Full download ebook at ebookgate.

com

Applied Optimization Methods for Wireless

Networks Hou Y.T.

https://ebookgate.com/product/applied-
optimization-methods-for-wireless-networks-hou-
y-t/

Download more ebook from https://ebookgate.com

More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Deploying wireless networks Wilton A.

https://ebookgate.com/product/deploying-wireless-networks-
wilton-a/

Security for Wireless Sensor Networks using Identity

Based Cryptography Patil

https://ebookgate.com/product/security-for-wireless-sensor-
networks-using-identity-based-cryptography-patil/

Wireless broadband networks David T. Wong

https://ebookgate.com/product/wireless-broadband-networks-david-
t-wong/

Protocols and architectures for wireless sensor

networks 1st Edition Holger Karl

https://ebookgate.com/product/protocols-and-architectures-for-
wireless-sensor-networks-1st-edition-holger-karl/
Advanced Wireless Sensing Techniques for 5G Networks
1st Edition Ashish Bagwari (Editor)

https://ebookgate.com/product/advanced-wireless-sensing-
techniques-for-5g-networks-1st-edition-ashish-bagwari-editor/

Building Wireless Community Networks Implementing the

Wireless Web 1st Edition Rob Flickenger

https://ebookgate.com/product/building-wireless-community-
networks-implementing-the-wireless-web-1st-edition-rob-
flickenger/

Wireless Sensor Networks 1st Edition Liam I. Farrugia

https://ebookgate.com/product/wireless-sensor-networks-1st-
edition-liam-i-farrugia/

Ambient Networks Co operative Mobile Networking for the

Wireless World 1st Edition Norbert Niebert

https://ebookgate.com/product/ambient-networks-co-operative-
mobile-networking-for-the-wireless-world-1st-edition-norbert-
niebert/

Wireless Broadband Networks Handbook 1st Edition John

R. Vacca

https://ebookgate.com/product/wireless-broadband-networks-
handbook-1st-edition-john-r-vacca/
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

Applied Optimization Methods for Wireless Networks

There is a growing trend in applying optimization approaches to study wireless

networks. This book is written to serve this need and is mainly targeted to
graduate students who are conducting theoretical research in wireless net-
works using optimization-based approaches. Specifically, the goal of this book
is to offer a course to graduate students by demonstrating a collection of
mathematical tools (with a focus on optimization techniques) and showing
how they can be used to address some challenging problems in wireless net-
works. Each chapter starts with a brief overview of a particular optimization
technique and is followed by a comprehensive coverage of a case study from
wireless networking. The choice of the case study in each chapter reflects state-
of-the-art wireless network research rather than discussing some classic but
outdated problems that are no longer of current research interest. There is a
strong emphasis on enhancing graduate students’ skills in mathematical mod-
eling, in particular, problem formulation, reformulation, and transformation.
For instructional purposes, each chapter contains a set of problems that are
suitable for homework exercises. A solution manual is available to instruc-
tors. PowerPoint slides for each chapter are available to both students and
instructors.

Y. Thomas Hou is a Professor in the Bradley Department of Electrical and

Computer Engineering, Virginia Tech, Blacksburg, Virginia, USA.
Yi Shi is a Research Scientist at Intelligent Automation Inc., Rockville,
Maryland, USA and an Adjunct Assistant Professor in the Bradley Department
of Electrical and Computer Engineering, Virginia Tech, Blacksburg, Virginia,
USA.
Hanif D. Sherali is a University Distinguished Professor Emeritus in the Grado
Department of Industrial and Systems Engineering, Virginia Tech, Blacksburg,
Virginia, USA. He is an elected member of the U.S. National Academy of
Engineering.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

Applied Optimization
Methods for Wireless
Networks

Y. Thomas Hou
Yi Shi
Hanif D. Sherali
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

University Printing House, Cambridge CB2 8BS, United Kingdom

Published in the United States of America by Cambridge University Press, New York

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107018808
© Cambridge University Press 2013
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2013
Printed in the United Kingdom by TJ International Ltd. Padstow Cornwall
A catalog record for this publication is available from the British Library
Library of Congress Cataloguing in Publication Data
ISBN 978110701880
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

To our parents
and
Our wives Tingting, Meiyu, and Semeen
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

Contents

Preface page xi
Acknowledgments xiv
Copyright permissions xvi

1 Introduction 1
1.1 Book overview 1
1.2 Book outline 3
1.3 How to use this book 7

Part I Methods for Optimal Solutions 9

2 Linear programming and applications 11
2.1 Review of key results in linear programming 11
2.2 Case study: Lexicographic max-min rate allocation and node lifetime
problems 13
2.3 System modeling and problem formulation 15
2.4 A serial LP algorithm based on parametric analysis 20
2.5 SLP-PA for the LMM node lifetime problem 25
2.6 A mirror result 27
2.7 Numerical results 30
2.8 Chapter summary 34
2.9 Problems 36

3 Convex programming and applications 38

3.1 Review of key results in convex optimization 38
3.2 Case study: Cross-layer optimization for multi-hop MIMO networks 40
3.3 Network model 41
3.4 Dual problem decomposition 46
3.5 Solving the Lagrangian dual problem 48
3.6 Constructing a primal optimal solution 50
3.7 Numerical results 51
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

viii Contents

3.8 Chapter summary 57

3.9 Problems 58

4 Design of polynomial-time exact algorithm 61

4.1 Problem complexity vs. solution complexity 61
4.2 Case study: Optimal cooperative relay node assignment 62
4.3 Cooperative communications: a primer 62
4.4 The relay node assignment problem 65
4.5 An optimization-based formulation 67
4.6 An exact algorithm 69
4.7 Proof of optimality 78
4.8 Numerical examples 82
4.9 Chapter summary 86
4.10 Problems 89

Part II Methods for Near-optimal and Approximation Solutions 93

5 Branch-and-bound framework and application 95
5.1 Review of branch-and-bound framework 95
5.2 Case study: Power control problem for multi-hop
cognitive radio networks 100
5.3 Mathematical modeling 101
5.4 Problem formulation 108
5.5 A solution procedure 110
5.6 Numerical examples 115
5.7 Chapter summary 119
5.8 Problems 119

6 Reformulation-linearization technique and applications 122

6.1 An introduction of reformulation-linearization
technique (RLT) 122
6.2 Case study: Capacity maximization for multi-hop cognitive radio
networks under the physical model 125
6.3 Mathematical models 126
6.4 Reformulation 129
6.5 A solution procedure 131
6.6 Numerical results 138
6.7 Chapter summary 144
6.8 Problems 146

7 Linear approximation 148

7.1 Review of linear approximation for nonlinear terms 148
7.2 Case study: Renewable sensor networks with wireless energy
transfer 151
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

ix Contents

7.3 Wireless energy transfer: a primer 153

7.4 Problem description 154
7.5 Renewable cycle construction 156
7.6 Optimal traveling path 162
7.7 Problem formulation and solution 164
7.8 Construction of initial transient cycle 174
7.9 Numerical examples 177
7.10 Chapter summary 179
7.11 Problems 179

8 Approximation algorithm and its applications - Part 1 191

8.1 Review of approximation algorithms 191
8.2 Case study: The base station placement problem 192
8.3 Network model and problem description 194
8.4 Optimal flow routing for a given base station location 196
8.5 Search space for base station location 197
8.6 Subarea division and fictitious cost points 199
8.7 Summary of algorithm and example 202
8.8 Correctness proof and complexity analysis 204
8.9 Numerical examples 207
8.10 Chapter summary 208
8.11 Problems 209

9 Approximation algorithm and its applications – Part 2 211

9.1 Introduction 211
9.2 Case study: The mobile base station problem 212
9.3 Problem and its formulation 213
9.4 From time domain to space domain 215
9.5 A (1 −  )-optimal algorithm 223
9.6 Numerical examples 233
9.7 Chapter summary 240
9.8 Problems 241

Part III Methods for Efficient Heuristic Solutions 243

10 An efficient technique for mixed-integer optimization 245
10.1 Sequential fixing: an introduction 245
10.2 Case study: Spectrum sharing for cognitive radio networks 246
10.3 Mathematical modeling and problem formulation 247
10.4 Deriving a lower bound 253
10.5 A near-optimal algorithm based on sequential fixing 254
10.6 Numerical examples 257
10.7 Chapter summary 258
10.8 Problems 260
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

x Contents

11 Metaheuristic methods 262

11.1 Review of key results in metaheuristic methods 262
11.2 Case study: Routing for multiple description video over wireless ad hoc
networks 264
11.3 Problem description 265
11.4 A metaheuristic approach 271
11.5 Numerical examples 274
11.6 Chapter summary 279
11.7 Problems 280

Part IV Other Topics 281

12 Asymptotic capacity analysis 283
12.1 Review of asymptotic analysis 283
12.2 Capacity scaling laws of wireless ad hoc networks 284
12.3 Case 1: Asymptotic capacity under the protocol model 287
12.4 Case 2: Asymptotic capacity under the physical model 295
12.5 Case 3: Asymptotic capacity lower bound under the generalized
physical model 300
12.6 Chapter summary 313
12.7 Problems 313

References 316
Index 327
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

Preface

Reasons for writing the book In recent years, there is a growing trend
in applying optimization approaches to study wireless networks. Such an
approach is usually necessary when the underlying goal is to pursue funda-
mental performance limits or theoretical results. This book is written to serve
this need and is mainly targeted to graduate students who are conducting the-
oretical research in wireless networks using optimization-based approaches.
This book will also serve as a very useful reference for researchers who wish
to explore various optimization techniques as part of their research methodolo-
gies.
To prepare a graduate student in either electrical and computer engineer-
ing (ECE) or computer science (CS) to conduct fundamental research in wire-
less networks, an ideal roadmap would include a series of graduate courses in
operations research (OR) and CS, in addition to traditional communications
and networking courses in ECE. These OR and CS courses would include
(among others) linear programming, nonlinear programming, integer program-
ming from OR, and complexity theory and algorithm design and analysis
from CS. Today, these courses are typically offered as core courses within the
respective disciplines. Instructors in OR and CS departments typically have
little knowledge of wireless networks and are unable to make a connection
between the mathematical tools and techniques in these courses and problem-
solving skills in wireless networks. ECE/CS students often find it difficult to
see how these courses would benefit their research in wireless networks. Due
to this gap between teaching scopes and learning expectations, we find that
the learning experience of our ECE/CS students in these courses is passive (or
“blind”) at best, as they do not have a clear picture of how these courses will
benefit their research.
An approach to bridge this gap is to offer a course that reviews a collection of
mathematical tools from OR and CS (with a focus on optimization techniques)
and shows how they can be used to address some challenging problems in
wireless networks. This book is written for this purpose.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

xii Preface

Each chapter in this book starts with a brief pointer to the underlying opti-
mization technique (with references to tutorials or textbooks so that students
can do an in-depth study in a formal course or on their own). The chapter then
immediately delves into a detailed case study in wireless networks to which
the technique will be applied. The focus in each chapter is to show how the
underlying technique can be used to solve a challenging problem in wireless
networks. To achieve this goal, we offer details on how to formulate a research
problem into a formal optimization model, reformulate or transform it in order
to improve mathematical tractability, and apply the underlying optimization
technique (with necessary customizations that are specific to the underlying
problem) to derive an optimal or near-optimal solution.
We have taught this course a number of times to ECE and CS graduate stu-
dents at Virginia Tech, using chapters from this book. The response from the
students has been overwhelmingly positive. In particular, we find that:

• For a graduate student (regardless of whether she has taken related OR or

CS courses), this course opens a new landscape or perspective on what opti-
mization techniques are available and how they can be applied to solve hard
problems in wireless networks;
• For those graduate students who are currently taking or will take the afore-
mentioned OR and CS courses, this course will help them better appreciate
the mathematical techniques in such OR and CS courses. The student will
also have a better purpose and a stronger motivation when she takes these
core courses in her future study.

We recognize that a single-volume book cannot possibly cover all tech-

niques exhaustively. Neither is it our intention to cover everything in one book.
Nevertheless, we have organized this book into four parts, where every chapter
focuses on a single technique. We hope this organization will serve our pur-
pose of offering a first course on this important subject of Applied Optimiza-
tion Methods in Wireless Networks. Our experience shows that after taking this
course, students become substantially more mature mathematically. Most of
them are able to consciously develop their learning paths into many areas in
OR and CS not covered in this book in order to further expand their own math-
ematical capabilities. This is an important ingredient in their life-long learning
and discovery.
Finally, the idea of having a book that offers a systematic coverage of opti-
mization techniques and their applications in wireless networks is a very nat-
ural one. Unfortunately (and quite surprisingly), after a rather thorough sur-
vey of the market (when we presented our initial proposal to our publisher),
we found that there were hardly any such books available. The closest book
that we can find that by Dimitri Bertsekas: Network Optimization: Continuous
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

xiii Preface

and Discrete Models (Athena Scientific, 1998). But that book still falls short
in showing students suitable case studies that are relevant to modern wireless
networks.
On the other hand, most other books addressing network optimization follow
a problem-oriented approach (vs. our method-oriented approach). They do not
offer a systematic treatment of the underlying optimization techniques like we
do in this book. To make this point clear, we quote the following text from the
preface of the book Combinatorial Optimization in Communication Networks,
edited by Maggie Xiaoyan Cheng, Yingshu Li, and Ding-Zhu Du (Springer,
2006), to explain why the problem-oriented approach was adopted by most
authors:
Two approaches were considered: optimization method oriented (starting from combi-
natorial optimization methods and finding appropriate network problems as examples)
and network problem oriented (focusing on specific network problems and seeking
appropriate combinatorial optimization methods to solve them). We finally decided to
use the problem-oriented approach, mainly because of the availability of papers: most
papers in the recent literature appear to address very specific network problems, and
combinatorial optimization comes as a convenient problem solver.

Such a problem-oriented approach offers a convenient way of composing a

book quickly (i.e., by assembling some research papers in the literature into
an edited volume). But books based on such a problem-oriented approach,
although useful as a reference book, do not teach graduate students optimiza-
tion techniques in a systematic manner. This critical dearth in the existing lit-
erature was our main motivation for writing this book and bringing it to the
community.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

Acknowledgments

This book is the fruit of close collaboration among the three authors for more
than ten years. We would like first to thank the former and current members of
our research group. In particular, many thanks to: Jia (Kevin) Liu, whose work
led to Chapter 3, Sushant Sharma, whose work led to Chapter 4, Liguang Xie,
whose work led to Chapter 7, Dr. Shiwen Mao, whose work led to Chapter 11.
We want to thank Huacheng Zeng, Liguang Xie, Xu Yuan, and Canming Jiang
for their help in proofreading some of the chapters. They also contributed to the
preparation and revision of the solution manual and Powerpoint Slides. With-
out their help, this book would not have reached its current shape. Some other
former and current members of our group, whose names were not mentioned
above but who contributed to this book in many other ways, include Sastry
Kompella, Cunhao Gao, Tong Liu, Xiaojun Wang, Xiaolin Cheng, Dr. Rongbo
Zhu, Dr. Lili Zhang, and Dr. Wangdong Qi.
We also want to thank the students in our ECE/CS 6570 class (Advanced
Foundations of Networking) over the years, who offered valuable feedback
to different versions of this book and helped us gauge the best match of such
materials for a graduate course in networking. In particular, those students who
took ECE/CS 6570 in Fall 2012 directly contributed to proofreading the final
book manuscript and their feedback is greatly appreciated.
We want to thank Dr. Philip Meyler, Acquisitions Editor of Cambridge Uni-
versity Press, who showed a genuine interest in the initial conception of this
book and encouraged us to move forward for a book proposal. He has also
been extremely patient with us when we requested a one year extension for
the final delivery of our manuscript. We thank him for his trust, patience, and
understanding, which allowed us to work on our schedule to bring this book
to reality. Looking back, we feel really lucky that we chose the best publisher
for this book. During the manuscript preparation stage, we worked with three
different Assistant Editors of Cambridge University Press – Elizabeth Horne,
Kirsten Bot, and Sarah Marsh. We thank all three of them, who worked dili-
gently with us at each step along the way to make this book a polished product.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

xv Acknowledgments

Tom Hou would like to thank Scott Midkiff, who recruited him to join the
Electrical and Computer Engineering Department at Virginia Tech in 2002.
Over the years, Scott has been a great colleague, a close friend, a resourceful
mentor, and, most recently, a supportive department head. The environment
that Scott and the department were able to offer has been instrumental to Tom’s
success in research and scholarship.
Finally, we would like to thank the National Science Foundation (NSF) and
the Office of Naval Research (ONR), whose funding support of our research
over the years led to this book.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

Copyright permissions

A highlight of this book is to include, in each chapter, a comprehensive presen-

tation of a case study that demonstrates how the particular optimization method
can be applied in solving a wireless networking problem. These case studies
are based on a number of papers written by the authors. The following list
acknowledges these publications and their respective journals and publishers
on a chapter-by-chapter basis. Portions of these papers have been adapted with
permission from the publishers as required. All rights are reserved as stipulated
by the various copyright agreements.

Chapter 2
Y.T. Hou, Y. Shi, and H.D. Sherali, “Rate allocation and network lifetime prob-
lems for wireless sensor networks,” IEEE/ACM Transactions on Networking,
vol. 16, no. 2, pp. 321–334, April 2008. Copyright © 2008 by, and with kind
permission from, IEEE.
Chapter 3
J. Liu, Y.T. Hou, Y. Shi, and H.D. Sherali, “Cross-layer optimization for
MIMO-based wireless ad hoc networks: routing, power allocation, and band-
width allocation,” IEEE Journal on Selected Areas in Communications, vol. 26,
no. 6, pp. 913–926, August 2008. Copyright © 2008 by, and with kind permis-
sion from, IEEE.
Chapter 4
S. Sharma, Y. Shi, Y.T. Hou, and S. Kompella, “An optimal algorithm for relay
node assignment in cooperative ad hoc networks,” IEEE/ACM Transactions on
Networking, vol. 19, issue 3, pp. 879–892, June 2011. Copyright © 2011 by,
and with kind permission from, IEEE.
Chapter 5
Y. Shi, Y.T. Hou, and H. Zhou, “Per-node based optimal power control
for multi-hop cognitive radio networks,” IEEE Transactions on Wireless
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

xvii Copyright Permissions

Communications, vol. 8, no. 10, pp. 5290–5299, October 2009. Copyright

© 2009 by, and with kind permission from, IEEE.
Chapter 6
Y. Shi, Y.T. Hou, S. Kompella, and H.D. Sherali, “Maximizing capacity in mul-
tihop cognitive radio networks under the SINR model,” IEEE Transactions on
Mobile Computing, vol. 10, no. 7, pp. 954–967, July 2011. Copyright © 2011
by, and with kind permission from, IEEE.
Chapter 7
L. Xie, Y. Shi, Y.T. Hou, and H.D. Sherali, “Making sensor networks immor-
tal: an energy-renewal approach with wireless power transfer,” IEEE/ACM
Transactions on Networking, vol. 20, issue 6, pp. 1748–1761, December 2012.
Copyright © 2012 by, and with kind permission from, IEEE.
Chapter 8
Y. Shi and Y.T. Hou, “Optimal base station placement in wireless sensor net-
works,” ACM Transactions on Sensor Networks, vol. 5, issue 4, article 32,
November 2009. Copyright © 2009 by, and with kind permission from, ACM.
Chapter 9
Y. Shi and Y.T. Hou, “Some fundamental results on base station movement
problem for wireless sensor networks,” IEEE/ACM Transactions on Network-
ing, vol. 20, issue 4, pp. 1054–1067, August 2012. Copyright © 2012 by, and
with kind permission from, IEEE.
Chapter 10
Y.T. Hou, Y. Shi, and H.D. Sherali, “Spectrum sharing for multi-hop network-
ing with cognitive radios,” IEEE Journal on Selected Areas in Communica-
tions, vol. 26, no. 1, pp. 146–155, January 2008. Copyright © 2008 by, and
with kind permission from, IEEE.
Chapter 11
S. Mao, Y.T. Hou, X. Cheng, H.D. Sherali, S.F. Midkiff, and Y.-Q. Zhang,
“On routing for multiple description video over wireless ad hoc networks,”
IEEE Transactions on Multimedia, vol. 8, no. 5, pp. 1063–1074, October 2006.
Copyright © 2006 by, and with kind permission from, IEEE.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

CHAPTER

1 Introduction

The present moment is the only moment available to us, and it is the door to
all moments.
Thich Nhat Hanh

1.1 Book overview

The goal of this book is to offer a course to graduate students by demon-

strating an important set of mathematical tools (with a focus on optimization
techniques) and to show how they can be used to address some challenging
problems in wireless networks.
Book organization This book consists of four parts.

• Part I, consisting of three chapters, is devoted to optimization and designing

algorithms that can offer optimal solutions.
• Part II, consisting of five chapters, is devoted to techniques that can offer
provably near-optimal solutions.
• Part III, consisting of two chapters, is devoted to some highly effective
heuristics.
• Part IV, consisting of only one chapter, is devoted to some miscellaneous
topics in the broader context of wireless network optimizations. This part
will be expanded in a future edition.

Structure of each chapter Each chapter starts with a brief overview of a

particular optimization technique and subsequently is followed by a compre-
hensive coverage of a case study in wireless networking. The goal of giving a
pointer to the underlying theory at the beginning of each chapter is to offer a
direction to students on what they should explore further in formal course work
or textbooks in these areas. These pointers are not meant to be comprehensive
tutorials, each of which could constitute a book on its own. The first section of

1
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

2 Introduction

each chapter is not meant to be a replacement of formal courses in operations

research (OR) or computer science (CS) and certainly does not offer a short
cut for students in their formal education in these subject areas.
Instead, the focus of each chapter is on how to apply the technique under dis-
cussion to solve a challenging problem in wireless networking. Each chapter
self-contained and shows all the details involved in problem formulation, refor-
mulation, and customization of optimization techniques in order to devise a
final solution. Our guiding principles in the choice of a case study in each
chapter are the following:

• To reflect state-of-the-art wireless network research rather than discuss some

classic but outdated problems that are no longer of current research interest.
In this spirit, the problems that we chose are mainly in the context of multi-
hop wireless networks, with the underlying wireless technologies being cog-
nitive radio (CR), multiple-input multiple-output (MIMO), and cooperative
communications (CC), among others.
• To offer a reasonable level of difficulty or challenge in each case study rather
than a simple problem that the students would hardly encounter in their
research. This is in contrast to simple and diverse examples that are typically
presented in standard optimization or algorithmic textbooks. In this book,
we want to show readers how the method introduced in each chapter can be
applied to solve a hard wireless networking problem, which they are likely to
encounter in research. For this purpose, each chapter is dominated by a case
study in terms of length coverage. We believe this approach will help readers
better appreciate the underlying method and to gain a better understanding
on its application in practice.
• To offer only essential background on the underlying wireless communi-
cation technology that is needed in formulating the problem rather than a
comprehensive overview of the technology. This is because the main goal
of this book is to learn various optimization techniques and apply each one
to solve a wireless networking problem as a case study rather than offering
a comprehensive tutorial on various wireless communication technologies.
Therefore, we decided to minimize the coverage on wireless technologies
and offer references that the readers can study further on their own.
• To help readers develop strong problem formulation and reformulation skills.
This can only be taught with examples with sufficient sophistication and
complexity. We believe such formulation/reformulation skills are important
for research and thus want to teach the readers such skills in detail in each
case study.

Key characteristics of the book

• Presents a collection of useful optimization techniques (one technique in

each chapter), with an emphasis on how each technique is put into action to
solve challenging wireless networking problems.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

3 1.2 Book outline

• Combines techniques from both operations research (OR) and computer sci-
ence (CS) disciplines, with a strong focus on solving optimization problems
in wireless networks.
• Shows various tricks and step-by-step details on how to develop optimiza-
tion models and reformulations, particularly in the context of cross-layer
optimization problems involving flow routing (network layer), scheduling
(link layer), and power control (physical layer).
• Discusses case studies that focus on multi-hop wireless networks (e.g., ad
hoc and sensor networks) and incorporates a number of advanced physical
layer technologies such as MIMO, cognitive radio (CR), and cooperative
communications (CC).
• Contains problem sets at the end of each chapter. PowerPoint slides for each
chapter are available to both the students and instructors. A solutions manual
is available to the instructors.

1.2 Book outline

This book has four parts. Part I of this book, consisting of Chapters 2 to 4,
is devoted to optimization and designing algorithms that can offer optimal
solutions.

• Chapter 2 reviews linear programming (LP) and shows how it can be

employed to solve certain problems in wireless networks. Although the LP
methodology itself is rather basic and straightforward, special care is still
needed to ensure that it is used correctly, as we demonstrate in the case study
in this chapter. The case study is rather interesting as it shows that even in
LP-based problem formulations, deep insights can be gained once we dig
deep into it. In particular, the case study considers lexicographic max-min
(LMM) rate allocation and node lifetime problems in a wireless sensor net-
work (WSN). We introduce the parametric analysis (PA) technique, which is
very useful in its own right. The concept of LMM is also important, and can
be employed as a fairness criterion for other problems in wireless network-
ing research. Through the case study in this chapter, the readers will gain a
rather deep understanding of LP and its applications in wireless networks.
• Chapter 3 reviews convex programming, which is a popular and power-
ful tool for studying nonlinear optimization problems. Once a problem is
shown to be a convex program, then there are standard solution techniques
and we may even directly apply a solver to obtain an optimal solution.
For many cross-layer convex optimization problems, the research commu-
nity is more interested in exploring a solution in its dual domain. There
are two reasons for this approach. First, many cross-layer problems, once
properly formulated in the dual domain, can be decomposed into sub-
problems, each of which may be decoupled from the other layers. Such a
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

4 Introduction

layering-based decomposition offers better insights and interpretations for

the underlying problems. Second, once a problem is decomposed in its dual
domain, the solution may be implemented in a distributed fashion, which is a
highly desirable feature for networking researchers. In the case study of this
chapter, we study a cross-layer optimization problem for a multi-hop MIMO
network, which involves variables at the transport, network, link, and physi-
cal layers. We show that the problem can be formulated as a convex program.
By studying the problem in its dual domain, we show that the dual problem
can be decomposed into two subproblems: one subproblem solely involving
variables at the transport and network layers and the other problem involving
variables at the link and physical layers. We describe how the dual problem
can be solved by a cutting-plane method and how the solution to the primal
problem can be recovered from the solution to the dual problem.
• Chapter 4 illustrates how an optimization problem can be solved by clever
algorithmic techniques from CS. For certain problems, general optimization
methods from OR may not always be the best approach. In fact, a formu-
lation following OR’s optimization approach may lead to a solution with
nonpolynomial-time complexity. But a solution with nonpolynomial-time
complexity does not mean that the problem is not in P. In fact, we may
well develop a different algorithm to solve the problem with polynomial-
time complexity. This is what we illustrate in this chapter, where we develop
a polynomial-time algorithm. In the case study, we consider a relay node
assignment problem in cooperative communications (CC). Our objective is
to assign a set of available relay nodes to different source–destination pairs
so as to maximize the minimum data rate among all the pairs. Following
the OR optimization approach, we show that the problem can be formulated
as a mixed-integer linear programming (MILP) problem, which is NP-hard
in general. But this does not mean that the problem is NP-hard. Instead, by
following a CS algorithm design approach, we develop a polynomial-time
exact algorithm for this problem. A novel idea in this algorithm is a linear
marking mechanism, which is able to achieve polynomial-time complexity
at each iteration. We give a formal proof of the optimality of the algorithm.

Part II of this book, consisting of Chapters 5 to 9, is devoted to techniques

that can offer provably near-optimal solutions.

• Chapter 5 presents the branch-and-bound framework and shows how it can

be applied to solve discrete and combinatorial optimization problems. Such
problems are typically considered most difficult in nonconvex optimization
and the branch-and-bound framework offers a general purpose and effec-
tive approach. The effectiveness of branch-and-bound resides in the care-
ful design of each component of its framework, such as computation of a
lower bound, local search of an upper bound, and selection of partitioning
variables (in the case of a minimization problem). It should be noted that
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

5 1.2 Book outline

the worst-case complexity of a branch-and-bound-based method remains

exponential, although a judicious design of each component could achieve
reasonable computational times in practice. In the case study, we consider a
per-node power control problem for a multi-hop CRN. This problem has a
large design space that involves a tight coupling relationship among power
control, scheduling, and flow routing, which is typical for a cross-layer opti-
mization problem. We develop a mathematical model and a problem formu-
lation, which is a mixed-integer nonlinear programming (MINLP) problem.
We show how to apply the branch-and-bound framework to design a solution
procedure.
• Chapter 6 presents the reformulation-linearization technique (RLT) for
deriving tight linear relaxations for any monomial. Simply put, RLT can be

applied to any polynomial term of the form ni=1 (xi )ci in variables xi , where
the ci -exponents are constant integers. Given such generality, RLT is a pow-
erful tool in deriving tight linear relaxations. In the case study, we consider
a throughput maximization problem in a multi-hop CRN under the signal-
to-interference-and-noise-ratio (SINR) model. We develop a mathematical
formulation for joint optimization of power control, scheduling, and flow
routing. We present a solution procedure based on the branch-and-bound
framework and apply RLT to derive tight linear relaxations for a product of
variables. In this case study, we also learn how to identify the core optimiza-
tion space for the underlying problem and how to exploit different physical
interpretations of the core variables in developing a solution.
• Chapter 7 presents a linear approximation algorithm, which is a powerful
method to tackle certain nonlinear optimization problems. We show how
such an approach could be employed to solve a nonlinear programming
(NLP) problem in a wireless sensor network (WSN). In addition to the linear
approximation technique, the problem in the case study is interesting on its
own, and shows how the so-called wireless energy transfer technology can
be employed to address network lifetime problems in a WSN.
• Chapter 8 shows how to design a polynomial-time approximation algorithm
to provide an (1 − ε)-optimal solution to a nonconvex optimization problem.
The case study focuses on a classic base station placement problem in a
WSN. The design of the (1 − ε)-optimal approximation algorithm is based
on several clever techniques such as discretization of cost parameters (and
distances), partitioning of the search space into a finite number of subareas,
and representation of subareas with fictitious points (with tight bounds on
costs). These three techniques can be exploited to develop approximation
algorithms for other problems. We prove that the approximation algorithm is
(1 − ε)-optimal.
• Chapter 9 is a sequel to Chapter 8. Again, our interest is on the design of a
(1 − ε)-optimal approximation algorithm for a mobile base station problem.
But the problem is much harder than that in the last chapter. By allowing
the base station to be mobile, both the location of the base station and the
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

6 Introduction

multi-hop flow routing in the network are time-dependent. To address this

problem, we show that as far as the network lifetime objective is con-
cerned, we can transform the time-dependent problem to a location (space)-
dependent problem. In particular, we show that flow routing only depends
on the base station location, regardless of when the base station visits this
location. Further, the specific time instances for the base station to visit a
location are not important, as long as the total sojourn time for the base
station to be present at this location is the same. This result allows us to
focus on solving a location-dependent problem. Based on the above result,
we further show that to obtain a (1 − ε)-optimal solution to the location-
dependent problem, we only need to consider a finite set of points within
the smallest enclosing disk for the mobile base station’s location. Here, we
follow the same approach as that in Chapter 8, i.e., discretization of energy
cost through a geometric sequence, division of a disk into a finite number
of subareas, and representation of each subarea with a fictitious cost point
(FCP). Then we can find the optimal sojourn time for the base station to stay
at each FCP (as well as the corresponding flow routing solution) so that the
overall network lifetime (i.e., sum of the sojourn times) is maximized via a
single LP Problem. We prove that the proposed solution can guarantee that
the achieved network lifetime is at least (1 − ε) of the maximum (unknown)
network lifetime. This chapter offers some excellent examples on how to
transform a problem from time domain to space domain and how to prove
results through construction.

Part III, consisting of Chapters 10 and 11, is devoted to some highly effective
heuristics.

• Chapter 10 presents an effective approach to address a class of mixed-

integer optimization problems. The technique, called sequential fixing (SF),
is designed to iteratively determine (fix) binary integer variables. It is a
heuristic procedure and has polynomial-time complexity. Its performance
is typically measured by comparing its solution value to some performance
bound, e.g., a lower bound for a minimization problem, or an upper bound
for a maximization problem. Based on our own experience, we find that that
this SF technique is very efficient and can offer highly competitive solutions.
As a case study, we study an optimization problem in a multihop CRN. Since
the problem formulation is an MINLP problem, we develop a lower bound to
estimate the optimal objective value. Subsequently, we present an SF algo-
rithm for this optimization problem. Numerical examples show that the solu-
tions produced by this SF algorithm can offer objective values that are very
close to the computed lower bounds, thus confirming their near-optimality.
• Chapter 11 presents metaheuristic methods, which are an important class of
heuristic methods and have been applied to solve some very complex prob-
lems in wireless networks. In this chapter, we give a review of some well-
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

7 1.3 How to use this book

known metaheuristic methods (e.g., basic local search, simulated annealing

(SA), tabu search (TS), and genetic algorithms (GA)). In the case study,
we focus on developing a GA-based method to solve a multi-path routing
problem for MD video. We find that a GA-based solution is eminently
suitable to address this particular problem, which involves complex objec-
tive functions and exponential solution space. By exploiting the survival-of-
the-fittest principle, a GA-based solution is able to evolve to a population
of better solutions after each iteration and eventually offers a near-optimal
solution.

Part IV, currently consisting of only Chapter 12, is devoted to some miscel-
laneous topics in the broader context of wireless network optimizations. This
part will be further expanded to include other topics in a future book edition.

• Chapter 12 presents an asymptotic capacity analysis for wireless ad hoc net-

works. Such an analysis addresses an achievable per-node throughput when
the number of nodes goes to infinity. We focus on so-called random net-
works, where each node is randomly deployed and each node has a randomly
chosen destination node. In this asymptotic capacity analysis, the results are
derived in the form of (·), O(·), and (·) and the underlying analysis is
very different from what we do in the other chapters, which focus on opti-
mization problems for finite-sized networks. We show that the asymptotic
capacity analysis heavily depends on the underlying interference model. In
this chapter, we consider three interference models (i.e., the protocol model,
the physical model, and the generalized physical model) and show how to
develop asymptotic capacity bounds for each model.

1.3 How to use this book

This book is written as a textbook and is mainly aimed at graduate stu-

dents (particularly students in electrical and computer engineering) pursuing
advanced research and study in wireless networks. The book could be adopted
for a second or third graduate course in networking. The prerequisites for this
course are a graduate course in networking.
At Virginia Tech, we have been this book in an Electrical and Computer
Engineering (ECE) and Computer Science (CS) cross-listed course titled
“ECE/CS 6570: Advanced Foundations of Networking.” The prerequisite for
this course is a graduate course in networking. We cover Chapters 2, 3, 4, 5,
8, 10, and 11 in a one semester course and the response from the students has
been overwhelmingly positive! For each chapter, we have prepared PowerPoint
slides, which are available to both the students and instructors. We have also
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

8 Introduction

prepared a solutions manual for all end-of-chapter problems, which is available

to the instructors.
In addition to its primary role as a graduate textbook, researchers in
academia who are active in conducting research in wireless networks will
find this book a very useful reference to expand their toolboxes in problem
solving. Further, researchers and engineers in industry and government labo-
ratories who perform active research in wireless networks will also find this
book to be a useful reference.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

PART

I Methods for Optimal Solutions

9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

CHAPTER

2 Linear programming and

applications

The real voyage of discovery consists not in seeking new landscapes but in
having new eyes.
Marcel Proust

2.1 Review of key results in linear programming

Linear programming (LP) is a problem consisting of a linear objective function

and a set of linear constraints (equations or inequalities) with real variables.
Such a problem aims to maximize or minimize a specific objective function
by systematically choosing the values of the real variables within an allowed
set (i.e., solution space). In an LP problem, both the objective function to be
optimized and all the constraints restricting the variables are linear.
A general form of an LP problem is as follows:
Maximize c1 x1 + c2 x2 + . . . + cn xn
subject to a11 x1 + a12 x2 + . . . + a1n xn ≤ b1
a21 x1 + a22 x2 + . . . + a2n xn ≥ b2
a31 x1 + a32 x2 + . . . + a3n xn = b3
..
.
am1 x1 + am2 x2 + . . . + amn xn = bm
xj ≥ 0 (1 ≤ j ≤ n).
The function c1 x1 + c2 x2 + . . . + cn xn is the objective function to be maxi-
mized, where c1 , c2 , . . . , cn are constant coefficients, and x1 , x2 , . . . , xn are
the so-called decision variables to be determined. The equality (or inequal-

ity) nj=1 aij xj = (or ≤ or ≥) bi is the ith constraint, where aij is a constant

11
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

12 Linear programming and applications

coefficient. The constant column vector with the ith component bi , 1 ≤ i ≤ m,

is the right-hand-side (RHS) vector. The restrictions xj ≥ 0, 1 ≤ j ≤ n are
called nonnegativity constraints. Sometimes, they may be replaced by general
lower and upper bounds on the decision variables. Observe that a maximization
problem can be easily converted into a minimization problem and conversely,
noting that
Maximum c1 x1 + c2 x2 + . . . + cn xn
= −Minimum − c1 x1 − c2 x2 − . . . − cn xn .
In a general form of an LP problem, both equalities and inequalities may
appear. Performing a simple manipulation on the inequalities (by adding or
subtracting nonnegative slack variables, see [10] for more details), any LP
problem can be transformed to an equivalent standard form as follows:
Maximize cT x
subject to Ax = b (2.1)
x ≥ 0,
where x is the vector of nonnegative continuous variables, c is the vector of
coefficients in the objective function, A is the matrix of coefficients in the con-
straints, and b is the RHS vector for the constraints. This simple transformation
is convenient for developing a general algorithmic procedure to solve and ana-
lyze LPs [10].
An LP problem can be solved optimally in a time complexity of O(nA 3 )
(e.g., using interior point methods [10]), where nA is the number of variables
in the standard form representation. Therefore, once we are able to formu-
late a problem as an LP, we can solve this problem in polynomial time by
employing open-source solvers (e.g., GLPK [52]) or commercial solvers (e.g.,
CPLEX [31] and Lindo [95]).
For an LP problem in standard form, there is another associated LP problem
called the dual problem, which is given as follows:
Minimize wT b
subject to AT w ≥ c,
where the vector of dual variables w is unrestricted in sign [10]. The dual
problem has some important properties and economic interpretation, and can
also be used to obtain the solution to the original LP problem (e.g., with the
dual simplex method or the primal-dual algorithm [10]). Based on the duality
relationship in LP theory, both the original LP and its dual LP can be solved
simultaneously by standard LP techniques in polynomial time.
We want to emphasize that all the variables in an LP problem should be real
(continuous) variables. If an optimization problem has a linear objective func-
tion, linear constraints, and some real variables and some integer/binary vari-
ables, then this optimization problem is a mixed-integer linear programming
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

13 2.2 Case study

(MILP), which is not an LP problem. A special case of MILPs are integer

linear programs (ILP), where all the decision variables are integer/binary
restricted. Unlike LP, there does not exist a polynomial-time algorithm to solve
an MILP or an ILP optimally.
In the rest of this chapter, we offer a case study on how LP can be applied to
address some interesting but difficult problems from wireless sensor networks.

2.2 Case study: Lexicographic max-min rate allocation and

node lifetime problems
Consider a wireless sensor network consisting of battery-powered nodes for
data collection. Although there have been significant improvements in pro-
cessor design and computing, advances in battery technology still lag behind,
making energy resource considerations the fundamental challenge in wire-
less sensor networks. Consequently, there have been active research efforts
on performance limits of wireless sensor networks. These performance limits
include, among others, network capacity (e.g., [78]) and network lifetime (e.g.,
[24]). Network capacity typically refers to the maximum amount of bit volume
that can be successfully delivered to the base station (“sink node”) by all the
nodes in the network, while network lifetime refers to the maximum time limit
for which all the nodes in the network remain alive until one or more nodes
drain up their energy.
In this chapter, we consider an overarching problem that encompasses both
performance metrics. In particular, we study the network capacity problem
under a given network lifetime requirement. Specifically, for a wireless sensor
network where each node is provisioned with an initial energy, if all nodes are
required to live up to a certain lifetime, what is the maximum amount of bit
volume that can be generated by the entire network? At first glance, it appears
desirable to maximize the sum of rates for all the nodes in the network, sub-
ject to the condition that each node can meet the network lifetime requirement.
Mathematically, this problem can be formulated as a linear programming (LP)
problem (see Section 2.3.2) within which the objective function is defined as
the sum of rates over all the nodes in the network and the constraints require
that (i) flow balance is preserved at each node, and (ii) the energy constraint
at each node is met for the given network lifetime requirement. However, the
solution to this problem shows (see Section 2.7) that there exists a severe bias
in rate allocation among the nodes, despite that the sum of bit rates is maxi-
mized. In particular, those nodes that consume the least amount of power on
their data path toward the base station are allocated with much more bit rates
than other nodes in the network. Consequently, the data collection behavior for
the entire network only favors certain nodes that have this property, while other
nodes will be unfavorably penalized with much smaller bit rates.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

14 Linear programming and applications

The fairness issue associated with the network capacity maximization objec-
tive calls for careful consideration in rate allocation among the nodes. In this
chapter, we study the rate allocation problem in an energy-constrained sen-
sor network for a given network lifetime requirement. Our objective is to
achieve a certain measure of optimality in the rate allocation that takes into
account both fairness and bit rate maximization. We employ the so-called Lex-
icographic Max-Min (LMM) criterion [99], which maximizes the bit rates
for all the nodes under the given energy constraint and network lifetime
requirement. At the first level, the smallest rate among all the nodes is max-
imized. Then, we continue to maximize the second smallest rate level and so
forth. The LMM rate allocation criterion is appealing since it addresses both
fairness and efficiency (i.e., bit rate maximization) in an energy-constrained
network.
A naive approach to the LMM rate allocation problem would be to apply
a max-min-like iterative procedure. Under this approach, successive LPs are
employed to calculate the maximum rate at each level based on the avail-
able energy for the remaining nodes, until all nodes use up their energy. We
call this approach “serial LP with energy reservation” (SLP-ER). We will
show that, although SLP appears intuitive, it is likely to offer an incorrect
solution. To understand how this could happen, we must understand a funda-
mental difference between the LMM rate allocation problem described here
and the classical max-min rate allocation in [14]. Under the LMM rate allo-
cation problem, the rate allocation is implicitly coupled with a flow routing
problem, while under the classical max-min rate allocation, there is no rout-
ing problem involved since the routes for all flows are given. As it turns out,
for the LMM rate allocation problem, any iterative rate allocation approach
that requires energy reservation at each iteration is incorrect. This is because,
unlike max-min, which addresses only the rate allocation problem with fixed
routes and yields a unique solution at each iteration, for the LMM rate allo-
cation problem, there usually exist nonunique flow routing solutions corre-
sponding to the same rate allocation at each level. Each of these flow routing
solutions will yield different available energy levels on the remaining nodes
for future iterations and so forth. This will lead to a different rate alloca-
tion vector, which may not coincide with the optimal LMM rate allocation
vector.
In this chapter, we show a correct approach to solve the LMM rate alloca-
tion problem. Our approach employs the so-called parametric analysis (PA)
technique [10] in LP to determine the minimum set of nodes at each rate level
that must deplete their energy. We call this approach serial LP with PA (SLP-
PA). We also extend the PA technique for the LMM rate allocation problem to
address the so-called maximum node lifetime curve problem in [20], which
we call the LMM node lifetime problem. More importantly, we show that
there exists a simple and elegant mirror relationship between the LMM rate
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

15 2.3 System modeling and problem formulation

allocation problem and the LMM node lifetime problem. As a result, it is suf-
ficient to solve only one of these two problems.
The remainder of this chapter is organized as follows. Section 2.3 intro-
duces the network and energy model, and formulates the LMM rate alloca-
tion problem. Section 2.4 describes the SLP-PA algorithm for the LMM rate
allocation problem. Section 2.5 introduces the LMM node lifetime problem
and shows how to apply the SLP-PA algorithm to solve it. Section 2.6 reveals
an interesting mirror relationship between the LMM rate allocation problem
and the LMM node lifetime problem. Section 2.7 presents numerical results.
Section 2.8 summarizes this chapter.

2.3 System modeling and problem formulation

We consider a two-tier architecture for wireless sensor networks. Figs. 2.1(a)

and (b) show the physical and hierarchical network topologies for such a net-
work, respectively. There are three types of nodes in the network – namely,
micro-sensor nodes (MSNs), aggregation and forwarding nodes (AFNs) – and
a base station (BS). The MSNs can be application-specific sensor nodes (e.g.,
temperature, chemical, video) and they constitute the lower-tier data collection
network. They are deployed in groups (or clusters) at strategic locations for
surveillance and monitoring applications. The MSNs are small and low cost.
The objective of an MSN is simple: once triggered by an event, it starts to
collect sensing data and sends it to the local AFN (in one hop).
For each cluster of MSNs, there is one AFN, which is different from an MSN
in terms of physical properties and functions. The primary functions of an AFN
are: (1) data aggregation (or “fusion”) for data flows from the local cluster of
MSNs, and (2) forwarding (or relaying) the aggregated information to the next
hop AFN (ultimately to the base station). Although an AFN is expected to be
provisioned with much more energy than an MSN, it also consumes energy
at a substantially higher rate (due to data transmission over longer distances).
Consequently, an AFN has a limited lifetime. Upon depletion of energy at an
AFN, we expect that the coverage for the particular area under surveillance is
lost, despite the fact that some of the MSNs within the cluster may still have
remaining energy.
The third component in the two-tier architecture is the base station. The
base station is, essentially, the sink node for data streams from all the AFNs
in the network. We assume that there is sufficient energy resource available
at the base station and thus there is no energy constraint at the base station.
In summary, from networking perspective, the lower-tier MSNs are for data
acquisition, while the upper-tier AFNs are for transporting data to the base
station.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

16 Linear programming and applications

Figure 2.1 Reference

architecture for two-tier wireless
sensor networks.
Base station
(BS)

Aggregation and forwarding node

(AFN) Micro-sensor node
(MSN)

(a) Physical topology.

Base station
(BS)

Upper tier
AFN

Lower tier

MSN

(b) A hierarchical view.

2.3.1 Energy model

Table 2.1 lists the notation used in this chapter. We focus on energy consump-
tion at AFNs. For AFN i, we assume that the aggregated bit rate collected
locally (after data fusion) is ri , i = 1, 2, . . . , N . These collected local bit
streams must be relayed to the base station. Our objective is to maximize the
ri -values according to the LMM criterion (see Definition 2.1) under a given
network lifetime requirement.
For an AFN, we assume that energy consumption due to transmission and
reception is the dominant source of energy consumption [4]. The power dissi-
pation at a radio transmitter can be modeled as

utij = Cij · fij , (2.2)

9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

17 2.3 System modeling and problem formulation

Table 2.1 Notation.

General notation for the LMM rate and LMM lifetime problems

B The base station

Cij (or CiB ) The power consumption coefficient for transmitting data from AFN
i to AFN j (or the base station B)
ei The initial energy at AFN i
fij (or fiB ) Data rate from AFN i to AFN j (or the base station B)
n The number of distinct elements in the sorted LMM-optimal
rate/lifetime vector
N Total number of AFNs in the network
Si The minimum set of nodes that reach their energy constraint limits at
the ith level
Ŝi The set of all possible AFNs that may reach their energy constraint
limits at the ith level, Si ⊆ Ŝi
Vij (or ViB ) Total data volume from AFN i to AFN j (or the base station B)
α Path-loss index
β 1 , β2 Constant terms in transmission power model
ρ The power consumption coefficient for receiving data
Symbols used for the LMM rate problem
ri The local bit rate collected at AFN i
gi The ith element in the sorted LMM-optimal rate vector, where g1 ≤
g2 ≤ . . . ≤ gN
T The network lifetime requirement
λi The ith rate level in the sorted LMM-optimal rate vector, i.e.,
λ1 (= g1 ) < λ2 < . . . < λn (= gN )
δi = λi − λi−1 , the difference between λi and λi−1
Symbols used for the LMM lifetime problem
ri The rate requirement at AFN i
ti The node lifetime at AFN i
τi The ith element in the sorted LMM-optimal lifetime vector, where
τ1 ≤ τ2 ≤ . . . ≤ τN
μi The ith drop point in the sorted LMM-optimal lifetime vector, i.e.,
μ1 (= τ1 ) < μ2 < . . . < μn (= τN )
ζi = μi − μi−1 , the difference between μi and μi−1

where utij is the power dissipated at AFN i when it is transmitting to node j ,

fij is the rate from AFN i to node j , Cij is the power consumption cost of
radio link i → j and is given by

Cij = β1 + β2 · dijα , (2.3)

where β1 and β2 are two constant terms, dij is the distance between these
two nodes, and α is the path-loss index. Typical values for these param-
eters are β1 = 50 nJ/b, β2 = 0.0013 pJ/b/m4 [67], and α = 4 [129]. Since
the power level of an AFN’s transmitter can be used to control the distance
coverage of an AFN (e.g., [127; 131; 166]), different network flow rout-
ing topologies can be formed by adjusting the power level of each AFN’s
transmitter.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

18 Linear programming and applications

The power dissipation at a receiver can be modeled as


uri = ρ · fki , (2.4)
k=i

where k=i fki (in b/s) is the rate of the received data stream at AFN i. A
typical value for the parameter ρ is 50 nJ/b [67].
To ensure data transmissions at different AFNs do not interfere with each
other, we assume that a scheduling mechanism has been employed. The details
of such a scheduling mechanism is beyond the scope of this chapter.

2.3.2 The LMM rate allocation problem

Before we formulate the LMM rate allocation problem, we revisit the maxi-
mum capacity problem (with “bias” in rate allocation) that was discussed in
Section 2.2. For a network with N AFNs, suppose that the rate of AFN i is
ri , and that the initial energy at this node is ei (i = 1, 2, . . . , N). For a given
network lifetime requirement T (i.e., each AFN must remain alive for at least
the time duration T ), the maximum sum of rates can be found by the following
LP problem:
N
MaxCap Maximize i=1 ri
 
subject to fiB + j =i fij − k=i fki = ri (1 ≤ i ≤ N ) (2.5)
 
k=i ρfki T + j =i Cij fij T + CiB fiB T ≤ ei (1 ≤ i ≤ N ) (2.6)
fij , fiB ≥ 0 (1 ≤ i, j ≤ N, j  = i).

The set of constraints in (2).5 are the flow balance equations. They state that
the total bit rate transmitted by AFN i is equal to the total bit rate received by
AFN i from other AFNs, plus the bit rate generated locally at this node (ri ).
Note that we allow flow splitting at a node so as to achieve more flexibility
in flow routing and load balancing in the network. The set of constraints in
(2).6 are the energy constraints. They state that for a given network lifetime
requirement T , the energy consumed for communications (i.e., transmitting
and receiving) cannot exceed the initial available energy.
Note that fki , fij , fiB , and ri are variables and T is a constant (the
given network lifetime requirement). Since MaxCap is an LP problem, it
can be solved in polynomial time [10]. Unfortunately, as we will see in the
numerical results (Section 2.7), the solution to this MaxCap problem favors
those AFNs whose data paths consume the least amount of power toward
the base station. Consequently, although the sum of rates is maximized over
T , the specific bit rate allocation among the AFNs (i.e., ri ’s) favors those
AFNs that have this property, while the other AFNs are unfavorably allocated
with much smaller (even close to 0) data rates. As a result of this unfair-
ness, the effectiveness of the sensor network in performing data collection is
questionable.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

19 2.3 System modeling and problem formulation

To address this fairness issue, we can employ the so-called lexicographic

max-min (LMM) rate allocation strategy [99], which bears some similarity to
the max-min rate allocation in data networks [14].1 Under LMM rate allo-
cation, we start with the objective of maximizing the bit rate of all nodes
until one or more nodes reach their limits for the given network lifetime
requirement. Given that the first level of the smallest rate allocation is max-
imized, we continue to maximize the second level of rate for the remain-
ing nodes that still have available energy, and so forth. More formally,
denote g = [g1 , g2 , . . . , gN ] as the sorted version (i.e., g1 ≤ g2 ≤ . . . ≤ gN )
of the rate vector r = [r1 , r2 , . . . , rN ], with ri corresponding to the rate
of node i. We then have the following definition for LMM-optimal rate
allocation:

Definition 2.1
LMM-optimal rate allocation For a given network lifetime require-
ment T , a sorted rate vector g = [g1 , g2 , . . . , gN ] yields an LMM-optimal
rate allocation if and only if for any other sorted rate allocation vector
ĝ = [ĝ1 , ĝ2 , . . . , ĝN ] with ĝ1 ≤ ĝ2 ≤ . . . ≤ ĝN there exists k, 1 ≤ k ≤ N ,
such that gi = ĝi for 1 ≤ i ≤ k − 1 and gk > ĝk .

2.3.3 Two incorrect approaches

Before we present a correct solution to the optimal LMM rate allocation prob-
lem, we discuss two incorrect solutions and explain why they cannot provide
an LMM-optimal solution.
Serial LP with energy reservation Based on the LMM-optimal definition,
we can calculate the first level optimal rate λ1 = g1 easily through the follow-
ing LP problem:

Maximize λ1
 
subject to fiB + j =i fij − k=i fki − λ1 = 0 (1 ≤ i ≤ N )
 
k=i ρTfki + j =i Cij Tfij + CiB TfiB ≤ ei (1 ≤ i ≤ N )
fij , fiB ≥ 0 (1 ≤ i, j ≤ N, j  = i).
Once we obtains a solution with the maximum λ1 , we can also calculate the
energy consumption at each node under this flow routing solution. Then we can
check whether or not a node has any remaining energy. If there are some nodes
that still have remaining energy, then we can construct another LP problem
to further increase their data rates (with a maximum rate increment of λ2 ).

1 However, there is a significant difference between max-min and LMM, which we will discuss
shortly.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

20 Linear programming and applications

This process terminates until all nodes use up their energy. Since energy is
reserved at each node after each iteration, we call this naive approach Serial
LP with Energy Reservation (SLP-ER).
SLP-ER cannot provide an LMM-optimal solution because there is a funda-
mental difference in the nature of the LMM rate allocation problem described
here and the classical max-min rate allocation problem in [14]. The LMM rate
allocation problem implicitly couples a flow routing problem (i.e., a determi-
nation of the fij and fiB for the entire network), while the classical max-min
rate allocation explicitly assumes that the routes for all the flows are given a
priori and remain fixed. For the LMM rate allocation problem, starting from
the first iteration, there usually exist nonunique flow routing solutions corre-
sponding to the same maximum rate level. Consequently, each of these flow
routing solutions, once chosen, will yield different remaining energy levels on
the nodes for future iterations and so forth. This will lead to a nonunique rate
allocation vector, which may not coincide with the LMM-optimal rate vector
(see numerical examples in Section 2.7).

Serial LP with rate reservation Instead of reserving energy based on a

flow routing solution at each rate level, another approach is to only deter-
mine the set of nodes that use up their energy at this level. The final flow
routing solution can be deferred to the last iteration. Since rate is reserved
during each iteration, we call this approach Serial LP with Rate Reservation
(SLP-RR).
SLP-RR cannot achieve an LMM-optimal rate vector either. This may be
harder to understand. But it is still due to the existence of non-unique flow
routing solutions at each rate level, whereby a node that uses up its energy
in one flow routing solution may still have remaining energy in another flow
routing solution. Therefore, if we reserve rates just based on one flow routing
solution, the set of nodes thought to have used up their energy at this rate level
may include some extra nodes that would otherwise be allocated with a larger
data rate in another flow routing solution. Consequently, a solution obtained
by SLP-RR may not be LMM-optimal.

2.4 A serial LP algorithm based on parametric analysis

In this section, we present a correct polynomial-time algorithm to solve the

LMM rate allocation problem. We first introduce some of the notation used.
Suppose that the rate vector g = [g1 , g2 , . . . , gN ] is LMM-optimal, with g1 ≤
g2 ≤ . . . ≤ gN . Note that the values of these N rates may not be distinct. To
focus on those distinct rate levels, we remove any repetitive elements in this
vector and rewrite it as [λ1 , λ2 , . . . , λn ] such that λ1 < λ2 < . . . < λn , where
λ1 = g1 , λn = gN , and n ≤ N . Now, for each λi , i = 1, 2, . . . , n, denote Si
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

21 2.4 A serial LP algorithm based on parametric analysis

as the corresponding set of nodes that use up their energy at this rate. Clearly,

we have ni=1 |Si | = |S| = N , where S denotes the set of all N nodes.
The essence to solving the LMM rate allocation problem is to find the correct
values for λ1 , λ2 , . . . , λn and the corresponding sets S1 , S2 , . . . , Sn , respec-
tively. This can be done iteratively. We first determine the rate level λ1 and the
corresponding set S1 , then determine the rate level λ2 and the corresponding
set S2 , and so on. In Section 2.4.1, we will show how to determine each rate
level and in Section 2.4.2, we will show how to determine the corresponding
node set.

2.4.1 Determining rate levels

Denote λ0 = 0 and S0 = ∅. For l = 1, 2, . . . , n, suppose that we have already
determined λ0 , λ1 , . . . , λl−1 and the corresponding sets S0 , S1 , . . . , Sl−1 .
The rate level λl can be found by the following optimization problem:

Maximize δl
  
l−1
subject to fiB + j =i fij − k=i fki − δl = λl−1 (i  ∈ Sh ) (2.7)
h=0
 
fiB + j =i fij − k=i fki = λh (i ∈ Sh , 1 ≤ h < l) (2.8)
  
l−1
( k=i ρfki + j =i Cij fij + CiB fiB )T ≤ ei (i  ∈ Sh ) (2.9)
h=0
 
( k=i ρfki + j =i Cij fij + CiB fiB )T = ei (i ∈ Sh , 1 ≤ h < l) (2.10)
fij , fiB ≥ 0 (1 ≤ i, j ≤ N, j  = i).

Note that for l = 1, constraints (2).8 and (2).10 do not exist. For 2 ≤ l ≤ n,
constraints (2).8 and (2).10 are for those nodes that have already reached their
LMM rate allocation during the previous (l − 1) iterations. In particular, the
set of constraints in (2).8 say that the sum of in-coming and local data rates
is equal to the out-going data rates for each node with its LMM-optimal rate
λh , 1 ≤ h < l. The set of constraints in (2).10 say that for those nodes that
have already reached their LMM-optimal rates, the total energy consumed for
communications has reached their initial energy provisioning. On the other
hand, the constraints in (2).7 and (2).9 are for the remaining nodes that have
not yet reached their LMM-optimal rates. Specifically, the set of constraints in
(2).7 state that, for those nodes that have not yet reached their energy constraint
levels, the sum of in-coming and local data rates is equal to the out-going data
rates. Note that the objective function is to maximize the additional rate δl for
those nodes. Furthermore, for those nodes, the set of constrains in (2).9 state
that the total energy consumed for communications should be upper bounded
by the initial energy provisioning.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

22 Linear programming and applications

To facilitate our later discussion on mirror results in Section 2.6, we fur-

ther re-formulate the foregoing LP problem. In particular, we multiply both
sides of (2).7 and (2).8 by T (which is a constant) and denote ViB = fiB T ,
Vij = fij T , and Vki = fki T . Intuitively, Vij and ViB represent the bit volume
that is transferred from node i to j and from node i to B, respectively, during
the lifetime T . We obtain the following LP formulation, which can be solved
using standard LP techniques:

LMM-rate Maximize δl
  
l−1
subject to ViB + j =i Vij − k=i Vki − δl T = λl−1 T (i  ∈ Sh ) (2.11)
h=0
 
ViB + j =i Vij − k=i Vki = λh T (i ∈ Sh , 1 ≤ h < l)

  
l−1

k=i ρVki + j =i Cij Vij + CiB ViB ≤ ei (i  ∈ Sh )

h=0
 
k=i ρVki + j =i Cij Vij + CiB ViB = ei (i ∈ Sh , 1 ≤ h < l)
Vij , ViB ≥ 0 (1 ≤ i, j ≤ N, j  = i).

Although a solution to the LMM rate problem gives the optimal solution for
δl at iteration l, it remains to determine the minimum set of nodes correspond-
ing to this δl , which is the key difficulty in the LMM rate allocation problem.
Fortunately, this problem can be solved by the so-called parametric analysis
(PA) technique in LP [10].

2.4.2 Determining minimum node set for a rate level

For the minimum node set for a rate level, we first need to know whether or
not this set is unique. This is affirmed in the following lemma. Its proof can be
found in [71].

Lemma 2.1
The minimum node set for each rate level under the LMM-optimal rate
allocation is unique.

With this lemma in place, we now discuss how to determine the minimum
node set Sl corresponding to the rate level λl . Denote Ŝl ( = ∅) as the set of
nodes for which the constraints (2).9 are binding at the lth iteration in LMM
rate, i.e., Ŝl include all the nodes that achieve equality in (2).9 at iteration l.
Although it is certain that at least one of the nodes in Ŝl belong to Sl (the
minimum node set for rate λl ), some nodes in Ŝl may still achieve greater rates
under other flow routing solutions. In other words, if |Ŝl | = 1, then we must
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

23 2.4 A serial LP algorithm based on parametric analysis

have Sl = Ŝl otherwise, we must determine the minimum node set Sl (⊆ Ŝl )
that achieves the LMM-optimal rate allocation.
The PA technique [10] is most suitable to address this problem. The main
idea of PA is to investigate how an infinitesimal perturbation on some param-
eters of the LMM rate problem can affect the objective function. Specifically,
node i belongs to the minimum node set Sl if and only if a small increase in
node i’s rate leads to a decrease in the objective value. In our problem, we con-
sider a small increase on the right-hand-side (RHS) of (2).11, i.e., changing bi
in the standard form (2).1 to bi + i , where i > 0. The physical meaning is
that node i’s data rate increased from δl + λl−1 to δl + λl−1 + Ti . We solve the
updated optimization problem with this new requirement on node i’s data rate.
Node i belongs to the minimum node set Sl if and only if the new δl is smaller,
+
i.e., ∂∂ δi l (0) < 0. The details of applying PA to determine the minimum node
set Sl can be found in [71].

2.4.3 Optimal flow routing for LMM rate allocation

After we solve the LMM rate allocation problem iteratively using the pro-
cedure described in Sections 2.4.1 and 2.4.2, the corresponding optimal flow
routing can be obtained by dividing the total bit volume on each link (Vij or
ViB ) by T , i.e.,
Vij ViB
fij = and fiB = . (2.12)
T T
Although the LMM-optimal rate allocation is unique, it is important to note
that the corresponding flow routing solution is not unique. This is because
upon the completion of the LMM rate allocation problem (i.e., upon find-
ing [λ1 , λ2 , . . . , λn ]), there usually exist nonunique bit volume solutions (Vij
and ViB values) corresponding to the same LMM-optimal rate allocation. This
result is summarized in the following lemma:

Lemma 2.2
The optimal flow routing solution corresponding to the LMM rate alloca-
tion may not be unique.

We use the following example to illustrate the nonuniqueness of the optimal

flow routing solution for an LMM rate allocation.

Example 2.1
Consider an eight-node network with a topology shown in Fig. 2.2. The
base station B is located at the origin (0, 0). There are two groups of nodes,
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

24 Linear programming and applications

G1 and G2 , in the network, with each group consisting of four nodes. Group
G1 consists of AFN1 at (100, 0), AFN3 at (0, 100), AFN5 at (−100, 0),
and AFN7 at (0, −100), respectively (all in meters); Group G2 consists of
AFN2 at (100, 100), AFN4 at (−100, 100), AFN6 at (−100, −100), and
AFN8 at (100, −100), respectively. Assume that all nodes have the same
initial energy e. For a network lifetime requirement of T , we can calculate
(via SLP-PA) that the final LMM-optimal rate allocation for all eight nodes
are identical (perfect fairness), i.e., r1 = r2 = . . . = r8 . We denote ri = g
for 1 ≤ i ≤ 8.
Y (m)

4 39g/62 − x 3 x 2
100
23

39g/62 − x
101g/ 62
g/

62
62

g/
x

23
5 101g/ 62 101g/ 62 1
X (m)
−100 0 100
101g/ 62
39g/62 − x

23
62

g/ x
g/

62
23

x 39g/62 − x

6 −100
7 8

Figure 2.2 A simple example showing that the optimal flow routing to the LMM rate allocation is not
unique. The range of x is 0 ≤ x ≤ 39g
62 .

Upon the completion of the SLP-PA algorithm, we also obtain an opti-

mal flow routing solution corresponding to this LMM-optimal rate g. This
optimal flow routing solution has the following flows: f21 = f43 = f65 =
f87 = 3962 g, f2B = f4B = f6B = f8B = 62 g, and f1B = f3B = f5B =
23

f7B = 62 g. We now show that the optimal flow routing solution is

101

nonunique. Since the network has a symmetrical property, it can be eas-

ily verified that for any x, 0 ≤ x ≤ 3962 g, the LMM-optimal rate allocation
can be achieved if the flow routing solution satisfies the following two con-
ditions: (i) each node in G2 (i.e., AFNs 2, 4, 6, and 8) sends a flow of
x and a flow of 39 62 g − x to its two neighboring G1 nodes as shown in
Fig. 2.2, and a remaining flow of 23 62 g directly to the base station; and (ii)
each node in G1 (i.e., AFNs 1, 3, 5, and 7) sends a total amount of 101 62 g
to the base station, which includes x and 39 62 g − x from its neighboring
nodes, plus g from itself. Clearly, there are infinitely many flow routing
solutions that meet these two conditions, each of which can be shown to
yield the LMM-optimal rate allocation g with the given network lifetime
requirement T .
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

25 2.5 SLP-PA for the LMM node lifetime problem

2.4.4 Complexity analysis

We now analyze the complexity of the SLP-PA algorithm. First, we consider
the complexity of finding each node’s rate and the total bit volume transmitted
along each link. At each stage, we solve an LP problem with a complexity of
O(nA 3 ) [10], where nA is the number of variables in the formulated LP (in the
standard form). Since the number of variables is O(N 2 ), the complexity of
solving the LP problem is O(N 6 ). After solving an LP problem at each stage,
we need to determine whether or not a node that just reached its energy binding
constraint belongs to the minimum node set for this stage. In [71], we showed
that the complexity of this is O(N 7 ). Hence, the complexity at each stage
is O(N 6 ) + O(N 7 ) = O(N 7 ). Since there are at most N stages, the overall
complexity is O(N 8 ).
The complexity in finding an optimal flow routing is bounded by the num-
ber of radio links in the network, which is O(N 2 ). Hence, the overall com-
plexity is O(N 8 ) + O(N 2 ) = O(N 8 ). Note that the analysis here gives a
worst-case time complexity. In practice, the run-time of SLP-PA is much
faster.

2.4.5 Extension to variable bit rate

So far, we have considered the case that each AFN generates data at a constant
rate. In practice, an AFN node may not always transmit data and may work in
an on–off mode to conserve energy. In this case, it is necessary to construct an
optimal flow routing solution for a variable bit rate source (where the on–off
mode is a special case). In [70], we have developed techniques to construct an
optimal flow routing solution for a variable bit rate, as long as the average rate
is known. Such an average rate corresponds to the constant rate in this chapter.
As a result, the case of the on–off mode (with a known average rate) can also
be handled using the techniques described in [70].

2.5 SLP-PA for the LMM node lifetime problem

In this section, we show that the SLP-PA algorithm can be used to solve the
so-called maximum node lifetime curve problem in [20], which we define as
the LMM node lifetime problem. We also show that the SLP-PA algorithm is
much more efficient than the one proposed in [20].

2.5.1 The LMM node lifetime problem

The LMM node lifetime problem can be described as follows. For a net-
work with N AFNs, with a given local bit rate ri (fixed) and initial energy
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

26 Linear programming and applications

ei for AFN i, i = 1, 2, . . . , N, how can we maximize the network lifetime

for all AFNs in the network? In other words, the LMM node lifetime prob-
lem not only considers how to maximize the lifetime until the first AFN
runs out of energy, but also the lifetimes for the rest of the AFNs in the
network.
More formally, denote the lifetime for each AFN i as ti , i = 1, 2, . . . , N.
Note that the ri ’s are fixed here, while the ti ’s are the optimization variables,
which are different from those in the LMM rate allocation problem in the pre-
vious section. Denote [τ1 , τ2 , . . . , τN ] as the sorted sequence of the ti -values
in nondecreasing order. Then, the LMM-optimal node lifetime can be defined
as follows:

Definition 2.2
LMM-optimal node lifetime A sorted node lifetime vector
[τ1 , τ2 , . . . , τN ] with τ1 ≤ τ2 ≤ . . . ≤ τN is LMM-optimal if and only if for
any other sorted node lifetime vector [τ̂1 , τ̂2 , . . . , τ̂N ], with τ̂1 ≤ τ̂2 ≤ . . . ≤
τ̂N , there exists a k, 1 ≤ k ≤ N such that τi = τ̂i for 1 ≤ i ≤ k − 1 and
τk > τ̂k .

2.5.2 Solution
It should be clear that, under the LMM-optimal node lifetime objective, we
must maximize the time until a set of nodes use up their energy (which is also
called a drop point in [20]), while minimizing the number of nodes that drain
up their energy at each drop point. We now show that the SLP-PA algorithm
developed for the LMM rate allocation problem can be applied to solve the
LMM node lifetime problem.
Suppose that [τ1 , τ2 , . . . , τN ] with τ1 ≤ τ2 ≤ . . . ≤ τN is LMM-optimal.
To keep track of distinct node lifetimes (or drop points) in this vector, we
remove all repetitive elements in the vector and rewrite it as [μ1 , μ2 , . . . , μn ]
such that μ1 < μ2 < . . . < μn , where μ1 = τ1 , μn = τN , and n ≤ N . Corre-
sponding to these drop points, denote S1 , S2 , . . . , Sn as the sets of nodes that
drain up their energy at drop points μ1 , μ2 , . . . , μn , respectively. Then |S1 | +
|S2 | + . . . + |Sn | = |S| = N, where S denotes the set of all N AFNs in the net-
work. The problem is to find the LMM-optimal values of μ1 , μ2 , . . . , μn and
the corresponding sets S1 , S2 , . . . , Sn .
Similar to the LMM rate allocation problem, the LMM node lifetime prob-
lem can be formulated as an iterative optimization problem as follows. Denote
μ0 = 0, S0 = ∅, and ζl = μl − μl−1 . Starting from l = 1, we solve the fol-
lowing LP problem iteratively:
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

27 2.6 A mirror result

LMM lifetime Maximize ζl

  
l−1
subject to ViB + Vij − Vki − ζl ri = μl−1 ri (i  ∈ Sh ) (2.13)
j =i k=i h=0
 
ViB + Vij − Vki = μh ri (i ∈ Sh , 1 ≤ h < l)
j =i k=i
  
l−1
ρVki + Cij Vij + CiB ViB ≤ ei (i  ∈ Sh )
k=i j =i h=0
 
ρVki + Cij Vij + CiB ViB = ei (i ∈ Sh , 1 ≤ h < l)
k=i j =i

Vij , ViB , ζl ≥ 0 (1 ≤ i, j ≤ N, j  = i).

Comparing the above LMM lifetime problem to the LMM rate problem in
Section 2.4.1, we find that they are mathematically identical. The only difference
is that under the LMM lifetime problem, the local bit rates ri ’s are constants and
the node lifetimes τi ’s are variables, while under the LMM rate problem, the
ri ’s are variables and the node lifetimes are all identical (T ), i = 1, 2, . . . , N.
Since the mathematical formulation for the two problems are identical, the
SLP-PA algorithm can be applied to solve the LMM node lifetime problem.
The only issue that we need to be concerned about is the optimal flow rout-
ing solution corresponding to the LMM-optimal lifetime vector. The optimal
flow routing solution here is not as simple as that for the LMM rate allocation
problem, which merely involves a simple division (see (2).12). We refer read-
ers to the appendix at the end of this chapter for an O(N 4 ) algorithm to obtain
an optimal flow routing solution for the LMM-optimal lifetime vector. Similar
to Lemma 2.2, the optimal flow routing solution corresponding to the LMM
node lifetime problem may not be unique.

2.6 A mirror result

In this section, we present an elegant result showing that there is a mirror rela-
tionship between the LMM rate allocation problem and the LMM node lifetime
problem. As a result, it is only necessary to solve only one of the two problems
and the results for the other problem can be obtained via simple algebraic cal-
culations.
To start with, we denote PR as the LMM rate allocation problem where we
have N AFNs in the network and all nodes have a common lifetime require-
ment T (a given constant). Denote ri as the LMM-optimal rate allocation for
node i under PR , i = 1, 2, . . . , N. Similarly, we denote PL as the LMM node
lifetime problem where all nodes have the same local bit rate R (constant).
Denote ti as the LMM node lifetime for node i under PL , i = 1, 2, . . . , N.
Then the following theorem shows how the solution to one problem can be
used to obtain the solution to the other.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

28 Linear programming and applications

Table 2.2 Mirror relationship between the LMM rate allocation problem PR and the
LMM node lifetime problem PL .

PR PL

ri (optimization variable) ri = R (constant)

ti = T (constant) ti (optimization variable)
Total bit volume at AFN i: ri · T = ti · R

Theorem 2.1
Mirror relation For a given node lifetime requirement T for all nodes
under Problem PR and a given local bit rate R for all nodes under Problem
PL , we have the following relationship between the solutions to the LMM
rate allocation problem PR and the LMM node lifetime problem PL .
(i) Suppose that we have solved Problem PR and obtained the LMM-
optimal rate allocation ri for each node i (i = 1, 2, . . . , N ). Then under
PL , the LMM node lifetime ti for node i is given by
ri T
ti =. (2.14)
R
(ii) Suppose that we have solved Problem PL and obtained the LMM-
optimal node lifetime ti for each node i (i = 1, 2, . . . , N). Then under PR ,
the LMM rate allocation ri for node i is given by
ti R
ri = . (2.15)
T

Table 2.2 shows the mirror relationship between solutions to Problems PR

and PL .

Proof. We prove (i) and (ii) in Theorem 2.1 separately.

(i) We organize our proof into two parts. First, we show that the ti ’s are feasible
node lifetimes in terms of flow balance and energy constraints at each node i
(i = 1, 2, . . . , N). Then we show that this solution indeed achieves the LMM-
optimal node lifetime.
Feasibility Since we have obtained the solution to Problem PR , we have one
feasible flow routing solution for sending bit rates ri , i = 1, 2, . . . , N, to the
base station. Under Problem PR , the bit volumes (Vij ’s and ViB ’s) must meet
the following equalities under the LMM-optimal rate allocation:
 
ViB + Vij − Vki = ri T ,
1≤j ≤N,j =i 1≤k≤N,k=i
 
ρVki + Cij Vij + CiB ViB = ei .
1≤k≤N,k=i 1≤j ≤N,j =i
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

29 2.6 A mirror result

Now, replacing ri T by ti R, we see that the same bit volume solution under PR
yields a feasible bit volume solution to the node lifetime problem under PL .
Consequently, we can use Algorithm 2.1 to obtain the flow routing solution to
Problem PL under the bit volume solution to Problem PL and this verifies that
ti , i = 1, 2, . . . , N, is a feasible solution to Problem PL .
Optimality The proof is based on contradiction. But first, let’s give some
notations. To prove that the ti -values obtained via (2).14 are indeed LMM-
optimal for Problem PL , we sort ri , i = 1, 2, . . . , N , under Problem PR in
nondecreasing order and denote it as [g1 , g2 , . . . , gN ]. We also introduce
a node index I = [i1 , i2 , . . . , iN ] for [g1 , g2 , . . . , gN ]. For example, i3 = 7
means that g3 actually corresponds to the rate of AFN 7, i.e., g3 = r7 .
Since ti is proportional to ri through the relationship (ti = R T
· ri ), listing
ti , i = 1, 2, . . . , N, according to I = [i1 , i2 , . . . , iN ] will yield a sorted (in
nondecreasing order) lifetime list, denoted as [τ1 , τ2 , . . . , τN ]. We now prove
that if [τ1 , τ2 , . . . , τN ] is not LMM-optimal for Problem PL , then we will have
that [g1 , g2 , . . . , gN ] is not LMM-optimal, which is a contradiction.
Suppose that [τ1 , τ2 , . . . , τN ] is not LMM-optimal for Problem PL . Assume
that the LMM-optimal lifetime vector to Problem PL is [τ̂1 , τ̂2 , . . . , τ̂N ]
(sorted in non-decreasing order) with the corresponding node index being
Iˆ = [î1 , î2 , . . . , îN ]. Then, by Definition 2.2, there exists a k such that τ̂j = τj
for 1 ≤ j ≤ k − 1 and τ̂k > τk .
We now claim that if tˆi , i = 1, 2, . . . , N, is a feasible solution to Problem
ˆ
PL , then r̂i obtained via r̂i = tiTR , i = 1, 2, . . . , N, is also a feasible solution
to Problem PR . The proof to this claim follows identically as above. Using
this result, we can obtain a corresponding feasible solution [ĝ1 , ĝ2 , . . . , ĝN ]
with ĝi = τ̂Ti R and the node index Iˆ for Problem PR . Hence, we have
τ̂ R τ R
ĝj = jT = jT = gj for 1 ≤ j ≤ k − 1 but ĝk = τ̂kTR > τkTR = gk . That is,
[g1 , g2 , . . . , gN ] is not LMM-optimal and this leads to a contradiction.
(ii) The proof for this part is similar to that for (i) and is left as a homework
exercise.

This mirror relationship offers important insights on system performance

issues, in addition to providing solutions to the LMM rate allocation and the
LMM node lifetime problems. For example, in Section 2.2, we pointed out the
potential bias (fairness) issue associated with the network capacity maximiza-
tion objective (i.e., sum of rates from all nodes). It is interesting to see that
a similar fairness issue exists in the node lifetime problem. In particular, the
objective of maximizing the sum of node lifetimes among all nodes also leads
to a bias (or fairness) problem because this objective would only favor those
nodes that consume energy at a small rate. As a result, certain nodes will have
much larger lifetimes, while some other nodes will be penalized with much
smaller lifetimes, although the sum of node lifetimes is maximized.
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

30 Linear programming and applications

Y (m) Y (m)
500 500

X (m) X (m)
−500 0 500 −500 0 500

−500 −500

(a) A 10-AFN network. (b) A 20-AFN network.

Figure 2.3 Network topologies

Table 2.3 Node coordinates for a 10-AFN network.
for the 10-AFN and 20-AFN
networks.
AFN i (xi , yi ) (in meters) AFN i (xi , yi ) (in meters)

1 (400, –320) 6 (–500, 100)

2 (300, 440) 7 (–400, 0)
3 (–300, –420) 8 (420, 120)
4 (320, –100) 9 (200, 140)
5 (–120, 340) 10 (220, –340)

2.7 Numerical results

In this section, we use numerical results to illustrate our SLP-PA algorithm

to the LMM rate allocation problem and compare it with other approaches.
We also use numerical results to illustrate the mirror relationship between the
LMM rate allocation problem and the LMM node lifetime problem.
We consider two network topologies, one with 10 AFNs and the other with
20 AFNs. Under both topologies, the base station B is located at the origin,
while the locations for the 10 or 20 AFNs are randomly generated over a
1000m × 1000m square area. See Figs. 2.3(a) and (b) and Tables 2.3 and 2.4
respectively for the two network topologies. We set β1 = 50 nJ/b, β2 = 0.0013
pJ/b/m4 [67], and α = 4 [129] in this study.

2.7.1 SLP-PA algorithm to the LMM rate allocation problem

We will compare SLP-PA with the naive approach SLP-ER (see Section 2.3.2)
that uses serial LP to “blindly” solve the LMM rate allocation problem and
9781107018808AR Engineering November 17, 2013 19:22 Char Count= 0

31 2.7 Numerical results

Table 2.4 Node coordinates for a 20-AFN network.

AFN i (xi , yi ) (in meters) AFN i (xi , yi ) (in meters)

1 (200, 130) 11 (110, –230)

2 (–400, –430) 12 (–210, 0)
3 (–100, 420) 13 (210, 320)
4 (0, 430) 14 (300, –480)
5 (–410, 440) 15 (–420, –470)
6 (–200, 230) 16 (–120, –240)
7 (400, –490) 17 (220, –440)
8 (410, –300) 18 (–220, –240)
9 (100, 310) 19 (–500, –110)
10 (10, 140) 20 (20, 330)

Table 2.5 Rate allocation under three approaches for the 10-AFN network.

i SLP-PA SLP-ER MaxCap

(sorted node gi AFN gi AFN gi AFN
index) (Kb/s) (Kb/s) (Kb/s)

1 0.1023 3 0.1023 1 0.0553 2

2 0.1023 6 0.1023 2 0.0627 3
3 0.1023 7 0.1023 3 0.0646 1
4 0.1536 5 0.1023 6 0.0658 6
5 0.2941 1 0.1023 7 0.1222 8
6 0.2941 2 0.1536 5 0.1653 10
7 0.2941 4 0.1536 8 0.1736 7
8 0.2941 8 0.1536 10 0.2628 5
9 0.2941 9 0.6563 4 0.3513 4
10 0.2941 10 0.6563 9 1.2398 9

performs energy reservation during each iteration. As discussed in Sec-

tion 2.3.2, the SLP-ER approach will not give the correct final solution to the
LMM rate allocation problem.
We will also compare our SLP-PA algorithm to the Maximum-Capacity
(MaxCap) approach (see Section 2.3.2). As discussed in the beginning of
Section 2.3.2, the rate allocation under the MaxCap approach can be extremely
biased and in favor of only those AFNs that consume the least power along
their data paths toward the base station.
10-AFN network We assume that the initial energy at each AFN is 50 KJ
and that the network lifetime requirement is 100 days under the LMM rate
allocation problem. The power consumption is for transmission and reception
defined in (2.2) and (2.4), respectively.
Table 2.5 shows the rate allocation for the AFNs under each approach,
which is also plotted in Fig. 2.4(a). The “sorted node index” corresponds to
the sorted rates among the AFNs in nondecreasing order. Clearly, among the
Another random document with
no related content on Scribd:
be in the most perfect condition as regards purity of matter,
composition of its different parts, size and position; no part of his
body must suffer from ill-health; he must in addition have studied
and acquired wisdom, so that his rational faculty passes from a state
of potentiality to that of actuality; his intellect must be as developed
and perfect as human intellect can be; his passions pure and equally
balanced; all his desires must aim at obtaining a knowledge of the
hidden laws and causes that are in force in the Universe; his
thoughts must be engaged in lofty matters; his attention directed to
the knowledge of God, the consideration of His works, and of that
which he must believe in this respect. There must be an absence of
the lower desires and appetites, of the seeking after pleasure in
eating, drinking, and cohabitation; and, in short, every pleasure
connected with the sense of touch. (Aristotle correctly says that this
sense is a disgrace to us, since we possess it only in virtue of our
being animals; and it does not include any specifically human
element, whilst enjoyments connected with other senses, as smell,
hearing, and sight, though likewise of a material nature, may
sometimes include [intellectual] pleasure, appealing to man as man,
according to Aristotle. This remark, although forming no part of our
subject, is not superfluous, for the thoughts of the most renowned
wise men are to a great extent affected by the pleasures of this
sense, and filled with a desire for them. And yet people are surprised
that these scholars do not prophesy, if prophesying be nothing but a
certain degree in the natural development of man.) It is further
necessary to suppress every thought or desire for unreal power and
dominion; that is to say, for victory, increase of followers, acquisition
of honour, and service from the people without any ulterior object.
On the contrary, the multitude must be considered according to their
true worth; some of them are undoubtedly like domesticated cattle,
and others like wild beasts, and these only engage the mind of the
perfect and distinguished man in so far as he desires to guard
himself from injury, in case of contact with them, and to derive some
benefit from them when necessary. A man who satisfies these
conditions, whilst his fully developed imagination is in action,
influenced by the Active Intellect according to his mental training,—
such a person will undoubtedly perceive nothing but things very
extraordinary and divine, and see nothing but God and His angels.
His knowledge will only include that which is real knowledge, and his
thought will only be directed to such general principles as would
tend to improve the social relations between man and man. [227]

We have thus described three kinds of perfection: mental perfection

acquired by training, perfection of the natural constitution of the
imaginative faculty, and moral perfection produced by the
suppression of every thought of bodily pleasures, and of every kind
of foolish or evil ambition. These qualities are, as is well known,
possessed by the wise men in different degrees, and the degrees of
prophetic faculty vary in accordance with this difference. Faculties of
the body are, as you know, at one time weak, wearied, and
corrupted, at others in a healthy state. Imagination is certainly one
of the faculties of the body. You find, therefore, that prophets are
deprived of the faculty of prophesying when they mourn, are angry,
or are similarly affected. Our Sages say, Inspiration does not come
upon a prophet when he is sad or languid. This is the reason why
Jacob did not receive any revelation during the period of his
mourning, when his imagination was engaged with the loss of
Joseph. The same was the case with Moses, when he was in a state
of depression through the multitude of his troubles, which lasted
from the murmurings of the Israelites in consequence of the evil
report of the spies, till the death of the warriors of that generation.
He received no message of God, as he used to do, even though he
did not receive prophetic inspiration through the medium of the
imaginative faculty, but directly through the intellect. We have
mentioned it several times that Moses did not, like other prophets,
speak in similes. This will be further explained (chap. xlv.), but it is
not the subject of the present chapter. There were also persons who
prophesied for a certain time and then left off altogether, something
occurring that caused them to discontinue prophesying. The same
circumstance, prevalence of sadness and dulness, was undoubtedly
the direct cause of the interruption of prophecy during the exile; for
can there be any greater misfortune for man than this: to be a slave
bought for money in the service of ignorant and voluptuous masters,
and powerless against them as they unite in themselves the absence
of true knowledge and the force of all animal desires? Such an evil
state has been prophesied to us in the words, “They shall run to and
fro to seek the word of God, but shall not find it” (Amos viii. 12);
“Her king and her princes are among the nations, the law is no
more, her prophets also find no vision from the Lord” (Lam. ii. 9).
This is a real fact, and the cause is evident; the pre-requisites [of
prophecy] have been lost. In the Messianic period—may it soon
commence—prophecy will therefore again be in our midst, as has
been promised by God.

[Contents]
 CHAPTER XXXVII
It is necessary to consider the nature of the divine influence, which
enables us to think, and gives us the various degrees of intelligence.
For this influence may reach a person only in a small measure, and
in exactly the same proportion would then be his intellectual
condition, whilst it may reach another person in such a measure
that, in addition to his own perfection, he can be the means of
perfection for others. The same relation may be observed
throughout the whole Universe. There are some beings so perfect
that they can govern other beings, but there are also beings that are
only perfect in so far as they can govern themselves and cannot
influence other beings. In some cases the [228]influence of the
[Active] Intellect reaches only the logical and not the imaginative
faculty; either on account of the insufficiency of that influence, or on
account of a defect in the constitution of the imaginative faculty, and
the consequent inability of the latter to receive that influence: this is
the condition of wise men or philosophers. If, however, the
imaginative faculty is naturally in the most perfect condition, this
influence may, as has been explained by us and by other
philosophers, reach both his logical and his imaginative faculties: this
is the case with prophets. But it happens sometimes that the
influence only reaches the imaginative faculty on account of the
insufficiency of the logical faculty, arising either from a natural
defect, or from a neglect in training. This is the case with statesmen,
lawgivers, diviners, charmers, and men that have true dreams, or do
wonderful things by strange means and secret arts, though they are
not wise men; all these belong to the third class. It is further
necessary to understand that some persons belonging to the third
class perceive scenes, dreams, and confused images, when awake,
in the form of a prophetic vision. They then believe that they are
prophets; they wonder that they perceive visions, and think that
they have acquired wisdom without training. They fall into grave
errors as regards important philosophical principles, and see a
strange mixture of true and imaginary things. All this is the
consequence of the strength of their imaginative faculty, and the
weakness of their logical faculty, which has not developed, and has
not passed from potentiality to actuality.

It is well known that the members of each class differ greatly from
each other. Each of the first two classes is again subdivided, and
contains two sections, namely, those who receive the influence only
as far as is necessary for their own perfection, and those who
receive it in so great a measure that it suffices for their own
perfection and that of others. A member of the first class, the wise
men, may have his mind influenced either only so far, that he is
enabled to search, to understand, to know, and to discern, without
attempting to be a teacher or an author, having neither the desire
nor the capacity; but he may also be influenced to such a degree
that he becomes a teacher and an author. The same is the case with
the second class. A person may receive a prophecy enabling him to
perfect himself but not others; but he may also receive such a
prophecy as would compel him to address his fellow-men, teach
them, and benefit them through his perfection. It is clear that,
without this second degree of perfection, no books would have been
written, nor would any prophets have persuaded others to know the
truth. For a scholar does not write a book with the object to teach
himself what he already knows. But the characteristic of the intellect
is this: what the intellect of one receives is transmitted to another,
and so on, till a person is reached that can only himself be perfected
by such an influence, but is unable to communicate it to others, as
has been explained in some chapters of this treatise (chap. xi.). It is
further the nature of this element in man that he who possesses an
additional degree of that influence is compelled to address his
fellow-men, under all circumstances, whether he is listened to or
not, even if he injures himself thereby. Thus we find prophets that
did not leave off speaking to the people until they were slain; it is
this divine influence that moves them, that does not allow them to
rest in any way, though they might [229]bring upon themselves great
evils by their action. E.g., when Jeremiah was despised, like other
teachers and scholars of his age, he could not, though he desired it,
withhold his prophecy, or cease from reminding the people of the
truths which they rejected. Comp. “For the Word of the Lord was
unto me a reproach and a mocking all day, and I said, I will not
mention it, nor will I again speak in His name; but it was in mine
heart as a burning fire, enclosed in my bones, and I was wearied to
keep it, and did not prevail” (Jer. xx. 8, 9). This is also the meaning
of the words of another prophet, “The Lord God hath spoken, who
shall not prophesy?” (Amos iii. 8) Note it.
[Contents]
 CHAPTER XXXVIII
Every man possesses a certain amount of courage, otherwise he
would not stir to remove anything that might injure him. This
psychical force seems to me analogous to the physical force of
repulsion. Energy varies like all other forces, being great in one case
and small in another. There are, therefore, people who attack a lion,
whilst others run away at the sight of a mouse. One attacks a whole
army and fights, another is frightened and terrified by the threat of a
woman. This courage requires that there be in a man’s constitution a
certain disposition for it. If man, in accordance with a certain view,
employs it more frequently, it develops and increases, but, on the
other hand, if it is employed, in accordance with the opposite view,
more rarely, it will diminish. From our own youth we remember that
there are different degrees of energy among boys.

The same is the case with the intuitive faculty; all possess it, but in
different degrees. Man’s intuitive power is especially strong in things
which he has well comprehended, and in which his mind is much
engaged. Thus you may yourself guess correctly that a certain
person said or did a certain thing in a certain matter. Some persons
are so strong and sound in their imagination and intuitive faculty
that, when they assume a thing to be in existence, the reality either
entirely or partly confirms their assumption. Although the causes of
this assumption are numerous, and include many preceding,
succeeding, and present circumstances, by means of the intuitive
faculty the intellect can pass over all these causes, and draw
inferences from them very quickly, almost instantaneously. This
same faculty enables some persons to foretell important coming
events. The prophets must have had these two forces, courage and
intuition, highly developed, and these were still more strengthened
when they were under the influence of the Active Intellect. Their
courage was so great that, e.g., Moses, with only a staff in his hand,
dared to address a great king in his desire to deliver a nation from
his service. He was not frightened or terrified, because he had been
told, “I will be with thee” (Exod. iii. 12). The prophets have not all
the same degree of courage, but none of them have been entirely
without it. Thus Jeremiah is told: “Be not afraid of them,” etc. (Jer. i.
8), and Ezekiel is exhorted, “Do not fear them or their word” (Ezek.
ii. 6). In the same manner, you find that all prophets possessed
great courage. Again, through the excellence of their intuitive
faculty, they could quickly foretell the future, but this excellence, as
is well known, likewise admits of different degrees. [230]

The true prophets undoubtedly conceive ideas that result from

premisses which human reason could not comprehend by itself; thus
they tell things which men could not tell by reason and ordinary
imagination alone; for [the action of the prophets’ mental capacities
is influenced by] the same agent that causes the perfection of the
imaginative faculty, and that enables the prophet thereby to foretell
a future event with such clearness as if it was a thing already
perceived with the senses, and only through them conveyed to his
imagination. This agent perfects the prophet’s mind, and influences
it in such a manner that he conceives ideas which are confirmed by
reality, and are so clear to him as if he deduced them by means of
syllogisms.

This should be the belief of all who choose to accept the truth. For
[all things are in a certain relation to each other, and] what is
noticed in one thing may be used as evidence for the existence of
certain properties in another, and the knowledge of one thing leads
us to the knowledge of other things. But [what we said of the
extraordinary powers of our imaginative faculty] applies with special
force to our intellect, which is directly influenced by the Active
Intellect, and caused by it to pass from potentiality to actuality. It is
through the intellect that the influence reaches the imaginative
faculty. How then could the latter be so perfect as to be able to
represent things not previously perceived by the senses, if the same
degree of perfection were withheld from the intellect, and the latter
could not comprehend things otherwise than in the usual manner,
namely, by means of premiss, conclusion, and inference? This is the
true characteristic of prophecy, and of the disciplines to which the
preparation for prophecy must exclusively be devoted. I spoke here
of true prophets in order to exclude the third class, namely, those
persons whose logical faculties are not fully developed, and who do
not possess any wisdom, but are only endowed with imaginative and
inventive powers. It may be that things perceived by these persons
are nothing but ideas which they had before, and of which
impressions were left in their imaginations together with those of
other things; but whilst the impressions of other images are effaced
and have disappeared, certain images alone remain, are seen and
considered as new and objective, coming from without. The process
is analogous to the following case: A person has with him in the
house a thousand living individuals; all except one of them leave the
house: when the person finds himself alone with that individual, he
imagines that the latter has entered the house now, contrary to the
fact that he has only not left the house. This is one of the many
phenomena open to gross misinterpretations and dangerous errors,
and many of those who believed that they were wise perished
thereby.

There were, therefore, men who supported their opinion by a dream

which they had, thinking that the vision during sleep was
independent of what they had previously believed or heard when
awake. Persons whose mental capacities are not fully developed,
and who have not attained intellectual perfection, must not take any
notice of these [dreams]. Those who reach that perfection may,
through the influence of the divine intellect, obtain knowledge
independent of that possessed by them when awake. They are true
prophets, as is distinctly stated in Scripture, ve-nabi lebab ḥokmah
(Ps. xc. 12), “And the true prophet possesseth a heart of wisdom.”
This must likewise be noticed. [231]

[Contents]
 CHAPTER XXXIX
We have given the definition of prophecy, stated its true
characteristics, and shown that the prophecy of Moses our Teacher
was distinguished from that of other prophets; we will now explain
that this distinction alone qualified him for the office of proclaiming
the Law, a mission without a parallel in the history from Adam to
Moses, or among the prophets who came after him; it is a principle
in our faith that there will never be revealed another Law.
Consequently we hold that there has never been, nor will there ever
be, any other divine Law but that of Moses our Teacher. According to
what is written in Scripture and handed down by tradition, the fact
may be explained in the following way: There were prophets before
Moses, as the patriarchs Shem, Eber, Noah, Methushelah, and
Enoch, but of these none said to any portion of mankind that God
sent him to them and commanded him to convey to them a certain
message or to prohibit or to command a certain thing. Such a thing
is not related in Scripture, or in authentic tradition. Divine prophecy
reached them as we have explained. Men like Abraham, who
received a large measure of prophetic inspiration, called their fellow-
men together and led them by training and instruction to the truth
which they had perceived. Thus Abraham taught, and showed by
philosophical arguments that there is one God, that He has created
everything that exists beside Him, and that neither the constellations
nor anything in the air ought to be worshipped; he trained his
fellow-men in this belief, and won their attention by pleasant words
as well as by acts of kindness. Abraham did not tell the people that
God had sent him to them with the command concerning certain
things which should or should not be done. Even when it was
commanded that he, his sons, and his servants should be
circumcised, he fulfilled that commandment, but he did not address
his fellow-men prophetically on this subject. That Abraham induced
his fellow-men to do what is right, telling them only his own will
[and not that of God], may be learnt from the following passage of
Scripture: “For I know him, because he commands his sons and his
house after him, to practise righteousness and judgment” (Gen. xix.
19). Also Isaac, Jacob, Levi, Kohath, and Amram influenced their
fellow-men in the same way. Our Sages, when speaking of prophets
before Moses, used expressions like the following: The bet-din (court
of justice) of Eber, the bet-din of Methushelah, and in the college of
Methushelah; although all these were prophets, yet they taught their
fellow-men in the manner of preachers, teachers, and pedagogues,
but did not use such phrases as the following: “And God said to me,
Speak to certain people so and so.” This was the state of prophecy
before Moses. But as regards Moses, you know what [God] said to
him, what he said [to the people], and the words addressed to him
by the whole nation: “This day we have seen that God doth talk with
man, and that he liveth” (Deut. v. 21). The history of all our
prophets that lived after Moses is well known to you; they
performed, as it were, the function of warning the people and
exhorting them to keep the Law of Moses, threatening evil to those
who would neglect it, and announcing blessings to those who would
submit to its guidance. This we believe will always be the case.
Comp. “It is not in the heavens that one might say,” etc. (ibid. xxx.
12); “For [232]us and for our children for ever” (ibid. xxix. 28). It is
but natural that it should be so. For if one individual of a class has
reached the highest perfection possible in that class, every other
individual must necessarily be less perfect, and deviate from the
perfect measure either by surplus or deficiency. Take, e.g., the
normal constitution of a being, it is the most proper composition
possible in that class; any constitution that deviates from that norm
contains something too much or too little. The same is the case with
the Law. It is clear that the Law is normal in this sense; for it
contains “Just statutes and judgments” (Deut. iv. 8); but “just” is
here identical with “equibalanced.” The statutes of the Law do not
impose burdens or excesses as are implied in the service of a hermit
or pilgrim, and the like; but, on the other hand, they are not so
deficient as to lead to gluttony or lewdness, or to prevent, as the
religious laws of the heathen nations do, the development of man’s
moral and intellectual faculties. We intend to discuss in this treatise
the reasons of the commandments, and we shall then show, as far
as necessary, the justice and wisdom of the Law, on account of
which it is said: “The Law of God is perfect, refreshing the heart”
(Ps. xix. 8). There are persons who believe that the Law commands
much exertion and great pain, but due consideration will show them
their error. Later on I will show how easy it is for the perfect to obey
the Law. Comp. “What does the Lord thy God ask of thee?” etc.
(Deut. x. 12); “Have I been a wilderness to Israel?” (Jer. ii. 31). But
this applies only to the noble ones; whilst wicked, violent, and
pugnacious persons find it most injurious and hard that there should
be any divine authority tending to subdue their passion. To low-
minded, wanton, and passionate persons it appears most cruel that
there should be an obstacle in their way to satisfy their carnal
appetite, or that a punishment should be inflicted for their doings.
Similarly every godless person imagines that it is too hard to abstain
from the evil he has chosen in accordance with his inclination. We
must not consider the Law easy or hard according as it appears to
any wicked, low-minded, and immoral person, but as it appears to
the judgment of the most perfect, who, according to the Law, are fit
to be the example for all mankind. This Law alone is called divine;
other laws, such as the political legislations among the Greeks, or
the follies of the Sabeans, are the works of human leaders, but not
of prophets, as I have explained several times.
[Contents]
 CHAPTER XL
It has already been fully explained that man is naturally a social
being, that by virtue of his nature he seeks to form communities;
man is therefore different from other living beings that are not
compelled to combine into communities. He is, as you know, the
highest form in the creation, and he therefore includes the largest
number of constituent elements; this is the reason why the human
race contains such a great variety of individuals, that we cannot
discover two persons exactly alike in any moral quality, or in external
appearance. The cause of this is the variety in man’s temperament,
and in accidents dependent on his form; for with every physical form
there are connected certain special accidents different from those
which are connected with the substance. Such a variety among the
individuals of a class [233]does not exist in any other class of living
beings; for the variety in any other species is limited; only man
forms an exception; two persons may be so different from each
other in every respect that they appear to belong to two different
classes. Whilst one person is so cruel that he kills his youngest child
in his anger, another is too delicate and faint-hearted to kill even a
fly or worm. The same is the case with most of the accidents. This
great variety and the necessity of social life are essential elements in
man’s nature. But the well-being of society demands that there
should be a leader able to regulate the actions of man; he must
complete every shortcoming, remove every excess, and prescribe for
the conduct of all, so that the natural variety should be
counterbalanced by the uniformity of legislation, and the order of
society be well established. I therefore maintain that the Law,
though not a product of Nature, is nevertheless not entirely foreign
to Nature. It being the will of God that our race should exist and be
permanently established, He in His wisdom gave it such properties
that men can acquire the capacity of ruling others. Some persons
are therefore inspired with theories of legislation, such as prophets
and lawgivers; others possess the power of enforcing the dictates of
the former, and of compelling people to obey them, and to act
accordingly. Such are kings, who accept the code of lawgivers, and
[rulers] who pretend to be prophets, and accept, either entirely or
partly, the teaching of the prophets. They accept one part while
rejecting another part, either because this course appears to them
more convenient, or out of ambition, because it might lead people to
believe that the rulers themselves had been prophetically inspired
with these laws, and did not copy them from others. For when we
like a certain perfection, find pleasure in it, and wish to possess it,
we sometimes desire to make others believe that we possess that
virtue, although we are fully aware that we do not possess it. Thus
people, e.g., adorn themselves with the poems of others, and
publish them as their own productions. It also occurs in the works of
wise men on the various branches of Science, that an ambitious, lazy
person sees an opinion expressed by another person, appropriates
it, and boasts that he himself originated it. The same [ambition]
occurs also with regard to the faculty of prophecy. There were men
who, like Zedekiah, the son of Chenaanah (1 Kings xxii. 11, 24)
boasted that they received a prophecy, and declared things which
have never been prophesied. Others, like Hananiah, son of Azzur
(Jer. xxviii. 1–5), claim the capacity of prophecy, and proclaim things
which, no doubt, have been said by God, that is to say, that have
been the subject of a divine inspiration, but not to them. They
nevertheless say that they are prophets, and adorn themselves with
the prophecies of others. All this can easily be ascertained and
recognized. I will, however, fully explain this to you, so that no doubt
be left to you on this question, and that you may have a test by
which you may distinguish between the guidance of human
legislation, of the divine law, and of teachings stolen from prophets.
As regards those who declare that the laws proclaimed by them are
their own ideas, no further test is required; the confession of the
defendant makes the evidence of the witness superfluous. I only
wish to instruct you about laws which are proclaimed as prophetic.
Some of these are truly prophetic, originating in divine inspiration,
some are of non-prophetic character, and some, though prophetic
originally, are the result of plagiarism. You will find that the sole
[234]object of certain laws, in accordance with the intention of their
author, who well considered their effect, is to establish the good
order of the state and its affairs, to free it from all mischief and
wrong; these laws do not deal with philosophic problems, contain no
teaching for the perfecting of our logical faculties, and are not
concerned about the existence of sound or unsound opinions. Their
sole object is to arrange, under all circumstances, the relations of
men to each other, and to secure their well-being, in accordance
with the view of the author of these laws. These laws are political,
and their author belongs, as has been stated above, to the third
class, viz., to those who only distinguish themselves by the
perfection of their imaginative faculties. You will also find laws
which, in all their rules, aim, as the law just mentioned, at the
improvement of the material interests of the people; but, besides,
tend to improve the state of the faith of man, to create first correct
notions of God, and of angels, and to lead then the people, by
instruction and education, to an accurate knowledge of the Universe:
this education comes from God; these laws are divine. The question
which now remains to be settled is this: Is the person who
proclaimed these laws the same perfect man that received them by
prophetic inspiration, or a plagiarist, who has stolen these ideas from
a true prophet? In order to be enabled to answer this question, we
must examine the merits of the person, obtain an accurate account
of his actions, and consider his character. The best test is the
rejection, abstention, and contempt of bodily pleasures; for this is
the first condition of men, and a fortiori of prophets; they must
especially disregard pleasures of the sense of touch, which,
according to Aristotle, is a disgrace to us; and, above all, restrain
from the pollution of sensual intercourse. Thus God exposes thereby
false prophets to public shame, in order that those who really seek
the truth may find it, and not err or go astray; e.g., Zedekiah, son of
Maasiah, and Ahab, son of Kolaiah, boasted that they had received a
prophecy. They persuaded the people to follow them, by proclaiming
utterances of other prophets; but all the time they continued to seek
the low pleasures of sensual intercourse, committing even adultery
with the wives of their companions and followers. God exposed their
falsehood as He has exposed that of other false prophets. The king
of Babylon burnt them, as Jeremiah distinctly states: “And of them
shall be taken up a curse by all the captivity of Judah, which are in
Babylon, saying, The Lord make thee like Zedekiah, and like Ahab,
whom the king of Babylon roasted in the fire. Because they have
committed villainy in Israel, and have committed adultery with their
neighbours’ wives, and have spoken lying words in my name, which
I have not commanded them” (Jer. xxix. 22, 23). Note what is meant
by these words.

[Contents]
 CHAPTER XLI
I need not explain what a dream is, but I will explain the meaning of
the term mareh, “vision,” which occurs in the passage: “In a vision
(be-mareh) do I make myself known unto him” (Num. xii. 6). The
term signifies that which is also called mareh ha-nebuah, “prophetic
vision,” yad ha-shem, “the hand of God,” and maḥazeh, “a vision.” It
is something terrible and fearful which the prophet feels while
awake, as is distinctly stated by [235]Daniel: “And I saw this great
vision, and there remained no strength in me, for my comeliness
was turned in me into corruption, and I retained no strength” (Dan.
x. 8). He afterwards continues, “Thus was I in deep sleep on my
face, and my face toward the ground” (ibid. ver. 9). But it was in a
prophetic vision that the angel spoke to him and “set him upon his
knees.” Under such circumstances the senses cease to act, and the
[Active Intellect] influences the rational faculties, and through them
the imaginative faculties, which become perfect and active.
Sometimes the prophecy begins with a prophetic vision, the prophet
greatly trembles, and is much affected in consequence of the perfect
action of the imaginative faculty, and after that the prophecy follows.
This was the case with Abraham. The commencement of the
prophecy is, “The word of the Lord came to Abraham in a vision”
(Gen. xv. 1); after this, “a deep sleep fell upon Abraham”; and at
last, “he said unto Abraham,” etc. When prophets speak of the fact
that they received a prophecy, they say that they received it from an
angel, or from God; but even in the latter case it was likewise
received through an angel. Our Sages, therefore, explain the words,
“And the Lord said unto her” that He spake through an angel. You
must know that whenever Scripture relates that the Lord or an angel
spoke to a person, this took place in a dream or in a prophetic
vision.
There are four different ways in which Scripture relates the fact that
a divine communication was made to the prophet. (1) The prophet
relates that he heard the words of an angel in a dream or vision; (2)
He reports the words of the angel without mentioning that they were
perceived in a dream or vision, assuming that it is well known that
prophecy can only originate in one of the two ways, “In a vision I
will make myself known unto him, in a dream I will speak unto him”
(Num. xii. 6). (3) The prophet does not mention the angel at all; he
says that God spoke to him, but he states that he received the
message in a dream or a vision. (4) He introduces his prophecy by
stating that God spoke to him, or told him to do a certain thing, or
speak certain words, but he does not explain that he received the
message in a dream or vision, because he assumes that it is well
known, and has been established as a principle that no prophecy or
revelation originates otherwise than in a dream or vision, and
through an angel. Instances of the first form are the following:
—“And the angel of the Lord said unto me in a dream, Jacob” (Gen.
xxxi. 11); “And an angel said unto Israel in a vision of night” (ibid.
xlvi. 2); “And an angel came to Balaam by night”; “And an angel said
unto Balaam” (Num. xxii. 20–22). Instances of the second form are
these: “And Elohim (an angel), said unto Jacob, Rise, go up to
Bethel” (Gen. xxxv. 1); “And Elohim said unto him, Thy name is
Jacob,” etc. (ibid. xxxv. 10); “And an angel of the Lord called unto
Abraham out of heaven the second time” (ibid. xxii. 15); “And
Elohim said unto Noah” (ibid. vi. 13). The following is an instance of
the third form: “The word of the Lord came unto Abraham in a
vision” (ibid. xv. 1). Instances of the fourth form are: “And the Lord
said unto Abraham” (ibid. xviii. 13); “And the Lord said unto Jacob,
Return,” etc. (ibid. xxxi. 3); “And the Lord said unto Joshua” (Josh. v.
9); “And the Lord said unto Gideon” (Judges vii. 2). Most of the
prophets speak in a similar manner: “And the Lord said unto me”
(Deut. ii. 2); “And the word of the Lord came unto me” [236](Ezek.
xxx. 1); “And the word of the Lord came” (2 Sam. xxiv. 11); “And
behold, the word of the Lord came unto him” (1 Kings xix. 9); “And
the word of the Lord came expressly” (Ezek. i. 3); “The beginning of
the word of the Lord by Hosea” (Hos. i. 2); “The hand of the Lord
was upon me” (Ezek. xxxvii. 1). There are a great many instances of
this class. Every passage in Scripture introduced by any of these four
forms is a prophecy proclaimed by a prophet; but the phrase, “And
Elohim (an angel) came to a certain person in the dream of night,”
does not indicate a prophecy, and the person mentioned in that
phrase is not a prophet; the phrase only informs us that the
attention of the person was called by God to a certain thing, and at
the same time that this happened at night. For just as God may
cause a person to move in order to save or kill another person, so
He may cause, according to His will, certain things to rise in man’s
mind in a dream by night. We have no doubt that the Syrian Laban
was a perfectly wicked man, and an idolater; likewise Abimelech,
though a good man among his people, is told by Abraham
concerning his land [Gerar] and his kingdom, “Surely there is no fear
of God in this place” (Gen. xx. 11). And yet concerning both of them,
viz., Laban and Abimelech, it is said [that an angel appeared to them
in a dream]. Comp. “And Elohim (an angel) came to Abimelech in a
dream by night” (ibid. ver. 3); and also, “And Elohim came to the
Syrian Laban in the dream of the night” (ibid. xxxi. 24). Note and
consider the distinction between the phrases, “And Elohim came,”
and “Elohim said,” between “in a dream by night,” and “in a vision by
night.” In reference to Jacob it is said, “And an angel said to Israel in
the visions by night” (Gen. xlvi. 2), but in reference to Laban and
Abimelech, “And Elohim came,” etc. Onkelos makes the distinction
clear; he translates, in the last two instances, ata memar min kodam
adonai, “a word came from the Lord,” and not ve-itgeli, “and the
Lord appeared.” The phrase, “And the Lord said to a certain person,”
is employed even when this person was not really addressed by the
Lord, and did not receive any prophecy, but was informed of a
certain thing through a prophet. E.g., “And she went to inquire of
the Lord” (Gen. xxv. 22); that is, according to the explanation of our
Sages, she went to the college of Eber, and the latter gave her the
answer; and this is expressed by the words, “And the Lord said unto
her” (ibid. ver. 23). These words have also been explained thus, God
spoke to her through an angel; and by “angel” Eber is meant here,
for a prophet is sometimes called “angel,” as will be explained; or
the angel that appeared to Eber in this vision is referred to, or the
object of the Midrash explanation is merely to express that wherever
God is introduced as directly speaking to a person, i.e., to any of the
ordinary prophets, He speaks through an angel, as has been set
forth by us (chap. xxxiv.).

[Contents]
 CHAPTER XLII
We have already shown that the appearance or speech of an angel
mentioned in Scripture took place in a vision or dream; it makes no
difference whether this is expressly stated or not, as we have
explained above. This is a point of considerable importance. In some
cases the account begins by stating that the prophet saw an angel;
in others, the account apparently introduces [237]a human being,
who ultimately is shown to be an angel; but it makes no difference,
for if the fact that an angel has been heard is only mentioned at the
end, you may rest satisfied that the whole account from the
beginning describes a prophetic vision. In such visions, a prophet
either sees God who speaks to him, as will be explained by us, or he
sees an angel who speaks to him, or he hears some one speaking to
him without seeing the speaker, or he sees a man who speaks to
him, and learns afterwards that the speaker was an angel. In this
latter kind of prophecies, the prophet relates that he saw a man who
was doing or saying something, and that he learnt afterwards that it
was an angel.

This important principle was adopted by one of our Sages, one of

the most distinguished among them, R. Ḥiya the Great (Bereshit
Rabba, xlviii.), in the exposition of the Scriptural passage
commencing, “And the Lord appeared unto him in the plain of
Mamre” (Gen. xviii.). The general statement that the Lord appeared
to Abraham is followed by the description in what manner that
appearance of the Lord took place; namely, Abraham saw first three
men; he ran and spoke to them. R. Ḥiya, the author of the
explanation, holds that the words of Abraham, “My Lord, if now I
have found grace in thy sight, do not, I pray thee, pass from thy
servant,” were spoken by him in a prophetic vision to one of the
men; for he says that Abraham addressed these words to the chief
of these men. Note this well, for it is one of the great mysteries [of
the Law]. The same, I hold, is the case when it is said in reference
to Jacob, “And a man wrestled with him” (Gen. xxxii. 25); this took
place in a prophetic vision, since it is expressly stated in the end
(ver. 31) that it was an angel. The circumstances are here exactly
the same as those in the vision of Abraham, where the general
statement, “And the Lord appeared to him,” etc., is followed by a
detailed description. Similarly the account of the vision of Jacob
begins, “And the angels of God met him” (Gen. xxxii. 2); then
follows a detailed description how it came to pass that they met
him; namely, Jacob sent messengers, and after having prepared and
done certain things, “he was left alone,” etc., “and a man wrestled
with him” (ibid. ver. 24). By this term “man” [one of] the angels of
God is meant, mentioned in the phrase, “And angels of God met
him”; the wrestling and speaking was entirely a prophetic vision.
That which happened to Balaam on the way, and the speaking of the
ass, took place in a prophetic vision, since further on, in the same
account, an angel of God is introduced as speaking to Balaam. I also
think that what Joshua perceived, when “he lifted up his eyes and
saw, and behold a man stood before him” (Josh. v. 13) was a
prophetic vision, since it is stated afterwards (ver. 14) that it was
“the prince of the host of the Lord.” But in the passages, “And an
angel of the Lord came up from Gilgal” (Judges ii. 1); “And it came
to pass that the angel of the Lord spake these words to all Israel”
(ibid. ver. 2); the “angel” is, according to the explanation of our
Sages, Phineas. They say, The angel is Phineas, for, when the Divine
Glory rested upon him, he was “like an angel.” We have already
shown (chap. vi.) that the term “angel” is homonymous, and
denotes also “prophet,” as is the case in the following passages:
—“And He sent an angel, and He hath brought us up out of Egypt”
(Num. xx. 16); “Then spake Haggai, the angel of the Lord, in the
Lord’s message” (Hagg. i. 13); “But they mocked the angels of
[238]God” (2 Chron. xxxvi. 16).—Comp. also the words of Daniel,
“And the man Gabriel, whom I had seen in the vision at the
beginning, being caused to fly swiftly, touched me about the time of
the evening oblation” (Dan. ix. 11). All this passed in a prophetic
vision. Do not imagine that an angel is seen or his word heard
otherwise than in a prophetic vision or prophetic dream, according to
the principle laid down:—“I make myself known unto him in a vision,
and speak unto him in a dream” (Num. xii. 6). The instances quoted
may serve as an illustration of those passages which I do not
mention. From the rule laid down by us that prophecy requires
preparation, and from our interpretation of the homonym “angel,”
you will infer that Hagar, the Egyptian woman, was not a
prophetess; also Manoah and his wife were no prophets; for the
speech they heard, or imagined they heard, was like the bat-kol
(prophetic echo), which is so frequently mentioned by our Sages,
and is something that may be experienced by men not prepared for
prophecy. The homonymity of the word “angel” misleads in this
matter. This is the principal method by which most of the difficult
passages in the Bible can be explained. Consider the words, “And an
angel of the Lord found her by the well of water” (Gen. xvi. 7),
which are similar to the words referring to Joseph—“And a man
found him, and behold, he was erring in the field” (ibid. xxxvii. 15).
All the Midrashim assume that by man in this passage an angel is
meant.

[Contents]
 CHAPTER XLIII
We have already shown in our work that the prophets sometimes
prophesy in allegories; they use a term allegorically, and in the same
prophecy the meaning of the allegory is given. In our dreams, we
sometimes believe that we are awake, and relate a dream to another
person, who explains the meaning, and all this goes on while we
dream. Our Sages call this “a dream interpreted in a dream.” In
other cases we learn the meaning of the dream after waking from
sleep. The same is the case with prophetic allegories. Some are
interpreted in the prophetic vision. Thus it is related in Zechariah,
after the description of the allegorical vision—“And the angel that
talked with me came again and waked me as a man that is
awakened from his sleep. And he said unto me, ‘What dost thou
see?’ ” etc. (Zech. iv. 1–2), and then the allegory is explained (ver. 6,
sqq.).

Another instance we find in Daniel. It is first stated there: “Daniel

had a dream and visions of his head upon his bed” (Dan. vii. 1). The
whole allegory is then given, and Daniel is described as sighing that
he did not know its interpretation. He asks the angel for an
explanation, and he received it in a prophetic vision. He relates as
follows: “I came near unto one of those that stood by, and asked
him the truth of all this. So he told me, and made me know the
interpretation of the things” (ibid. ver. 16). The whole scene is called
ḥazon (vision), although it was stated that Daniel had a dream,
because an angel explained the dream to him in the same manner
as is mentioned in reference to a prophetic dream. I refer to the
verse: “A vision appeared to me Daniel, after that which appeared to
me at the first” (ibid. viii. 1). This is clear, for ḥazon (vision) is
derived from ḥaza, “to see,” and mareh, “vision,” from raah, “to
see”; and ḥaza and raah are [239]synonymous. There is therefore no
difference whether we use mareh, or maḥazeh, or ḥazon, there is no
other mode of revelation but the two mentioned in Scripture: “In a
vision I make myself known to him, in a dream I will speak unto
him” (Num. xii. 6). There are, however, different degrees [of
prophetic proficiency], as will be shown (chap. xlv.).

There are other prophetic allegories whose meaning is not given in a

prophetic vision. The prophet learns it when he awakes from his
sleep. Take, e.g., the staves which Zechariah took in a prophetic
vision.

You must further know that the prophets see things shown to them
allegorically, such as the candlesticks, horses, and mountains of
Zechariah (Zech. iv. 2; vi. 1–7), the scroll of Ezekiel (Ezek. ii. 9), the
wall made by a plumb-line (Amos vii. 7), which Amos saw, the
animals of Daniel (Dan. vii. and viii.), the seething pot of Jeremiah
(Jer. i. 13), and similar allegorical objects shown to represent certain
ideas. The prophets, however, are also shown things which do not
illustrate the object of the vision, but indicate it by their name
through its etymology or homonymity. Thus the imaginative faculty
forms the image of a thing, the name of which has two meanings,
one of which denotes something different [from the image]. This is
likewise a kind of allegory. Comp. Makkal shaked, “almond staff,” of
Jeremiah (i. 11–12). It was intended to indicate by the second
meaning of shaked the prophecy, “For I will watch” (shoked), etc.,
which has no relation whatever to the staff or to almonds. The same
is the case with the kelub ḳayiẓ, “a basket of summer fruit,” seen by
Amos, by which the completion of a certain period was indicated,
“the end (ha-ḳeẓ) having come” (Amos viii. 2). Still more strange is
the following manner of calling the prophet’s attention to a certain
object. He is shown a different object, the name of which has
neither etymologically nor homonymously any relation to the first
object, but the names of both contain the same letters, though in a
different order. Take, e.g., the allegories of Zechariah (chap. xi. 7,
sqq.). He takes in a prophetic vision staves to lead the flock; he calls
the one Noʻam (pleasure), the other ḥobelim. He indicates thereby
that the nation was at first in favour with God, who was their leader
and guide. They rejoiced in the service of God, and found happiness
in it, while God was pleased with them, and loved them, as it is said,
“Thou hast avouched the Lord thy God,” etc., and “the Lord hath
avouched thee,” etc. (Deut. xxvi. 17, 18). They were guided and
directed by Moses and the prophets that followed him. But later a
change took place. They rejected the love of God, and God rejected
them, appointing destroyers like Jeroboam and Manasse as their
rulers. Accordingly, the word ḥobelim has the same meaning [viz.,
destroying] as the root ḥabal has in Meḥabbelim keramim,
“destroying vineyards” (Song of Sol. ii. 15). But the prophet found
also in this name Ḥobelim the indication that the people despised
God, and that God despised them. This is, however, not expressed
by the word ḥabal, but by a transposition of the letters Ḥet, Bet, and
Lamed, the meaning of despising and rejecting is obtained. Comp.
“My soul loathed them, and their soul also abhorred me” [baḥalah]
(Zech. xi. 8). The prophet had therefore to change the order of the
letters in ḥabal into that of Baḥal. In this way we find very strange
things and also mysteries (Sodot) in the words neḥoshet, Kalal,
regel, ʻegel, and ḥashmal of the Mercabah, and in other terms in
other [240]passages. After the above explanation you will see the
mysteries in the meaning of these expressions if you examine them
thoroughly.

[Contents]
 CHAPTER XLIV
Prophecy is given either in a vision or in a dream, as we have said so
many times, and we will not constantly repeat it. We say now that
when a prophet is inspired with a prophecy he may see an allegory,
as we have shown frequently, or he may in a prophetic vision
perceive that God speaks to him, as is said in Isaiah (vi. 8), “And I
heard the voice of the Lord saying, Whom shall I send, and who will
go for us?” or he hears an angel addressing him, and sees him also.
This is very frequent, e.g., “And the angel of God spake unto me,”
etc. (Gen. xxxi. 11); “And the angel that talked with me answered
and said unto me, Dost thou not know what these are” (Zech. iv. 5);
“And I heard one holy speaking” (Dan. viii. 13). Instances of this are
innumerable. The prophet sometimes sees a man that speaks to
him. Comp., “And behold there was a man, whose appearance was
like the appearance of brass, and the man said to me,” etc. (Ezek. xl.
3, 4), although the passage begins, “The hand of the Lord was upon
me” (ibid. ver. 1). In some cases the prophet sees no figure at all,
only hears in the prophetic vision the words addressed to him; e.g.,
“And I heard the voice of a man between the banks of Ulai” (Dan.
viii. 16); “There was silence, and I heard a voice” (in the speech of
Eliphaz, Job iv. 16); “And I heard a voice of one that spake to me”
(Ezek. i. 28). The being which Ezekiel perceived in the prophetic
vision was not the same that addressed him; for at the conclusion of
the strange and extraordinary scene which Ezekiel describes
expressly as having been perceived by him, the object and form of
the prophecy is introduced by the words, “And I heard a voice of a
man that spake to me.” After this remark on the different kinds of
prophecy, as suggested by Scripture, I say that the prophet may
perceive that which he hears with the greatest possible intensity,
just as a person may hear thunder in his dream, or perceive a storm