one sees with startling
clearness the enormously powerful and rapid effect of residence in
the tropics on the mental balance of a foreign race. It does not belong
to my office to point to the—to say the least of it, curious—
excrescences of our German class and caste spirit, which here, in a
circle of Europeans numbering a dozen or less, brings forth
singularly unpleasant fruits. I need not relate how the military
element, recently “dethroned” by the establishment of a civil
administration, looks down with a superior smile on the officials of
that administration, or how the intrusion of the personal element
into affairs cuts off every possibility of social intercourse, and, what
is worse, of cordial cooperation in common work. To the new-comer,
expressing his astonishment at such a state of things, old residents
say (with a coolness contrasting strangely with their usual state of
chronic irritation): “What do you expect?—this is not the only place
where things are so—you will find it the same everywhere!” So it
seems to be, if I may judge by all I have heard during these
instructive weeks; but one may hope that this disagreeable
phenomenon is only one of the many infantile diseases incidental to
the early stages in the life of every colony. One thing, however, which
I absolutely fail to understand is the furious fits of rage to which
every white man who has lived long in the country appears to be
subject. I am doing my best in the meantime to go on my way
without calling of names or boxing of ears, but everyone is agreed in
assuring me that I shall learn better in the course of the next few
months. I cannot judge for the present whether life is really
impossible without thrashing people—but I hope it is not the case.
In order not to dwell exclusively on the darker traits characteristic
of Europeans in the tropics, I must mention the admirable gifts of
household management possessed by most of them. Dar es Salam is
so far a centre of civilization as to possess bakers, butchers, and
shops of all kinds in plenty, yet even there I fancy that the office of
mess president is by no means a sinecure. But who shall describe
how the unlucky bachelor in a remote coast town has to rack his
brains in order to set before his messmates—not merely something
new, but anything at all! Only experience can teach how far in
advance one has to provide for all the thousand-and-one trifles
which are inseparable from our housekeeping. The price alone makes
it impossible to depend to any great extent on tinned goods, and it
becomes necessary to have sufficient stores on hand to last for days—
sometimes for weeks and months, and, in addition, to concoct
eatable dishes out of the wild herbs which the cook and kitchen-boy
bring in. On the coast some variety is secured by the abundance of
good fish; in the interior this resource fails. And when it happens—as
it does just now—that even the standard typical bird of Africa, the
domestic fowl, and its product, the egg, are not to be had, then the
case is desperate indeed, and catering for a large number of people
becomes a serious problem.
It is remarkable, however, how skilled even the most inveterate
bachelors among the German residents are in solving this problem—
not always with elegance, and certainly not always to the satisfaction
of their critical predecessors in office, but yet so as to fill the novice
at any rate with astonished admiration. Dr. Franz Stuhlmann, who
accompanied Emin Pasha on his last disastrous journey—a
thoroughly competent ethnographer and the guardian and cherisher
of the African plant-world, so far as it can be adapted to the service of
man—has long been a celebrity in the culinary department
throughout the whole Colony. Stuhlmann has the reputation of being
able to prepare a dainty dish from every weed that grows beside the
native path; he is a walking encyclopædia of tropical cookery. Others
are less proficient than this, but I cannot yet get over my
astonishment at the way in which Captain Seyfried, for instance, can
produce something eatable out of the most elementary ingredients,
at his achievements in salting and pickling, at the unimpeachable
jellies he contrives to serve up even at the present temperature, and
at the variety which always characterizes his bill of fare.
I must here make an end, once for all, of one fallacy prevalent at
home. “Why, you surely cannot eat anything in that heat!” is a
remark which never fails to occur in any conversation having the
tropics for its subject, but which betrays a complete misconception of
the conditions. In the first place, the heat is not so unendurable as
commonly supposed by us—at any rate during the dry season, on the
coast, where a fresh sea-breeze always blows by day. But, in addition
to this, the waste of tissue goes on much more rapidly in tropical
than in temperate climates. Not even the new-comer is surprised to
see “old Africans” consuming an extensive “first breakfast” at a very
early hour, in which various preparations of meat figure, though fruit
is also conspicuous. At midday even a minor official never thinks of
less than two courses and dessert, and in the evening after office
hours, all ranks and professions go in for a repast which at home
would certainly rank as a public banquet. This seemingly luxurious
mode of life, however, by no means deserves the reprehension one
may feel inclined to bestow on it. On the contrary, it is
physiologically both justifiable and necessary, if the body is to offer
permanent resistance to the deleterious influence of the climate. The
new-comer is not surprised by the appetite of others because,
unconsciously, he shares it. Personally, though I wield quite a
creditable knife and fork at home, my performances out here would
make me the terror of most German housewives.
The only article of diet I do not get on with is alcohol. At home I
can appreciate a glass of beer or wine, and on board the Prinzregent
we passengers levied a pretty heavy toll on the supplies of
“Münchener” and “Pilsener”; but since I landed in this country I have
taken no beer at all and wine only in very small quantities, while I
have been quite unable to acquire a taste for whisky and soda, the
national drink of all Germans in East Africa. Such abstinence is
easily understood at Lindi, where there is no ice to be had; but even
at Dar es Salam, where Schultz’s brewery supplies the whole town
with ice every day, I found I had no taste for alcoholic beverages.
This is a great advantage as regards my journey into the interior, as I
am saved the inconvenience of taking loads of bottles with me.
I am glad to say that my enforced detention on the coast is nearing
its end. Commissioner Ewerbeck, who returned from the interior a
few days ago, is most kindly willing to start again with me to-
morrow, so as to escort me with a detachment of police through the
Wamwera country—the scene of the late rising—as far as Masasi. He
has still work to do in the Central Lukuledi Valley, for, though most
of the insurgent leaders have long ago been captured and adorn the
streets of Lindi in the shape of chain-gangs, the pursuit of others is
still going on and will yet cost many a shauri. From Masasi, Mr.
Ewerbeck will have to return immediately to Lindi, in time for the
formal reception of the delegates from the Reichstag, who are to visit
the south of the Colony next month, on their much-discussed tour
through East Africa.
My first glimpse of the interior, by the bye, has hardly been a
pleasant one. In the course of the riding-lessons which Captain
Seyfried has been giving me, we one evening made an excursion to
the Kitulo. This is a long, fairly precipitous range of heights, about
570 feet above sea-level, rising immediately behind Lindi and
separating the narrow sandy plain on which the town stands from
the back country. A landmark of our civilization—a tower built for
the sake of the view—was, some years ago, erected on the top of this
Kitulo. When I ascended it by the help of a somewhat decrepit
ladder, the sun had already set, and the whole western landscape—
precisely the part of the Dark Continent which I wish to penetrate
within the next few days—lay extended before me as a dark,
menacing shadow. For one moment my mind was clouded by gloomy
forebodings, but I speedily recalled my old luck which has never yet
forsaken me. “Never mind—I’ll get the better of you yet!” I
exclaimed, sotto voce, as I lit a new cigar with the utmost philosophy,
and mounted my mule for the return journey.
THE LIKWATA DANCE BEING PHOTOGRAPHED BY THE
AUTHOR. DRAWN BY PESA MBILI, THE MNYAMWEZI
HEADMAN
 CHAPTER IV
FIRST IMPRESSIONS OF THE INTERIOR

Masasi, July 20, 1906.

Few people, I fancy, will know where Masasi is, yet those interested
in the Colonies might well be acquainted with its situation, for in its
own small way it is quite a civilizing centre. The English Mission[6]
has been at work here for nearly the third of a century, and, since the
suppression of the rebellion, a native corporal with a dozen black
German soldiers has been gallantly maintaining his ground, in a
boma specially built for the purpose, in case of any renewed warlike
impulses on the part of the interior tribes.
I preferred to take up my quarters with the soldiers, not from any
hostility to religion, but because the two clergymen at the mission
station, about an hour’s walk from us, are both advanced in years,
and it would be unfair to trouble them with visitors. Besides their
station was burnt down during the rebellion, so that they are leading
for the moment a more idyllic than agreeable life in their former
cattle-shed. In spite of this, the two old gentlemen, as I had every
opportunity of convincing myself in the course of two long visits,
enjoy extraordinarily good health. Archdeacon Carnon, the younger
of the two, in particular, took as lively an interest in the German
Emperor and his family as if he lived in a London suburb, instead of
in a negro village at the ends of the earth. Canon Porter seems to be
failing a little, but this is only to be expected as he is getting on for
eighty and has been in the country nearly thirty years.[7] In former
days I understand that he studied the ethnology of his district
(inhabited by Wanyasa, Wayao, and Wamakonde) very thoroughly,
so that up to yesterday I had great hopes of profitable results from
my intercourse with him and his more active colleague. But in this I
was disappointed. At the ceremonious, and, I must say, sumptuous
breakfast which the two clerical gentlemen set before us two
worldlings, Ewerbeck and me, whenever I began to speak about the
inhabitants of the neighbourhood and their tribal affinities, the
conversation was invariably diverted towards the Emperor and his
family! He must have made a truly extraordinary impression on
other nations.
However, our business is with the native African, not with the
white intruder, even though he should come in the peaceful guise of
the missionary.
My landing at Lindi of itself implied the main course of my
journey. A glance at the map of East Africa shows that the extreme
south-eastern corner of our colony, considered with regard to
population, stands out like an island from the almost uninhabited
country surrounding it. The region north of the Middle, and partly
also of the Upper Rovuma is (as Lieder, the geologist, whose early
death is such a loss to science, described it) a silent pori for hundreds
of miles, extending far beyond the Umbekuru and into the hinterland
of Kilwa—an uninhabited wilderness, where not a single native
village speaks of the large and peaceable population found here by
Roscher, Livingstone and Von Der Decken nearly half-a-century ago.
Only a narrow strip running parallel to the coast some distance
inland connects this island of population with the north, while
another, much more scantily peopled, runs up the Rovuma to the
Nyasa country.

MAKUA WOMEN FROM THE LUKULEDI VALLEY

Being thus cut off from surrounding tribes, the south-east—i.e., the
Makonde Plateau, the Lukuledi Valley north of it, and the wide plain
to the west of these highlands—forms a compact, well-defined whole,
 an ideal sphere of work for one who, like
myself, has only a limited time at his disposal,
but wishes the work done in this time to be as
far as possible complete. The Wamwera,
whom I had in view in the first instance, have
had, to my great regret, to be postponed for
the present. I left Lindi on July 11th, with the
Imperial District Commissioner, Mr.
Ewerbeck. Ngurumahamba, the first
noticeable place on the Lukuledi road, still
bears the impress of the Coast—there is even a
stone house among the huts of the Waswahili;
but on the second day we reach the Yao tribe
at Mtua. Here we first come in touch with the
far interior, for these are the advance guard of
the great migration which brought this
vigorous and energetic race about the middle
of the last century from its old home south-
A MAN OF THE
MWERA TRIBE AND A east of Lake Nyasa towards the shores of the
YAO Indian Ocean, and which is still going on. As
to the way in which these migrations are
accomplished, we are apt to be misled by the
picture—no doubt a very incorrect one—which has remained in our
minds from our school-days, in connection with the migration par
excellence—the great westward movement of our own forefathers.
We think of men, horses, and waggons, a dense, compact wave of
people, rolling on slowly but irresistibly across the countries lying in
its track. Here we find nothing of the sort. It is true that these Mtua
Yaos are not typical of their tribe in this respect, as they were rescued
from the Wangoni, further north, on the eastern shore of Nyasa,
about ten years ago by Captain Engelhardt, and transferred to this
settlement. But otherwise the immigration of foreign (though still
African) elements takes place, here in the south, quietly and almost
imperceptibly—a band, a horde, a group of families, sometimes, but
not always, under the command of a chief, appears one fine day, hoes
a piece of land at a suitable place in the pori, builds a few airy huts,
and the immigration is complete. Conflicts, more or less sanguinary,
between the aborigines and the intruders may have occurred—may
even have been the rule—in former times; nothing of the kind seems
to happen to-day. Whether the native has become more tolerant, or
the firm hand of the German Government, to whom every accession
of population must be welcome, has produced a change in his views,
I am compelled to leave undecided.
In outward appearance these Yaos can scarcely be distinguished
from the Swahilis of the coast. The women are dressed in precisely
the same kind of kanga (calico printed in brightly-coloured patterns,
and manufactured in Holland), as the Coast women, though not so
neatly and fashionably as the girls at Dar es Salam, where the
patterns in vogue change faster than even at Paris. They also wear
the same coquettish little pin in the left nostril as the Coast ladies. Of
Indian origin, this kipini, called chipini in Yao, has conquered the
whole east coast of Africa, and is spreading, as a symbol of higher
culture and refinement, among the more progressive tribes of the
interior. In its simplest form a mere cylinder of pith, the better
specimens are made—according to the means of the wearer—of
ebony, tin, or silver. The ebony pins are almost always very tastefully
inlaid with tin. To our notions, the chipini hardly beautifies the
human countenance; but once the beholder is accustomed to its
effect, it becomes quite pretty and attractive, lending a coquettish
touch to the brown face it adorns.

RUINS OF NYANGAO MISSION STATION

 The more distant hinterland inhabited by the Wamwera contrasts
very unfavourably with the well-cultivated zone near the coast. The
condition of Nyangao, the Benedictine Mission station, is a symptom
of all the misery which the rebellion so short-sightedly conjured up
by the natives has brought on this part of Africa. Up to the summer
of 1905, the Fathers and Sisters here were peacefully engaged in their
work of evangelizing and teaching, when the poison of the majimaji
(magic water) idea spread to the Rondo Plateau and the central
Lukuledi Valley. Before the unsuspicious missionaries had even any
thought of coming disaster, it was already upon them. After fighting
desperately for their lives, and losing one of the Sisters, the whole
staff had to fly, and all the extensive buildings were destroyed by the
rebels. The present state of Nyangao is shown in the accompanying
photograph. Three of the Fathers (whose acquaintance I had the
pleasure of making on board the Prinzregent), have ventured back to
their old station, and, living in the house formerly occupied by the
Sisters, surrounded by heaps of ruins, have courageously and
indefatigably taken up their work once more.
The Majimaji rebellion still forms the principal topic of
conversation at native camp-fires, though the Lindi District has long
been at peace again. Its origin belongs to the most interesting
phenomena in military history, showing, as it does, the general and
almost instantaneous amalgamation of the severed fragments of a
race under the influence of a superstitious notion, once it has gained
a hold and welded them into a unit animated by a common and
fervid enthusiasm. So far as one can gather at present, the idea
underlying the rising was that of shaking off the white man’s yoke by
means of a concerted effort on the part of the whole native
population. Without dawa, i.e., charms of some sort, such a rising
would have been difficult, if not impossible to bring about, and thus
the instigators of this disastrous war had recourse to the dawa of the
“magic water.” As to this, several versions are current. According to
one, the real ringleader was a man living near the Pangani Rapids on
the Rufiji, who taught that he was commissioned by the Almighty,
and communicated with Him by means of a serpent which had its
abode in the river. This serpent had told him to make all the men
drink the water of the hot springs at Kimambare, which would give
them strength and courage to drive the Germans into the sea, and at
the same time render them invulnerable to European bullets.