Get the full ebook with Bonus Features for a Better Reading Experience on ebookluna.

com

(eBook PDF) A Guide to Algorithm Design:

Paradigms, Methods, and Complexity Analysis

https://ebookluna.com/product/ebook-pdf-a-guide-to-
algorithm-design-paradigms-methods-and-complexity-analysis/

OR CLICK HERE

DOWLOAD NOW

Download more ebook instantly today at https://ebookluna.com

 Instant digital products (PDF, ePub, MOBI) ready for you
Download now and discover formats that fit your needs...

(eBook PDF) Research Methods, Design, and Analysis 13th

Edition

https://ebookluna.com/product/ebook-pdf-research-methods-design-and-
analysis-13th-edition/

ebookluna.com

(eBook PDF) Algorithm Design by Jon Kleinberg

https://ebookluna.com/product/ebook-pdf-algorithm-design-by-jon-
kleinberg/

ebookluna.com

(Original PDF) Algorithm Design and Applications by

Michael T. Goodrich

https://ebookluna.com/product/original-pdf-algorithm-design-and-
applications-by-michael-t-goodrich/

ebookluna.com

Modern Control: State-Space Analysis and Design Methods

1st Edition - eBook PDF

https://ebookluna.com/download/modern-control-state-space-analysis-
and-design-methods-ebook-pdf/

ebookluna.com
(eBook PDF) Handbook of Research Methods in Complexity
Science: Theory and Applications

https://ebookluna.com/product/ebook-pdf-handbook-of-research-methods-
in-complexity-science-theory-and-applications/

ebookluna.com

(eBook PDF) Data Structures & Algorithm Analysis in C++

4th Edition

https://ebookluna.com/product/ebook-pdf-data-structures-algorithm-
analysis-in-c-4th-edition/

ebookluna.com

Data Structures & Algorithm Analysis in C++ 4th Edition

(eBook PDF)

https://ebookluna.com/product/data-structures-algorithm-analysis-
in-c-4th-edition-ebook-pdf/

ebookluna.com

(eBook PDF) Qualitative Psychology: A Practical Guide to

Research Methods 3rd Edition

https://ebookluna.com/product/ebook-pdf-qualitative-psychology-a-
practical-guide-to-research-methods-3rd-edition/

ebookluna.com

(eBook PDF) Research Design and Methods: A Process

Approach 11th Edition

https://ebookluna.com/product/ebook-pdf-research-design-and-methods-a-
process-approach-11th-edition/

ebookluna.com
 Computer Science/Computer Engineering/Computing Chapman & Hall/CRC
Applied Algorithms and Data Structures Series

ALGORITHM DESIGN
“This book is a great technical arsenal for every graduate student and post-
graduate researcher. By providing a treasure trove of concrete algorithmic A GUIDE TO
ALGORITHM
examples, the book trains the reader to recognize clues that indicate the
complexity of a broad range of algorithmic problems, while supplying

A GUIDE TO
a battery of techniques for solving a particular problem in hand. …”
—Umit Catalyurek, Professor, Ohio State University

“This book is unique among texts on algorithmics in its emphasis on how

to ‘think algorithmically’ rather than just how to solve specific (classes of)
DESIGN
algorithmic problems. The authors skillfully engage the reader in a journey Paradigms, Methods, and Complexity Analysis
of algorithmic self-discovery as they cover a broad spectrum of issues …
I shall be very happy to have this text on my bookshelf as a reference on
methods as well as results.”
—Arnold L. Rosenberg, Research Professor, Northeastern University, and
Distinguished University Professor Emeritus, University of Massachusetts
Amherst

“This book presents a well-balanced approach to theory and algorithms

and introduces difficult concepts using rich motivating examples. It
demonstrates the applicability of fundamental principles and analysis
techniques to practical problems facing computer scientists and engineers.
You do not have to be a theoretician to enjoy and learn from this book.”
—Rami Melhem, Professor of Computer Science, University of Pittsburgh

Presenting a complementary perspective to standard books on algorithms,

A Guide to Algorithm Design: Paradigms, Methods, and Complexity

Benoit, Robert,
Analysis provides a roadmap for readers to determine the difficulty of an

and
algorithmic problem by finding an optimal solution or proving complexity
results. It gives a practical treatment of algorithmic complexity and guides Vivien
readers in solving algorithmic problems. The book offers a comprehensive
set of problems with solutions as well as in-depth case studies that
demonstrate how to assess the complexity of a new problem.

Anne Benoit, Yves Robert,

K11226 and Frédéric Vivien

K11226_Cover.indd 1 5/31/13 9:17 AM

 vii

Solution to Exercise 3.5: Scheduling independent tasks with

deadlines . . . . . . . . . . . . . . . . . . . . . . . . . 73
Solution to Exercise 3.6: Edge matroids . . . . . . . . . . . . 74
Solution to Exercise 3.7: Huffman code . . . . . . . . . . . . . 75
3.7 Bibliographical notes . . . . . . . . . . . . . . . . . . . . . . 79

4 Dynamic programming 81
4.1 The coin changing problem . . . . . . . . . . . . . . . . . . . 81
4.2 The knapsack problem . . . . . . . . . . . . . . . . . . . . . 84
4.3 Designing dynamic-programming algorithms . . . . . . . . . 86
4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Exercise 4.1: Matrix chains . . . . . . . . . . . . . . . . . . . 87
Exercise 4.2: The library . . . . . . . . . . . . . . . . . . . . . 88
Exercise 4.3: Polygon triangulation . . . . . . . . . . . . . . . 88
Exercise 4.4: Square of ones . . . . . . . . . . . . . . . . . . . 89
Exercise 4.5: The wind band . . . . . . . . . . . . . . . . . . 89
Exercise 4.6: Ski rental . . . . . . . . . . . . . . . . . . . . . . 89
Exercise 4.7: Building set . . . . . . . . . . . . . . . . . . . . 90
4.5 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . 90
Solution to Exercise 4.1: Matrix chains . . . . . . . . . . . . . 90
Solution to Exercise 4.2: The library . . . . . . . . . . . . . . 91
Solution to Exercise 4.3: Polygon triangulation . . . . . . . . 93
Solution to Exercise 4.4: Square of ones . . . . . . . . . . . . 96
Solution to Exercise 4.5: The wind band . . . . . . . . . . . . 98
Solution to Exercise 4.6: Ski rental . . . . . . . . . . . . . . . 98
Solution to Exercise 4.7: Building set . . . . . . . . . . . . . . 102
4.6 Bibliographical notes . . . . . . . . . . . . . . . . . . . . . . 103

5 Amortized analysis 105

5.1 Methods for amortized analysis . . . . . . . . . . . . . . . . . 105
5.1.1 Running examples . . . . . . . . . . . . . . . . . . . . 105
5.1.2 Aggregate analysis . . . . . . . . . . . . . . . . . . . . 106
5.1.3 Accounting method . . . . . . . . . . . . . . . . . . . 106
5.1.4 Potential method . . . . . . . . . . . . . . . . . . . . . 107
5.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Exercise 5.1: Binary counter . . . . . . . . . . . . . . . . . . . 108
Exercise 5.2: Inserting and deleting . . . . . . . . . . . . . . . 108
Exercise 5.3: Stack . . . . . . . . . . . . . . . . . . . . . . . . 109
Exercise 5.4: Deleting half the elements . . . . . . . . . . . . 109
Exercise 5.5: Searching and inserting . . . . . . . . . . . . . . 109
Exercise 5.6: Splay trees . . . . . . . . . . . . . . . . . . . . . 110
Exercise 5.7: Half perimeter of a polygon . . . . . . . . . . . 112
5.3 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . 112
Solution to Exercise 5.1: Binary counter . . . . . . . . . . . . 112
Solution to Exercise 5.2: Inserting and deleting . . . . . . . . 113

© 2014 by Taylor & Francis Group, LLC

viii

Solution to Exercise 5.3: Stack . . . . . . . . . . . . . . . . . 114

Solution to Exercise 5.4: Deleting half the elements . . . . . . 115
Solution to Exercise 5.5: Searching and inserting . . . . . . . 116
Solution to Exercise 5.6: Splay trees . . . . . . . . . . . . . . 117
Solution to Exercise 5.7: Half perimeter of a polygon . . . . . 119
5.4 Bibliographical notes . . . . . . . . . . . . . . . . . . . . . . 122

II NP-completeness and beyond 123

6 NP-completeness 125
6.1 A practical approach to complexity theory . . . . . . . . . . 125
6.2 Problem classes . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.2.1 Problems in P . . . . . . . . . . . . . . . . . . . . . . 127
6.2.2 Problems in NP . . . . . . . . . . . . . . . . . . . . . 129
6.3 NP-complete problems and reduction theory . . . . . . . . . 132
6.3.1 Polynomial reduction . . . . . . . . . . . . . . . . . . 132
6.3.2 Cook’s theorem . . . . . . . . . . . . . . . . . . . . . . 133
6.3.3 Growing the class NPC of NP-complete problems . . . 134
6.3.4 Optimization problems versus decision problems . . . 135
6.4 Examples of NP-complete problems and reductions . . . . . 136
6.4.1 3-SAT . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.4.2 CLIQUE . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.4.3 VERTEX-COVER . . . . . . . . . . . . . . . . . . . . 139
6.4.4 Scheduling problems . . . . . . . . . . . . . . . . . . . 140
6.4.5 Other famous NP-complete problems . . . . . . . . . . 142
6.5 Importance of problem definition . . . . . . . . . . . . . . . . 143
6.6 Strong NP-completeness . . . . . . . . . . . . . . . . . . . . 145
6.7 Why does it matter? . . . . . . . . . . . . . . . . . . . . . . 146
6.8 Bibliographical notes . . . . . . . . . . . . . . . . . . . . . . 146

7 Exercises on NP-completeness 149

7.1 Easy reductions . . . . . . . . . . . . . . . . . . . . . . . . . 149
Exercise 7.1: Wheel . . . . . . . . . . . . . . . . . . . . . . . 149
Exercise 7.2: Knights of the round table . . . . . . . . . . . . 149
Exercise 7.3: Variants of CLIQUE . . . . . . . . . . . . . . . 149
Exercise 7.4: Path with vertex pairs . . . . . . . . . . . . . . 150
Exercise 7.5: VERTEX-COVER with even degrees . . . . . . 150
Exercise 7.6: Around 2-PARTITION . . . . . . . . . . . . . . 150
7.2 About graph coloring . . . . . . . . . . . . . . . . . . . . . . 151
Exercise 7.7: COLOR . . . . . . . . . . . . . . . . . . . . . . 151
Exercise 7.8: 3-COLOR . . . . . . . . . . . . . . . . . . . . . 151
Exercise 7.9: 3-COLOR-PLAN . . . . . . . . . . . . . . . . . 152
7.3 Scheduling problems . . . . . . . . . . . . . . . . . . . . . . . 152
Exercise 7.10: Scheduling independent tasks with p processors 152

© 2014 by Taylor & Francis Group, LLC

 ix

Exercise 7.11: Scheduling with two processors . . . . . . . . . 152

7.4 More involved reductions . . . . . . . . . . . . . . . . . . . . 153
Exercise 7.12: Transitive subchain . . . . . . . . . . . . . . . 153
Exercise 7.13: INDEPENDENT SET . . . . . . . . . . . . . . 153
Exercise 7.14: DOMINATING SET . . . . . . . . . . . . . . . 153
Exercise 7.15: Carpenter . . . . . . . . . . . . . . . . . . . . . 153
Exercise 7.16: k-center . . . . . . . . . . . . . . . . . . . . . . 153
Exercise 7.17: Variants of 3-SAT . . . . . . . . . . . . . . . . 154
Exercise 7.18: Variants of SAT . . . . . . . . . . . . . . . . . 154
7.5 2-PARTITION is NP-complete . . . . . . . . . . . . . . . . . 155
Exercise 7.19: SUBSET-SUM . . . . . . . . . . . . . . . . . . 155
Exercise 7.20: NP-completeness of 2-PARTITION . . . . . . 155
7.6 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . 155
Solution to Exercise 7.1: Wheel . . . . . . . . . . . . . . . . . 156
Solution to Exercise 7.2: Knights of the round table . . . . . 156
Solution to Exercise 7.3: Variants of CLIQUE . . . . . . . . . 157
Solution to Exercise 7.4: Path with vertex pairs . . . . . . . . 158
Solution to Exercise 7.5: VERTEX-COVER with even degrees 158
Solution to Exercise 7.6: Around 2-PARTITION . . . . . . . 159
Solution to Exercise 7.7: COLOR . . . . . . . . . . . . . . . . 160
Solution to Exercise 7.8: 3-COLOR . . . . . . . . . . . . . . . 162
Solution to Exercise 7.9: 3-COLOR-PLAN . . . . . . . . . . . 163
Solution to Exercise 7.10: Scheduling independent tasks with
p processors . . . . . . . . . . . . . . . . . . . . . . . . 166
Solution to Exercise 7.11: Scheduling with two processors . . 166
Solution to Exercise 7.12: Transitive subchain . . . . . . . . . 167
Solution to Exercise 7.13: INDEPENDENT SET . . . . . . . 168
Solution to Exercise 7.14: DOMINATING SET . . . . . . . . 169
Solution to Exercise 7.15: Carpenter . . . . . . . . . . . . . . 170
Solution to Exercise 7.16: k-center . . . . . . . . . . . . . . . 171
Solution to Exercise 7.17: Variants of 3-SAT . . . . . . . . . . 172
Solution to Exercise 7.18: Variants of SAT . . . . . . . . . . . 174
Solution to Exercise 7.19: SUBSET-SUM . . . . . . . . . . . 175
Solution to Exercise 7.20: NP-completeness of 2-PARTITION 177
7.7 Bibliographical notes . . . . . . . . . . . . . . . . . . . . . . 178

8 Beyond NP-completeness 179

8.1 Approximation results . . . . . . . . . . . . . . . . . . . . . . 179
8.1.1 Approximation algorithms . . . . . . . . . . . . . . . . 180
8.1.2 Vertex cover . . . . . . . . . . . . . . . . . . . . . . . 181
8.1.3 Traveling salesman problem (TSP) . . . . . . . . . . . 182
8.1.4 Bin packing . . . . . . . . . . . . . . . . . . . . . . . . 183
8.1.5 2-PARTITION . . . . . . . . . . . . . . . . . . . . . . 187
8.2 Polynomial problem instances . . . . . . . . . . . . . . . . . 192
8.2.1 Partitioning problems . . . . . . . . . . . . . . . . . . 193

© 2014 by Taylor & Francis Group, LLC

x

8.2.2 Assessing problem complexity . . . . . . . . . . . . . . 194

8.3 Linear programming . . . . . . . . . . . . . . . . . . . . . . . 195
8.3.1 Formal definition . . . . . . . . . . . . . . . . . . . . . 195
8.3.2 Relaxation and rounding . . . . . . . . . . . . . . . . . 197
8.4 Randomized algorithms . . . . . . . . . . . . . . . . . . . . . 200
8.4.1 The algorithm . . . . . . . . . . . . . . . . . . . . . . 201
8.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 201
8.5 Branch-and-bound and backtracking . . . . . . . . . . . . . . 202
8.5.1 Backtracking: The n queens . . . . . . . . . . . . . . . 203
8.5.2 Branch-and-bound: The knapsack . . . . . . . . . . . 204
8.5.3 Graph algorithms . . . . . . . . . . . . . . . . . . . . . 206
8.6 Bibliographical notes . . . . . . . . . . . . . . . . . . . . . . 209

9 Exercises going beyond NP-completeness 211

9.1 Approximation results . . . . . . . . . . . . . . . . . . . . . . 211
Exercise 9.1: Single machine scheduling . . . . . . . . . . . . 211
Exercise 9.2: SUBSET-SUM . . . . . . . . . . . . . . . . . . . 212
Exercise 9.3: SET-COVER . . . . . . . . . . . . . . . . . . . 213
Exercise 9.4: VERTEX-COVER . . . . . . . . . . . . . . . . 213
Exercise 9.5: Scheduling independent tasks in parallel . . . . 215
Exercise 9.6: Point clustering . . . . . . . . . . . . . . . . . . 215
Exercise 9.7: k-center . . . . . . . . . . . . . . . . . . . . . . 216
Exercise 9.8: Knapsack . . . . . . . . . . . . . . . . . . . . . . 217
9.2 Dealing with NP-complete problems . . . . . . . . . . . . . . 218
Exercise 9.9: Mixed integer linear program for replica place-
ment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
Exercise 9.10: A randomized algorithm for independent set . 218
Exercise 9.11: Branch-and-bound applied to MAX-SAT . . . 219
9.3 Solutions to exercises . . . . . . . . . . . . . . . . . . . . . . 219
Solution to Exercise 9.1: Single machine scheduling . . . . . . 219
Solution to Exercise 9.2: SUBSET-SUM . . . . . . . . . . . . 221
Solution to Exercise 9.3: SET-COVER . . . . . . . . . . . . . 223
Solution to Exercise 9.4: VERTEX-COVER . . . . . . . . . . 224
Solution to Exercise 9.5: Scheduling independent tasks in par-
allel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
Solution to Exercise 9.6: Point clustering . . . . . . . . . . . 228
Solution to Exercise 9.7: k-center . . . . . . . . . . . . . . . . 229
Solution to Exercise 9.8: Knapsack . . . . . . . . . . . . . . . 231
Solution to Exercise 9.9: Mixed integer linear program for
replica placement . . . . . . . . . . . . . . . . . . . . . 234
Solution to Exercise 9.10: A randomized algorithm for inde-
pendent set . . . . . . . . . . . . . . . . . . . . . . . . 237
Solution to Exercise 9.11: Branch-and-bound applied to MAX-
SAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
9.4 Bibliographical notes . . . . . . . . . . . . . . . . . . . . . . 238

© 2014 by Taylor & Francis Group, LLC

 xi

III Reasoning on problem complexity 239

10 Reasoning to assess a problem complexity 241
10.1 Basic reasoning . . . . . . . . . . . . . . . . . . . . . . . . . 241
10.1.1 Polynomial instances . . . . . . . . . . . . . . . . . . . 241
10.1.2 NP-complete instances . . . . . . . . . . . . . . . . . . 242
10.2 Set of problems with polynomial-time algorithms . . . . . . . 243
10.3 Set of NP-complete problems . . . . . . . . . . . . . . . . . . 244
10.3.1 Numbers . . . . . . . . . . . . . . . . . . . . . . . . . 245
10.3.2 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . 246

11 Chains-on-chains partitioning 249

11.1 Optimal algorithms for homogeneous resources . . . . . . . . 249
11.1.1 Dynamic-programming algorithm . . . . . . . . . . . . 250
11.1.2 Binary search algorithm . . . . . . . . . . . . . . . . . 250
11.1.3 Improved algorithms . . . . . . . . . . . . . . . . . . . 250
11.2 Variants of the problem . . . . . . . . . . . . . . . . . . . . . 252
11.2.1 Communication costs . . . . . . . . . . . . . . . . . . 252
11.2.2 Chain of heterogeneous resources . . . . . . . . . . . . 253
11.3 Extension to a clique of heterogeneous resources . . . . . . . 254
11.3.1 NP-completeness . . . . . . . . . . . . . . . . . . . . . 254
11.3.2 Practical solutions . . . . . . . . . . . . . . . . . . . . 257
11.3.3 Integer linear program . . . . . . . . . . . . . . . . . . 257
11.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

12 Replica placement in tree networks 261

12.1 Access policies . . . . . . . . . . . . . . . . . . . . . . . . . . 262
12.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . 262
12.1.2 Impact of the policies on the existence of a solution . 263
12.1.3 Impact of the policies on the cost of a solution . . . . 264
12.2 Complexity results . . . . . . . . . . . . . . . . . . . . . . . . 266
12.2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . 266
12.2.2 MinNb problem . . . . . . . . . . . . . . . . . . . . . 267
12.2.3 MinCost problem . . . . . . . . . . . . . . . . . . . . 273
12.2.4 Integer linear program . . . . . . . . . . . . . . . . . . 275
12.3 Variants of the replica placement problem . . . . . . . . . . . 279
12.3.1 Enforcing a quality of service . . . . . . . . . . . . . . 280
12.3.2 Power-aware replica placement . . . . . . . . . . . . . 282
12.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286

13 Packet routing 287

13.1 MEDP: Maximum edge-disjoint paths . . . . . . . . . . . . 288
13.1.1 Problem statement . . . . . . . . . . . . . . . . . . . . 288
13.1.2 Naive greedy algorithm . . . . . . . . . . . . . . . . . 289
13.1.3 Short-requests-first greedy algorithm . . . . . . . . . . 291

© 2014 by Taylor & Francis Group, LLC

xii

13.1.4 Inapproximability result . . . . . . . . . . . . . . . . . 292

13.2 PRVP: Packet routing with variable-paths . . . . . . . . . . 294
13.2.1 Problem statement . . . . . . . . . . . . . . . . . . . . 294
13.2.2 Bounding optimal makespan via linear programming . 295
13.2.3 Routing algorithm . . . . . . . . . . . . . . . . . . . . 297
13.2.4 Steady-state approach . . . . . . . . . . . . . . . . . . 300
13.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301

14 Matrix product, or tiling the unit square 303

14.1 Problem motivation . . . . . . . . . . . . . . . . . . . . . . . 304
14.2 NP-completeness . . . . . . . . . . . . . . . . . . . . . . . . . 307
14.3 A guaranteed heuristic . . . . . . . . . . . . . . . . . . . . . 311
14.3.1 The ColPeriSum(s) problem . . . . . . . . . . . . . 312
14.3.2 Performance guarantee . . . . . . . . . . . . . . . . . . 316
14.3.3 Looking for a better solution . . . . . . . . . . . . . . 317
14.4 Related problems . . . . . . . . . . . . . . . . . . . . . . . . 320

15 Online scheduling 321

15.1 Flow time optimization . . . . . . . . . . . . . . . . . . . . . 322
15.2 Competitive analysis . . . . . . . . . . . . . . . . . . . . . . 324
15.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 324
15.2.2 Method to establish a competitive analysis result . . . 327
15.3 Makespan optimization . . . . . . . . . . . . . . . . . . . . . 334
15.3.1 List scheduling algorithms . . . . . . . . . . . . . . . . 335
15.3.2 Randomized optimization of makespan . . . . . . . . . 338
15.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347

References 349

Index 359

© 2014 by Taylor & Francis Group, LLC

List of exercises

1.1 Longest balanced section . . . . . . . . . . . . . . . . . . . . . 8

1.2 Find the star . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Breaking boxes . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Maximum of n integers . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Maximum and minimum of n integers . . . . . . . . . . . . . 10
1.6 Maximum and second maximum of n integers . . . . . . . . . 11
1.7 Merging two sorted sets . . . . . . . . . . . . . . . . . . . . . 11
1.8 The toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.9 Sorting a small number of objects . . . . . . . . . . . . . . . . 12

2.1 Product of two polynomials . . . . . . . . . . . . . . . . . . . 39

2.2 Toeplitz matrices . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3 Maximum sum . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4 Boolean matrices: The Four-Russians algorithm . . . . . . . . 41
2.5 Matrix multiplication and inversion . . . . . . . . . . . . . . . 42

3.1 Interval cover . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.2 Memory usage . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.3 Scheduling dependent tasks on several machines . . . . . . . . 65
3.4 Scheduling independent tasks with priorities . . . . . . . . . . 66
3.5 Scheduling independent tasks with deadlines . . . . . . . . . . 66
3.6 Edge matroids . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.7 Huffman code . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4.1 Matrix chains . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.2 The library . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.3 Polygon triangulation . . . . . . . . . . . . . . . . . . . . . . 88
4.4 Square of ones . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.5 The wind band . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.6 Ski rental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.7 Building set . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

5.1 Binary counter . . . . . . . . . . . . . . . . . . . . . . . . . . 108

5.2 Inserting and deleting . . . . . . . . . . . . . . . . . . . . . . 108
5.3 Stack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.4 Deleting half the elements . . . . . . . . . . . . . . . . . . . . 109
5.5 Searching and inserting . . . . . . . . . . . . . . . . . . . . . 109

xiii

© 2014 by Taylor & Francis Group, LLC

xiv

5.6 Splay trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

5.7 Half perimeter of a polygon . . . . . . . . . . . . . . . . . . . 112

7.1 Wheel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

7.2 Knights of the round table . . . . . . . . . . . . . . . . . . . . 149
7.3 Variants of CLIQUE . . . . . . . . . . . . . . . . . . . . . . . 149
7.4 Path with vertex pairs . . . . . . . . . . . . . . . . . . . . . . 150
7.5 VERTEX-COVER with even degrees . . . . . . . . . . . . . . 150
7.6 Around 2-PARTITION . . . . . . . . . . . . . . . . . . . . . . 150
7.7 COLOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.8 3-COLOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.9 3-COLOR-PLAN . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.10 Scheduling independent tasks with p processors . . . . . . . . 152
7.11 Scheduling with two processors . . . . . . . . . . . . . . . . . 152
7.12 Transitive subchain . . . . . . . . . . . . . . . . . . . . . . . . 153
7.13 INDEPENDENT SET . . . . . . . . . . . . . . . . . . . . . . 153
7.14 DOMINATING SET . . . . . . . . . . . . . . . . . . . . . . . 153
7.15 Carpenter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.16 k-center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.17 Variants of 3-SAT . . . . . . . . . . . . . . . . . . . . . . . . 154
7.18 Variants of SAT . . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.19 SUBSET-SUM . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.20 NP-completeness of 2-PARTITION . . . . . . . . . . . . . . . 155

9.1 Single machine scheduling . . . . . . . . . . . . . . . . . . . . 211

9.2 SUBSET-SUM . . . . . . . . . . . . . . . . . . . . . . . . . . 212
9.3 SET-COVER . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
9.4 VERTEX-COVER . . . . . . . . . . . . . . . . . . . . . . . . 213
9.5 Scheduling independent tasks in parallel . . . . . . . . . . . . 215
9.6 Point clustering . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.7 k-center . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.8 Knapsack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
9.9 Mixed integer linear program for replica placement . . . . . . 218
9.10 A randomized algorithm for independent set . . . . . . . . . . 218
9.11 Branch-and-bound applied to MAX-SAT . . . . . . . . . . . . 219

© 2014 by Taylor & Francis Group, LLC

Preface

Objective
YABA? Yet Another Book on Algorithms?
No thanks. There are so many good books on the design of algorithms that
it is hard to choose and pick one. If asked to name our two favorite refer-
ences, we would recommend Introduction to Algorithms by Cormen, Leiserson,
Rivest, and Stein [27] and Algorithms by Dasgupta, Papadimitriou, and Vazi-
rani [30]. For sure, this book does not intend to compete with such established
monuments.
Instead, this book proposes a complementary perspective. It aims at guid-
ing students and researchers who need to solve problems, either by finding
optimal algorithms or by assessing new complexity results. In a nutshell, the
main objective of this book is to outline the roadmap to follow, and to prac-
tice all the corresponding steps, in order to determine the complexity of a
problem.

Intended audience and use

The target audience for this book is graduate students and postgraduate re-
searchers in computer science and related fields.
This book does have prerequisites: We expect the reader to have some
experience with the design of algorithms, maybe through following an under-
graduate course in the field, or through reading a few chapters of the reference
books quoted above. In particular, we assume that the reader is familiar with
classic algorithms, such as comparison-based sorting (e.g., quick sort or merge
sort), and has a good knowledge of elementary graph theory, including:
• traversals (depth-first, breadth-first, connected components);
• shortest paths (one-source, such as Dijkstra, all-pairs, such as Floyd–
Warshall);
• maximum matchings in bipartite graphs.
In fact, one needs to know only that efficient algorithms exist to solve these
graph theory problems, but, of course, it is better to understand how they
work. Because excellent external sources already cover these topics, we refer
to them.
Also, we assume that readers have already been exposed, at least up to
some extent, to the basic paradigms of algorithm design: divide-and-conquer,
greedy algorithms, dynamic programming, and amortized analysis. But here,
rather than (or in addition to) referring to external sources, this book provides

xv

© 2014 by Taylor & Francis Group, LLC

xvi

extensive material so that the readers can assess their skills by solving the
many exercises in Part I.
Part II of this book can be used to teach an undergraduate or graduate class
on NP-completeness, with a focus on polynomial reductions, and a survey of
approaches that go beyond NP-completeness.
Part III of this book can be used to teach a graduate class on advanced
algorithms, either in the form of a series of classes presenting the case studies,
or in the form of projects assigned to students.

Book content and organization

The book is composed of three main parts:
• Part I: Polynomial-time algorithms: Exercises
• Part II: NP-completeness and beyond
• Part III: Reasoning on problem complexity

Part I aims at training the reader to design efficient algorithms. To do

so, we provide a comprehensive set of problems to investigate. Problems are
organized along the main design principles, which we each revisit through a
brief introduction and a series of related exercises. This leads to five chapters:
1. Introduction to complexity
2. Divide-and-conquer
3. Greedy algorithms
4. Dynamic programming
5. Amortized analysis
All solutions to exercises are provided.

Part II deals with NP-completeness and beyond. Our coverage of NP-

completeness focuses on polynomial reductions. We deliberately ignore Tur-
ing machines and the theoretic arsenal. The (small) price to pay is to admit
Cook’s theorem, the existence of the canonical NP-complete problem, for-
mula satisfiability, a.k.a. SAT. In Part II, we also cover approaches that go
beyond NP-completeness: identifying polynomial instances, approximation
algorithms, linear programming, randomized algorithms, branch-and-bound,
and backtracking. Part II consists of four chapters:
1. NP-completeness
2. Exercises on NP-completeness
3. Beyond NP-completeness
4. Exercises going beyond NP-completeness
All solutions to exercises are provided.

Part III constitutes the main originality of the book. It is devoted to case
studies whose goal is to provide the reader with tools and techniques to assess
problem complexity: which instances are polynomial, and which are NP-hard,
and what do to for the latter. Part III consists of an introduction summarizing

© 2014 by Taylor & Francis Group, LLC

 xvii

how to assess the complexity of a new problem, and it is illustrated with five
case studies:
1. Chains-on-chains partitioning
2. Replica placement in tree networks
3. Packet routing
4. Matrix product, or tiling the unit square
5. Online scheduling

Thanks
The content of this book, or at least preliminary versions of it, has been
used to teach courses at École Normale Supérieure de Lyon. We are grate-
ful to the students for their feedback and suggestions. We also thank all
our colleagues who helped gather the problems of Part I. The teaching assis-
tants when Yves Robert was teaching the Algorithms course were (ordering by
year) Odile Millet-Botta, Tanguy Risset, Alain Darte, Bruno Durand, Frédéric
Vivien, Jean-Christophe Dubacq, Olivier Bodini, Daniel Hirschkoff, Matthieu
Exbrayat, Natacha Portier, Emmanuel Hyon, Eric Thierry, Michel Morvan,
and Yves Caniou. The teaching assistants when Anne Benoit took over were
(ordering by year) Victor Poupet, Damien Regnault, Benjamin Depardon,
Jean-François Pineau, Clément Rezvoy, Christophe Mouilleron, Fanny Du-
fossé, and Anne-Cécile Orgerie.
We also wish to thank the following people who have contributed to some
of the content by their insightful suggestions, their own previously published
work, or their help reviewing draft chapters: Guillaume Aupy, Marin Bougeret,
Jean-Yves l’Excellent, Arnaud Legrand, Loris Marchal, Paul Renaud-Goud,
Veronika Sonigo, and Bora Uçar.
Finally, a word of caution on bibliographical notes: Some exercises have ap-
peared in many sources, and the references that we give may well not be the
original ones. Also, the absence of any reference is not a claim for originality!
However, all solutions are ours, and they have been tested and verified by the
students at ENS Lyon, the teaching assistants, and ourselves (but we keep the
sole responsibility for errors). We welcome comments and suggestions to our
e-mail addresses.

Anne Benoit, [email protected]

Yves Robert, [email protected]
Frédéric Vivien, [email protected]

© 2014 by Taylor & Francis Group, LLC

© 2014 by Taylor & Francis Group, LLC
 Part I

Polynomial-time
algorithms: Exercises

© 2014 by Taylor & Francis Group, LLC

© 2014 by Taylor & Francis Group, LLC
Another Random Scribd Document
with Unrelated Content
matter,[33] including those elements whose reversed
spectra correspond with the Fraunhofer lines—namely,
sodium, iron, hydrogen, lithium, calcium, magnesium,
&c. Thus in spectrum analysis we find a means of
determining the composition of the inaccessible
heavenly luminaries, and much has been done in this
respect since Kirchhoff's theory was formulated. By
observations on the spectra of many heavenly bodies,
changes have been discovered going on in them,[34] and
many of the elements known to us have been found
with certainty in them.[35] From this it must be
concluded that the same elements which exist on the
earth occur throughout the whole universe, and that at
that degree of heat which is proper to the sun those
simple substances which we accept as the elements in
chemistry are still undecomposed and remain
unchanged. A high temperature forms one of those
conditions under which compounds most easily
decompose; and if sodium or a similar element were a
compound, in all probability it would be decomposed
into component parts at the high temperature of the
sun. This may indeed be concluded from the fact that in
ordinary spectroscopic experiments the spectra obtained
often belong to the metals and not to the compounds
taken; this depends on the decomposition of these
compounds in the heat of the flame. If common salt be
introduced into the flame of a gas-burner, a portion of it
is decomposed, first forming, in all probability, with
water, hydrochloric acid and sodium hydroxide, and the
latter then becoming partially decomposed by the
hydrocarbons, giving metallic sodium, whose
incandescent vapour emits light of a definite
refrangibility. This conclusion is arrived at from the
following experiment:—If hydrochloric acid gas be
introduced into a flame coloured by sodium it is
observed that the sodium spectrum disappears, owing
to the fact that metallic sodium cannot remain in the
flame in the presence of an excess of hydrochloric acid.
The same thing takes place on the addition of sal-
ammoniac, which in the heat of the flame gives
hydrochloric acid. If a porcelain tube containing sodium
chloride (or sodium hydroxide or carbonate), and closed
at both ends by glass plates, be so powerfully heated
that the salt volatilises, then the sodium spectrum is not
observable; but if the salt be replaced by sodium, then
either the bright line or the absorption spectra is
obtained, according to whether the light emitted by the
incandescent vapour be observed, or light passing
through the tube. Thus the above spectrum is not given
by sodium chloride or other sodium compound, but is
proper to the metal sodium itself. This is also the case
with other analogous metals. The chlorides and other
halogen compounds of barium, calcium, copper, &c.,
give independent spectra which differ from those of the
metals. If barium chloride be introduced into a flame, it
gives a mixed spectrum belonging to metallic barium
and barium chloride. If besides barium chloride,
hydrochloric acid or sal-ammoniac be introduced into
the flame, then the spectrum of the metal disappears,
and that of the chloride remains, which differs distinctly
from the spectrum of barium fluoride, barium bromide,
or barium iodide. A certain common resemblance and
certain common lines are observed in the spectra of two
different compounds of one and the same element
obtained in the above-described manner, and also in the
spectrum of the metal, but they all have their
peculiarities. The independent spectra of the
compounds of copper are easily observed (fig. 75). Thus
certain compounds which exist in a state of vapour, and
are luminous at a high temperature, give their
independent spectra. In the majority of cases the
spectra of compounds are composed of indistinct
luminous lines and complete bright bands, whilst
metallic elements generally give a few clearly-defined
spectral lines.[36] There is no reason for supposing that
the spectrum of a compound is equal to the sum of the
spectra of its elements—that is, every compound which
is not decomposed by heat has its own proper
spectrum. This is best proved by absorption spectra,
which are essentially only reversed spectra observed at
low temperatures. If every salt of sodium, lithium, and
potassium gives one and the same spectrum, this must
be ascribed to the presence in the flame of the free
metals liberated by the decomposition of their salts.
Therefore the phenomena of the spectrum are
determined by molecules, and not by atoms—that is,
the molecules of the metal sodium, and not its atoms,
produce those particular vibrations which determine the
spectrum of a sodium salt. Where there is no free
metallic sodium there is no sodium spectrum.
Spectrum analysis has not only endowed science with
a knowledge of the composition of distant heavenly
bodies (of the sun, stars, nebulæ, comets, &c.), but has
also given a new method for studying the matter of the
earth's surface. With its help Bunsen discovered two
new elements belonging to the group of the alkali
metals, and thallium, indium, and gallium were
afterwards discovered by the same means. The
spectroscope is employed in the study of rare metals
(which in solution often give distinct absorption
spectra), of dyes, and of many organic substances, &c.
[37]
With respect to the metals which are analogous to
sodium, they all give similar very volatile salts and such
very characteristic spectra that the least traces of
them[38] are discovered with great ease by means of the
spectroscope. For instance, lithium gives a very brilliant
red coloration to a flame and a very bright red spectral
line (wave-length, 670 millionths mm.), which indicates
the presence of this metal in admixture with compounds
of other alkali metals.
Lithium, Li, is, like potassium and sodium, somewhat
widely spread in siliceous rocks, but only occurs in small
quantities and as mere traces in considerable masses of
potassium and sodium salts. Only a very few rather rare
minerals contain more than traces of it,[39] for example,
spodumene and lithia mica.
Many compounds of lithium are
in all respects closely analogous
to the corresponding
compounds of sodium and
potassium; but the carbonate is
sparingly soluble in cold water,
which fact is taken advantage of
for separating lithium from Fig. 77.—
potassium and sodium. This Preparation of
salt, Li2CO3, is easily converted lithium by the
into the other compounds of action of a
lithium. Thus, for instance, the galvanic current
lithium hydroxide, LiHO, is on fused lithium
obtained in exactly the same chloride.
way as caustic soda, by the
action of lime on the carbonate,
and it is soluble in water and crystallises (from its
solution in alcohol) as LiHO,H2O. Metallic lithium is
obtained by the action of a galvanic current on fused
lithium chloride; for this purpose a cast-iron crucible,
furnished with a stout cover, is filled with lithium
chloride, heated until the latter fuses, and a strong
galvanic current is then passed through the molten
mass. The positive pole (fig. 77) consists of a dense
carbon rod C (surrounded by a porcelain tube P fixed in
an iron tube BB), and the negative pole of an iron wire,
on which the metal is deposited after the current has
passed through the molten mass for a certain length of
time. Chlorine is evolved at the positive pole. When a
somewhat considerable quantity of the metal has
accumulated on the wire it is withdrawn, the metal is
collected from it, and the experiment is then carried on
as before.[39 bis] Lithium is the lightest of all metals, its
specific gravity is 0·59, owing to which it floats even on
naphtha; it melts at 180°, but does not volatilise at a
red heat. Its appearance recalls that of sodium, and, like
it, it has a yellow tint. At 200° it burns in air with a very
bright flame, forming lithium oxide. In decomposing
water it does not ignite the hydrogen. The characteristic
test for lithium compounds is the red coloration which
they impart to a colourless flame.[40]
Bunsen in 1860 tried to determine by means of the
spectroscope whether any other as yet unknown metals
might not occur in different natural products together
with lithium, potassium, and sodium, and he soon
discovered two new alkali metals showing independent
spectra. They are named after the characteristic
coloration which they impart to the flame. One which
gives a red and violet band is named rubidium, from
rubidius (dark red), and the other is called cæsium,
because it colours a pale flame sky blue, which depends
on its containing bright blue rays, which appear in the
spectrum of cæsium as two blue bands (table on p.
565). Both metals accompany sodium, potassium, and
lithium, but in small quantities; rubidium occurs in larger
quantity than cæsium. The amount of the oxides of
cæsium and rubidium in lepidolite does not generally
exceed one-half per cent. Rubidium has also been found
in the ashes of many plants, while the Stassfurt
carnallite (the mother-liquor obtained after having been
treated for KCl) forms an abundant source for rubidium
and also partly for cæsium. Rubidium also occurs,
although in very small quantities, in the majority of
mineral waters. In a very few cases cæsium is not
accompanied by rubidium; thus, in a certain granite on
the Isle of Elba, cæsium has been discovered, but not
rubidium. This granite contains a very rare mineral
called pollux, which contains as much as 34 per cent. of
cæsium oxide. Guided by the spectroscope, and aided
by the fact that the double salts of platinic chloride and
rubidium and cæsium chlorides are still less soluble in
water than the corresponding potassium salt, K2PtCl6,[41]
Bunsen succeeded in separating both metals from each
other and from potassium, and demonstrated the great
resemblance they bear to each other. The isolated
metals,[42] rubidium and cæsium, have respectively the
specific gravities 1·52 and 2·366, and melting points 39°
and 27° as N. N. Beketoff showed (1894), he having
obtained cæsium by heating CsAlO2 with Mg([42 bis]).
Judging by the properties of the free metals, and of
their corresponding and even very complex compounds,
lithium, sodium, potassium, rubidium, and cæsium
present an indubitable chemical resemblance. The fact
that the metals easily decompose water, and that their
hydroxides RHO and carbonates R2CO3 are soluble in
water, whilst the hydroxides and carbonates of nearly all
other metals are insoluble, shows that these metals
form a natural group of alkali metals. The halogens and
the alkali metals form, by their character, the two
extremes of the elements. Many of the other elements
are metals approaching the alkali metals, both in their
capacity of forming salts and in not forming acid
compounds, but are not so energetic as the alkali
metals, that is, they form less energetic bases. Such are
the common metals, silver, iron, copper, &c. Some other
elements, in the character of their compounds,
approach the halogens, and, like them, combine with
hydrogen, but these compounds do not show the
energetic property of the halogen acids; in a free state
they easily combine with metals, but they do not then
form such saline compounds as the halogens do—in a
word, the halogen properties are less sharply defined in
them than in the halogens themselves. Sulphur,
phosphorus, arsenic, &c. belong to this order of
elements. The clearest distinction of the properties of
the halogens and alkali metals is expressed in the fact
that the former give acids and do not form bases, whilst
the latter, on the contrary, only give bases. The first are
true acid elements, the latter clearly-defined basic or
metallic elements. On combining together, the halogens
form, in a chemical sense, unstable compounds, and the
alkali metals alloys in which the character of the metals
remains unaltered, just as in the compound ICl the
character of the halogens remains undisguised; thus
both classes of elements on combining with members of
their own class form non-characteristic compounds,
which have the properties of their components. On the
other hand, the halogens on combining with the alkali
metals form compounds which are, in all respects,
stable, and in which the original characters of the
halogens and alkali metals have entirely disappeared.
The formation of such compounds is accompanied by
evolution of a large amount of heat, and by an entire
change of both the physical and chemical properties of
the substances originally taken. The alloy of sodium and
potassium, although liquid at the ordinary temperature,
is perfectly metallic, like both its components. The
compound of sodium and chlorine has neither the
appearance nor the properties of the original elements;
sodium chloride melts at a higher temperature, and is
more difficultly volatile, than either sodium or chlorine.
With all these qualitative differences there is,
however, an important quantitative resemblance
between the halogens and the alkali metals. This
resemblance is clearly expressed by stating that both
orders of elements belong to those which are univalent
with respect to hydrogen. It is thus correct to say that
both the above-named orders of elements replace
hydrogen atom for atom. Chlorine is able to take the
place of hydrogen by metalepsis, and the alkali metals
take the place of hydrogen in water and acids. As it is
possible to consecutively replace every equivalent of
hydrogen in a hydrocarbon by chlorine, so it is possible
in an acid containing several equivalents of hydrogen to
replace the hydrogen consecutively equivalent after
equivalent by an alkali metal; hence an atom of these
elements is analogous to an atom of hydrogen, which is
taken, in all cases, as the unit for the comparison of the
other elements. In ammonia, and in water, chlorine and
sodium are able to bring about a direct replacement.
According to the law of substitution, the formation of
sodium chloride, NaCl, at once shows the equivalence of
the atoms of the alkali metals and the halogens. The
halogens and hydrogen and the alkali metals combine
with such elements as oxygen, and it is easily proved
that in such compounds one atom of oxygen is able to
retain two atoms of the halogens, of hydrogen, and of
the alkali metals. For this purpose it is enough to
compare the compounds KHO, K2O, HClO, and Cl2O,
with water. It must not be forgotten, however, that the
halogens give, with oxygen, besides compounds of the
type R2O, higher acid grades of oxidation, which the
alkali metals and hydrogen are not capable of forming.
We shall soon see that these relations are also subject
to a special law, showing a gradual transition of the
properties of the elements from the alkali metals to the
halogens.[43]
The atomic weights of the alkali metals, lithium 7,
sodium 23, potassium 39, rubidium 85, and cæsium
133, show that here, as in the class of halogens, the
elements may be arranged according to their atomic
weights in order to compare the properties of the
analogous compounds of the members of this group.
Thus, for example, the platinochlorides of lithium and
sodium are soluble in water; those of potassium,
rubidium, and cæsium sparingly soluble, and the greater
the atomic weight of the metal the less soluble is the
salt. In other cases the reverse is observed—the greater
the atomic weight the more soluble are the
corresponding salts. The variation of properties with the
variation in atomic weights even shows itself in the
metals themselves; thus lithium volatilises with difficulty,
whilst sodium is obtained by distillation, potassium
volatilises more easily than sodium, and rubidium and
cæsium as we have seen, are still more volatile.

Footnotes:
[1] Tutton's researches (1894) upon the analogy of the crystalline
forms of K2SO4, Rb2SO4 and Cs2SO4 may be taken as a typical example
of the comparison of analogous compounds. We cite the following data
from these excellent researches: the sp. gr. at 20°/4° of K2SO4 is 2·6633
of Rb2SO4, 3·6113, and of Cs2SO4, 4·2434. The coefficient of cubical
expansion (the mean between 20° and 60°) for the K salt is 0·0053, for
the Rb salt 0·0052, for the Cs salt 0·0051. The linear expansion (the
maximum for the vertical axis) along the axis of crystallisation is the
same for all three salts, within the limits of experimental error. The
replacement of potassium by rubidium causes the distance between the
centres of the molecules in the direction of the three axes of
crystallisation to increase equally, and less than with the replacement of
rubidium by cæsium. The index of refraction for all rays and for every
crystalline path (direction) is greater for the rubidium salt than for the
potassium salt, and less than for the cæsium salt, and the differences
are nearly in the ratio 2 : 5. The lengths of the rhombic crystalline axes
for K2SO4 are in the ratio 0·5727 : 1 : 0·7418, for Rb2SO4, 0·5723 : 1 :
0·7485, and for Cs2SO4, 0·5712 : 1 : 0·7521. The development of the
basic and brachy-pinacoids gradually increases in passing from K to Rb
and Cs. The optical properties also follow the same order both at the
ordinary and at a higher temperature. Tutton draws the general
conclusion that the crystallographic properties of the isomorphic rhombic
sulphates R2SO4 are a function of the atomic weight of the metals
contained in them (see Chapter XV.) Such researches as these should do
much towards hastening the establishment of a true molecular
mechanics of physico-chemical phenomena.
[1 bis] The origin of the primary rocks has been mentioned in Chapter
X., Note 2.
[2] Carnallite belongs to the number of double salts which are directly
decomposed by water, and it only crystallises from solutions which
contain an excess of magnesium chloride. It may be prepared artificially
by mixing strong solutions of potassium and magnesium chlorides, when
colourless crystals of sp. gr. 1·60 separate, whilst the Stassfurt salt is
usually of a reddish tint, owing to traces of iron. At the ordinary
temperature sixty-five parts of carnallite are soluble in one hundred
parts of water in the presence of an excess of the salt. It deliquesces in
the air, forming a solution of magnesium chloride and leaving potassium
chloride. The quantity of carnallite produced at Stassfurt is now as much
as 100,000 tons a year.
[3] The method of separating sodium chloride from potassium chloride
has been described in Chapter I. On evaporation of a mixture of the
saturated solutions, sodium chloride separates; and then, on cooling,
potassium chloride separates, owing to the difference of rate of variation
of their solubilities with the temperature. The following are the most
trustworthy figures for the solubility of potassium chloride in one
hundred parts of water (for sodium chloride, see Chapter X., Note 13):—
10° 20° 40° 60° 100°
32 35 40 46 57
When mixed with solutions of other salts the solubility of potassium
chloride naturally varies, but not to any great extent.
 [3 bis] The specific gravity of the solid salt is 1·99—that is, less than
that of sodium chloride. All the salts of sodium are specifically heavier
than the corresponding salts of potassium, as are also their solutions for
equal percentage compositions. If the specific gravity of water at 4° =
10,000, then at 15° the specific gravity of a solution of p p.c. potassium
chloride = 9,992 + 63·29p + 0·226p2, and therefore for 10 p.c. =
1·0647, 20 p.c. = 1·1348, &c.
Potassium chloride combines with iodine trichloride to form a
compound KCl + ICl3 = KICl4, which has a yellow colour, is fusible, loses
iodine trichloride at a red heat, and gives potassium iodate and
hydrochloric acid with water. It is not only formed by direct combination,
but also by many other methods; for instance, by passing chlorine into a
solution of potassium iodide so long as the gas is absorbed, KI + 2Cl2 =
KCl,ICl3. Potassium iodide, when treated with potassium chlorate and
strong hydrochloric acid, also gives this compound; another method for
its formation is given by the equation KClO3 + I + 6HCl = KCl,ICl3 + 3Cl
+ 3H2O. This is a kind of salt corresponding with KIO2 (unknown) in
which the oxygen is replaced by chlorine. If valency be taken as the
starting-point in the study of chemical compounds, and the elements
considered as having a constant atomicity (number of bonds)—that is, if
K, Cl, and I be taken as univalent elements—then it is impossible to
explain the formation of such a compound because, according to this
view, univalent elements are only able to form dual compounds with
each other; such as, KCl, ClI, KI, &c., whilst here they are grouped
together in the molecule KICl4. Wells, Wheeler, and Penfield (1892)
obtained a large number of such poly-haloid salts. They may all be
divided into two large classes: the tri-haloid and the penta-haloid salts.
They have been obtained not only for K but also for Rb and Cs, and
partially also for Na and Li. The general method of their formation
consists in dissolving the ordinary halogen salt of the metal in water, and
treating it with the requisite amount of free halogen. The poly-haloid salt
separates out after evaporating the solution at a more or less low
temperature. In this manner, among the tri-haloid salts, may be
obtained: KI3, KBr2I, KCl2I, and the corresponding salts of rubidium and
cæsium, for instance, CsI3, CsBrI2, CsBr2I, CsClBrI,CsCl2I, CsBr3,
CsClBr2, CsCl2Br, and in general MX3 where X is a halogen. The colour of
the crystals varies according to the halogen, thus CsI3 is black, CrBr3
yellowish red, CrBrI2 reddish brown, CsBr2I red, CsCl2Br yellow. The
cæsium salts are the most stable, and those of potassium least so, as
also those which contain Br and I separately or together; for cæsium no
compounds containing Cl and I were obtained. The penta-haloid salts
form a smaller class; among these salts potassium forms KCl4I, rubidium
RbCl4I, cæsium CsI5, CsBr, CsCl4I, lithium LiCl4I (with 4H2O) and sodium
NaCl4I (with 2H2O). The most stable are those salts containing the
metal with the greatest atomic weight—cæsium (see Chapter XI., Note
63).
[4] It is possible to extract the compounds of potassium directly from
the primary rocks which are so widely distributed over the earth's
surface and so abundant in some localities. From a chemical point of
view this problem presents no difficulty; for instance, by fusing
powdered orthoclase with lime and fluor spar (Ward's method) and then
extracting the alkali with water (on fusion the silica gives an insoluble
compound with lime), or by treating the orthoclase with hydrofluoric acid
(in which case silicon fluoride is evolved as a gas) it is possible to
transfer the alkali of the orthoclase to an aqueous solution, and to
separate it in this manner from the other insoluble oxides. However, as
yet there is no profit in, nor necessity for, recourse to this treatment, as
carnallite and potash form abundant materials for the extraction of
potassium compounds by cheaper methods. Furthermore, the salts of
potassium are now in the majority of chemical reactions replaced by
salts of sodium, especially since the preparation of sodium carbonate
has been facilitated by the Leblanc process. The replacement of
potassium compounds by sodium compounds not only has the
advantage that the salts of sodium are in general cheaper than those of
potassium, but also that a smaller quantity of a sodium salt is needed
for a given reaction than of a potassium salt, because the combining
weight of sodium (23) is less than that of potassium (39).
[5] It has been shown by direct experiment on the cultivation of
plants in artificial soils and in solutions that under conditions (physical,
chemical, and physiological) otherwise identical plants are able to thrive
and become fully developed in the entire absence of sodium salts, but
that their development is impossible without potassium salts.
[6] If herbaceous plants contain much sodium salts, it is evident that
these salts mainly come from the sodium compounds in the water
absorbed by the plants.
[7] As plants always contain mineral substances and cannot thrive in a
medium which does not contain them, more especially in one which is
free from the salts of the four basic oxides, K2O, CaO, MgO, and Fe2O3,
and of the four acid oxides, CO2, N2O5, P2O5, and SO3, and as the
amount of ash-forming substances in plants is small, the question
inevitably arises as to what part these play in the development of plants.
With the existing chemical data only one answer is possible to this
question, and it is still only a hypothesis. This answer was particularly
clearly expressed by Professor Gustavson of the Petroffsky Agricultural
Academy. Starting from the fact (Chapter XI., Note 55) that a small
quantity of aluminium renders possible or facilitates the reaction of
bromine on hydrocarbons at the ordinary temperature, it is easy to
arrive at the conclusion, which is very probable and in accordance with
many data respecting the reactions of organic compounds, that the
addition of mineral substances to organic compounds lowers the
temperature of reaction and in general facilitates chemical reactions in
plants, and thus aids the conversion of the most simple nourishing
substances into the complex component parts of the plant organism.
The province of chemical reactions proceeding in organic substances in
the presence of a small quantity of mineral substances has as yet been
but little investigated, although there are already several disconnected
data concerning reactions of this kind, and although a great deal is
known with regard to such reactions among inorganic compounds. The
essence of the matter may be expressed thus—two substances, A and B,
do not react on each other of their own accord, but the addition of a
small quantity of a third particularly active substance, C, produces the
reaction of A on B, because A combines with C, forming AC, and B reacts
on this new compound, which has a different store of chemical energy,
forming the compound AB or its products, and setting C free again or
retaining it.
 It may here be remarked that all the mineral substances necessary for
plants (those enumerated at the beginning of the note) are the highest
saline compounds of their elements, that they enter into the plants as
salts, that the lower forms of oxidation of the same elements (for
instance, sulphites and phosphites) are harmful to plants (poisonous),
and that strong solutions of the salts assimilated by plants (their osmotic
pressure being great and contracting the cells, as De Vries showed, (see
Chapter I., Note 19)) not only do not enter into the plants but kill them
(poison them).
[8] Besides which, it will be understood from the preceding paragraph
that the salts of potassium may become exhausted from the soil by long
cultivation, and that there may therefore be cases when the direct
fertilisation by salts of potassium may be profitable. But manure and
animal excrements, ashes, and, in general, nearly all refuse which may
serve for fertilising the soil, contain a considerable quantity of potassium
salts, and therefore, as regards the natural salts of potassium
(Stassfurt), and especially potassium sulphate, if they often improve the
crops, it is in all probability due to their action on the properties of the
soil. The agriculturist cannot therefore be advised to add potassium
salts, without making special experiments showing the advantage of
such a fertiliser on a given kind of soil and plant.
The animal body also contains potassium compounds, which is
natural, since animals consume plants. For example, milk, and especially
human milk, contains a somewhat considerable quantity of potassium
compounds. Cow's milk, however, does not contain much potassium salt.
Sodium compounds generally predominate in the bodies of animals. The
excrement of animals, and especially of herbivorous animals, on the
contrary, often contains a large proportion of potassium salts. Thus
sheep's dung is rich in them, and in washing sheep's wool salts of
potassium pass into the water.
The ash of tree stems, as the already dormant portion of the plant
(Chapter VIII., Note 1), contains little potash. For the extraction of
potash, which was formerly carried on extensively in the east of Russia
(before the discovery of the Stassfurt salt), the ash of grasses, and the
green portions of potatoes, buckwheat, &c., are taken and treated with
water (lixiviated), the solution is evaporated, and the residue ignited in
order to destroy the organic matter present in the extract. The residue
thus obtained is composed of raw potash. It is refined by a second
dissolution in a small quantity of water, for the potash itself is very
soluble in water, whilst the impurities are sparingly soluble. The solution
thus obtained is again evaporated, and the residue ignited, and this
potash is then called refined potash, or pearlash. This method of
treatment cannot give chemically pure potassium carbonate. A certain
amount of impurities remain. To obtain chemically pure potassium
carbonate, some other salt of potassium is generally taken and purified
by crystallisation. Potassium carbonate crystallises with difficulty, and it
cannot therefore be purified by this means, whilst other salts, such as
the tartrate, acid carbonate, sulphate, or nitrate, &c., crystallise easily
and may thus be directly purified. The tartrate is most frequently
employed, since it is prepared in large quantities (as a sediment from
wine) for medicinal use under the name of cream of tartar. When ignited
without access of air, it leaves a mixture of charcoal and potassium
carbonate. The charcoal so obtained being in a finely-divided condition,
the mixture (called ‘black flux’), is sometimes used for reducing metals
from their oxides with the aid of heat. A certain quantity of nitre is
added to burn the charcoal formed by heating the cream of tartar.
Potassium carbonate thus prepared is further purified by converting it
into the acid salt, by passing a current of carbonic anhydride through a
strong solution. KHCO3 is then formed, which is less soluble than the
normal salt (as is also the case with the corresponding sodium salts),
and therefore crystals of the acid salt separate from the solution on
cooling. When ignited, they part with their water and carbonic
anhydride, and pure potassium carbonate remains behind. The physical
properties of potassium carbonate distinguish it sufficiently from sodium
carbonate; it is obtained from solutions as a powdery white mass,
having an alkaline taste and reaction, and, as a rule, shows only traces
of crystallisation. It also attracts the moisture of the air with great
energy. The crystals do not contain water, but absorb it from the air,
deliquescing into a saturated solution. It melts at a red heat (1045°),
and at a still higher temperature is even converted into vapour, as has
been observed at glass works where it is employed. It is very soluble. At
the ordinary temperature, water dissolves an equal weight of the salt.
Crystals containing two equivalents of water separate from such a
saturated solution when strongly cooled (Morel obtained K2CO33H2O in
well-formed crystals at +10°). There is no necessity to describe its
reactions, because they are all analogous to those of sodium carbonate.
When manufactured sodium carbonate was but little known, the
consumption of potassium carbonate was very considerable, and even
now washing soda is frequently replaced for household purposes by
‘lye’—i.e. an aqueous solution obtained from ashes. It contains
potassium carbonate, which acts like the sodium salt in washing tissues,
linen, &c.
A mixture of potassium and sodium carbonates fuses with much
greater ease than the separate salts, and a mixture of their solutions
gives well-crystallised salts—for instance (Marguerite's salt),
K2CO3,6H2O,2Na2CO3,6H2O. Crystallisation also occurs in other multiple
proportions of K and Na (in the above case 1 : 2, but 1 : 1 and 1 : 3 are
known), and always with 6 mol. H2O. This is evidently a combination by
similarity, as in alloys, solutions, &c.
[8 bis] About 25,000 tons of potash annually are now prepared from
KCl by this method at Stassfurt.
[9] Potassium sulphate, K2SO4, crystallises from its solutions in an
anhydrous condition, in which respect it differs from the corresponding
sodium salt, just as potassium carbonate differs from sodium carbonate.
In general, it must be observed that the majority of sodium salts
combine more easily with water of crystallisation than the potassium
salts. The solubility of potassium sulphate does not show the same
peculiarities as that of sodium sulphate, because it does not combine
with water of crystallisation; at the ordinary temperature 100 parts of
water dissolve about 10 parts of the salt, at 0° 8·3 parts, and at 100°
about 26 parts. The acid sulphate, KHSO4, obtained easily by heating
crystals of the normal salt with sulphuric acid, is frequently employed in
chemical practice. On heating the mixture of acid and salt, fumes of
sulphuric acid are at first given off; when they cease to be evolved, the
acid salt is contained in the residue. At a higher temperature (of above
600°) the acid salt parts with all the acid contained in it, the normal salt
being re-formed. The definite composition of this acid salt, and the ease
with which it decomposes, render it exceedingly valuable for certain
chemical transformations accomplished by means of sulphuric acid at a
high temperature, because it is possible to take, in the form of this salt,
a strictly definite quantity of sulphuric acid, and to cause it to act on a
given substance at a high temperature, which it is often necessary to do,
more especially in chemical analysis. In this case, the acid salt acts in
exactly the same manner as sulphuric acid itself, but the latter is
inefficient at temperatures above 400°, because it all evaporates, while
at that temperature the acid salt still remains in a fused state, and acts
with the elements of sulphuric acid on the substance taken. Hence by its
means the boiling-point of sulphuric acid is raised. Thus the acid
potassium sulphate is employed, where for conversion of certain oxides,
such as those of iron, aluminium, and chromium, into salts, a high
temperature is required.
Weber, by heating potassium sulphate with an excess of sulphuric acid
at 100°, observed the formation of a lower stratum, which was found to
contain a definite compound containing eight equivalents of SO3 per
equivalent of K2O. The salts of rubidium, cæsium, and thallium give a
similar result, but those of sodium and lithium do not. (See Note 1.)
[10] The bromide and iodide of potassium are used, like the
corresponding sodium compounds, in medicine and photography.
Potassium iodide is easily obtained in a pure state by saturating a
solution of hydriodic acid with caustic potash. In practice, however, this
method is rarely had recourse to, other more simple processes being
employed although they do not give so pure a product. They aim at the
direct formation of hydriodic acid in the liquid in the presence of
potassium hydroxide or carbonate. Thus iodine is thrown into a solution
of pure potash, and hydrogen sulphide passed through the mixture, the
iodine being thus converted into hydriodic acid. Or a solution is prepared
from phosphorus, iodine, and water, containing hydriodic and phosphoric
acid; lime is then added to this solution, when calcium iodide is obtained
in solution, and calcium phosphate as a precipitate. The solution of
calcium iodide gives, with potassium carbonate, insoluble calcium
carbonate and a solution of potassium iodide. If iodine is added to a
slightly-heated solution of caustic potash (free from carbonate—that is,
freshly prepared), so long as the solution is not coloured from the
presence of an excess of iodine, there is formed (as in the action of
chlorine on a solution of caustic potash) a mixture of potassium iodide
and iodate. On evaporating the solution thus obtained and igniting the
residue, the iodate is destroyed and converted into iodide, the oxygen
being disengaged, and potassium iodide only is left behind. On
dissolving the residue in water and then evaporating, cubical crystals of
the anhydrous salt are obtained, which are soluble in water and alcohol,
and on fusion give an alkaline reaction, owing to the fact that when
ignited a portion of the salt decomposes, forming potassium oxide. The
neutral salt may be obtained by adding hydriodic acid to this alkaline salt
until it gives an acid reaction. It is best to add some finely-divided
charcoal to the mixture of iodate and iodide before igniting it, as this
facilitates the evolution of the oxygen from the iodate. The iodate may
also be converted into iodide by the action of certain reducing agents,
such as zinc amalgam, which when boiled with a solution containing an
iodate converts it into iodide. Potassium iodide may also be prepared by
mixing a solution of ferrous iodide (it is best if the solution contain an
excess of iodine) and potassium carbonate, in which case ferrous
carbonate FeCO3, is precipitated (with an excess of iodine the precipitate
is granular, and contains a compound of the suboxide and oxide of iron),
while potassium iodide remains in solution. Ferrous iodide, FeI2, is
obtained by the direct action of iodine on iron in water. Potassium iodide
considerably lowers the temperature (by 24°), when it dissolves in water,
100 parts of the salt dissolve in 73·5 parts of water at 12·5°, in 70 parts
at 18°, whilst the saturated solution which boils at 120° contains 100
parts of salt per 45 parts of water. Solutions of potassium iodide dissolve
a considerable amount of iodine; strong solutions even dissolving as
much or more iodine than they contain as potassium iodide (see Note 3
bis and Chapter XI., Note 64).
[11] Caustic potash is not only formed by the action of lime on dilute
solutions of potassium carbonate (as sodium hydroxide is prepared from
sodium carbonate), but by igniting potassium nitrate with finely-divided
copper (see Note 15), and also by mixing solutions of potassium
sulphate (or even of alum, KAlS2O8) and barium hydroxide, BaH2O2. It is
sometimes purified by dissolving it in alcohol (the impurities, for
example, potassium sulphate and carbonate, are not dissolved) and then
evaporating the alcohol.
 The specific gravity of potassium hydroxide is 2·04, but that of its
solutions (see Chapter XII., Note 18) at 15° S = 9,992 + 90·4p + 0·28p2
(here p2 is +, and for sodium hydroxide it is -). Strong solutions, when
cooled, yield a crystallo-hydrate, KHO,4H2O, which dissolves in water,
producing cold (like 2NaHO,7H2O), whilst potassium hydroxide in
solution develops a considerable amount of heat.
[12] When the yellow prussiate is heated to redness, all the cyanogen
which was in combination with the iron is decomposed into nitrogen,
which is evolved as gas, and carbon, which combines with the iron. In
order to avoid this, potassium carbonate is added to the yellow prussiate
while it is being fused. A mixture of 8 parts of anhydrous yellow
prussiate and 3 parts of pure potassium carbonate is generally taken.
Double decomposition then takes place, resulting in the formation of
ferrous carbonate and potassium cyanide. But by this method, as by the
first, a pure salt is not obtained, because a portion of the potassium
cyanide is oxidised at the expense of the iron carbonate and forms
potassium cyanate, FeCO3 + KCN = CO2 + Fe + KCNO; and the
potassium cyanide very easily forms oxide, which acts on the sides of
the vessel in which the mixture is heated (to avoid this iron vessels
should be used). By adding one part of charcoal powder to the mixture
of 8 parts of anhydrous yellow prussiate and 3 parts of potassium
carbonate a mass is obtained which is free from cyanate, because the
carbon absorbs the oxygen, but in that case it is impossible to obtain a
colourless potassium cyanide by simple fusion, although this may be
easily done by dissolving it in alcohol. Cyanide of potassium may also be
obtained from potassium thiocyanate, which is formed from ammonium
thiocyanate obtained by the action of ammonia upon bisulphide of
carbon (see works upon Organic Chemistry). Potassium cyanide is now
prepared in large quantities from yellow prussiate for gilding and
silvering. When fused in large quantities the action of the oxygen of the
air is limited, and with great care the operation may be successfully
conducted, and therefore, on a large scale, very pure salt is sometimes
obtained. When slowly cooled, the fused salt separates in cubical
crystals like potassium chloride.
Pure KCN is obtained by passing CNH gas into an alcoholic solution of
KHO. The large amount of potassium cyanide which is now required for
the extraction of gold from its ores, is being replaced by a mixture
(Rossler and Gasslaker, 1892) of KCN and NaCN, prepared by heating
powdered and dried yellow prussiate with metallic sodium: K4Fe(CN)6 +
2Na = 4KCN + 2NaCN + Fe. This method offers two advantages over
the above methods: (1) the whole of the cyanide is obtained, and does
not decompose with the formation of N2; and (2) no cyanates are
formed, as is the case when carbonate of potash is heated with the
prussiate.
[13] A considerable quantity of potassium cyanide is used in the arts,
more particularly for the preparation of metallic solutions which are
decomposed by the action of a galvanic current; thus it is very
frequently employed in electro-silvering and gilding. An alkaline solution
is prepared, which is moderately stable owing to the fact that potassium
cyanide in the form of certain double salts—that is, combined with other
cyanides—is far more stable than when alone (yellow prussiate, which
contains potassium cyanide in combination with ferrous cyanide, is an
example of this).
[13 bis] A dilute solution of KCN is taken, not containing more than 1
per cent. KCN. MacLaurin explains this by the fact that strong solutions
dissolve gold less rapidly, owing to their dissolving less air, whose
oxygen is necessary for the reaction.
[14] Besides which Schloesing and Müntz, by employing similar
methods to Pasteur, showed that the formation of nitre in the
decomposition of nitrogenous substances is accomplished by the aid of
peculiar micro-organisms (ferments), without which the simultaneous
action of the other necessary conditions (alkalis, moisture, a
temperature of 37°, air, and nitrogenous substances) cannot give nitre.
[14 bis] Before fusing, the crystals of potassium nitrate change their
form, and take the same form as sodium nitrate—that is, they change
into rhombohedra. Nitre crystallises from hot solutions, and in general
under the influence of a rise of temperature, in a different form from
that given at the ordinary or lower temperatures. Fused nitre solidifies to
a radiated crystalline mass; but it does not exhibit this structure if
metallic chlorides be present, so that this method may be taken
advantage of to determine the degree of purity of nitre.
 Carnelley and Thomson (1888) determined the fusing point of
mixtures of potassium and sodium nitrates. The first salt fuses at 339°
and the second at 316°, and if p be the percentage amount of
potassium nitrate, then the results obtained were—
p = 10 20 30 40 50 60 70 80 90
298° 283° 268° 242° 231° 231° 242° 284° 306°
which confirms Shaffgotsch's observation (1857) that the lowest
fusing point (about 231°) is given by mixing molecular quantities (p =
54·3) of the salts—that is, in the formation of the alloy, KNO3,NaNO3.
A somewhat similar result was discovered by the same observers for
the solubility of mixtures of these salts at 20° in 100 parts of water.
Thus, if p be the weight of potassium nitrate mixed with 100 - p parts by
weight of sodium nitrate taken for solution, and c be the quantity of the
mixed salts which dissolves in 100, the solubility of sodium nitrate being
85, and of potassium nitrate 34, parts in 100 parts of water, then—
p = 10 20 30 40 50 60 70 80 90
c = 110 136 136 138 106 81 73 54 41
The maximum solubility proved not to correspond with the most
fusible mixture, but to one much richer in sodium nitrate.
Both these phenomena show that in homogeneous liquid mixtures the
chemical forces that act between substances are the same as those that
determine the molecular weights of substances, even when the mixture
consists of such analogous substances as potassium and sodium
nitrates, between which there is no direct chemical interchange. It is
instructive to note also that the maximum solubility does not correspond
with the minimum fusing point, which naturally depends on the fact that
in solution a third substance, namely water, plays a part, although an
attraction between the salts, like that which exists between sodium and
potassium carbonates (Note 8), also partially acts.
[15] Fused nitre, with a further rise of temperature, disengages
oxygen and then nitrogen. The nitrite KNO2 is first formed and then
potassium oxide. The admixture of certain metals—for example, of
finely-divided copper—aids the last decomposition. The oxygen in this
case naturally passes over to the metal.