(Download PDF) Inequalities in Analysis and Probability 3Rd Edition Odile Pons Ebook Online Full Chapter

Inequalities In Analysis And Probability
3rd Edition Odile Pons

Visit to download the full and correct content document:
https://ebookmeta.com/product/inequalities-in-analysis-and-probability-3rd-edition-odil
e-pons/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Analysis and Differential Equations 2nd Edition Odile

Pons
https://ebookmeta.com/product/analysis-and-differential-
equations-2nd-edition-odile-pons/
Functional Estimation for Density Regression Models and

Processes 2nd Edition Odile Pons
https://ebookmeta.com/product/functional-estimation-for-density-
regression-models-and-processes-2nd-edition-odile-pons/
Geometric Analysis of Quasilinear Inequalities on

Complete Manifolds Frontiers in Mathematics Bruno
Bianchini
https://ebookmeta.com/product/geometric-analysis-of-quasilinear-
inequalities-on-complete-manifolds-frontiers-in-mathematics-
bruno-bianchini/
Introduction to Linear Regression Analysis (Wiley

Series in Probability and Statistics) 6th Edition
Montgomery
https://ebookmeta.com/product/introduction-to-linear-regression-
analysis-wiley-series-in-probability-and-statistics-6th-edition-
montgomery/
Probability and Statistical Inference 3rd Edition
Magdalena Niewiadomska-Bugaj
https://ebookmeta.com/product/probability-and-statistical-
inference-3rd-edition-magdalena-niewiadomska-bugaj/
Probability and Statistical Inference 3rd Edition

Magdalena Niewiadomska-Bugaj
https://ebookmeta.com/product/probability-and-statistical-
inference-3rd-edition-magdalena-niewiadomska-bugaj-2/
Probability (Schaum's Outline Series, 3rd ed.) 3rd

Edition Seymour Lipschutz
https://ebookmeta.com/product/probability-schaums-outline-
series-3rd-ed-3rd-edition-seymour-lipschutz/
Women in Science and Technology: Confronting

Inequalities First Edition Gupta
https://ebookmeta.com/product/women-in-science-and-technology-
confronting-inequalities-first-edition-gupta/
Statistics for Data Scientists: An Introduction to

Probability, Statistics, and Data Analysis Maurits
Kaptein
https://ebookmeta.com/product/statistics-for-data-scientists-an-
introduction-to-probability-statistics-and-data-analysis-maurits-
kaptein/
INEQUALITIES IN
ANALYSIS AND
PROBABILITY
Third Edition
B1948 Governing Asia
This page intentionally left blank
B1948_1-Aoki.indd 6 9/22/2014 4:24:57 PM

INEQUALITIES IN
ANALYSIS AND
PROBABILITY
Third Edition
Odile Pons
National Institute for Agronomical Research, France
World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TAIPEI • CHENNAI • TOKYO
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
Library of Congress Cataloging-in-Publication Data

Names: Pons, Odile, author.
Title: Inequalities in analysis and probability / Odile Pons,
National Institute for Agronomical Research, France.
Description: Third edition. | New Jersey : World Scientific Publishing Co. Pte. Ltd., [2022] |
Includes bibliographical references and index.
Identifiers: LCCN 2021031175 | ISBN 9789811231346 (hardcover) |
ISBN 9789811231353 (ebook)
Subjects: LCSH: Inequalities (Mathematics)
Classification: LCC QA295 .P66 2022 | DDC 512.9/7--dc23
LC record available at https://lccn.loc.gov/2021031175
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
Copyright © 2022 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
For any available supplementary material, please visit

https://www.worldscientific.com/worldscibooks/10.1142/12135#t=suppl
Printed in Singapore
LaiFun - 12135 - Inequalities in Analysis and Probability.indd 1 6/8/2021 9:06:25 am

September 6, 2021 8:31 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page v
Preface
The third edition proves new inequalities for martingales and their supre-
mum; thus improving the classical inequalities. These are presented in two
different chapters, for discrete and time-continuous martingales. They are
applied to the weak convergence of (local) martingales and semi-martingales
and to Gaussian and point processes. The chapter on stochastic integration
is completed and new results are established for local times.
February 2021
Preface of the Second Edition. The most important changes made

in this edition are the insertion of two chapters: Chapter 5 on stochastic
calculus with first-order differentiation and exponential (sub)-martingales,
and Chapter 7 on time-continuous Markov processes, the renewal equations
and the Laplace transform of the processes. I have added to Chapter 4 a
section on the p-order variations of a process and modified the notations
to extend the inequalities to martingales with a discontinuous process of
quadratic variations, the notations of the previous edition were not modi-
fied for a continuous martingale and for a point process with a continuous
predictable compensator. I have also corrected a few misprints and errors.
Examples of Poisson and Gaussian processes illustrate the text and their
investigation leads to general results for processes with independent incre-
ments and for semi-martingales.
February 2016
Preface of the First Edition. The inequalities in vector spaces and

functional Hilbert spaces are naturally transposed to random variables,
martingales and time indexed stochastic processes with values in Banach
spaces. The inequalities for transforms by convex functions are examples
v
April 30, 2021 13:24 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page vi
vi Inequalities in Analysis and Probability (Third Edition)
of the diffusion of simple arithmetic results to a wide range of domains

in mathematics. New inequalities have been developed independently in
these fields. This book aims to give an account of inequalities in analysis
and probability and to complete and extend them.
The introduction gives a survey of classical inequalities in several fields
with the main ideas of their proofs and applications of the analytic inequal-
ities to probability. This is not an exhaustive list. They are compared
and sometimes improved with simple proofs. Further developments in the
literature are mentioned. The book is organized according to the main con-
cepts and it provides new inequalities for sums of random variables, their
maximum, martingales, Brownian motions and diffusion processes, point
processes and their suprema.
The emphasis on the inequalities is aimed at graduate students and re-
searchers having the basic knowledge of courses in Analysis and Probability.
The concepts of integration theory and of probabilities are supposed to be
known, so the fundamental inequalities in these domains are acquired and
references to other publications are added to complete the topic whenever
possible. The book contains many proofs, in particular basic inequalities
for martingales with discrete or continuous parameters in detail and the
progress in several directions are easily accessible to the readers. They are
illustrated by applications in probability.
I undertook this work in order to simplify the approach of uniform
bounds for stochastic processes in functional classes. In the statistical ap-
plications, the assumptions for most results of this kind are specific to
another distance than the uniform distance. Here, the results use inequal-
ities of Chapter 4 between the moments of martingales and those of their
predictable variations, then the conditions and the constants of the proba-
bilistic bound differ from those of the other authors. During the preparation
of the book, I added other inequalities while reading papers and books con-
taining errors and unproved assertions; it should therefore fill some gaps.
It does not cover the convex optimization problems and the properties of
their solutions. It can be used as an introduction to more specific domains
of the functional analysis or probability theory and as a reference for new
applications to the asymptotic behaviour of non-standard empirical pro-
cesses in statistics. Several applications to the tail behaviour of processes
are developed in the following chapters.
Odile M.-T. Pons
April 2012
April 30, 2021 13:24 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page vii
Contents
Preface v
1. Preliminaries 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Cauchy and Hölder inequalities . . . . . . . . . . . . . . . 2
1.3 Inequalities for transformed series and functions . . . . . . 6
1.4 Applications in probability . . . . . . . . . . . . . . . . . . 9
1.5 Hardy’s inequality . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Inequalities for discrete martingales . . . . . . . . . . . . . 16
1.7 Martingales indexed by continuous parameters . . . . . . 21
1.8 Large deviations and exponential inequalities . . . . . . . 27
1.9 Functional inequalities . . . . . . . . . . . . . . . . . . . . 31
1.10 Content of the book . . . . . . . . . . . . . . . . . . . . . 33
2. Inequalities for Means and Integrals 37

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2 Inequalities for means in real vector spaces . . . . . . . . . 37
2.3 Hölder and Hilbert inequalities . . . . . . . . . . . . . . . 42
2.4 Generalizations of Hardy’s inequality . . . . . . . . . . . . 44
2.5 Carleman’s inequality and generalizations . . . . . . . . . 53
2.6 Minkowski’s inequality and generalizations . . . . . . . . . 54
2.7 Inequalities for the Laplace transform . . . . . . . . . . . 58
2.8 Inequalities for multivariate functions . . . . . . . . . . . . 62
3. Analytic Inequalities 67
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 67
vii
April 30, 2021 13:24 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page viii
viii Inequalities in Analysis and Probability (Third Edition)
3.2 Bounds for series . . . . . . . . . . . . . . . . . . . . . . . 69

3.3 Cauchy’s inequalities and convex mappings . . . . . . . . 72
3.4 Inequalities for the mode and the median . . . . . . . . . 76
3.5 Mean residual time . . . . . . . . . . . . . . . . . . . . . . 80
3.6 Functional equations . . . . . . . . . . . . . . . . . . . . . 82
3.7 Carlson’s inequality . . . . . . . . . . . . . . . . . . . . . . 87
3.8 Functional means . . . . . . . . . . . . . . . . . . . . . . . 91
3.9 Young’s inequalities . . . . . . . . . . . . . . . . . . . . . . 94
3.10 Entropy and information . . . . . . . . . . . . . . . . . . . 96
4. Inequalities for Discrete Martingales 103

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.2 Inequalities for sums of independent random variables . . 104
4.3 Inequalities for discrete martingales . . . . . . . . . . . . . 115
4.4 Inequalities for first passage and maximum . . . . . . . . 124
4.5 Inequalities for p-order variations . . . . . . . . . . . . . . 126
4.6 Weak convergence of discrete martingales . . . . . . . . . 131
5. Inequalities for Time-Continuous Martingales 133

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.2 Inequalities for martingales indexed by R+ . . . . . . . . . 134
5.3 Inequalities for the maximum . . . . . . . . . . . . . . . . 141
5.4 Inequalities for p-order variations . . . . . . . . . . . . . . 143
5.5 Weak convergence of martingales and point processes . . . 146
5.6 Poisson and renewal processes . . . . . . . . . . . . . . . . 152
5.7 Brownian motion . . . . . . . . . . . . . . . . . . . . . . . 162
5.8 Diffusion processes . . . . . . . . . . . . . . . . . . . . . . 170
5.9 Martingales in the plane . . . . . . . . . . . . . . . . . . . 174
6. Stochastic Calculus 177

6.1 Stochastic integration . . . . . . . . . . . . . . . . . . . . 177
6.2 Exponential solutions of differential equations . . . . . . . 182
6.3 Exponential martingales, submartingales . . . . . . . . . . 184
6.4 Gaussian processes . . . . . . . . . . . . . . . . . . . . . . 191
6.5 Processes with independent increments . . . . . . . . . . . 197
6.6 Semi-martingales . . . . . . . . . . . . . . . . . . . . . . . 200
6.7 Level crossing probabilities . . . . . . . . . . . . . . . . . . 202
6.8 Local times . . . . . . . . . . . . . . . . . . . . . . . . . . 207
April 30, 2021 13:24 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page ix
Contents ix
7. Functional Inequalities 215

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 215
7.2 Exponential inequalities for functional empirical processes 217
7.3 Inequalities for functional martingales . . . . . . . . . . . 223
7.4 Weak convergence of functional processes . . . . . . . . . 230
7.5 Differentiable functionals of empirical processes . . . . . . 233
7.6 Regression functions and biased length . . . . . . . . . . . 237
7.7 Regression functions for processes . . . . . . . . . . . . . . 243
7.8 Functional inequalities and applications . . . . . . . . . . 244
8. Markov Processes 247

8.1 Ergodic theorems . . . . . . . . . . . . . . . . . . . . . . . 247
8.2 Inequalities for Markov processes . . . . . . . . . . . . . . 251
8.3 Convergence of diffusion processes . . . . . . . . . . . . . 252
8.4 Branching process . . . . . . . . . . . . . . . . . . . . . . 253
8.5 Renewal processes . . . . . . . . . . . . . . . . . . . . . . 257
8.6 Maximum variables . . . . . . . . . . . . . . . . . . . . . . 263
8.7 Shock process . . . . . . . . . . . . . . . . . . . . . . . . . 268
8.8 Laplace transform . . . . . . . . . . . . . . . . . . . . . . . 273
8.9 Time-space Markov processes . . . . . . . . . . . . . . . . 278
9. Inequalities for Processes 283

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 283
9.2 Stationary processes . . . . . . . . . . . . . . . . . . . . . 285
9.3 Ruin models . . . . . . . . . . . . . . . . . . . . . . . . . . 287
9.4 Comparison of models . . . . . . . . . . . . . . . . . . . . 292
9.5 Moments of the processes at Ta . . . . . . . . . . . . . . . 294
9.6 Empirical process in mixture distributions . . . . . . . . . 297
9.7 Integral inequalities in the plane . . . . . . . . . . . . . . 300
9.8 Spatial point processes . . . . . . . . . . . . . . . . . . . . 301
9.9 Spatial Gaussian processes . . . . . . . . . . . . . . . . . . 309
10. Inequalities in Complex Spaces 315

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 315
10.2 Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . 318
10.3 Fourier and Hermite transforms . . . . . . . . . . . . . . . 322
10.4 Inequalities for the transforms . . . . . . . . . . . . . . . . 328
10.5 Inequalities in C . . . . . . . . . . . . . . . . . . . . . . . 330
April 30, 2021 13:24 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page x
x Inequalities in Analysis and Probability (Third Edition)
10.6 Complex spaces of higher dimensions . . . . . . . . . . . . 331

10.7 Stochastic integrals . . . . . . . . . . . . . . . . . . . . . . 335
Appendix A Probability 339

A.1 Definitions and convergences in probability spaces . . . . . 339
A.2 Boundary-crossing probabilities . . . . . . . . . . . . . . . 344
A.3 Distances between probabilities . . . . . . . . . . . . . . . 345
A.4 Expansions in L2 (R) . . . . . . . . . . . . . . . . . . . . . 348
Bibliography 351
Index 357
April 30, 2021 13:24 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page 1
Chapter 1
Preliminaries
1.1 Introduction
The origin of the inequalities for convex functions is the inequalities in real
vector spaces which have been extended to functional spaces by limits in
Lebesgue integrals. They are generalized to inequalities for the tail distri-
bution of sums of independent or dependent variables, under conditions for
the convergence of their variance, and to inequalities for the distribution
of martingales indexed by discrete or continuous sets. These inequalities
are the decisive arguments for bounding series, integrals or moments of
transformed variables and for proving other inequalities.
The convergence rate of sums of variables with mean zero is determined
by probability inequalities which prove that a sum of variables normalized
by the exact convergence rate satisfies a compactness property. If the nor-
malization has a smaller order than its convergence rate, the upper bound
of the inequality is one and it tends to zero if the order of the normalization
is larger.
Many probability results are related to the Laplace transform, such
as Chernoff’s large deviations theorem, Bennett’s inequalities and other
exponential inequalities for sums of independent variables. This subject
has been widely explored since the review papers of the Sixth Berkeley
Symposium in Mathematical Statistics and Probability (1972) which covers
many inequalities for martingales, Gaussian and Markov processes and the
related passage problems and sojourn times. Some of them are revisited
and extended after a brief review in this chapter. The upper bounds for the
tail probability of the maximum of n variables depend on n, in the same
way, the tail probability of the supremum of functional sums have upper
bounds depending on the dimension of the functional classes.
1
2 Inequalities in Analysis and Probability (Third Edition)
1.2 Cauchy and Hölder inequalities
Inequalities for finite series were first written as inequalities in a vector

space V provided with an Euclidean norm kxk, for x in V . The scalar
product of x and y in V is defined from the norm by
1
< x, y > = {kx + yk2 − kxk2 − kyk2 } (1.1)
2
and, conversely, an Euclidean norm is a `2 -norm related to the scalar prod-
uct as
1
kxk = < x, x > 2 .
From the definition (1.1), the norms of vectors x and y of an Euclidean
vector space V satisfy the geometric equalities
kx + yk2 + kx − yk2 = 2(kxk2 + kyk2 ) (1.2)
2 2
kx + yk − kx − yk = 4 < x, y > .
The space `2 (V ) is the space of series of V with a finite Euclidean norm.
An orthonormal basis (ei )1≤i of V is defined by the orthogonality property
< ei , ej > = 0 for i 6= j and by the normalization kei k = 1 for i ≥ 1. Let
Vn be a vector space of dimension n, for example Vn = Rn , for an integer n
and V∞ be its limit as n tends to infinity. Every vector x of `2 (Vn ), n ≥ 1,
is the sum of its projections in the orthonormal basis
Xn
x= < x, ei > ei ,
i=1
its coordinates in the basis are xi = < x, ei >, i = 1, . . . , n, and its norm is
Pn 1
kxk2 = ( i=1 x2i ) 2 . In `2 (V∞ ), a vector x is the limit as n tends to infinity
Pn Pn 1
of i=1 < x, ei > ei and its norm is the finite limit of ( i=1 x2i ) 2 as n
tends to infinity. The space `p (Vn ), 1 ≤ p < ∞, is defined with respect to
the norm
X n p1
kxkp = |xi |p
i=1
and the space `∞ (Vn ) is the space of vector with a finite uniform norm
kxk∞ = max1≤i≤n |xi |. In `p (V∞ ) and `∞ (V∞ ), the norms are defined
as the limits of the norms of `p (Vn ) as n tends to infinity. The norms
(kxkp )0<p≤∞ are an increasing sequence for every x in a vector space V .
The triangular inequality is kx + yk ≤ kxk + kyk with equality if and
only if < x, y > = 0. Consequently, for all x and y in a vector space
|kxk − kyk| ≤ kx − yk.
Preliminaries 3
The Cauchy inequality (1821) in the vector space Vn = Rn , for an

integer n, or V∞ is
< x, y > ≤ kxk2 kyk2
for all x and y in Vn , with equality if and only if x and y are proportional.
It is established recursively from the triangular inequality in V2 .
All norms are equivalent in an n-dimensional vector space Vn : For every
x in Rn and for 1 ≤ p, q ≤ ∞, there exist constants cp,q,n and Cp,q,n
depending only on n, p and q, such that cp,q,n kxkp ≤ kxkq ≤ Cp,q,n kxkp
1
kxk∞ ≤ kxkp ≤ n p kxk∞ ,
n−1 kxk1 ≤ kxk∞ ≤ kxk1 ,
1 1
n− p kxkp ≤ kxk1 ≤ n p0 kxkp , (1.3)
− p10 1
n kxk1 ≤ kxkp ≤ n kxk1 , p
− p10 − q1 1
+ q10
n kxkq ≤ kxkp ≤ n p kxkq ,
0 0
with p−1 + p −1 = 1 and q −1 + q −1 = 1.
Extensions of Cauchy’s inequality to bilinear series have been studied by
Hilbert, who proved that for positive real series (xn )n≥1 in `p and (ym )m≥1
in `q
X X xn ym π X X
≤ −1
k xi kp k y j k p0
n+m sin(p π)
n≥1 m≥1 i≤n j≤m
0
where p and p0 are conjugates such that p−1 + p −1 = 1. Other examples
are given by Hardy, Littlewood and Pólya (1952).
Cauchy’s inequality is extended to an inequality for integrals with re-
spect to the Lebesgue measure on R. Let f and g be square-integrable
functions in L2 (R), the Cauchy-Schwarz inequality is
Z Z 12 Z 12
f (x)g(x) dx ≤ f 2 (x) dx g 2 (x) dx ,
R R R
with equality if and only if f and g are proportional. Let µ be a positive

measure on R and w be a positive weighting function, the Cauchy-Schwarz
weighted inequality is
Z Z 21 Z 12
2
wf g dµ ≤ wf dµ wg 2 dµ ,
R R R
with equality under the same R condition. A simple proof for both inequali-
ties relies on the inequality R (tf (x)−g(x))2 w(x) dµ(x) ≥ 0 which develops
as an equation
R of the2 second-order with
R respect to t, with a negative dis-
2 2
R
criminant ( wf g dµ) − ( wf dµ)( wg dµ).
It is extended to the Hölder inequalities for Lp -integrable real functions
with a finite or infinite support [a,Rb], where −∞ ≤ a < b ≤ ∞. For every
b 1
real p > 1, let Lp,µ (a, b) = {f : ( a |f |p dµ) p < ∞}, if 1 ≤ p < ∞, and
L∞ (a, b) = {f : sup(a,b) |f | < ∞}. Let p and q be conjugates and let f
in Lp,µ (a, b) and g in Lq,µ (a, b), Hölder inequality is a consequence of the
inequality uv ≤ p−1 up + q −1 v p for (u, v) in R2+ such that kukLp = 1 and
kvkLq = 1
Z Z p1 Z q1
|f g| dµ ≤ |f |p dµ |g|q dµ . (1.4)
R R R
Let r ≥ 1 and let p and q be conjugates such that p−1 + q −1 = r−1 Hölder
inequality generalizes as
Z r1 Z p1 Z q1
|f g|r dµ ≤ |f |p dµ |g|q dµ . (1.5)
R R R
p
The L norms are increasing. This implies Khintchine’s inequality for p ≥ 1:
Let (rk )k≥0 be the Rademacher functions on [0, 1], for every measurable
2
P P
function f = k≥0 ak rk on [0, 1] satisfying k≥0 ak < ∞, there exist
constants Ap > 0 and Bp such that
X 12 X 1
Ap a2k ≤ kf kp ≤ Bp (a2k ) 2 . (1.6)
k≥0 k≥0
Since the constants do not depend on the dimension of the projection of the
function onto the basis, (1.6) implies the equivalence between the norms of
the function, Ap kf k2 ≤ kf kp ≤ Bp kf k2 .
Inequalities for a countable sequence x = (xi )i≥0 in the space `p (R∞ )
are deduced from functional inequalities for Rpiecewise constant integrable
ai
functions of a space Lp,µ or by setting xi = ai−1 f dµ for some function f
p,µ
of L and ∪i≥0 (ai−1 , ai ) = (a, b). Let r < p and let q be the conjugate of
p such that r−1 = p−1 + q −1 , then kxykr ≤ kxkp kykq .
The Cauchy and Hölder inequalities are recursively generalized to more
than two series or functions. Let k ≥ 2 and let p1 , . . . , pk ≥ 1 such that
Pk −1 −1
i=1 pk = r and let f1 , . . . , fk be functions in Lp1 ,µ (a, b), . . . , Lpk ,µ (a, b)
respectively, the Hölder inequality for the functions f1 , . . . , fk is
Z b Y k r1 k
Y
| fi |r dµ ≤ kfi kpi ,µ (1.7)
a i=1 i=1
with equality if and only if the functions fi are proportional.

Preliminaries 5
The analogous Cauchy inequality for a set of k series (xij )1≤i≤n,1≤j≤k

of Lpk , respectively, is obtained with piecewise constant functions
Xn k
Y r1 k
Y
| xij |r ≤ kxj kpj . (1.8)
i=1 j=1 j=1
It is extended to infinite arrays (xij )1≤i<∞,1≤j<∞ .
The Hölder inequality is a consequence of the concavity of the function
ϕα (u, v) = uα v 1−α with α = p−1 (Neveu 1970) and the same argument
applies for the other inequalities. More generally, let f be a real function of
L1 (a, b), then for every convex function ϕ from R to R, Jensen’s inequality
for integrals is
Z b Z b
ϕ f dµ ≤ ϕ(f ) dµ
a P a
n Pn
and it extends the property ϕ( i=1 αi ai ) ≤ i=1 αi ϕ(ai ) for convex func-
Pn
tions in vector spaces, with i=1 αi = 1, and the inequality is an equality
if and only if the ai are proportional. The Minkowski inequality is the
additivity property of the norms in the spaces `p (Rn ), 1 ≤ n ≤ ∞
kx + ykp ≤ kxkp + kykp , 1 ≤ p ≤ ∞
and it is extended to Lp (R)
kf + gkp ≤ kf kp + kgkp , 1 ≤ p ≤ ∞,
with equality if and only if x and y and, respectively, f and g are propor-
tional. In `p (Rn ), it is a consequence of the concavity and the homogeneity
of the function ϕp (x) = kxkp and it generalizes to Lp (R).
In the complex space C, the Euclidean norm of x = x1 + ix2 is
1 1
kxk = (xx̄) 2 = (x21 + x22 ) 2 ,
with the complex conjugate x̄ = x1 − ix2 , and the scalar product < x, y >
of x and y is still defined by (1.1). In the product spaces Cn and C∞ ,
let x = (xi )i∈I and y = (yi )i∈I be vectors of finite or countable dimension,
where I is a subset of integers I = {1, . . . , n}, I = N or I = Z, its Euclidean
P 1
norm is kxk = ( i∈I xi x̄i ) 2 and < x, y > follows by (1.1). The inequalities
of Cauchy and Minkowski are written in the same form as in R2 .
The Fourier transform of a variable
R∞ X with density function fX is de-
fined as fbX (ω) = Ee2πiωX = −∞ e2πiωx fX (x) dx. Its norms Lp ([0, 1])
satisfy the properties
kfbX kp ≤ kfX kp ,
0
kfbX k1 ≤ kfbX kp kfbX kp0 , p−1 + p0−1 = 1.
p p
It is generalized to functions of Lp (Rd ). The Fourier transform of the sum of
independent variables is the product of their transforms, hence by Hölder’s
inequality kfbX+Y kr ≤ kfbX kp kfbY kp0 , r−1 = p−1 + p0−1 .
1.3 Inequalities for transformed series and functions
The Cauchy inequality has been extended by convex transformations of the

series. A convex real function f on!R satisfies
Pk Pk
i=1 αi xi αi f (xi )
f Pk ≤ i=1 Pk
i=1 αi i=1 αi
Pk
for every linear combination i=1 αi xi with all αi > 0, or equivalently
Pk Pk
f ( i=1 αi xi ) ≤ i=1 αi f (xi ) for every linear combination such that αi > 0
Pk
for every i and i=1 αi = 1. With k = 2 and α > 0, consider x > y and f
increasing at y then
0 ≤ f (y + α(x − y)) − f (y) ≤ α{f (x) − f (y)} → 0
as α tends to zero. Otherwise, let f be decreasing at x, 0 ≤ f (x − (1 −
α)(x − y)) − f (x) ≤ (1 − α){f (y) − f (x)} which tends to zero as α tends
to 1, so every convex function is continuous.
Let f be increasing and belong to C 2 (R) and let y satisfy f 0 (y) = 0.
Let α in [0, 1] and x > y. As α tends to zero, a Taylor expansion of f in a
2
neighbourhood of y is written f (y+α(x−y))−f (y) = α2 (x−y)2 f 00 (y)+(α2 )
therefore f 00 (y) > 0. If f is a decreasing convex function of C 2 (R), its second
derivative at y such that f 0 (y) = 0 is also strictly positive, so the second
derivative of a convex function is strictly positive where its first derivative
is zero and it is a minimum of the function. Conversely, a function f is
concave if −f is convex, its second derivative is strictly negative where its
first derivative is zero and it is a maximum of the function. The polygons
generate the space of convex functions.
The Hadamard inequality for a convex function f on a finite interval
[a, b] is
a + b Z b
1 f (a) + f (b)
f ≤ f (x) dx ≤ .
2 b−a a 2
Cauchy (1821) proved other inequalities for convex functions of a countable
number of variables, for example let (xi )i≥1 be a sequence of real numbers,
then Y nX o
(1 + xi ) < exp xi .
i≥1 i≥1
Let f be a positive and concave function on a subinterval [a, b] of R+
Pk
and (xi )i=1,...,k be k points in [a, b] and let x̄α = i=1 αi xi be a weighted
Pk
mean of these points, with i=1 αi = 1. By concavity
Yk
{f (xi )}αi ≤ f (x̄α ), (1.9)
i=1
Preliminaries 7
with equality if α1 x1 = · · · = αk xk . The exponential function is convex and

the equality is fulfilled for all sequences (xi )i=1,...,k and (αi )i=1,...,k then
conditions for equality in (1.9) depend on the function f . The inequality
Pk
(1.9) is also written i=1 αi log f (xi ) ≤ log f (x̄α ) and it is satisfied by
Jensen’s inequality, due to the concavity of the logarithm function.
With f (x) = 1 + x, (1.9) implies
k
Y
(1 + xi ) ≤ (1 + x̄k )k ,
i=1
−1
Pk
with x̄k = k i=1 xi . Applying this inequality with xi = 2i provides a
bound for the 2k-th moments m2k = 1.3 . . . (2k − 1) = {2k (k)!}−1 (2k)! of
Qk−1
the normal distribution, i=0 (1 + 2i) ≤ k k . This is a mean value bound
but it is not very sharp. By the convexity of the exponential function, the
Qk
converse inequality at points (t1 , . . . , tk ) yields (1 + et̄k )k ≤ i=1 (1 + eti ),
which it is equivalent to
k k
Y 1 Y 1
1+ xi ≤
k
(1 + xi ) k .
i=1 i=1
Qk 1
Replacing xi by xi yi−1 and multiplying both sides by i=1 yik implies
k k k
Y 1 Y 1 Y 1
xik + yik ≤ (xi + yi ) k
i=1 i=1 i=1
for all sequences (xi )i=1,...,k and (yi )i=1,...,k . More generally, let (α1 , . . . , αk )
Pk
be positive numbers such that i=1 αi = 1, then
k
Y k
Y k
Y
xα
i +
i
yiαi ≤ (xi + yi )αi .
i=1 i=1 i=1
For an array (xij )i=1,...,k,j=1,...,n , a recursive argument implies
n Y
X k k
Y
n−1 xα
ij ≤
i
x̄α
i
i
(1.10)
j=1 i=1 i=1
Pn
with the partial means x̄i = n−1 j=1 xij . This inequality differs from the
inequality (1.8) and from the Minkowski inequality for a sum of n series in
Pn
Rk , kx̄kp ≤ n−1 j=1 kxj kp , for p ≥ 1.
With the convex function f (t) = log(1 + et ), Equation (1.9) implies that
for every xi > 0, for i = 1, . . . , k
Y k k
Y
αi
log 1 + xi ≤ {log(1 + xi )}αi .
i=1 i=1
Replacing xi by xi yi−1 , with yi > 0 for i = 1, . . . , k, and adding to the left

Qk
member log i=1 yiαi ≤ log ȳα yields
k
Y k
Y k
Y
log{(xi + yi )αi } ≤ log (xi + yi )αi − log yiαi + log ȳα .
i=1 i=1 i=1
For every convex function ϕ from R to R, Jensen’s inequality for the

integral of a function f with respect to the positive measure µ on [a, b] is
Z b Z b
1 1
ϕ{ f dµ} ≤ ϕ ◦ f dµ,
M (b) − M (a) a M (b) − M (a) a
with M (a) = µ(] − ∞, a]) and M (b) = µ(] − ∞, b]). As a consequence, for
every real function f on [a, b]
Z b Z b
n 1 o 1
exp log f dµ ≤ f dµ.
M (b) − M (a) a M (b) − M (a) a
Other integral inequalities for convex function can be found in Maurey
(2004).
An integral equation similar to Equation (1.9) is obtained from Jensen’s
inequality for every concave function f > 0
n Z y o n Z y o
−1 −1
exp y log f dµ ≤ f y x dµ(x) , y > 0. (1.11)
0 0
The bound
Z y n Z y o
y −1 f dµ ≤ f y −1 x dµ(x)
0 0
is sharper and it is an equality for affine functions f > 0.
For example, with f (x) = 1 + x, Equation (1.11) is written
Z y y
log(1 + x) dx ≤ y log 1 +
0 2
for
R y every y > 0. With f (x) = xα , with x > 0 and 0 < α < 1, (1.10) implies
0
log x dx ≤ y log 2 and Jensen’s inequality yields y α ≤ α(α + 1) log y2 .
y
Most inequalities for bounding finite series are intervals for the error of
their approximation by a Taylor expansion or they are obtained by induc-
tion. For example, Cauchy (1833) defined the exponential function as
ex = lim (1 + α)n = lim (1 − α)−n
n→∞,nα→x n→∞,nα→x
which provided him the following interval

x x
(1 + α) α < ex < (1 − α)− α ,
Preliminaries 9
for every real α > 0. This definition entails immediately the main properties
of the exponential. Reciprocally, the unique function φ which satisfies
φ(x + y) = φ(x)φ(y),
for all x and y, is the exponential function φ(x) = {φ(1)}x ≡ ex .
Other classical bounds for functions have been established for the first
n terms of their Taylor expansions. Alzer (1990b) studied a lower bound
of the form In−1 (x)In+1 (x) > cn In2 (x) for the sum In (x) of the first n
terms in the Taylor expansion of the exponential function. More generally,
Darboux’s sums for a function f on an interval [a, b] are defined using a
partition πn = (ak )k=0,...,an of increasing numbers of this interval such that
a = a0 and an = b
Xn
S(f, πn ) = mk (ak − ak−1 ),
k=1
Xn
T (f, πn ) = Mk (ak − ak−1 ),
k=1
with mi = inf ak−1 ≤x<ak f (x) and Mi = supak−1 ≤x<ak f (x) and the con-
vergence
R b of the sum T (f, πn ) implies the convergence of the integral
I = a f (x) dx, conversely the convergence of I implies the convergence
of the sum S(f, πn ). The sum 1 + 3 + · · · + (2k − 1) has then the
Pk−1
bounds k 2 − k − 1 ≤ i=0 (1 + 2i) ≤ k 2 + k − 1 and the product
Pk−1
m2k = 1.3 . . . (2k − 1) = exp{ i=0 log(1 + 2i)} satisfies
k−1
Y 2
(1 + 2i) ≤ (1 + 2k)1+2k e−2k ≤ e(2k) = 2ke2k , (1.12)
i=0
the inequality m2k ≤ k k obtained from (1.9) is therefore better than the
upper bound of (1.12) for m2k . Close approximations by Darboux’s sums
require a thin partition πn .
1.4 Applications in probability
Let (Ω, F, P ) be a probability space and let X be a real variable defined

on (Ω, F, P ), with distribution function F respectively. The norms Lp (R)
define norms for variables
1
nZ o p1
kXkp = (E|X|p ) p = |x|p dF (x) ,
R
kXk∞ = sup{x : P (|X| > x) > 0}.
Let X be a random vector of Rn+ , by concavity of the norm, we have

EkXkp ≤ kEXkp for every p ≥ 1. The sequence (kXkp )0<p≤∞ is in-
creasing and the additive and multiplicative properties of the norms are
consequences of the Hölder and Minkowski inequalities. For dependent
variables X and Y , the Cauchy-Schwarz inequality entails
E(XY ) ≤ kXkp kY kq
for all conjugates p, q ≥ 2. Let (Xi )i≥1 be a sequence of random variables in

a vector space V , k maxi=1,...,n Xi k ≤ maxi=1,...,n kXi k and the inequality
extends as n tends to infinity.
The density of the sum X +Y of independent variables with distribution
functions F and G and densities f and g is the convolution
Z Z
f ∗ g(t) = f (t − s) dG(s) = g(t − s) dF (s).
R R
The density of the ratio XY −1 of independent variables X and Y > 0

−1
R
is the multiplicative convolution h(t) = R f (s t) dG(s), its norms has
the bound khkr ≤ kf kp kgkq , for all p, q, r such that p−1 + q −1 = r−1 ,
1 ≤ p ≤ ∞.R If the variables haveR no densities, the convolution is P (X +
Y ≤ t) = R G(t − s) dF (s) = R F (t − s) dG(s) and it is sufficient that
F or G is continuous (respectively, has a density) to ensure the continuity
(respectively, the derivability) of the distribution function of their sum.
Their ratio has the distribution function R F (s−1 t) dG(s).
R
The inequalities of Section 1.3 imply the following inequalities for the
sequence of variables. Let (Xi )i=1,...,n be a sequence of random variables
Pn
and let X̄n = n−1 i=1 Xi be their empirical mean, Cauchy’s inequality
implies that for every integer n
n n
X o
E n−1 log(1 + Xi ) < E{X̄n }.
i=1
Pn
Let (αi )i=1,...,n be a real sequence such that i=1 αi = 1. For every positive
concave function f on R+ , inequality (1.9) entails
n
nX o k
X
E αi log f (Xi ) ≤ E log f αi Xi
i=1 i=1
−1
Pn
in particular E{n i=1 log f (Xi )} ≤ E log f (X̄n ). By Jensen’s inequal-
ity, for every convex function ϕ and for every variable X
ϕ(EX) ≤ Eϕ(X).
Preliminaries 11
Bienaymé-Chebychev’s inequality has been formulated for a variable X of

L2 (R) as P (|X| > a) ≤ a−2 E|X|2 , a > 0. For a variable X in Lp (R),
1 ≤ p < ∞, and such that EX = m, it is
P (|X − m| > a) ≤ a−2 VarX, for every a > 0, (1.13)
where VarX = E(X − EX)2 is the variance of X. It extends to p ≥ 2
P (X − m > a) ≤ a−p E|X − m|p , for every a > 0.
Another Chebychev’s inequality for the mean of n variables with mean m
and variance σ 2 is
1
P (|X̄ − m| > a) ≤ , a > 0.
1 + σ −2 na2
Pn
Kolmogorov’s inequalities for the sum Sn = i=1 Xi of n independent
random variables (X1 , . . . , Xn ) are

P sup |Sk − E(Sk )| ≥ a ≤ a−2 VarSn , if Xi ∈ L2 ,
1≤k≤n
(a + k)2
P sup |Sk − E(Sk )| ≥ a ≤ 1 − , if kXi k∞ ≤ k.
1≤k≤n E(Sn2 )
The Laplace transform of a variable X with distribution function F is
LX (t) = Ee−tX = LfX (t)
where fX is the density of X and by convexity, LX (t) ≥ e−tEX . Its moment
generating function is defined as GX (t) = E(tX ) or ϕX (t) = EetX . The
moments µk = EX k of the variable X are obtained from the value at
(k) (k)
zero of the derivatives of LX or GX , µk = (−1)k LX (0) = ϕX (0) and
(k)
GX (0) = E{X(X − 1) · · · (X − k + 1)}, k ≥ 1. The inversion formula is the
same for ϕX and LX . As a consequence of the expansion of the exponential,
at every point of continuity x of FX
X (−1)k λk (k)
FX (x) = lim ϕX (λ).
λ→∞ k!
k≤λx
Let (Xi )i=1,...,n be a sequence of independent random variables with

Pn
means zero and respective variances σi2 and let Sn = i=1 Xi , X̄n = n−1 Sn
Pn
and Vn = i=1 σi2 . If σ̄n2 = n−1 Vn converges to a finite limit σ 2 > 0, then
for every β > 0 and a > 0, the inequality (1.13) implies
β
P (n 2 |X̄n | > a) ≤ nβ a−2 E X̄n2 = nβ−1 σ̄n2 .
As the upper bound tends to zero for β < 1, the inequality cannot be
satisfied for every a. The upper bound being infinity for every β > 1, the
Pn
convergence rate of the mean X̄n = n−1 i=1 Xi to zero is β = 1. For
independent variables with respective means µi , the convergence rate of
n−α Sn to a limit µ is determined by the convergence rates of the mean
Pn
µn = n−α i=1 µi to µ and of the mean variance σn2 = n−β Vn to a limit
σ 2 > 0. For every a > 0
β
P (n 2 |n−α Sn − µ| > a) ≤ n2(β−α) a−2 n−β Vn ,
therefore the convergence rates α and β are equal.
Chow and Lai (1971) proved the following equivalences for a sequence
of independent and identically distributed random variables (Xi )i≥1 with
mean EX1 = 0, for α in ]0, 1[
E exp(t|X1 |α ) < ∞, t > 0,
−1
lim (log n)α Xn = 0, a.s.,
n→∞
n
−1 X
lim (log n)α cn−i Xi = 0, a.s.,
n→∞
i=1
for a sequence of weighting constants such that cn = O(n−ν ), with ν > 21 .

Other equivalences between convergences proved by Chow and Lai (1971)
are given in the appendix.
For every a > 0, Bienaymé-Chebychev’s inequality for the maximum of

n positive centered variables of L2 with the same variance σ 2 is
1 σ2
lim sup P n− 2 max |Sk | > a ≤ 2 .
n→∞ k=1,...,n a
Billingsley (1968) proved many inequalities for partial sums and their mo-
ments and for the empirical process of independent random variables. In
particular E(|Sn − Sk |2 |Sk − Sj |2 ) = (Vn − Vk )(Vk − Vj ), 0 ≤ j ≤ k ≤ n,
the variable
Mn0 = max min{|Sk |, |Sn − Sk |}
0≤k≤n
satisfies the inequalities

Mn0 ≤ Mn ≤ Mn0 + |Sn |, Mn0 ≤ Mn ≤ Mn0 + 3 max |Xk |
0≤k≤n
and for every λ > 0, there exists a constant K such that

P (Mn0 > λ) ≤ Kλ−4 Vm4 .
For dependent variables, the variance of Sn is bounded by a sum which
also depends on the means E|Xi Xj | for every i 6= j and a mixing coefficient
Preliminaries 13
ηj−i determines the convergence rate of this sum through the inequality
E|Xi Xj | ≤ ηj−i σi σj . If there exist constants α > 0 and β > 0 such that
Pn
µn = n−α i=1 µi converges to µ and n−β i=1,...,n j=1,...,n ηj−i σi σj
P P
converges to a limit σ 2 > 0, then the convergence rate of n−α Sn to µ

is still n−β .
The Berry-Essen inequality for independent variables concerns the con-
−1
vergence of the distribution of a normalized sum of variables Vn 2 Sn to
normal variable with distribution function Φ. If E|Xi |3 is finite, there ex-
ists a constant K such that
Pn
− 12 E|Xi − EX|3
sup |P (Vn (Sn − EX) ≤ x) − Φ(x)| ≤ K i=1 3 .
x∈R Vn2
For dependent variables with mixing coefficients such that k>0 k 2 ηk is
P
1
finite, the convergence rate of Sn is n 2 (Billingsley, 1968) and the Berry-
Essen inequality is satisfied.
The Fourier transform of the correlation function RX (t) = E{X0 Xt } of
a stationary process X is defined as its spectral density
Z ∞
fbX (ω) = RX (t)e2πωt dt.
−∞
Inverting the mean and the integrals, its norms satisfy the properties
kfbX kp ≤ kRX kp ,
0
kfbX k1 ≤ kfbX kpp kfbX kpp0 , where p−1 + p0−1 = 1.
1.5 Hardy’s inequality
Hardy’s inequalities have been developed in different forms presented in

the monograph by Hardy, Littlewood and Pólya (1952). The functional
inequality is a consequence of the Hölder inequality. Let p > 1 and
let f be a positive function of Lp (R+ ), with primitive F and such that
1 1
limx→0 x p −1 F (x) = 0 and limx→∞ x p −1 F (x) = 0. The integral
Z yn Z yn
F (x) op p F (x) op−1 1 F p (y)
Ip (y) := dx = f (x) dx − ,
0 x p−1 0 x p − 1 y p−1
(1.14)
and Ip = Ip (∞) have the bounds
p p Z ∞
Ip ≤ f p (x) dx, 1 < p < ∞, (1.15)
p−1 0
and the inequality is strict except if f is a constant. With p = 1, for every

function f > 0 on R+ having a primitive F such that limx→0 f (x) = 0,
limx→∞ x−1 F (x) = 0, the integral cannot be calculated in the same way,
however we have
Z ∞ Z ∞
F (x) f (x)
I1 = 2
dx = dx.
0 x 0 x
If 0 < p < 1, integrating by parts yields
Z yn
1 h F (x) op−1 i
Ip (y) = p f (x) dx − y 1−p F p (y)
1−p 0 x
where
Z yn
F (x) op−1 1− 1
f (x) dx ≥ Ip p (y)kf kp .
0 x
Since y 1−p F p (y) > 0 and tends to infinity as y tends to infinity, the in-
equality (1.15) cannot be inversed for p < 1.
Pn
For positive real series (ai )i≥1 , let An = i=1 ai and let p > 1 such
that (ai )i≥1 belongs to `p . Hardy’s inequality is
X An p p p X p
< ai .
n p−1
n≥1 i≥1
2
A special case is Euler’s limit of the series k≥1 k −2 = π6 (Euler, 1735)
P
which was obtained as the coefficient of x2 in the expansion of the function

x−1 sin x as a polynomial with roots ±kπ. Integrating the approximation
of (1 − x)−1 by (1 − xn )(1 − x)−1 , with normalization, it appears that
Z 1 Z x
1 1 − tn π2
dt dx =
0 x 0 1−t 6
(Bradley, d’Antonio and Sandifer, 2007; Dunham, 1999; Sandifer, 2007). It
Pn
follows that n≥1 n−2 ( k=1 k −1 )2 < 23 π 2 . Hardy and Littlewood (1930)
P
extended the inequality to a function f (z) = f (reiθ ) with maximum on the

radius F (θ) = supr>0 |f (z)|, for p > 1
F p (θ)
Z p p Z
dθ ≤ f p (z) dz.
xp p−1
Bickel et al. (1993) mentioned a statistical application of the inequality to
the derivative of the density of a randomly right-censored time variable.
Let S be a positive real random variable with distribution function
R H and
density h, for every square integrable function a such that a dH = 0
R ∞ a dH 2
E S ≤ 4Ea2 (S).
H(S)
July 7, 2021 9:22 ws-book9x6 Inequalities in Analysis and Probability (Third Edition) 12135-main page 15
Preliminaries 15
Kufner and Persson (2003) and others later proved weighted inequalities of
the Hardy type. Pachpatte (2005) also provided a review of the classical
inequalities
Z xZ y of the same type √ and Zseveral generalizations Z y of the inequality
f (s)g(t) xy n x 0 0
o 21
ds dt ≤ pq (x − s)f 2 (s) ds (y − t)g 2 (t) dt .
0 0 s+t 2 0 0
Let f be a complex function defined on Rd , with the Lebesgue measure
λd and let Br (x) be the ball of Rd , centred at x and with radius r. The
maximum average value of f is Z
1
M f (x) = sup |f | dλd .
r>0 λd (Br (x)) Br (x)
Hardy-Littelwood’s weak inequality for M f is
λd {x ∈ Rd : M f (x) > a} ≤ a−1 Cd kf kL1 (Rd ) , for every a > 0,
where Cd is a constant. A stronger inequality for M f is due to
Marcinkiewicz for every p > 1
kM f kLp (Rd ) ≤ Cd,p kf kLp (Rd )
where Cd,p is a constant.
Another type of analytical inequalities for integrals is the Mean Value
Theorem (Lagrange) and its extensions to higher order approximations.
Hardy, Littlewood and Pólya (1952) proved a bound for the difference be-
tween the integral of the square of a continuously differentiable function f
on [0, 1] and the square of its integral
Z 1 nZ 1 1 1
o2 Z
0
0≤ f 2 (x) dx − f (x) dx ≤ x(1 − x)f 2 (x) dx.
0 0 2 0
Ostrowski’s type inequalities (Sahoo and Riedel, 1998, Mitrinović, Pecarić
and Fink 1957, Dragomir and Sofo 2000, Dragomir and Rassias 2002) have
been written for the approximation of a continuously differentiable function
f defined in an interval (a, b), having a derivative bounded by a constant
M , in the form
1
Z b h 1 {x − 1 (a + b)}2 i
2
f (x) − f (t) dt ≤ (b − a)M + .
b−a a 4 (b − a)2
Gruss’s inequalities are bounds for the difference between the integral of
a product of bounded functions and the product of their integrals. Let
ϕ ≤ f (x) ≤ Φ and γ ≤ g(x) ≤ Γ
Z b
1 n b o 1 nZ b
Z
1 o
f (x)g(x) dx − f (x) dx g(x) dx
b−a a b−a a b−a a
1
≤ (Φ − ϕ)(Γ − γ).
4
These inequalities have been developed for several classes of functions
and extended to inequalities of higher order approximations (Barnett and
Dragomir, 2001, 2002).
1.6 Inequalities for discrete martingales
On a probability space (Ω, F, P ), let (Fn )n≥0 be a filtration of F, i.e. a

nondecreasing sequence of subsigma-algebras of F, and let (Xn )n≥0 be a
sequence of Fn -measurable real variables. It is a martingale with respect
to (Fn )n if E(Xm |Fn ) = Xn for every m > n ≥ 0. Then EXn = EX0 and
E{(Xm − Xn )Yn } = 0, for every Fn -measurable variable Yn . It follows that
the conditional variance of a square integrable (Fn )n -martingale (Xn )n≥0
is E{(Xm − Xn )2 |Fn } = E(Xm 2
|Fn ) − Xn2 .
The sequence (Xn )n≥0 is a submartingale if E(Xm |Fn ) ≥ Xn for every
m > n ≥ 0 and it is a supermartingale if E(Xm |Fn ) ≤ Xn . Let (Xn )n≥0
be a real martingale and let ϕ be a function defined from R to R. If ϕ is
convex, E{ϕ(Xn+1 )|Fn } ≥ ϕ(E{Xn+1 |Fn }) = ϕ(Xn ) then (ϕ(Xn ))n≥0 is
a submartingale. If ϕ is concave, then (ϕ(Xn ))n≥0 is a supermartingale.
Proposition 1.1 (Neveu, 1972). Let (Xn )n≥0 be a positive super-

martingale then for every c > 0

P sup Xn > c | F0 ≤ min(c−1 X0 , 1)
n≥0
and supn≥0 Xn is a.s. finite on {X0 < ∞}.
Theorem 1.1. Every positive supermartingale (Xn )n≥0 converges a.s. to

X∞ = limn→∞ Xn and E(X∞ | Fn ) ≤ Xn . Every positive submartingale
(Xn )n≥0 such that supn∈N E(Xn+ ) is finite converges a.s. to a limit X∞ in
L1 such that E(X∞ |Fn ) ≥ Xn for every integer n.
The following examples of discrete martingales are classical.

Pn
(1) A random walk Xn = i=1 ζi is defined by a sequence of independent
and identically distributed random variables (ζi )i = (Xi −Xi−1 )i=1,...,n .
Let µ = Eζi be the mean increment of Xn , if µ = 0 then Xn is a
martingale, if µ > 0 then Xn is a submartingale and if µ < 0 then Xn
is a supermartingale.
(2) For every increasing sequence of random variables (An )n≥1 such that
An is Fn -measurable and E(A∞ |F0 ) < ∞ a.s., Xn = E(A∞ |Fn ) − An
is a supermartingale and EMn = (A∞ |Fn ) is a martingale.
(3) Let
n
X
Vn (X) = X02 + (Xi − Xi−1 )2
i=1
Preliminaries 17
be the quadratic variations of a L2 (P, Fn )n ) martingale (Xn )n≥1 with

respect to a filtration (Fn )n≥1 , this is an increasing sequence. For every
Pn Pn
n ≥ 0, EXn2 = i=1 E(Xn2 − Xn−1 2
) + EX02 = E{ i=1 (Xi − Xi−1 )2 } +
EX02 = EVn (X). Let
E{Vn+1 (X)|Fn } = Vn (X) + E{(Xn+1 − Xn )2 |Fn } ≥ Vn (X),
(Vn (X))n≥1 is a submartingale and it converges a.s. to a limit V (X) in
L1 (P ).
Theorem 1.2. On a filtered probability space (Ω, F, (Fn )n≥0 , P ), let

(Xn )n≥1 be a sequence of Fn -measurable variables with X0 = 0 and let
X
Xen = E(Xn − Xn−1 | Fn−1 ),
i=1
then (Xen )n≥1 is the unique predictable process such that (Xn − X
en )n≥1 is
a (Fn )n -martingale.
Theorem 1.3. Let (Xn )n≥1 be a square integrable martingale on a fil-

tered probability space (Ω, F, (Fn )n≥0 , P ) then (Xn2 − Vn (X))n≥1 and
Pn
( k=1 {E(Xk2 | Fk−1 ) − Xk−1
2
})n≥1 are (Fn )n -martingales.
This is proved by Theorem 1.2 and the equality

E(Xn2 − Xn−1
2
|Fn−1 ) = E{(Xn − Xn−1 )2 |Fn−1 }
= E(Vn − Vn−1 |Fn−1 ).
A transformed martingale on (Ω, F, (Fn )n≥0 , P ) is defined by two sequences
of L1 (P ) random variables (Xn )n≥1 and (An )n≥1 by Y0 = X0 and
Yn+1 = Yn + An (Xn+1 − Xn ), n ≥ 1, (1.16)
where An is Fn−1 -measurable, for every integer n, and Xn is a (Fn )-
martingale, then E{Yn+1 |Fn } = Yn + An E(Xn+1 − Xn |Fn ), so Yn is a
(Fn )-martingale. If Xn is a Fn -submartingale (respectively supermartin-
gale), then Yn is a Fn -submartingale (respectively supermartingale). The
quadratic variations Vn (Y ) of (Yn )n≥1 satisfy
E{Vn+1 (Y )|Fn } − Vn (Y ) = A2n E{(Xn+1 − Xn )2 |Fn } ≥ 0, (1.17)
n−1
hence the process Vn (Y ) = X02 + i=1 A2i {Vi+1 (X) − Vi (X)} defines a
P
convergent submartingale.
Kolmogorov’s inequality for a (Fn )-martingale (Xn )n≥1 of Lp (R), for

1 ≤ p < ∞, is similar to Bienaymé-Chebychev’s inequality for independent
variables
P (|Xn+1 | > a|Fn ) ≤ a−p |Xn |p , for every a > 0. (1.18)
A stopping time T of a uniformly integrable martingale sequence (Xn )n≥1
defined on a filtered probability space (Ω, F, (Fn )n≥0 , P ) satisfies the next
measurability property: {T ≤ n} is Fn -measurable, for every integer n.
Doob established that E(XT | FS ) = XS for all stopping times S and
T > S, let Xn = E(X∞ | Fn )
∞
X
E(XT | FS ) = E(Xi 1{T =i} | FS )
i=S+1
X∞
= E{E(X∞ | Fi )1{T =i} | FS }
i=S+1
X∞
= E(X∞ 1{T =i} | FS ) = E(X∞ | FS ),
i=S+1
therefore E(XT | FS ) = XS∧T and EXT = EX0 , for all S and T .

Let (Xn )n≥1 be a sequence of (Fn )n -adapted random variables with val-
ues in [0, 1], for every stopping time τ , the martingale Mn = E(Xn+1 |Fn ),
for every n, satisfies the property (Freedman, 1973)
τ
X τ
X b a
P Xn ≤ a, Mn ≥ b ≤ ea−b
i=1 i=1
a
n (a − b)2 o
≤ exp − ,
2c
Xτ Xτ b a
P Xn ≥ a, Mn ≤ b ≤ ea−b ,
i=1 i=1
a
where 0 ≤ a ≤ b and c = max(a, b), and the bound reduced to 1 if a = b = 1.

The inequalities for the maximum of variables extend to martingales. Let
Xn∗ = max |Xi |
1≤i≤n
be the maximal variable of (X1 , . . . , Xn ) and Xn+ = max(Xn , 0).
Theorem 1.4. Let τ1 = 1 and for k > 1, let

τk = min{n > τk−1 : Xn∗ = Xn }
Another random document with
no related content on Scribd:
“It was a close shave, girl,” he said, looking at the avenger, whom he
now encountered for the first time. “I owe the preservation of my life
to the fact that you fired downward; had I been on a level with you
there were no need of my talking now.”
“No,” answered the She-wolf. “I would have sworn that you were an
Indian—Wacomet.”
“You are not the first person who has recognized Wacomet in me
within the last twenty-four hours, and I had better counterfeit another
red gentleman than he. But, girl, we must seek shelter; there may be
sharp eyes nearing us, and then here’s one who needs rest, to live.”
As he finished, the spy glanced at the Briton, and a minute later the
boat was moored to the bank.
It was no child’s play for the quartette, though Effie could lend but
little assistance, to bear the helpless major up the loose and ragged
rocks to Nanette’s cave home. But the Herculean task was bravely
undertaken and accomplished, and the wounded man found himself
almost buried in a pile of soft skins, that seemed to him a bed of
down.
Then his wounds were more carefully examined, and found to be of
a less serious nature than was at first supposed; but still he was far
from safety. The irritation caused by the journey up the rocks might
speedily prove fatal, and terminate a life not without guilt.
The day passed quickly to the inmates of the cave, and when night
came again, Mark Morgan announced his intention of carrying out
the wishes of his commander before dawn—viz: to enter the Indian
village and ascertain the numbers, etc., of the red-men and their
white allies who were to meet Wayne on the banks of the Maumee.
He knew almost to a certainty that the conflict would take place near
Nanette’s cave, and he resolved to leave Effie under the protection
of the young avenger, until he returned from the American forces.
She would be safer there than while being conducted through
forests, swarming with red and white foes.
The young spy now doffed the dress peculiar to Wacomet, which he
had worn, and adopted that of an Ottawa sub-chief, in which he
would be more likely to carry out his plans satisfactorily, both to
himself and Mad Anthony.
Nanette resolved to accompany the scout to the suburbs of the
“town,” and there await his return, leaving Effie and the Briton under
the watchful eye of Kenowatha.
While the boy—for boy Kenowatha may well be called—inwardly
chafed at being left to play an inactive part in the red drama that was
being enacted, he submitted with good grace when Nanette told him
that soon he should tread with her the path of vengeance, from dawn
till dawn.
Disguised as an Indian girl, yet bearing her rifle, the young She-wolf
—as the Girl Avenger had been styled by the savages—left the cave
with the scout, and, after a rapid walk of two hours, parted with him
on the suburbs of the Ottawa village, he promising to return against
midnight.
The girl had chosen a position a short distance from the river, and
within thirty feet of Turkey-foot’s lodge, the entrance of which she
faced.
The curtains of skins that formed the door were raised, thus
exposing the well-lighted interior of the spacious wigwam to the girl.
Presently six dark figures, gliding as noiseless as serpents over the
meadow, passed Nanette and entered the lodge.
After awhile a solitary figure, which she recognized as Joe Girty,
approached and walked among the warriors. Then followed the
drawing of the totems, and when the face of Stomah, the red artist,
was revealed to the avenger, her rifle flew to her shoulder, and,
before the gust of revenge left the girl, Stomah was ebbing out his
life-blood, as the reader has seen, over the totems. Stomah had paid
the penalty he had incurred upon a certain stormy November night,
years prior to the inauguration of our romance.
After the fatal shot, the avenger crept nearer the bank of the stream
and noiselessly reloaded her rifle. Then gliding back she listened to
the oath of the Death League, and saw its members leave the
Ottawa’s lodge.
She knew that she would not be followed that night, for Turkey-foot
had said as much, and suddenly, while she waited for the spy’s
return, Joe Girty’s whoop, indicating an important capture, rent the
air.
The cry seemed to have alarmed the entire village, for the lodges
near her poured forth their human contents, that hurried toward the
center of the “town.” With almost throbless heart, and rifle at half-
cock, the disguised avenger darted forward with natural caution, and
presently her worst fears were confirmed.
A multitude of torches illuminated a large space, that might be
termed a well-defined square, and around one man, secure in the
grip of the renegade, howled fully four hundred mad representatives
of six red nations.
Eager to witness what would follow, and confident of the
trustworthiness of her disguise, the Indian’s enemy placed her gun
against a wigwam and boldly joined the assembly.
In a few words Girty described the spy’s capture, and calmly Mark
Morgan awaited his doom.
“To the stake at once!” cried the loud voice of Wacomet, who saw in
the young scout the accepted lover of the girl he admired. “At once
to the tree! and we’ll send the white dog’s ashes to his master.”
This was greeted with shouts of approval, which still echoed down
the Maumee, when a pale-face sprung from the crowd and paused
before the prisoner.
It was Mitre St. Pierre!
“Mark Morgan,” he shrieked, with flashing eyes, as his bony fingers
closed on the spy’s throat, “where’s my gal? Tell me where she is
this minute, or, by the God that created us I’ll scatter your brains over
these braves.”
“Ye’ve axed the chap a question which he can’t answer while you
continner to choke his wind off,” said Girty, clutching the exasperated
trader’s arm. “Take yer fingers away, an’ we’ll see what he sez to yer
question.”
Sullenly the Frenchman complied, for he saw Girty’s hand touch the
ornamented hilt of a huge Spanish dirk that glistened in his girdle.
“Now answer my question, white dog!” he cried, stepping a pace
from the prisoner, who eyed him with something of a look of triumph
mingled with defiance. “Where’s Effie—my gal!”
“Where I left her, Mitre St. Pierre!”
The scout’s answer drove a yell of rage from the trader’s throat; he
shot forward, and before Joe Girty could interpose his hand, again
griped Mark’s throat, and a pistol-barrel glittered in the starlight.
There was murder in St. Pierre’s eyes.
To prevent the deed, several braves and Simon Girty darted forward;
but their assistance was not needed for the White Ottawa had
knocked the weapon from the Frenchman’s hand, and hurled its
frenzied owner to the earth. And when he rose again, looks told him
that his personal safety depended on quietude.
“Now what shall we do with the spy?” demanded Joe Girty.
A majority of the Indians cried aloud for immediate execution by the
terrible ordeal of fire; but the whites, exclusive of Joe Girty, overruled
them.
“Wait until the braves return with May and M’Lellan,” said Captain
McKee, addressing the savages. “Our men are sure to catch the two
spies who were prowling around Fort Miami, and when they return
we’ll burn all together.”
This speech was seconded by Simon Girty, Elliott and others, and
Nanette saw the spy dragged to a strong log cabin in the center of
the village, there to await the capture of May and M’Lellan, two
brother scouts, upon whose trail a band of savages had been for
several days, and whose capture, by the sanguine enemies of
Wayne, was regarded certain.
The crowd that accompanied Mark Morgan to the prison was
immense, and the cries that soared starward proclaimed the true
wishes of the Indians.
“I’ll wring her whereabouts from you yet,” Nanette heard Mitre St.
Pierre howl, “and if you give me any more such answers as you did
awhile ago I’ll tear your heart out and cram it into your mouth.”
The spy said nothing, but smiled faintly at the Frenchman, which
exasperated him the more.
All at once Nanette began to force herself through the crowd toward
the scout, and at last she found herself at his side.
“To-morrow night!” she whispered in his ears, while the crowd
greeted an outburst of anger from St. Pierre with hideous cries.
“We’ll come for you then—Kenowatha and I. They won’t catch May
and M’Lellan.”
He did not betray the avenger’s presence, and when the door closed
on him, and a triple guard was thrown around the cabin, the young
She-wolf hurried toward her home in the rocks, which she reached
an hour or more before dawn.
After telling Effie that her lover was on his journey to Wayne—for she
did not wish to inform the girl of the scout’s peril, she drew
Kenowatha aside and communicated the true state of affairs.
“I told him that we would come to-morrow night,” she said.
“We will come!” cried Kenowatha, eager for action. “We will enter the
village, and if any red-men cross our paths we’ll mark them to the
terror of their brothers. Kai Ja Manitou shall look down upon a new
mark—soon to be as terrible as the bloody half-moon. I have chosen
my mark, Nanette—a cross!”
“Oh, may the cross become as terrible as the crescent!” said
Nanette.
“It shall! it shall!” and Kenowatha’s hand stole to Nanette’s, and then
they returned to the fire.
When another day had faded, the two avengers glided from the
cavern, by an entrance seldom used, and started toward the Ottawa
village—intent upon the rescue of Wayne’s intrepid spy.
Effie St. Pierre, than whom woman never possessed a braver heart,
was content to remain with the wounded major until their return.
The two avengers fell confident of crowning their extremely
hazardous enterprise with success; but could they have foreseen the
events which transpired in the cave before dawn, it is doubtful
whether they would have went forth.
CHAPTER IX.
THE DEATH LEAGUE AT WORK.
A trio of phantom figures gliding cautiously up iron-gray rocks,
wearing a ghastly hue in the dim light of the stars. Now and then
they pause in the shade of a crag, and listen with upturned faces.
But, not a sound comes to their ears; the night overhead is as silent
as that which sleeps upon the bosom of the stream a hundred feet
below.
At last, when within perhaps thirty feet of the top of the cliff, they halt
at the mouth of a cave which seems to nestle far within the rocks,
and whose gloomy aspect, now relieved by the song of a nightingale
perched over the aperture, is as foreboding as such a spot can well
be.
Now the trio place their heads together, and this is what their lips
say:
“Who shall lead the way?”
The question is quickly answered.
“Wacomet.”
“It is well,” was the whispered answer. “Let Wacomet lead the way
into the She-wolf’s den, and Leather-lips and the Speckled Snake
will crawl at his heels.”
“Wacomet is sure that we have found the She-wolf’s lair, and our
brothers will do nothing but frighten lizards, owls and bats from their
holes. The red-man’s Terror, dreaming not of a foe, sleeps beside
the White Fox, and they will wake in the Ottawa’s arms.”
Again, but with Wacomet in the van, the trio move forward. The
Death League was at work.
Below the trio whom we have seen enter the Girl Avenger’s home,
the remainder of the band, headed by Joe Girty and Turkey-foot,
were exploring other cavernous openings, leading inward from the
cliff, for, to a certainty, Nanette’s cave was not known to her
enemies. Time and again she had been pursued by the Indians, but
she had in the end eluded them by darting down the precipice, and
disappearing in one of the openings I have described.
The place of Stomah had been filled by the bloodthirsty Speckled
Snake, one of the braves who accompanied Wacomet, and a savage
who had lost three brothers by the delicate hand of the girl demon.
He entered the League with a zest born of the grossest revenge, and
took the terrible oath with a vindictiveness that sent a shudder to his
co-swearers’ hearts.
The low stone roof of the passage in which the trio found themselves
compelled them to crawl forward, slowly groping their way like the
blind. Every now and then Wacomet would halt, and the three would
listen intently for several minutes, and, hearing nothing, glide forward
again.
Not a word was uttered by the trio, for the worn surface of the ground
over which they crawled told them that the cave at the termination of
the labyrinthine corridor was or had been inhabited.
All at once, as Wacomet groped his way around a gigantic rock,
which lay in the passage, the murmur of voices brought the Ottawa
to a sudden halt.
He had recognized a tone which had sent a thrill to his heart.
Wacomet knew all regarding the tragedy beneath the cottonwood—
he had listened to the narrative from the lips of Mitre St. Pierre,
scarce six hours before, and he had wished, from the depths of his
heart, that he might get possession of the girl his passions yearned
to possess. Then, and the Ottawa’s heart throbbed exultingly at the
thought, the trader would see her no more, until she had become his
pale-faced squaw—his slave. Ah! he knew a spot along the very
stream that sung its way to the Miami of the Lake, far below them,
which no feet save but one other’s than his had ever pressed—a
spot as difficult to find as the Holy Grail or the heart of Byron. Yes, he
rejoiced in the knowledge of such a spot as this, and thither, yet that
night, he would convey the owner of the dulcet tones he had first and
was still listening to, while he crouched beside the loose rock.
When Leather-lips and Speckled Snake reached Wacomet’s side,
and before they could hear the murmurs far ahead, the young
Ottawa turned suddenly upon them, and in low tones commanded
them to retrace their reptatory movements.
Wondering, yet not daring to seek by questions the cause the
strange command, they obeyed, and when they had reached an
acute angle, lately passed, Wacomet, full of plots and artifices,
spoke:
“Not far ahead,” he said, in the lowest of whispers, while the trio’s
heads touched in the almost palpable gloom, “is the She-wolf’s den;
but other animals surround her.”
A grunt of surprise escaped the listeners’ lips.
“While Leather-lips and Speckled Snake tarried behind, Wacomet
heard the voice of a pale-face for whose scalp Blue-Jacket would
give his own—Wells, the Black Snake’s big spy. And not only does
he sit at the She-wolf’s fire, but with him sit the two spies for whom
our braves now hunt.”
The two red auditors uttered ejaculations of astonishment that told
how welcome the intelligence was to them.
“Then let us glide forward and spring upon the white snakes,” said
Leather-lips, to whom fear was a stranger. “They watch not for the
red-man to-night, and when the Black Snake crawls down the
Maumee, we’ll throw their scalps into his teeth.”
These words found favor in the eyes of Speckled Snake, and when
the sorcerer finished, the twain drew their knives and made a motion
to resume their work; but Wacomet’s hand gently checked their
progress.
“We are not strong enough,” he said. “Think! the She-wolf,
Kenowatha, and the three pale snakes. There are but three of us—
Leather-lips, Speckled Snake and Wacomet. They will fight—fight to
the door of Manitou’s lodge. They possess the little guns (i. e.,
pistols); we the long shooting-irons and our knives. Brothers, there
must be more of us—the other members of our dread band must be
with us, and then—then, we keep the words we have given to the
Great Spirit. Go and seek them—our brothers. Wacomet will squat
here like the toad until you return, with other tomahawks that glitter in
girdles not our own. Go, hasten, before the darkness flies. Our
brothers will not be hard to find.”
While the two chiefs would fain have signalized themselves by a
conflict with the hated spies whom they believed to be in the cave,
they concluded to obey Wacomet, a chief superior to them in
appointment; so they glided away, and the Ottawa was left alone in
the darkness.
By adroit lying he had gained his ends, and prepared to carry his
plans into execution. The wily Ottawa knew that the three spies were
not in the cave, which also he knew to be tenanted by but two
persons—Effie St. Pierre and a man whose voice he could not
recognize, though he felt certain that he had heard it before. He
knew, too, that his brother chiefs would experience trouble in beating
up the other members of the Death League, and prided himself that
ere they could return to the cave with the help sent for, he would be
far on his journey to the “secret spot,” with the young white girl.
Again the Ottawa crept forward, and at length the turning of an angle
brought him in full view of the inmates of the cave.
In the center of the underground apartment a bright fire leaped
ceilingward, and bathed the entire chamber in a ruddy light. Upon a
couch of skins lay the form of a man, whose face the Ottawa at once
recognized, and an ejaculation of surprise and triumph, entirely
unexpected, bubbled to his lips.
“Not only will Wacomet take the girl,” muttered the Indian, as he
shrunk from the glare of the fire, “but he will take the skins, rifles and
gold-pieces, that the great red-coat at the fort offers for the pale-
face, who struck his young soldier. Ha! how came the pale soldier
here? for the trader shot him, and he fell into the stream.”
Then the lips grew still, and for several minutes the Ottawa watched
the inmates of the cave, himself as motionless as a statue. Frequent
companionship with the whites had made Wacomet, to a great
extent, a master of the English language, and every word that fell
from the lips of those whom he watched was intelligible to him. Effie
St. Pierre sat on the floor of the cave, near the British major’s couch,
braiding the wealth of hair which she had drawn over her shoulders.
With upturned face, Rudolph Runnion was breathing into her ears
the story of his twice-told passion, to which the girl was listening
calmly and in silence. Still there lurked around her lips a sneer, for
the tale to which she was listening; no doubt her mind flitted back to
the scene beneath the cottonwood, and her narrow escape from the
brutal lusts of the minion of an imbecile king who spoke.
By and by Wacomet ceased to listen to the conversation which had
informed him that Major Runnion’s wound was healing rapidly, and
turned his dark orbs upon the beautiful girl.
Yes, yes, she should be his; of her three lovers—two white, the third
red—he should be successful, and unable to restrain himself longer,
he crept forward.
Unsuspicious of the red serpent and wolf combined that approached
them, the twain remained motionless until—when Wacomet was very
near the mouth of the corridor—Effie suddenly darted to her feet,
and faced the intruder. Cursing in his bitter tongue, the alarm which
his foot had sprung, the Ottawa darted forward with a tiger-like
spring, and a moment later the cave was filled with smoke and a
deafening report.
A cry, half-shriek, half-groan, burst from Wacomet’s lips, and he
staggered like a drunken man, then sunk upon his knees.
Still clutching the smoking double-barreled pistol, Effie St. Pierre
waited for the smoke to clear away that she might witness the result
of her shot. She knew that it had not been without effect, for she had
seen the savage sink to the stones; but yet he might not be dead—
only wounded, and, like the bear, a greater terror when wounded
than before.
Despite his wounds the pistol-shot had forced Rudolph Runnion from
his couch, and now he looked around for something with which to
defend himself, for he believed that other braves had followed the
stricken chief into the cave. And for him to be taken prisoner by the
red-men now, was death, not at their hands, but by those of his
countrymen—his comrades in arms.
The smoke had not begun to clear away when a form, bleeding from
a frightful wound in the cheek, rushed through the thick volumes, and
knocked the pistol from Effie’s hand, before she could bring it to bear
upon him, so unexpected was the wounded Indian’s recovery.
A shriek escaped her when she found herself in the grasp of the
devil, who dashed the Briton to earth with a blow with his tomahawk,
as he advanced, for his own as well as the girl’s safety.
“Wacomet’s squaw at last!” hissed the savage, a brutal expression of
long-sought triumph lighting up his swarthy face. “The White Star is
Wacomet’s now! He thought to find She-wolf and White Fox here,
but ah! he has discovered better prizes than they. Where they
gone?”
Effie returned no answer which might furnish the savages with a clue
to the whereabouts of her young friends, and cause them to fall into
the hands of the red avengers.
The Ottawa did not press the question, but quickly bound the girl’s
hands, the while gloating over his triumph, and taunting her with the
poor result of her shot.
“When next I shoot I will take better aim,” said Effie, looking into the
Indian’s eyes, “and Wacomet must watch the white girl close that she
gets not another shot.”
“When she is Wacomet’s squaw she will not think of using the little
gun,” said the red victor, turning to bind Runnion’s hands. “Yes, in the
hidden hole, when the White Star sings songs to Wacomet’s
pappooses, she will forget how to use the little guns, and build the
Ottawa’s fire when the sun comes over the hill.”
In a short space of time Rudolph Runnion’s arms were pinioned at
his side, and a blow drew him to his feet. His face wore the ghastly
pallor that belongs to the dead, for while securing him, the Indian
had told him that a price was set upon his head, and that he
intended to deliver him up to the insatiate hounds of justice. For late
investigation had proved to the British at Fort Miami that St. Pierre’s
shot had not immediately proved fatal, if at all, and it was generally
believed that the slayer of Firman Campbell yet lived.
While Wacomet bound the Briton, Effie St. Pierre had obtained a
piece of keil, and, as well as her pinioned hands would permit, traced
these words upon the gray limestone wall of the cave:
“We are the prisoners of Wacomet the Ottawa, destined
for a hidden place somewhere.
Effie.”
The chief did not notice the “handwriting on the wall,” when he
turned to the girl, and pointed to the corridor with his tomahawk.
“We go to Wacomet’s home under the ground,” he said. “Come!”
The prisoners had stepped forward, when the Ottawa glanced
overhead and espied the long string of scalps taken from his red
brethren by the hands of the Terror of the Maumee. Beside them
hung rifles, tomahawks, and richly ornamented wampums—the
trophies of the Girl Avenger’s battles. A cry of indignation burst from
his lips, when his eye fell upon all these, and a minute later they
composed a confused heap at his feet. He caught up the wampums,
and threw them on the fire, the scalps quickly followed, and the rifles
and other weapons of Indian warfare crowned the crackling heap.
“Won’t the She-wolf howl when she returns to her den!” he cried,
with fiendish anticipation, as he gazed upon the work of his
revengeful hands.
Effie St. Pierre and Rudolph Runnion looked upon the Indian’s
revenge with different thoughts. They knew that the red tribes would
suffer terribly for that night’s work—that the Girl Avenger would not
rest until every dried scalp was replaced by a fresh one.
Suddenly Wacomet turned toward the corridor again, and presently
he was conducting his prisoners down the gloomy passage. He
forced them near his half-naked body by a rope of sinews attached
to their wrists, and secured to his girdle, and accompanied his
commands for silence by threats of a death too horrible to be
mentioned here.
At length the cool night-air fanned the faces of the trio, and looking
over Wacomet’s broad shoulders Effie saw the scintillating celestial
worlds. But a moment later the current of air was interrupted, and
Wacomet forced his prisoners into a natural niche before which he
had paused. In the deeper gloom of the niche the trio remained as
motionless as rocks, and presently they felt six figures,
undistinguishable in the blackness, move past, like giant animals. No
noise accompanied the new-comers to indicate their identity, but
Wacomet knew each one as he passed. They were the members of
the Death League, and Joe Girty crept at their head.
When the sounds died away in the gloomy distance, Wacomet drew
a breath of relief and again glided toward the opening. Soon the
three stood upon the flat rock just beyond the corridor, and faster
than was requisite for safety the red chief hurried the two whites
down the rocks.
In safety, however, the base of the cliff was reached, and along the
bank, toward the head of the little stream, the Indian bounded,
compelling his captives to keep pace with him.
He would seek the hidden place with his captives, and then return to
the Death League with a mouth full of lies.
Already he had planned a deceptive story, and his wounded cheek
would lend confirmation to his words.
Wacomet was a wily dog.
CHAPTER X.
THE RESCUE AND WHAT FOLLOWED.
The night was not far advanced when Kenowatha and the Girl
Avenger reached the vicinity of the Indian village, and concealed
themselves in the tall weeds that flourished there on the banks of the
Maumee. Beside the flambeaus, they had borne disguises down
from the cave, which they now proceeded to don. Kenowatha had
laid aside the gray fox-tails that had formed a portion of his fantastic
head-dress, divested himself of all the characteristic ornaments that
were wont to distinguish him as the renegade’s young protege, and
disguised himself as a common Indian boy—one of the many little
red vagabonds that run about the Indian villages deep in mischief,
long after their staider parents had yielded to the wooings of the
somnolent god. He realized the peril attached to the expedition he
had undertaken, he knew that a decree of outlawry had been passed
against him, that the common warrior who took his scalp should
forthwith be promoted to a full chief, and lead three hundred braves
against the Black Snake, who at that hour was marching from
Defiance to Victory.
Never before had he inaugurated action so perilous as the present
one, and it seemed the pinnacle of rashness for two children—for in
stature and years the avengers might be styled thus—to attempt to
rescue a white spy from the midst of a tribe that could muster five
thousand warriors.
The young twain felt no fear, and prepared for work as calmly as
though getting ready for a great festival. They resolved before
entering the woods to submit to no captivity.
“Better,” said Nanette, “to fall striking dead those who spared not our
own parents, than to undergo the terrible tortures inflicted by the
Indians and their scarlet queen.”
“Ay, ay,” returned Kenowatha, “we will not fall into their hands alive. If
they get us at all, it will be bereft of life. But they will not capture us,
girl. As yet I have not struck a single blow of vengeance. I am not to
die thus, no! no! It is decreed by high Heaven that my chosen mark
—the red cross—shall become as terrible as your bloody crescent.”
When the twain rejoined each other in the woods—for they had
separated to don their disguises—a silent pressure of hands,
followed, accompanied by a look that told how inseparably their fates
were linked together, by the strangest circumstances that ever
existed in the untraveled forest wilds.
Their guns were concealed in the grass, and armed but with their
knives, concealed but ready for use, about their persons, they stood
erect, listened a moment, then marched boldly forward. The most
careful observer would have proclaimed the couple what they
seemed to be—what they counterfeited—an Indian boy and girl—for
though they kept together they roamed about the village with a
nonchalance that would deceive the most suspicious. Still, as a
precautionary measure, they kept in the shadows as much as
possible, yet did not shrink from walking into the glare of the ground
fires whenever necessary. No suspicion seemed to attach itself to
the young would-be-rescuers, and the lazy savages little dreamed
that their deadliest foes were in their very midst.
By-and-by in their seemingly aimless saunterings they approached
the prison cabin wherein a noble man waited for them—waited
behind strong logs and a guard of stalwart braves, whose eyes noted
every thing that came near in the fitful starlight.
The avengers were prepared to find the prison strongly guarded, and
knew that to rescue the young spy they must have recourse to a
strategy seldom if ever used in red villages. The strategy, as perilous
and startling in its character as it was unique, found birth in the Girl
Avenger’s brain, and Kenowatha joined in the scheme. Crouching in
the gloomy shadow of a lodge, the twain remained silent until the
opaque clouds that crept over the western horizon had completely
clothed the sky in blackness, and vailed the shining faces of the
stars.
The prison-hut was built of strong beech logs, dove-tailed, after the
usual mode of constructing cabins in the West, and the roof was
composed of three layers of bark, each two inches thick, and
secured by strong wooden pins. The hut had been built for the
express purpose of securing prisoners, under the eye of the Girty
brothers, and other renegades, and many a noble red and white
captive had marched, in the noon of life, from its gloomy recesses to
the fatal stake, beneath a pellucid sky. The inmates of the prison-hut
were secured to a stake in the center of the structure, thus
preventing them from self-escape.
Kenowatha and the Girl Avenger crept to the rear of the hut without
attracting the attention of the guards who stood at the door, and with
the noiseless and ghostly movements of the lizard, as it ascends the
slimy wall, Nanette began to ascend the cabin to its roof. Breathless,
and with drawn knife Kenowatha waited below, ready to give the
signal of danger or drive the keen blade into the heart of the first red-
man who approached.
The rough exterior of the logs aided the young She-wolf in her
perilous undertaking, and presently she found herself grasping the
bark coverings of the hut.
For a moment she paused, but no signal or sound coming up from
below, she went to work. Bracing herself against the jutting logs at
the corner of the structure, her hands, by dint of toil, withdrew
several of the wooden pins, which she thrust into her bosom, and a
strip of bark was laid aside. Another and another noiselessly followed
it, and then, like a cat, the girl dropped from the roof into the interior
of the hut. The blackness of palpable darkness surrounded her, and
in the lowest of whispers she whispered a name:
“Mark Morgan!”
A noise—no voice—guided her to the captive, and a moment later
her knife noiselessly severed his bonds.
“I thought you never were coming,” whispered the young spy; “but
better late than never, you know, girl. The guards are dead? You
found the door too well secured for your little hands, eh?”
“The guards are as watchful as ever,” was the response. “We’ve
stolen a march on them—Kenowatha and I.”
Mark Morgan could scarcely repress an exclamation of astonishment
at the daring feat.
“Come!” whispered the Girl Avenger, “our work is but half-done. You
are not free yet. Go up first; I’ll follow.”
In silence Morgan groped his way to the corner, and clambered
toward the opening in the rock. While he was performing this action,
Nanette crept back to the post, upon whose worn surface she hastily
cut her dreaded crescent which she reddened with her own blood,
drawn with the point of her knife.
“They will know who’s been here when they open the door,” she
murmured, as she completed her work. “In my native country—La
Belle France—visiting people leave their card, and this is mine.
They’ve seen it before,” she added, with an inaudible chuckle, “for I
am a frequent, though not very welcome, visitor in these parts.”
The rescued spy reached the ground in safety; but as Nanette was
about to descend, a signal of danger came up from Kenowatha. She
paused on the roof, and awaited developments. Vainly she tried to
pierce the gloom, as she listened attentively. Suddenly, just the
slightest noise floated up to her, and then followed Kenowatha’s
signal for her to recommence her descent. The danger was passed.
A moment later, she stood on terra firma, and Kenowatha grasped
her hand.
“Stoop!” he whispered in her ear, while he drew the little stained
hand nearer the earth.
She obeyed.
Lower and lower the boy thrust her hand, until it touched a man’s
face, still warm, and covered with what she well knew to be blood.
Kenowatha allowed the girl’s hand to rest on the dead face for a
moment: then he took her finger and ran it through two deep gashes
on the bloody forehead, which formed his red mark—the cross.
“He came from the lodges, no doubt, to assassinate our spy,”
whispered Kenowatha, proud of his work. “But his work is ended.
They who wear my cross never admire it. Let us go; the guards may
scent their brother’s gore.”
The Girl Avenger rose to her feet, and a moment later the trio were
speeding toward the hidden guns, which reached, they resumed
their journey until they paused on the banks of the Maumee several
miles below the Ottawa town.
“You may tell her all now,” said Mark Morgan, as he took his
rescuer’s hands to say farewell. “I owe you a life, and if you guard
the girl—my white flower—well, I shall owe you another. I will not be
gone long. I will meet my General marching this way, and you shall
receive his thanks for what you have done to-night. Now mind you,
watch the girl well; keep sleepless eyes on these persons—
Wacomet the trader and Rudolph Runnion. When his wound permits,
unless watched, that red-coat will attempt something desperate, for I
tell you he has sworn that Effie shall not become my bride—before
that happens he will take her life, or worse.”
“We will watch those whom you mention as we watch over our own
lives against the Death League,” said Kenowatha. “Have no fears on
that score; the first hand that is raised against the girl drops in
death.”
Again Mark Morgan wrung the hands of the hunted twain, promised
to return in a few days with Wayne, with whom they yearned to fight,
and was gone.
“Now for the She-wolf’s den,” said Kenowatha, looking into Nanette’s
face. “We have not broken our word. We told Effie that we were
going deer-hunting, and if we haven’t hunted and saved one dear, to
her, then I’ve no judgment. Won’t the girl open her eyes when we tell
her all about her lover, and won’t that red-coat wish that the red-
skins had scalped him before we found him? Yes, yes, girl, that man
must be watched. Mark was right; he’s a dog. I’ve a mind to give him
over to them who want him. I’ll warrant you that he shot young
Campbell in cool blood, for I’ve seen the very devil in his black eyes.
I tell you, girl—Nanette—I’m in for giving him up. I can get to the fort
before dawn; the English will not give me over to those who seek my
heart’s blood. I will appoint a place for a small portion of the garrison
to meet me, say, to-morrow night, and then we will deliver the
murderer over to justice.”
The White Fox’s proposition found favor in Nanette’s eyes. She felt
that Rudolph Runnion was a murderer, deserving a murderer’s
reward, and then he was the enemy of those who were her friends.
Under these circumstances she believed that they would be justified
in delivering him over to the avengers of youthful blood, and thus
spoke to Kenowatha.
They were nearer the fort now than they would be when at the
hidden home, and intent upon accomplishing the work they had
resolved upon, Kenowatha glided from his companion’s side, and a
moment later was lost in the gloom. He promised to meet her in the
cave some time during the coming day.
Freed from the renegade’s protege, the young She-wolf resumed her
journey to her home, and at length crept into the gloomy opening
from whence they had executed their exit a few hours before. She
never dreamed of the thrilling events that had transpired during her
absence, and glided along the underground corridor entirely
unsuspicious of danger. When near the cave proper, she noted a
smoldering fire ahead, and very naturally concluded that Effie and
the British criminal had fallen asleep.
On, on, she glided, and at length entered the large apartment
wrapped in demi-gloom, despite the efforts of the dying blaze to the
contrary. The walls, in which a score of natural niches or holes
remained, were shrouded in almost impenetrable gloom, while the
deathlike silence boded ill for the hunted beauty. When she crossed
the threshold, she turned to the couches; but before her eyes could
greet them, a terrific yell broke the stillness, and the niches poured
forth the members of the Death League!
For hours Joe Girty and his band had waited for the coming of the
hunted ones, nor had they darted from their concealment until they
were satisfied that Kenowatha was not with the Girl Avenger. The
bloodthirsty band hemmed her in on all sides save one, for with the

(Download PDF) Inequalities in Analysis and Probability 3Rd Edition Odile Pons Ebook Online Full Chapter

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Download PDF) Inequalities in Analysis and Probability 3Rd Edition Odile Pons Ebook Online Full Chapter

Uploaded by

Copyright:

Available Formats

Inequalities In Analysis And Probability

3rd Edition Odile Pons

Analysis and Differential Equations 2nd Edition Odile

Functional Estimation for Density Regression Models and

Geometric Analysis of Quasilinear Inequalities on

Introduction to Linear Regression Analysis (Wiley

Probability and Statistical Inference 3rd Edition

Probability (Schaum's Outline Series, 3rd ed.) 3rd

Women in Science and Technology: Confronting

Statistics for Data Scientists: An Introduction to

This page intentionally left blank

B1948_1-Aoki.indd 6 9/22/2014 4:24:57 PM

Library of Congress Cataloging-in-Publication Data

British Library Cataloguing-in-Publication Data

Copyright © 2022 by World Scientific Publishing Co. Pte. Ltd.

For any available supplementary material, please visit

LaiFun - 12135 - Inequalities in Analysis and Probability.indd 1 6/8/2021 9:06:25 am

Preface of the Second Edition. The most important changes made

Preface of the First Edition. The inequalities in vector spaces and

vi Inequalities in Analysis and Probability (Third Edition)

of the diffusion of simple arithmetic results to a wide range of domains

2. Inequalities for Means and Integrals 37

viii Inequalities in Analysis and Probability (Third Edition)

3.2 Bounds for series . . . . . . . . . . . . . . . . . . . . . . . 69

4. Inequalities for Discrete Martingales 103

5. Inequalities for Time-Continuous Martingales 133

6. Stochastic Calculus 177

7. Functional Inequalities 215

8. Markov Processes 247

9. Inequalities for Processes 283

10. Inequalities in Complex Spaces 315

x Inequalities in Analysis and Probability (Third Edition)

10.6 Complex spaces of higher dimensions . . . . . . . . . . . . 331

Appendix A Probability 339

2 Inequalities in Analysis and Probability (Third Edition)

1.2 Cauchy and Hölder inequalities

Inequalities for finite series were first written as inequalities in a vector

The Cauchy inequality (1821) in the vector space Vn = Rn , for an

with equality if and only if f and g are proportional. Let µ be a positive

4 Inequalities in Analysis and Probability (Third Edition)

with equality if and only if the functions fi are proportional.

The analogous Cauchy inequality for a set of k series (xij )1≤i≤n,1≤j≤k

6 Inequalities in Analysis and Probability (Third Edition)

1.3 Inequalities for transformed series and functions

The Cauchy inequality has been extended by convex transformations of the

with equality if α1 x1 = · · · = αk xk . The exponential function is convex and

8 Inequalities in Analysis and Probability (Third Edition)

Replacing xi by xi yi−1 , with yi > 0 for i = 1, . . . , k, and adding to the left

For every convex function ϕ from R to R, Jensen’s inequality for the

which provided him the following interval

1.4 Applications in probability

Let (Ω, F, P ) be a probability space and let X be a real variable defined

10 Inequalities in Analysis and Probability (Third Edition)

Let X be a random vector of Rn+ , by concavity of the norm, we have

for all conjugates p, q ≥ 2. Let (Xi )i≥1 be a sequence of random variables in

The density of the ratio XY −1 of independent variables X and Y > 0

Bienaymé-Chebychev’s inequality has been formulated for a variable X of

Let (Xi )i=1,...,n be a sequence of independent random variables with

12 Inequalities in Analysis and Probability (Third Edition)

for a sequence of weighting constants such that cn = O(n−ν ), with ν > 21 .

For every a > 0, Bienaymé-Chebychev’s inequality for the maximum of

satisfies the inequalities

and for every λ > 0, there exists a constant K such that