美国大学生数学建模竞赛题解析与研究第5辑

内容提要
本系列丛书是以美国大学生数学建模竞赛（MCM/ICM）赛题为主
要研究对象，结合竞赛特等奖的优秀论文，对相关的问题做深入细致的
解析与研究。本辑针对 2013 年 MCM/ICM 竞赛的 3 个题目：最佳巧克
力蛋糕烤盘问题、淡水资源的调配问题以及地球生态环境的健康临界点
问题等进行了解析与研究。由于参赛论文需要用英语书写，同时考虑到
不同层次读者的需要，故本书包括两部分：一部分是地道的英文，另一
部分为与之对应的中文，便于读者在学习建模方法的同时，逐步培养用
英文写作及思考的习惯。
本书内容新颖、实用性强，可作为指导学生参加美国大学生数学建
模竞赛的主讲教材，也可作为本科生、研究生学习和准备全国大学生、
研究生数学建模竞赛的参考书，同时也可供研究相关问题的科研人员参
考使用。
　　图书在版编目（C I P）数据
　　美国大学生数学建模竞赛题解析与研究 . 第 5 辑 / 王
杰等编著 . -- 北京 : 高等教育出版社，2014. 3
　　（美国 MCM/ICM 竞赛指导丛书 / 王杰主编）
　　ISBN 978-7-04-033914-7
　　Ⅰ. ①美… Ⅱ. ①王… Ⅲ. ①数学模型 - 竞赛题 - 研

究 Ⅳ . ① O141.4-44
　　中国版本图书馆 CIP 数据核字（2014）第 039903 号
策划编辑刘英　　　　责任编辑　刘英　　　封面设计　李卫青　　　　版式设计　杜微言

插图绘制　尹　莉　　　　责任校对　陈　杨　　　　责任印制　韩刚

出版发行高等教育出版社咨询电话 400-810-0598
社址北京市西城区德外大街4号网址 http://www.hep.edu.cn
邮政编码 100120 http://www.hep.com.cn
印刷涿州市星河印刷有限公司网上订购 http://www.landraco.com
开本 787mm×1092mm 1/16 http://www.landraco.com.cn
印张 17.25 版次 2014 年 6 月第 1 版
字数 330千字印次 2014 年 6月第 1 次印刷
购书热线 010-58581118 定价 45.00元
本书如有缺页、倒页、脱页等质量问题，请到所购图书销售部门联系调换
版权所有侵权必究
物料号 33914-00
审图号 GS（2013）2902号
“美国ＭＣＭ／
ＩＣＭ竞赛指导丛书”
编审委员会
顾　问
Ｓｏ
ｌＧａ
ｒｆｕｎｋｅ
ｌ　美国数学及应用联合会（ＣＯＭＡＰ）
Ｃｈｒ
ｉｓＡｒ
ｎｅｙ西点军校
主　编
王　杰麻省大学罗威尔分校／
ＣＯＭＡＰ中国合作部
副主编
叶其孝北京理工大学
秘　书
毛紫阳国防科学技术大学
委　员（按姓氏拼音排序）
Ｊ
ａｙＢｅ
ｌａｎｇ
ｅｒ杜鲁门州立大学
Ｊ
ｏｓｅ
ｐｈＭｙ
ｅｒｓ美国陆军研究部
陈秀珍乔治华盛顿大学
龚维博麻省大学阿默斯特分校
韩中庚解放军信息工程大学
贺明峰大连理工大学
贺祖国北京邮电大学
刘深泉华南理工大学
鲁习文华东理工大学
谭　忠厦门大学
王嘉寅圣路易斯华盛顿大学
吴孟达国防科学技术大学
叶正麟西北工业大学
张存权西弗吉尼亚大学
COMAP
the Mathematical Contest in Modeling, MCM

30 90
25
5 000
COMAP
COMAP 3 MCM, ICMthe Interdisciplinary Con-
test in Modeling HiMCMthe High School Mathematical Contest in Modeling

——

COMAP

COMAP

MCM/ICM

· ,
COMAP
2012 11
Forward by Sol Garfunkel
While it is hard for me to believe, the Mathematical Contest in Modeling (MCM)

is fast approaching its 30th year. During this time we have grown from 90 US teams
to over 5,000 teams representing 25 countries from all across the globe. We have
been especially buoyed by the enthusiasm shown by our Chinese colleagues and the
rapid growth in Chinese participation. COMAP welcomes your involvement with
open arms.
COMAP runs three contests in mathematical modeling; they are MCM, ICM
(the Interdisciplinary Contest in Modeling), and HiMCM (the High School Mathe-
matical Contest in Modeling). The purpose of all of these contests has never been
simply to reward student efforts — as important as that is. Rather, our objective
from the beginning has been to increase the presence of applied mathematics and
modeling in education systems at all levels worldwide. Modeling is an attempt to
learn how the world works and the use of mathematics can help us produce better
models. This is not a job for one country, but for all. The COMAP modeling
contests were conceived and evolved to be strong instruments to help achieve this
much larger goal.
It is my supreme hope that through this excellent book series the Chinese
students will learn more about COMAP contests and more about the process of
mathematical modeling. I hope that you will begin to work on the exciting and im-
portant problems you see here, and that you will join the MCM/ICM contests and
the rewarding work of increasing the awareness of the importance of mathematical
modeling.
Sol Garfunkel, PhD

Executive Director
COMAP
November 2012
ICM

MCM/ICM

· ,

ICM
2011 10
Forward by Chris Arney
Undergraduate students who receive instruction and experiences in mathematical

modeling become better and more creative problem solvers and graduate students.
This book series is being published to prepare and educate students on the topics
and concepts of mathematical modeling to help them establish a problem solving
foundation for a successful career.
Mathematical modeling is both a process and a mindset or philosophy. As
a process, students need instruction and experience in understanding and using
the modeling process or framework. As part of their experience, they need to see
various levels of sophistication and complexity, along with various types of mathe-
matical structures (discrete, continuous, linear, nonlinear, deterministic, stochastic,
geometric, and analytic). As a mindset, students need to see problems that are
relevant, challenging, and interesting so they build a passion for the process and
its utility in their lives. A major goal in modeling is for students to want to model
problems and find their solutions. Recipes for structured or prescribed problem
solving (canned algorithms and formulas) do exist in the real world, but mathe-
matical modelers can do much more than execute recipes or formulas. Modelers
are empowered to solve new, open, unsolved problems.
In order to build sufficient experience in modeling, student exposure must be-
gin as early as possible – definitely by the early undergraduate years. Then the
modeling process can be reinforced and used throughout their undergraduate pro-
gram. Since modeling is interdisciplinary, students from all areas of undergraduate
study benefit from this experience.
The articles and chapters in this series expose the readers to model construc-
tion, model analysis, and modeling as a research tool. All these areas are important
and build the students’ modeling skills. Modeling is a challenging and advanced
skill, but one that is empowering and important in student development. In to-
day’s world, models are often complex and require sophisticated computation or
simulation to provide solutions or insights into model behavior. Now is an exciting
ii Forward by Chris Arney
time to be a skilled modeler since methodology to provide visualization and find

solutions are more prevalent and more powerful than ever before.
I wish the students well in their adventure into modeling and I likewise wish
faculty well as they use the examples and techniques in this book series to teach the
modeling process to their students. My advice to all levels of modelers is to build
your confidence and skills and use your talents to solve society’s most challenging
and important problems. Good luck in modeling!
Chris Arney, PhD

United States Military Academy at West Point
Professor of Mathematics
Director of the Interdisciplinary Contest in Modeling
October 2011

the Mathematical Contest in Modeling, MCM/

the Interdisciplinary Contest in Modeling, ICM

1985

3 72 4 96

MCM/ICM

158
90
2012 MCM/ICM 5 026

MCM 3 697 ICM 1 329 MCM/ICM

Outstanding Winners

MCM/ICM

MCM/ICM

COMAP

MCM/ICM
ii

MCM/ICM

COMAP Sol
Garfunkel ICM Chris Arney

MCM/ICM

2013 MCM/ICM 1985

Outstanding Winners
2014
2013 MCM/ICM

! 5 124 MCM
" " #

"
"

$

2013 MCM/ICM MCM/ICM

MCM/ICM #

2013 MCM A
B

C
C
C % C

1

2

3 4

A B

5 6 C

5
6
3 4
5 6

ii

MCM/ICM

;

wang@
comap.com

COMAP

Jay Belanger
2013 10

COMAP
Forward by Sol Garfunkel
ICM
Forward by Chris Arney
I Mathematical Modeling for MCM/ICM 2013
Chapter 1 Introduction · · · · · · · · · · · · · · · · · · · · · · · · · 2
Chapter 2 General Methodology · · · · · · · · · · · · · · · · · · · · 6

2.1 Before the Contest · · · · · · · · · · · · · · · · · · · · · · · · · · 6
2.2 Reading the Problem· · · · · · · · · · · · · · · · · · · · · · · · · 7
2.3 Literature Search· · · · · · · · · · · · · · · · · · · · · · · · · · · 9
2.4 Creating Models · · · · · · · · · · · · · · · · · · · · · · · · · · · 11
2.5 Finding Solutions · · · · · · · · · · · · · · · · · · · · · · · · · · 12
2.6 Stability and Sensitivity · · · · · · · · · · · · · · · · · · · · · · · 13
2.7 Writing up the Solution · · · · · · · · · · · · · · · · · · · · · · · 14
Exercises· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 15
Chapter 3 The Ultimate Brownie Pan · · · · · · · · · · · · · · · · 16

3.1 Problem Description · · · · · · · · · · · · · · · · · · · · · · · · · 16
ii
3.2 How to Approach the Problem · · · · · · · · · · · · · · · · · · · · 17

3.2.1 Deciding What Needs to Be Done· · · · · · · · · · · · · · · 17
3.2.2 Searching the Literature· · · · · · · · · · · · · · · · · · · · 19
3.2.3 Creating a Model · · · · · · · · · · · · · · · · · · · · · · · 21
3.2.4 Finding Solutions · · · · · · · · · · · · · · · · · · · · · · · 22
3.2.5 Stability Testing · · · · · · · · · · · · · · · · · · · · · · · · 22
3.3 Judging· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 23
3.3.1 Judging Criteria · · · · · · · · · · · · · · · · · · · · · · · · 23
3.3.2 Outstanding Winners · · · · · · · · · · · · · · · · · · · · · 24
3.4 Model 1: The Best Rounded Rectangle for Ultimate Brownies · · · 25
3.4.1 Packing the Pans in Standard Ovens · · · · · · · · · · · · · 25
3.4.2 Maximizing the Even Distribution of Heat · · · · · · · · · · 28
3.4.3 Optimizing the Perimeters of Pans · · · · · · · · · · · · · · 34
3.5 Model 2: An Outstanding Brownie Pan · · · · · · · · · · · · · · · 35
3.5.1 Analyzing a Circular Pan · · · · · · · · · · · · · · · · · · · 36
3.5.2 Examining Two-dimensional Shape · · · · · · · · · · · · · · 37
3.5.3 Finding the Best Shape · · · · · · · · · · · · · · · · · · · · 39
Exercises· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 41
Chapter 4 Water, Water, Everywhere · · · · · · · · · · · · · · · · · 43

4.2.1 Searching the Literature· · · · · · · · · · · · · · · · · · · · 44
4.2.2 Creating a Model · · · · · · · · · · · · · · · · · · · · · · · 45
4.2.3 Finding Solutions · · · · · · · · · · · · · · · · · · · · · · · 46
4.2.4 Stability Test · · · · · · · · · · · · · · · · · · · · · · · · · 46
4.3 Judging· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 46
4.4 Model 1: A Highly Regarded Water Strategy · · · · · · · · · · · · 48
4.4.1 Water Supply and Demand · · · · · · · · · · · · · · · · · · 48
4.4.2 Modeling Water Transfer · · · · · · · · · · · · · · · · · · · 52
4.4.3 Water Transfer and Desalinization · · · · · · · · · · · · · · 53
4.4.4 Including Water Conservation· · · · · · · · · · · · · · · · · 55
iii
4.5 Model 2: An Outstanding Water Strategy · · · · · · · · · · · · · · 55

4.5.1 Water Supply and Demand · · · · · · · · · · · · · · · · · · 56
4.5.2 Water Distribution Costs · · · · · · · · · · · · · · · · · · · 57
4.5.3 Modeling Water Transfer and Desalinization · · · · · · · · · 58
4.5.4 Using the Model to Determine a Strategy· · · · · · · · · · · 59
4.5.5 The Best Strategies · · · · · · · · · · · · · · · · · · · · · · 60
4.6 Model 3: Meeting the Demands of Saudi Arabia’s Water Needs · · · 61
4.6.1 Modeling Desalinization · · · · · · · · · · · · · · · · · · · · 61
4.6.2 A Water Distribution Strategy · · · · · · · · · · · · · · · · 62
4.6.3 Treating Waste Water · · · · · · · · · · · · · · · · · · · · · 65
4.6.4 Comments · · · · · · · · · · · · · · · · · · · · · · · · · · · 66
Exercises· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 66
Chapter 5 Network Modeling of Earth’s Health I · · · · · · · · · · 68

5.2.1 Select Measures · · · · · · · · · · · · · · · · · · · · · · · · 73
5.2.2 Devise Network Models · · · · · · · · · · · · · · · · · · · · 73
5.2.3 Validate Models and Perform Sensitivity Analysis · · · · · · 75
5.3 Judging Criteria · · · · · · · · · · · · · · · · · · · · · · · · · · · 75
5.4 Model 1: A Two-level Network Model Using Differential Equations · 78
5.4.1 Network Model · · · · · · · · · · · · · · · · · · · · · · · · 79
5.4.2 Sub-level Integrations as Predator and Preys · · · · · · · · · 80
5.4.3 Reaction Diffusions in the Top-level Network · · · · · · · · · 81
5.4.4 The Final Equation · · · · · · · · · · · · · · · · · · · · · · 82
5.4.5 Tipping Points · · · · · · · · · · · · · · · · · · · · · · · · 83
5.4.6 Sensitivity Analysis · · · · · · · · · · · · · · · · · · · · · · 85
5.4.7 Network Structures · · · · · · · · · · · · · · · · · · · · · · 86
5.4.8 Adding Disturbances to Parameters· · · · · · · · · · · · · · 88
5.4.9 Comments · · · · · · · · · · · · · · · · · · · · · · · · · · · 88
5.5 Model 2: A Network Model on Ocean Pollution Diffusion · · · · · · 89
5.5.1 An Ocean Pollution Network · · · · · · · · · · · · · · · · · 89
iv
5.5.2 Model 2.1: Pollution Diffusion · · · · · · · · · · · · · · · · 91

5.5.3 Prediction of Ocean Pollution Trends · · · · · · · · · · · · · 92
5.5.4 Model 2.2: A Transmission Control System · · · · · · · · · · 94
5.5.5 Prediction of Pollution Trends · · · · · · · · · · · · · · · · 95
5.5.7 Network Structures and Critical Nodes · · · · · · · · · · · · 98
5.5.8 Comments · · · · · · · · · · · · · · · · · · · · · · · · · · · 100
Exercises· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 100
Chapter 6 Network Modeling of Earth’s Health II · · · · · · · · · 103

6.1 Model 1: A Two-level Network over Forrester’s World Model · · · · 103
6.1.1 The Network · · · · · · · · · · · · · · · · · · · · · · · · · 103
6.1.2 Measuring Environmental Health · · · · · · · · · · · · · · · 106
6.1.3 Sub-level Interactions as Bipartite Bayesian Belief
Networks · · · · · · · · · · · · · · · · · · · · · · · · · · · 107
6.1.4 Tipping Point for Environmental Health · · · · · · · · · · · 108
6.1.5 Top-level Interactions as Communication Networks · · · · · · 109
6.1.6 Global Health and the Tipping Point · · · · · · · · · · · · · 111
6.1.7 Prediction of Future Trends · · · · · · · · · · · · · · · · · · 113
6.1.8 Comments · · · · · · · · · · · · · · · · · · · · · · · · · · · 113
6.2 Model 2: A Technology-diffusion Network Model · · · · · · · · · · 115
6.2.1 Technology Diffusion Model · · · · · · · · · · · · · · · · · · 115
6.2.2 CO2 Regression Model · · · · · · · · · · · · · · · · · · · · 117
6.2.3 CO2 Absorbtion Model · · · · · · · · · · · · · · · · · · · · 117
6.2.4 Model Validation · · · · · · · · · · · · · · · · · · · · · · · 120
6.2.5 Global Health Prediction · · · · · · · · · · · · · · · · · · · 120
6.2.6 Structure Analysis· · · · · · · · · · · · · · · · · · · · · · · 121
6.2.7 Comments · · · · · · · · · · · · · · · · · · · · · · · · · · · 122
6.3 Model 3: A Data-driven Network Model · · · · · · · · · · · · · · · 123
6.3.1 Environmental Effects by Earth Damage Score · · · · · · · · 123
6.3.2 Human Actions · · · · · · · · · · · · · · · · · · · · · · · · 123
6.3.3 Assumptions· · · · · · · · · · · · · · · · · · · · · · · · · · 124
6.3.4 The Network Model · · · · · · · · · · · · · · · · · · · · · · 125
6.3.5 Measuring Centrality with PageRank · · · · · · · · · · · · · 128
v

6.3.7 Comments · · · · · · · · · · · · · · · · · · · · · · · · · · · 131
Exercises· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 131
II 2013
1 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 134
2 · · · · · · · · · · · · · · · · · · · · · · · · · · · 137

2.1 · · · · · · · · · · · · · · · · · · · · · · · · · · 137
2.2 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 138
2.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 139
2.4 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 140
2.5 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 141
2.6 · · · · · · · · · · · · · · · · · · · · · · · · · · · 142
2.7 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 143
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 143
3 · · · · · · · · · · · · · · · · · · · · · · · 144

3.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 144
3.2 · · · · · · · · · · · · · · · · · · · · · · · · · · · · 145
3.2.1 · · · · · · · · · · · · · · · · · · · · · · · · · 145
3.2.2 · · · · · · · · · · · · · · · · · · · · · · · · · · · 146
3.2.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · 147
3.2.4 · · · · · · · · · · · · · · · · · · · · · · · · · · · 148
3.2.5 · · · · · · · · · · · · · · · · · · · · · · · · · · 149
3.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 149
3.3.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · 149
3.3.2 · · · · · · · · · · · · · · · · · · · · · · 150
3.4 1: · · · · · · · · · · · · · · · 151
3.4.1 · · · · · · · · · · · · · · · · · · · · 151
3.4.2 · · · · · · · · · · · · · · · · · · · · · 153
3.4.3 · · · · · · · · · · · · · · · · · · · · · · · · · 159
vi
3.5 2: · · · · · · · · · · · · · · · · · · 160

3.5.1 · · · · · · · · · · · · · · · · · · · · · · 160
3.5.2 · · · · · · · · · · · · · · · · · · · · · · · · · · · 161
3.5.3 · · · · · · · · · · · · · · · · · · · · · · 163
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 165
4 , , · · · · · · · · · · · · · · · · · · · · · · · · · · 167

4.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 167
4.2 · · · · · · · · · · · · · · · · · · · · · · · · · · 167
4.2.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · 168
4.2.2 · · · · · · · · · · · · · · · · · · · · · · · · · · · 168
4.2.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · 169
4.2.4
· · · · · · · · · · · · · · · · · · · · · · · · · · 169
4.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 169
4.3.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · 169
4.3.2 · · · · · · · · · · · · · · · · · · · · · · 170
4.4 1: · · · · · · · · · · · · · · · · · · · · 171
4.4.1 · · · · · · · · · · · · · · · · · · · · · · · 171
4.4.2 · · · · · · · · · · · · · · · · · · · · · · · · · 174
4.4.3 · · · · · · · · · · · · · · · · · · · · · 175
4.4.4 · · · · · · · · · · · · · · · · · · · · · · · · · 176
4.5 2: · · · · · · · · · · · · · · · · · · · · · 177
4.5.1 · · · · · · · · · · · · · · · · · · · · · · 177
4.5.2 · · · · · · · · · · · · · · · · · · · · · · · · · 178
4.5.3 · · · · · · · · · · · · · · · · · · · · · 179
4.5.4 · · · · · · · · · · · · · · · · · · · · · 180
4.5.5 · · · · · · · · · · · · · · · · · · · · · · · · · · · 180
4.6 3:
· · · · · · · · · · · · · 181
4.6.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · 181
4.6.2 · · · · · · · · · · · · · · · · · · · · · · · · · · · 182
4.6.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · 184
4.6.4 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 185
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 185
vii
5 I · · · · · · · · · · · · · · · · · · · · · · · · 187

5.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 187
5.2
· · · · · · · · · · · · · · · · · · · · · · · 190
5.2.1
· · · · · · · · · · · · · · · · · · · · · · · · · 191
5.2.2 · · · · · · · · · · · · · · · · · · · · · · · · · 191
5.2.3 · · · · · · · · · · · · · · · · · · · · 192
5.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 192
5.3.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · 193
5.3.2 · · · · · · · · · · · · · · · · · · · · · · 194
5.4 1: · · · · · · · · · · · · · · · 195
5.4.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · 195
5.4.2 · · · · · · · · · · · 197
5.4.3

· · · · · · · · · · · · · · · · · · · · · 197
5.4.4 · · · · · · · · · · · · · · · · · · · · · · · · · · · 199
5.4.5 · · · · · · · · · · · · · · · · · · · · · · · · · · · · 199
5.4.6 · · · · · · · · · · · · · · · · · · · · · · · · · · 201
5.4.7 · · · · · · · · · · · · · · · · · · · · · · · · · · · 202
5.4.8
· · · · · · · · · · · · · · · · · · · · · · · 203
5.4.9 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 204
5.5 2:
· · · · · · · · · · · · · · · · · · · 204
5.5.1
· · · · · · · · · · · · · · · · · · · · · · · · · 204
5.5.2 2.1:
· · · · · · · · · · · · · · · · · · · 206
5.5.3
· · · · · · · · · · · · · · · · · · · · · · 207
5.5.4 2.2: · · · · · · · · · · · · · · · · · · · 208
5.5.5
· · · · · · · · · · · · · · · · · · · · · · · 209
5.5.6 · · · · · · · · · · · · · · · · · · · · · · · · · · 210
5.5.7 · · · · · · · · · · · · · · · · · · · · · 212
5.5.8 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 214
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 214
6 II · · · · · · · · · · · · · · · · · · · · · · · 216

6.1 1: Forrester · · · · · · · · · · 216
6.1.1 · · · · · · · · · · · · · · · · · · · · · · · · · · · 216
6.1.2 · · · · · · · · · · · · · · · · · · · · · · · · · 218
viii
6.1.3 · · · · · 219

6.1.4 · · · · · · · · · · · · · · · · · · · · · · 221
6.1.5 · · · · · · · · · · · 221
6.1.6 · · · · · · · · · · · · · · · · · · · · · · 223
6.1.7 · · · · · · · · · · · · · · · · · · · · · · · 224
6.1.8 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 225
6.2 2:
· · · · · · · · · · · · · · · · · · · · · 226
6.2.1
· · · · · · · · · · · · · · · · · · · · · · · · · 226
6.2.2 CO2 · · · · · · · · · · · · · · · · · · · · · · · · · 227
6.2.3 CO2 · · · · · · · · · · · · · · · · · · · · · · · · · 229
6.2.4 · · · · · · · · · · · · · · · · · · · · · · · · · · · 230
6.2.5 · · · · · · · · · · · · · · · · · · · · · · · · · 230
6.2.6 · · · · · · · · · · · · · · · · · · · · · · · · · · · 231
6.2.7 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 232
6.3 3: · · · · · · · · · · · · · · · · · · · · · 232
6.3.1 · · · · · · · · · · · · · · · · · · 232
6.3.2 · · · · · · · · · · · · · · · · · · · · · · · · · · · 233
6.3.3 · · · · · · · · · · · · · · · · · · · · · · · · · · · 234
6.3.4 · · · · · · · · · · · · · · · · · · · · · · · · · · · 234
6.3.5 PageRank · · · · · · · · · · · · · · · · · 237
6.3.6 · · · · · · · · · · · · · · · · · · · · · · · · · · 237
6.3.7 · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 239
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 240
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 241
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 244
· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 247
Ⅰ
Mathematical Modeling
for MCM/ICM 2013
Chapter 1 Introduction
The Mathematical Contest in Modeling (MCM) and the Interdisciplinary Contest

in Modeling (ICM) are annual mathematical contests held at the same time orga-
nized by the Consortium for Mathematics and Its Applications (COMAP), where
students work in teams to solve non-trivial real world problems. As the name im-
plies, the teams are expected to create and use mathematical models to describe
the problem and use them to come up with their solutions. Ben Fusaro, the founder
of the MCM, said: “Students generally like a challenge and probably are attracted
by the opportunity, for perhaps the first time in their mathematical lives, to work
as a team on a realistic applied problem.”[1]
The MCM was motivated by an older contest; the William Lowell Putnam
Mathematical Competition (Putnam), which began in 1927. For the Putnam exam,
the contestants have six hours, broken up into two three hour sessions, to solve
twelve problems. The students do the work individually with no resources; solving
the problems typically relies on cleverness rather than a knowledge of advanced
mathematics. Low scores are the norm for the Putnam, and it is not unusual for a
contestant to get no points.
In recent years there has been a greater emphasis on applied mathematics in
the undergraduate curriculum. This is at least partially due to the rise of accessi-
ble computing, which invigorated mathematical modeling and made it feasible for
students to work on non-trivial mathematical modeling problems.
The MCM began in 1984, the same year that the GNU project started and
the Apple Macintosh was introduced. Ben Fusaro thought that there should be
an applied counterpart to the theoretical Putnam contest. He originally suggested
having a contest with two problems, one involving continuous mathematics and one
involving discrete mathematics, and having the contestants work on both problems
in a day, but later changed this to having a team of students work on one problem
over a weekend. Sol Garfunkel, the director of COMAP, told Ben Fusaro that he
should ask for funding from the Department of Education’s Fund for the Improve-
Chapter 1 Introduction 3
ment of Postsecondary Education (FIPSE). This resulted in a three year grant.

From the proposal[2] :
The purpose of this competition is to involve students and faculty in

clarifying, analyzing, and proposing solutions to open-ended problems.
We propose a structure which will encourage widespread participation
and emphasize the entire modeling process. Major features include:
• The selection of realistic open-ended problems chosen with the advice
of working mathematicians in industry and government.
• An extended period for teams to prepare solution papers within clearly
defined format.
• The ability of participants to draw on outside resources including com-
puters and texts.
• An emphasis on clarity of exposition in determining final awards with
the best papers published in professional mathematics journals.
As the contest becomes established in the mathematics community, new
courses, workshops, and seminars will be developed to help students and
faculty gain increased experience with mathematical modeling.
The MCM began as a national contest but quickly became international. The
contest was founded in 1984, but the first contest was in 1985 and had less than
200 teams. In 2013, there were over 6,000 teams participating in the MCM/ICM
contests, most of which were international, and over 90% of them were from China.
The MCM consists of two problems called Problem A and Problem B. Contin-
uous mathematics should be used to model Problem A and discrete mathematics
for Problem B. The ICM consists of one problem called Problem C, which can be
modeled using either continuous mathematics, or discrete mathematics, or both.
Each team is only required to complete one problem of their choice. The MCM/ICM
papers are graded into categories by judges, with the following possible results:
Unsuccessful Participant This indicates that the paper did not significantly
address the problem or violated the rules of the contest.
Successful Participant This indicates that the paper addressed the problem and
was reasonably well written. More than half of the papers are Successful Par-
ticipants.
Meritorious A little over ten percent of the papers are Meritorious.
Honorable Mention A little over one fourth of the papers get Honorable Men-
tion.
4 Chapter 1 Introduction
Finalist About one percent of the papers are Finalists.

Outstanding Winner This indicates that the judges deemed the paper to be the
best written and use the best models. About one percent of the papers are
Outstanding Winners.
In addition, the Fusaro Award, named after Ben Fusaro, is given to an MCM team
which demonstrated a particularly creative approach and the Frank R. Giordano
Award, names after the man who directed the MCM for twenty years, is given to
an MCM team which demonstrated a particularly good modeling process.
The judging takes place over a weekend three weeks after the end of the contest.
The first part of judging is “Triage”. For triage, there is about one judge for each
twenty-five papers, as well as a head judge. Each paper is read by two judges, who
independently give the paper a score. The scores are usually pretty close; if they
are not, the two judges discuss the paper to resolve the discrepancy. If they cannot
agree, a third judge is consulted to help them. Once the paper has two close scores,
the scores are added together to give the paper an overall score. The head judge
and contest director decide on a cutoff score; papers at or below the cutoff score
will be classified as Successful Participants unless they don’t meet the standards
of the competition, in which case they are Unsuccessful Participants. About half
of the papers will have scores above the cutoff; these papers will go on to further
judging for a higher ranking.
In 2013, the MCM papers submitted by the Chinese teams were, for the first
time, also judged simultaneously by the Chinese judges in the triage. The purpose
was to compare how the Chinese judges and the US judges would judge the papers.
The Chinese and US judges do not necessarily judge the papers in the same way.
For instance, none of the papers submitted by the Chinese teams that were ranked
the highest by the Chinese triage judges was made Outstanding Winners. On the
surface, the Chinese judges seemed to focus more on technical details while the US
judges would focus more on creativity and innovation. These differences, however,
may be caused by deeper reasons rooted in the different education systems and
philosophical viewpoints, which would deserve further investigations.
The triage judges have little time to read a lot of papers; they only get to spend
about ten minutes on each paper. This restricts them to base the score largely on
the summary, which makes the summary the most important part of the paper.
The papers that go on to further judging are judged by three different sets of
judges, one for Problem A, one for Problem B, and one for Problem C. The final
judges for Problem A and Problem B often meet in the same location, while the final
Chapter 1 Introduction 5
judges for Problem C often meet in a different location (typically in the COMAP
headquarters in Bedford, Massachusetts). There will then be several rounds of
judging; for each round, a paper either passes and goes on to the next round or
is pulled out. The ranking of a paper depends on how many rounds it passes. The
judges will spend more time reading these papers, but still they will need to read
them quickly. The papers will have to have their papers well organized to impress
the judges; the parts of the paper that the judges look for, such as the assumptions
and their justifications, a clearly stated model, sensitivity analysis, and explicit
conclusions, should be clearly delineated.
The Outstanding Winners are chosen in the final round of judging. In this
round, the remaining papers are read by all the judges. All of the judges, as well
as the director and associate director, need to agree that a paper deserved to be
an Outstanding Winner before it gets that honor.
In this book, we will discuss what to do when working on an MCM/ICM
problem and what the judges look for in a good paper. We then look at some of the
Outstanding Winner papers from the 2013 MCM and all the Outstanding Winner
papers from the 2013 ICM.
For convenience, we will often use “you” to mean “you” or “your team”.
“We” will typically mean “you and the authors of this book”, although when we
are working our way through some of the Outstanding Winner papers, “we” will
also include the author of those papers.
Chapter 2 General Methodology
The chapter discusses what you should do before the contest starts and how to ap-
proach the problem that you choose to solve when the contest begins. In particular,
you should keep in mind the following key points:
• You should be on a team with people you can work well with.
• It is better to have a careful analysis of a simple model than a cursory discus-
sion of a complicated model.
2.1 Before the Contest
Before the contest begins, you will not know the exact problem that you will be
working on and may not even know the type of problem that you will be solving.
Even so, there is a lot of important work that can take place before the beginning
of the contest.
The first part of joining the contest is becoming part of a team. The emphasis
here needs to be on team. The founder of the MCM, Ben Fusaro, once said “It is not
the three best that make the team, it is the best three”. The members of your team
need, first and foremost, to be able to work well together, and then to complement
each other’s abilities. The contest involves mathematics, computing, and writing,
and your team should have an expert in each. Each team member should have a
specific job to work on as soon as you decide on a problem; one member should
begin a literature search, one member should begin computer programming, and
one member should begin writing the paper (yes, the writing should begin right
away).
Teamwork needs to be emphasized throughout the contest. Nobody should
be idle during the contest, and all members of your team should be working on a
coordinated strategy. Everybody must agree to what should be done; any disagree-
ments should be resolved by discussing it until everybody agrees, never by voting.
In a group with three people, it is very easy for one member to feel left out or
2.2 Reading the Problem 7
ignored; if any decisions are left to a vote and one person is outvoted by the other
two, then that person might feel that their efforts are not appreciated.
Well before the contest begins, your team should meet regularly to practice.
It would be a good idea to look over problems from previous contests and discuss
how you might approach them. By the time the contest begins, your team should
have practiced the skills that you will need for the contest, including the following:
Mathematics Looking over problems from previous years will give you some idea
of techniques that can be useful. If you might choose the continuous problem,
you should be familiar with differential equations, particularly the wave equa-
tion and the heat equation. As well as analytic techniques for solving math
problems, you will need to be familiar with approximation techniques. To find
optimal solutions for many of the MCM/ICM problems, stochastic methods,
such as simulated annealing for graph and other discrete problems and genetic
algorithms would be useful to know.
Computers You will almost certainly need to use a computer to help you with
your problem and you don’t want to have to learn how to use a new program
when you are working on the contest. The most commonly used programs are
GNU Octave and MATLAB; they are similar programs and either is a good
choice to deal with numeric issues. They each also have add-ons to help with
specific types of problems. To help with non-numeric issues, programs such as
Maxima or Mathematica can be useful. It would also be useful to have a team
member who can quickly write a program in a general purpose programming
language. A good way to familiarize yourself with useful programs and how
to use them would be to look at Outstanding Winner papers from previous
years in the UMAP Journal and recreate their computer results.
Writing During the contest you cannot have your writing critiqued, so you should
write some practice papers before the contest and have them looked at by
experts. You should practice writing papers containing math and make sure
that the paper is largely exposition and not simply pages of equations and
symbols.
2.2 Reading the Problem
You and your teammates will be working on a problem for four days. It is important
that you work on a problem which catches your interest and excites you. Whether
you expect to find continuous or discrete math more interesting, be sure to read
8 Chapter 2 General Methodology
all problems carefully. Then discuss them with your teammates. If everyone agrees
what problem to work on, great. If not, then you need to discuss it. Recall that
you need to come to an agreement, not a vote. Everybody needs to be excited by
the problem.
It should not take long to decide on a problem, and once you have done that
you need to start deciding what needs to be done for the problem. Some contest
problems require little interpretation, but for the most part you should realize
that when you read the problem, you are also interpreting the problem and to
some extent creating the problem. For example, one of the 2010 problems involved
was to “Explain the ‘sweet spot’ on a baseball bat.” The problem statement
continues: “Every hitter knows that there is a spot on the fat part of a baseball
bat where maximum power is transferred to the ball when hit.” Successful teams
nonetheless offered other interpretations of “sweet spot”, many of which can be
found in the literature, and teams that did not offer more than one did not go far
in the competition; teams which did not actively search for other meanings often
didn’t make it past triage. Some other interpretations that were offered include the
spot on the bat which will cause the ball to travel farthest, the center of percussion,
and the spot on the bat which will transfer the least vibration to the hitter’s hands.
(These last interpretations may be worth mentioning, but since they do not affect
the outcome of a baseball game probably would not be the best interpretation
to investigate.) While different interpretations may end up being the same thing
under the surface, they can affect how you model the problem.
Most problems will have key English words that must be given exact mathe-
matical meaning, particularly words that describe “optimization”. A problem from
2009 required the students to determine “how best to control traffic flow” around
a traffic circle. It was up to the investigators to decide what “best” meant in that
case. (The problem explicitly stated “State clearly the objective(s) you use”, but
that should be implicit in any problem that you do.) “Best” in this case could refer
to getting the cars through the traffic circle as quickly as possible; at least one team
considered “best” to mean getting the cars into the circle as quickly as possible.
However you interpret the problem, you need to make it clear (to yourselves and to
your readers) what it means to you. For another example, one of the 2004 problems
was to investigate the likelihood of misidentifying a person using fingerprints. The
investigators had to decide what was meant by “fingerprint” (is it part of a finger
or residue left behind when somebody touches a smooth object, for example) and
then decide how to give it a precise mathematical interpretation.
2.3 Literature Search 9
As well as determining what you are going to model, you also have to determine
what you need to do with the model. Poorer teams will create a model and not do
anything with it. Be sure to understand (and interpret, if necessary) what questions
are being asked. Usually the goal is pretty clear, but be sure to carefully read the
problem and understand what questions you are expected to answer. It is easy
to overlook even clearly stated questions; for example, after describing the issue,
the 2005 Tollbooth problem states: “Make a model to help you determine the
optimal number of tollbooths to deploy in a barrier-toll plaza. Explicitly consider
the scenario where there is exactly one tollbooth per incoming travel lane.” While
the tollbooths can be modeled using queuing theory, working in the details of a
tollbooth plaza is not easy. It may have been because the first part of the problem
was so difficult, but many teams overlooked the second part; that is, they did not
consider the situation where there is one tollbooth per lane. Even if this had not
been explicitly brought up in the problem statement, it would have been a good
idea to consider it to see if the modeling techniques provide reasonable solutions
in this relatively simple case.
After reading and discussing the problem that you are going to work on, you
should have a pretty good idea of
• what you will be modeling;

• what you want to use the model to find out;
• how you might come up with solutions;
• how you can compare solutions.
2.3 Literature Search
Once you have chosen a problem, there is still work to do before you create your
model. You will need to get as much information on the problem as you can.
You will need to look for problems that are similar to yours for inspiration. The
Internet is a good place to start. The UMAP Journals with problems and papers
from previous years might be helpful. You should expect to spend some time in the
library. And you will need to keep track of the references that you use. A paper
which has not done a thorough literature search will probably not make it past
triage, and a paper that does not cite any references will be disqualified.
You will need some information on the problem. For the “sweet spot” problem
mentioned above, the investigators should have gotten information on the size and
compressibility of a baseball, the shape and material of a baseball bat, etc. For the
“fingerprint” problem, the investigators would need to find out the characteristics
of fingerprints and how they are distinguished. (The FBI has a set of criteria that
it uses; anybody working on this problem should know about it.)
You should also search for previous work on the problem. While you may not
find the exact problem, you can and are expected to find related problems. There
are several models of varying sophistication in the literature for working with the
“sweet spot” of a baseball bat; while none of these models should be used “as is”,
they provide a starting point for the investigators to build their own model. While
there does not seem to be any previous work on the tollbooth problem mentioned
above, a literature search should be done to find standard techniques for queuing
theory problems. At this point, you are not looking for a model that you can use
(it is extremely unlikely that you would find one), but a model that you can adapt.
You should not expect to either find exactly what you need in a literature search
nor create a model from scratch; you should expect to take previous work and add
your creativity to it. A big part of your literature search should be to find a starting
point for your model.
When you write up your paper, you should have something to say about your
search. In “The Sweet Spot: A Wave Model of Baseball Bats”[3] , the introduction
contains a discussion of previous attempts in the literature to investigate the sweet
spots of baseball bats (and similar things in other sports). The authors were familiar
enough with previous work on the subject to point out the shortcomings and to
build upon and improve previous models for their work. At the other end, in “A
Half-Blood Half-Pipe, A Perfect Performance”, the authors admit “We find no
analysis of the ‘best’ shape for a half-pipe”, although they do go on and mention
that they did find some information on the typical shape for a half-pipe, which
they used in their paper.
Many modeling problems are instances of a larger class of problems. For exam-
ple, many (if not most) of the A problems are applications of differential equations,
and the Tollbooth problem is easily thought of as a queuing theory problem. You
will need to determine what type of problem you have and learn the standard
techniques for that area. For this, you should look in textbooks for tried and true
methods rather than cutting edge techniques.
2.4 Creating Models 11
2.4 Creating Models
Your “model” might well be a series of models. You should begin with a simple
model, possibly one that you can work on with pencil and paper alone. This could
be something taken from your literature search. This not only provides a starting
point for creating a more sophisticated model, it also provides you with a baseline
to compare your later results to. For example, in “The Sweet Spot: A Wave Model
of Baseball Bats”[3], the authors begin with a model from an older paper which
assumed that a baseball bat was rigid. They point out that this is not a realistic
assumption, but use this as a starting point to create a more sophisticated model
(also drawing on other models they found in the literature). They demonstrated
how the non-rigidity of the bat plays a role in the sweet spot and noted that in a
non-rigid bat the sweet spot is farther from the handle than it would be if the bat
were completely rigid.
Even if you do not base your model on something you found, starting with
a simple model and gradually making it more sophisticated is a good idea. For
example, in “One Ring to Rule Them All: The Optimization of Traffic Circles”[4],
the authors investigated how to control traffic near a traffic circle. They began
with a model where they assumed the traffic rates into and out of the traffic circle
queues are constant. They gradually made it more sophisticated by taking into
account the number of cars that can fit into the circle, how that affects the rate
into the circle, and how it affects the rate out of the circle. They finally created a
computer simulation of a traffic circle to take into account even more complicated
issues. They showed how different assumptions lead to different recommendations
of how to control the traffic, and their ultimate recommendation was broken down
into how much an effect the additional assumptions they made have.
Whatever assumptions you make for coming up with your solutions should
be thoroughly discussed. The better papers include a careful discussion of the
assumptions and simplifications used to create their model(s); both the rationale
and the effect they have on the model should be brought up in your paper.
Keep in mind that more important than the sophistication of your model is
what you do with it; some outstanding papers are well-written analyses of simple
models. Your paper should not include a lot of tedious computation; that should be
left to the computer. Your paper should focus on the description of your model and
analysis of the results. In your writing you need to make it clear that you completely
understand the justifications and mathematics of what you are doing. You need to
impress the judges not just with your model but with your understanding of how it
works (and how it does not work). A poor paper might devise a decent model but
not discuss the assumptions used to create it. Given a choice between a model and
a less comprehensive model that you can analyze better, choose the latter. K.S.
Cline, in “Secrets of the Mathematical Contest in Modeling”[5] , wrote
The best papers are often ones that do something fairly simple, fairly
basic, but they do things and explain things so carefully that it is obvious
that they really understand the methods they’re using. That is the mark
of a great paper: Use the right mathematical tool for the problem, and
make it clear that you understand this tool inside and out, its strengths,
its weaknesses, its limitations, why it is useful, and where it will fail.
2.5 Finding Solutions
The goal of the modeling contest is not to create a model or even just solve the
given problem, but to use a model that you create to solve an interesting problem.
Some teams devised a model and stopped. Many teams devised a models for their
problems and then discussed solutions which were not connected to their models.
These teams did not go far in the judging process. You are expected to use the
model (or models) you have created and use it (or them) to gain insights into the
problem. And you need to finish it up with solutions—not one solution, but more
than one.
In the traffic circle problem mentioned above, the authors of “One Ring to
Rule Them All” created a sequence of increasingly more sophisticated models, and
they used them to come up with appropriate methods to control the traffic in the
circle. In their simpler models, where the capacity of the traffic circle is ignored,
their model indicated that getting the cars into the circle as fast as possible is the
best choice, and so yield signs would be the best way to control traffic. In the more
complicated models, when congestion and the capacity of the traffic circle become
an issue, then more control, such as traffic lights, would be better. The important
thing isn’t just their recommendations, but that their recommendations are based
on their models.
Generally, your model will have a number of parameters, and you should see
what happens when you vary them. (This will also be useful for stability and
sensitivity analysis.) In “The Sweet Spot” paper mentioned above, the authors
2.6 Stability and Sensitivity 13
created a model that depended on the density of the bat, the stiffness of the bat,
and the shape of the bat. While there are standard values (for a baseball bat) of
these parameters, the authors investigate what happens when they are changed.
They concluded that the density and shape of a bat are special.
Keep in mind when you are working on your model that you need to come up
with a solution to the given problem based on this model, and if at all possible you
should come up with more than one.
2.6 Stability and Sensitivity
An important aspect of the contest that is too often overlooked by the students is
stability. It will not be overlooked by the judges, however, and many years a team
can distinguish itself from the competition by investigating the stability of their
results.
Checking for stability can be as simple as varying the parameters within your
model and checking the results. If a small change in a parameter yields a significant
change in conclusion, this needs to be noticed and addressed. In “A Half-Blood Half-
Pipe, A Perfect Performance”[6], the authors determined the height a snowboarder
can get from a half-pipe based on parameters such as the width and depth of the
half-pipe. They showed that if the parameters are changed by as much as 10% then
the height changes by less than 10%, indicating that their solution is robust. (This
is not the same as accurate, of course, and they also compared their solution to
the actual heights achieved by world class snowboarders to indicate they are also
accurate.) This analysis also shows that the height is not very sensitive to the
size of the flat portion of the half-pipe. Whatever parameters your model depends
on, you should be sure to investigate what happens when these parameters vary.
Your paper should mention how stable your solution is under changes in these
parameters and which parameters your model is sensitive to.
If you test your model on randomly generated data, you should vary the data
as well as the method of generating it. A 2011 problem asked the investigators to
determine how many “repeaters” (stations which pick up weak radio signals and
amplify them) would be necessary to ensure all the radio users in a given area
can be covered. Once the investigators decide how to place the repeaters, they
should test it on randomly placed users. They should at a minimum test that their
placement works on several sets of uniformly distributed users; even better would
be if they also tested their placement scheme for non-uniformly distributed users.
2.7 Writing up the Solution
While the MCM/ICM is a contest to solve a given problem with a model you create,
your solutions need to be written up for judging. In many ways, the writing is just
as important as the modeling; a well-written paper with a lesser model will often
be received better than a poorly written paper with a better model. Some member
of your team should start writing the paper right away. For an elaboration how to
write for MCM/ICM, the reader is referred to [7].
The judges will have little time to read a lot of papers, so your paper will need
to be easily readable and easily skimmable. There should be a table of contents,
and the sections and section names should provide an outline of the paper. The
important points should be highlighted with such things as boldface and displayed
equations. There should be sections devoted to the following:
• A restatement of the problem, as you interpret it.

• The assumptions you make, clearly stated.
• Justification of the model being used.
• The model itself.
• A definite conclusion.
• Strengths and weaknesses.
• Suggestions for further exploration.
• A complete list of all sources used.
When you discuss the assumptions that you use, be sure to do more than state
them, you need to justify your use of them. If necessary, your justification can be
that using an assumption makes the model simpler. Do not bring up any assump-
tions that are not used. Somewhere in the paper you should mention how your
assumptions affect your results. You should have a section devoted to the strengths
and weaknesses of your work; you should be paying attention to this throughout
your work. You will need to make sure that your sources are clear; uncredited work
is grounds for disqualification.
The last thing that you work on should be the executive summary, and this
may well be the most important part of your paper. It will determine whether or
not your paper makes it past triage. This summary should be succinct; it should
only be half to two-thirds of a page, but needs to clearly explain the highlights of
your paper, discuss your model, and answer the questions.
Exercises 15
Exercises
Find a copy of the UMAP Journal or CD that contains Outstanding Winner pa-
pers from previous contests. For each problem from that year, read one of the
winning papers carefully, paying close attention to the structure of the paper, the
assumptions that are made and how they are justified, and the summary. Look for
weaknesses in the papers, and compare them to the weaknesses that the authors
mention.
Chapter 3 The Ultimate Brownie
Pan
Each year, Problem A of the MCM can be attacked using continuous mathemat-
ics, which typically means differential equations. For 2013, the problem involved
cooking brownies.
3.1 Problem Description
The exact statement of Problem A is given below:
When baking in a rectangular pan heat is concentrated in the 4 corners and the
product gets overcooked at the corners (and to a lesser extent at the edges).
In a round pan the heat is distributed evenly over the entire outer edge and the
product is not overcooked at the edges. However, since most ovens are rectangular
in shape using round pans is not efficient with respect to using the space in an
oven. Develop a model to show the distribution of heat across the outer edge of
a pan for pans of different shapes–rectangular to circular and other shapes in
between.
Assume
1. A width to length ratio of W/L for the oven which is rectangular in shape.
2. Each pan must have an area of A.
3. Initially two racks in the oven, evenly spaced.
Develop a model that can be used to select the best type of pan (shape) under
the following conditions:
1. Maximize number of pans that can fit in the oven (N ).
2. Maximize even distribution of heat (H) for the pan.
3. Optimize a combination of conditions (1) and (2) where weights p and
(1 − p) are assigned to illustrate how the results vary with different values of
W/L and p.
3.2 How to Approach the Problem 17
In addition to your MCM formatted solution, prepare a one to two page adver-
tising sheet for the new Brownie Gourmet Magazine highlighting your design and
results.
3.2 How to Approach the Problem
The first thing you are asked to do is to create a model for the heat distribution
on the edges of a pan, which presumably should be used for the latter parts of
the problem. At the end of the problem you are asked to create an advertisement
for your pans. The advertisement should highlight the type of pan you came up
with, although many teams simply advertised a generic brownie pan. A possible
danger when creating the advertisement is addressing it to math majors; the given
audience should be readers of Brownie Gourmet Magazine, whose readers would
presumably not be as mathematically inclined as you are.
3.2.1 Deciding What Needs to Be Done

Although the problem statement looks pretty clear, there are still a few items that
need to be clarified before you can begin the problem.
What is a brownie pan? A standard brownie pan is a familiar object to
most western people, where ovens are standard cooking appliances. It consists of a
flat metal base, typically a rectangle or a circle, with relatively short sides (See Fig.
3.1). Mathematically speaking, it can be viewed as a short hollow metal cylinder
with a bottom but no top. As a quick search on the Internet shows there are other
types of brownie pans. There are soft-sided pans made from silicone. Some people
like brownie edges, and various types of pans have been created to accommodate
them. There are brownie pans with “walls” in the middle to add extra edges, and
pans which are subdivided into smaller pans to ensure that all sides of every brownie
is at an edge of a pan. While most teams worked with a standard brownie pan,
some teams chose to work with one of the other kinds.
Figure 3.1 Standard brownie pans
The standard brownie pan shapes are rectangles and circles, but you are asked
18 Chapter 3 The Ultimate Brownie Pan
to consider different shapes. The problem statement suggests “rectangular to circu-

lar and other shapes in between”. You need to figure out what shapes “in between”
are. One reasonable interpretation would be to consider rectangles with rounded
corners. Some teams decided to consider squares, circles and then regular polygons
as the in-between shapes.
What is N , the number of pans which can fit in the oven?
The problem seems to ask how to design a pan so that as many as possible will fit
into the oven, and then determine how many will fit. By itself, this is a straight-
forward problem. Some teams, with an eye towards fitting pans which provide
good heat distribution, considered fitting the maximum number of pans of various
shapes.
How do you measure H, the distribution of heat for the pan?
Probably the simplest way to measure how even the heat is on the outer edge of
the pan is to subtract the minimum temperature from the maximum. Some teams
wanted to create a measure that will be larger when the temperature is more even
so they could later maximize
pN + (1 − p)H.
Some measurements that were used are listed below:

Tf − Tmax
• − 1,
Tf − Tmin
where Tmax is the maximum temperature on the outer edge of the pan, Tmin
is the minimum temperature, and Tf is the oven temperature (presumably
larger than Tmax and Tmin )[8] .
• The variance of the norm of the heat gradient over the outer edge of the pan[9] .
• C/P for pans which are rectangles with rounded corners, where P is the
perimeter of the circumscribed rectangle and C is the circumference of the
circle formed by the rounded corners[10]. The rationale for this measure is pre-
sumably that the uniformity of the heat corresponds to the radius of curvature
of the corners of the pan (which is proportional to C) and the authors showed
that this measure gives 1 for circular pans and 0 for rectangular pans, although
they should have shown a more direct connection between this measure and
the heat distribution.
Whatever means a team used to measure the heat distribution for the pan, it was
important that the team clearly and explicitly state their choice and then stick to
it.
3.2.2 Searching the Literature

While you probably have a pretty good idea of what a brownie pan is, beginning
the problem with a search for brownie pans can bring up other possibilities. As
mentioned above, one team chose a non-standard brownie pan to base their model
on. The properties of pans (such as the material that they are made out of, the
thermal conductivity of the material, etc.) will be necessary to know, but you may
wish to know exactly what properties you will need to know before you look them
up.
A big part of your literature search should be trying to find a model to build
on. While the chances of finding a previous competition problem that can be used
as a starting point are small, the advantages of finding one makes it worth a quick
search. (If there is such a problem, each year some outstanding entries are published
in the UMAP Journal.) For the brownie pan problem, the only problem that looks
like it might have potential is the Concrete Slab Floors problem from 1994, which
involves the temperatures of concrete slabs. This problem probably is not close
enough to use as a starting point for the brownie pan problem, however.
Depending on the library that you have access to, there may be some relevant
journals worth looking at. Examples of such journals are Journal of Food Science
and International Journal of Heat and Mass Transfer. A more efficient way of
searching might be to use an Internet search. Web pages typically are not as well-
written or edited (for style and accuracy) as published works, but there are some
very good sources on the Internet and, more importantly, a search will likely yield
references to more traditional sources. While a list of references should not consist
completely, or even mostly, of web references, the Internet might be a good place
to start a search.
A web search for “brownie pans” or something more specific (such as “heat
brownie pans”) can be used to find some key properties of the pans that will be
necessary for a good model, but is not likely to lead to a starting model. A search for
something more general, such as “physics baking” is more likely to yield a starting
point. The problem is about temperature, so even models involving cooling could
be a starting point, and some teams did begin with cooling. Finally, a web search
will likely lead to a recipe for brownies, which you could use to bake delicious
post-contest treats.
There are many papers discussing the mathematics of baking and cooling
foods, and either by a web search or browsing through journals, you should come up
with some. Some examples found by the teams are “Modeling of Simultaneous Heat
and Water Transport in the Baking Process”[11] and “Modeling Three Conventional
Cooling Processes of Cooked Meat by Finite Element Method.”[12] . The former of
these papers emphasized the role of evaporation in the baking process, something
that many teams overlooked—another reason why building on an existing model
is a good idea.
A search for “baking mathematical model” brings up many more papers on
baking models, some of which also discuss the effects of crust formation in the bak-
ing of bread (to which there is an obvious analog in brownie baking). Searching for
“food temperature mathematical model” will bring up a broader choice, although
“temperature mathematical model” will bring up too broad a range of articles.
Unfortunately, many references found this way will be to articles in journals your
library may not have, and given the short time for the contest there probably would
not be enough time to have your library order a copy. Still, given the abundance of
references that can be found this way, it is very possible that some will be locally
available.
If you do not find an existing model that matches what you need and you need
to modify a model (as will almost certainly be the case), then you should make
sure that you know all the processes and details involved in baking brownies to
make sure they are addressed. Some of the things that affect baking the brownies
include the following:
• Convection heat from the air

• Radiated heat from the oven coils
• Radiated heat from the other pans
• Evaporation of the water
• Chemical changes in the brownies.
Ultimately, you should rely on textbooks and the like for most of your model
building. Even if you are basing your model on something found in the literature,
you will need to add to it and justify the components in your paper. It should not
be necessary to use any cutting edge techniques, but rather careful applications of
standard techniques that can be found in most textbooks.
3.2.3 Creating a Model

There are many processes you can take into account when creating a model of
baking brownies. For example,
• Convection from the air in the oven

• Heat radiated from the oven coils
• Heat radiated from other pans
• Heat loss from evaporation of water in the brownie batter
• Other chemical changes that can affect the heat of the brownies
• Volume and shape changes of the brownies that occur during cooking
• The different heat transfer that occurs at the top of the brownies (where the
brownies meet the air) and the sides of the brownies (where the brownies meet
the pan)
• Any differences between heating in the different oven levels and different places
in the oven.
That does not mean that you should take them all into account; in fact, you almost
certainly should not. An overly complicated model will be hard to analyze.
It is more important to have a well-justified model and a well-discussed model
than a comprehensive model. Any simplifications you make should be justified
as well as possible, or at least mentioned. If nothing else, a simplification can be
justified simply by saying that it is needed to create a model, and indeed you should
begin by creating simplified models and adding conditions as you can. Many teams
began by simply considering brownie batter heating in the oven, with the only heat
coming from the air which was assumed to have constant temperature, not taking
into account heat radiated from other pans and the effect that two racks can have.
In fact, some outstanding teams did not go beyond that; it bears repeating that
having a clearly explained and well analyzed simple model is more important than
a complicated model. This does not mean that effects that are not part of your
model should be ignored; if you do not take into account the chemical changes of
the brownie batter you should mention why. You not only need to know when you
are simplifying the problem, you also need to let the readers know that you know.
Discussing the mathematics of baking brownies will necessarily involve using
the heat equation to model the transfer of heat within the brownie batter. This
will require getting information on the batter and pan material (heat conductivity
of stainless steel, for example) as well as assumptions on the cooking conditions
(oven temperature, etc.). Other effects, such as the different boundary conditions
where the brownies meet the air and the pan, the heat transfer from the air to the
pan, heat radiated by the oven heating coils and other pans, moisture loss from the
brownies, etc., can be added to the model to make it more realistic, but only after
you have carefully taken care of the simpler situations.
3.2.4 Finding Solutions

Once the model is created, there are three parts of the problem.
The first part, Maximize number of pans that can fit in the oven (N ) does not
involve the model.
The second part, Maximize even distribution of heat (H) for the pan, has the
answer given in the problem itself; assuming that the conditions in the oven are
isotropic (which may not be completely true), a circular pan will provide the most
uniform heat distribution along the sides. You should show that your model gives
the correct answer in this case.
The most difficult component of the problem is the third part: Optimize a
combination of conditions (1) and (2) where weights p and (1 − p) are assigned
to illustrate how the results vary with different values of W/L and p. Finding
a solution for this part will require using the model (and a definition of H) to
determine the heat distribution for various pan shapes, and then finding the best
balance between N and H. Some teams did this by maximizing pN + (1 − p)H,
while other teams took a less direct approach to the joint optimization. For teams
which maximized pN +(1−p)H, they must choose reasonable measure for H. If the
chosen value of H is too small, then H becomes insignificant compared to N ; if it
is too large, then N becomes insignificant. Some teams normalized their measures
for the temperature gradient and the number of pans that could fit in the oven, so
that neither value carried more weight.
As with finding a solution to most MCM/ICM problems, optimizing the com-
bination is not straightforward. Using stochastic methods is often the best way to
come up with an approximately optimal solution, and for this problem many teams
used genetic algorithms.
3.2.5 Stability Testing

The proper pan shape may likely depend on the oven size and shape, but should
not depend very much on the oven temperature (given that it is reasonable), the
type of brownie batter, and other considerations. If you determine the shape using
a genetic algorithm, you should rerun it several times to make sure that you get
3.3 Judging 23
(nearly) the same result every time. You should also see what happens to the pan
shape when the oven temperature changes slightly, when the heat conductivity of
the batter changes slightly, etc. The shape of a good brownie pan should not heavily
depend on these considerations.
3.3 Judging
There are five parts to this problem:
• Developing a model for the heat distribution on the outside of the pan.
• Maximizing the number of pans that can fit in the oven.
• Maximize the even distribution of heat for the pan.
• Optimize a combination of the number of pans and the even distribution of
heat.
• Write an advertisement for the pan.
The judges consider each part to be crucial.
3.3.1 Judging Criteria

The model needs to be a model of a reasonable pan. Without a model based on
realistic materials, the paper will not score well[13] .
Next, maximizing the number of pans which can fit in the oven is trivial. It is
still a necessary part of the paper, but if this is the only issue addressed the paper
will score poorly.
Making sure the heat is evenly distributed along the sides of the pan is more
important. More than one shape needs to be analyzed, however, and the shapes
compared. Working with only one shape will result in a low score.
A decent paper needs to address simultaneously maximizing the number of
pans and ensuring the heat is evenly distributed. As a minimum, this should be
done for at least one value of p between 0 and 1. A good paper will consider a
range of values for p, find values that give good results, and explain why some
values might only give bad solutions.
The last part of the problem is writing an advertisement for the pan. The
advertisement should convince a consumer that your pan is preferable to others by
stressing the advantage of the shape, and it needs to do that without being too
technical for the average magazine reader.
3.3.2 Outstanding Winners

The judges selected four Chinese teams and two US teams to receive the highest
honor as Outstanding Winners, listed below in alphabetical order of school names.
• Bethel University, USA. (MAA Award Recipient).

Michael Tetzlaff, Jake Smith and Anthony J. Burand, Jr.
Title: Cooking Up the Optimal Baking Strategy.[14]
Advisor: Nathan M. Gossett.
• Fudan University, China.
Kunrui Wang, Jing Xu, Wei Zheng.
Title: Optimize the Shape for a Brownie Pan.[15]
Advisor: Ling Yang.
• Peking University, China.
Meng Wu, Chong Jin and Bowen Liu.
Title: Applied Mathematicians Bake the Best Brownies: The Ultimate Pan
Design.[16]
Advisor: Xufeng Liu.
• Shandong University, China. (INFORMS Award Recipient).
Yankan Song, Ke Xu and Fan Yi.
Title: The Design of the Ultimate Brownie Pan.[17]
Advisor: Hengxu Zhang.
• Shanghai Jiaotong University, China.
Libin Wen, Jingyuan Wu and Cong Wang.
Title: Solving the Brownie Pan Problem.[18]
Advisor: Yuehui Zhang.
• University of Colorado, Boulder, USA. (SIAM Award Recipient).
Christopher V. Aicher, Tracy Babb and Fiona Pigott.
Title: The Best Rounded Rectangle for Ultimate Brownies.[19]
Advisor: Anne M. Dougherty.
While it was not chosen as an Outstanding Winner, Tongji University in China

(Control number 17617)[20] won the Ben Fusaro Award.
3.4 Model 1: The Best Rounded Rectangle for Ultimate Brownies 25
3.4 Model 1: The Best Rounded Rectangle for

Ultimate Brownies
The Outstanding-Winner team at the University of Colorado, Boulder[19] worked

on creating brownie pans that are rectangles with rounded corners (“rounded rect-
angles”), and carefully analyzed a simplified model of baking brownies. We will
work through their solution. The paper had an early section introducing the no-
tation and explicitly stating the assumptions used throughout the paper; the best
papers should do this. Since here we are trying to develop the model rather than ex-
plaining a developed model, we will not assume any foresight and we will introduce
the assumptions as we use them.
3.4.1 Packing the Pans in Standard Ovens

The first part of the problem is to maximize the number of pans of area A can
fit into an oven with dimensions W × L. Since the suggested shape of the pans is
“between” a rectangle and a circle, it makes sense to begin by looking at packing
rectangles and circles.
Packing rectangles
Given many identical rectangles and a larger rectangle, packing the smaller rect-
angles into the larger is a classic problem. The Pallet Loading Problem restricts
the positions of the identical rotations to translations and 90◦ rotations, which are
reasonable restrictions for placing pans in an oven. However, for many rectangle
shapes, packing rectangles into a not-too-much larger rectangle is not efficient. We
should attack this problem from a different angle; namely, we would like to de-
termine the shape of the rectangle that will allow us to pack as many rectangular
pans into the oven as possible.
Taken by itself, this first part of the problem has the simple solution of N ∗ =
W L/A pans. If the pans have width W and length A/W , then they will have
area A and N ∗ of them can fit nicely into the oven (with left over area less than
A). (See Fig. 3.2.) While at first glance we might think this would result in very
skinny pans, if we use a width of 47 cm for the oven (which is the width of a typical
Kenmore oven) and an area of 760 square centimeters for the pan (which is the
area of a common large brownie pan), then the pans will have dimensions of about
47 cm by 16.2 cm. Since small brownie pans are often 15.2 cm wide, our current
pan is not necessarily too narrow.
Figure 3.2 N ∗ pans fitting into the oven
While the above pans do completely answer the first part of the problem, we
may wish to think more about this part of the problem while considering how the
shape might affect the heating. If the oven is particularly large or given area of the
pans is fairly small, then the pan from our first solution may not be wide enough
to cook the brownies well. We can modify the above solution by making the pans
twice as long and half as wide (see Fig. 3.3). We can also make them k times as
long and 1/k times as wide, for small values of k. This gives us more flexibility to
create pans that might give better baking results.
Figure 3.3 N ∗ pans fitting into the oven
Packing circles

There is only one circle with area A (namely the circle with radius r = A/π),
so we do not have to consider different shapes. Finding the densest packing of
the circles in a given rectangle is a complicated problem, but we can examine two
reasonable and reasonably simple packings; namely, we can take the hexagonal and
square tiling the plane (see Fig. 3.4) and restrict them to the rectangle determined
by the oven. Not surprisingly, these packings are not as effective as rectangle
packings.
Figure 3.4 The hexagonal and square fitting of circular pans
Packing rounded rectangles
For our intermediate shapes, we will consider rounded rectangles, which are rectan-
gles with round corners. Such a rectangle can be specified by w, l, and r, where w
and l are the semi-width and semi-length of the inner rectangle and r is the radius
of the corners (see Fig. 3.5). In terms of these variables, the area will be
A = πr2 + 4(lw + lr + wr).
Notice that such a rounded rectangle will be inscribed in a rectangle with width
2(w + r) and length 2(l + r); the difference between the areas will be WBR =
(4 − π)r2 .
Figure 3.5 A rounded rectangle
Our optimal solution with rectangles (see Fig. 3.2) will likely have some oven
space left over, and in this case it is possible to round the corners so that the width
remains the same (i.e., W = 2(w + r)), the length (2(l + r)) increases, but the same
number (N ∗ ) of pans will still fit in the oven. To find the value of r, we can equate
the space that is left over from our optimal rectangular solution, W L − N ∗ A, with
the total extra space our rounded rectangular pans will use, N ∗ WBR ; this will give
us
W L − AN ∗
r= .
N ∗ (4 − π)
L
If r > , then the rounded rectangular pans will not fit into the oven (the overall
2N ∗
length of the pans will be too large), and so our optimal value of r will actually be

W L − AN ∗ L
r = min , .
N ∗ (4 − π) 2N ∗
Since 2(w + r) = W and 2(l + r)N ∗ = L, we can find the corresponding values of
l and w.
3.4.2 Maximizing the Even Distribution of Heat

We need to develop a model that will tell us the distribution of heat.
Creating a model
The problem specifically asks about the heat on the outer edge of the pan, but
presumably the temperature of the outer edge of the pan will match the outer edge
of the brownies. So we will model the heat throughout the brownie. In particular,
we will maximize the even distribution of the heat throughout the brownie rather
than the edge. While a fairly even distribution of heat throughout the brownie
still allows temperature spikes along the edge of the brownie, an even distribution
throughout the brownie should still result in a nicely cooked treat.
Let T represent the temperature of the brownie batter and T (x, y, z, t) the
temperature at position (x, y, z) and time t. Then T will satisfy
T
= ∇ · (α∇T ),
t
where α is the thermal diffusivity of the batter. If α is constant, then this takes on
the standard form of the heat equation:
T
= α∇2 T.
t
In our case, the batter will change as it is cooked and so α will not be constant. The
diffusivity of the brownie batter, we can assume that it is between the diffusivity
of water (1.5 × 10−7 m2 /s) and that of cooked brownies (1 × 10−7 m2 /s). To begin
with, we will assume that α is constant.

While the heat equation will describe the movement of heat within the brownie
batter, we need boundary conditions to describe how the brownie batter is heated.
If the pan is made out of a metal such as aluminum, then it will quickly heat to
the air temperature. Thus, we will assume that the boundary of the brownies is
constantly at air temperature, denoted by Tb . We will also assume that the interior
of the brownie batter begins at a uniform temperature of Ti .
Let Ω represent the set of points making up the brownie batter. Then our
model is
T
= ∇ · (α∇T ), (3.1)
t
with boundary conditions
T (x, y, z, t) = Tb for (x, y, z) ∈ Ω

T (x, y, z, 0) = Ti for (x, y, z) ∈ Ω.
An analytic solution
In general, it is not feasible to find an exact solution to our differential equations

and so we will need to use numerical techniques to approximate a solution. We
can, however, find explicit solutions for certain special cases, and it would be
useful to compare an exact solution to our numeric approximation. We will solve
the differential equation explicitly for the case of a rectangular pan with width
W , length L, and height H (in this section, W and L will represent the width
and length of our rectangular pan rather than the oven) and a constant thermal
diffusivity for the batter. Our region Ω is given by
0 x W,
0 y L,
0 z H.
Letting u(x, y, z, t) = T (x, y, z, t)−Tb to give us homogeneous boundary conditions,

our problem becomes
u
= α∇2 u
t
u(x, y, z, t) = 0 for (x, y, z) ∈ Ω
u(x, y, z, 0) = Ti − Tb for (x, y, z) ∈ Ω.
We can look up a solution to this or solve it with separation of variables; from the
first two equations we get
∞
∞
∞ mπx nπy
lπz 2
u(x, y, z, t) = Am,n,l sin sin sin e−λm,n,l αt ,
m=1 n=1 l=1
W L H
where
2
mπ 2 nπ 2 lπ
λ2l,m,n = + + .
W L H
We can use the remaining equation u(x, y, z, 0) = Ti − Tb for (x, y, z) ∈ Ω (and the
orthogonality of the sine functions) to get
⎧
⎨ (T − T ) −64 if m, n, l are all odd
i b
Am,n,l = mnlπ3
⎩0 otherwise.
This gives us an explicit solution.
Numeric solutions
For more complicated domains, we will have to approximate solutions. In some

cases, this means using a special-purpose software package, but we will do it our-
selves (with the help of a computer, but not necessarily a special purpose program)
using the finite difference method. (This method is something that everybody who
works on an A problem should be familiar with.)
With numeric techniques, we can consider dropping the condition that the
thermal diffusivity of the brownie batter is constant. The heat equation
T
= ∇ · (α∇T )
t
then becomes
T
= (αx Tx + αTxx ) + (αy Ty + αTyy ) + (αz Tz + αTzz ). (3.2)
t
For the rounded corners, we will want to use cylindrical coordinates; in cylindrical
coordinates, our heat equation becomes

T 1 1
= αTr + αr Tr + αTrr + (αθ Tθ + αTθθ ) + (αz Tz + αTzz ). (3.3)
t r r2
For our finite difference approximation, we will use time steps of Δt, steps in the
x-direction of Δx, and similarly for y and z. For convenience, we will write T for the
temperature T (x, y, z, t), T (t+) for T (x, y, z, t + Δt), T (x+) for T (x + Δx, y, z, t),
T (x−) for T (x − Δx, y, z, t), and similarly for T (y+), T (y−), T (z+) and T (z−).
We can take the approximation
T T (t+) − T
≈
t Δt
and the three point approximations
T T (x+) − T (x−)
≈
x Δx
2
T T (x+) − 2T + T (x−)
≈
x2 Δx2
and similarly for y, z and α. Replace the values in Equation 3.2 with these approx-
imations and solve for T , we will get the update equation

1 1 1
T (t+) = T · 1 − 2Δtα · + + +
Δx2 Δy 2 Δz 2

Δt α(x+) − α(x−) Δt α(x+) − α(x−)

T (x+) α+ + T (x−) α− +
Δx2 4 Δx2 4

Δt α(y+) − α(y−) Δt α(y+) − α(y−)

T (y+) 2 α+ + T (y−) 2 α− +
Δy 4 Δy 4

Δt α(z+)−α(z−) Δt α(z+)−α(z−)
T (z+) 2 α+ +T (z−) 2 α− . (3.4)
Δz 4 Δz 4
For the rounded corners, we can get a similar update equation in cylindrical coor-
dinates. We will then need to develop an update equation for points on the border
between the rounded corners and the rectangular parts of the brownies.
We will use the boundary condition T (x, y, z, t) = Tb for (x, y, z) ∈ Ω. (Re-

mark: The authors of the paper also briefly discussed a convective boundary con-
T
dition, = −c(Tb − T ) for appropriate c, where n is the outward normal at the
n
boundary. They pointed out that the finite difference method gave poor results in
this case.)
Finally, to take into account the changing thermal diffusivity of the brownies,
we will use the simplifying assumption that the diffusivity is that of water when
the brownie is less than the boiling point of water and it is that of cooked brownies
otherwise:
αwater if T < 100,
α(T ) =
αcake if T 100.
When cooking the brownies, we will need to specify the pan by giving the pan
thickness and values of w, l, r, an oven temperature Tb , an initial temperature for
the brownie batter Ti , and a target temperature for the brownies. For our numeric
approximation, we also need to specify the values of Δt, Δx, Δy, and Δz. To begin
our approximation, we set

Ti if (x, y, z) ∈ Ω,
T (x, y, z, 0) =
Tb if (x, y, z) ∈ Ω.
Given the temperature at time t, we find the temperature at time t+ = t + Δt by
• updating T in the rectangular region using Equation 3.4;

• updating T in the cylindrical region;
• updating T on the boundary between the rectangular and cylindrical regions;
• updating T according to the boundary conditions; namely, T (x, y, z, t+) = Tb
for (x, y, z) ∈ Ω;
• updating α.
We can continue stepping forward in time until the brownies reach the desired
temperature, at which point we will end up with a cooking time and temperature
distribution for the brownies.
Stability and sensitivity of the numeric solutions
To ensure the stability of our finite difference numeric approximation, we can use
von Neumann analysis. Note that being able to check on the stability of numeric
approximations to finite difference schemes is also something that anybody who
works on an A problem should be familiar with. To ensure the stability of our
procedure (where we will use Δx = Δy), we should ensure that Δt, Δx, and Δz
satisfy
2 1 1
αΔt + .
Δx2 Δz 2 2
We also want to make sure that our solutions do not depend too much on
our choice of certain values. To do this, we may carry out the numeric calculations
with varying parameters on a rectangular pan (9 inch by 11 inch) and a circular
pan with the same area. The conclusions are given below:
Pan height Increasing the pan height significantly affects the baking time.
Temperature Increasing the oven temperature or the initial temperature of the
brownie batter will cook the brownies more quickly but less evenly.
Diffusivity (α) If α is constant, then changing the value changes the time needed
to bake the brownies changes but doesn’t affect the heat distribution.
If α changes value from a smaller value to a larger value at the boiling
point of water, then there will be greater variation in the brownie temperature.
This should be expected, since increasing the diffusivity of the hotter parts of
the brownie will cause the hotter parts to absorb more heat.
Comparing the analytic and numeric solutions
We found an analytic solution for the case where the brownie pan is rectangular
and the diffusivity α is constant. Our numeric technique will also work in this case,
by letting r = 0 and not letting α change value. If we compute both for a pan with
width 20 inch, length 30 inch, and height 3 inch; and look at the heat distributions
of the middle cross-sections, we will get Fig. 3.6. The heat distributions are similar,
giving credence to our techniques, although our numeric approximation tends to
put more of the heat near the edges.
Figure 3.6 The heat distributions for the analytic (left) and numeric (right) solutions
Measuring the evenness of the heat distribution
As mentioned earlier, the problem asks about the heat distribution over the edge of
the brownies, but we will consider the distribution throughout the entire brownie
and throughout the entire baking time even though this may minimize some tem-
perature spikes on the edge of the brownie at the end of the baking.
For our heat distribution, we will use the values computed in the finite differ-
ence calculations, which will be a discrete set of points on a grid. This means we
will be summing values of T rather than integrating. To measure the temperature

at a point, we will consider the sum of the temperatures over all of the times used
in our calculations: t = 0, Δt, 2Δt, · · · , total baking time.

Let T̄ be the mean of T (t) over all points in our grid, where the sum is
t
taken over all times listed above. We will consider two possible measures of the
even distribution of heat H.
• Cumulative variance:
2

HV = T (x, y, z, t) − T̄
(x,y,z) t
• Cumulative absolute deviation:

HA = T (x, y, z, t) − T̄
t
(x,y,z)
A rounded rectangular pan of a fixed area can be described in terms of stretch-

iness γ1 , which is the following ratio of the length to width of the bounding rect-
angle:
l+r
γ1 = ,
w+r
and the following circleness γ2 , which measures how close the shape is to a circle:
r
γ2 = .
l+w
We would expect a larger stretchiness to be better for packing the pans and a
larger circleness to result in a better heat distribution. Fig. 3.7 graphs our two
temperature distributions (HV in units of 108 and HA in units of 104 ) compared
to stretchiness and circleness.
3.4.3 Optimizing the Perimeters of Pans

We have discussed maximizing N , the number of pans which will fit into the oven;
and minimizing H, the distribution of heat. The last part of the problem is to
optimize the combination, where they are combined by putting a weight of p on
N and a weight of 1 − p on H. Since we are only considering rounded rectangular
pans, a pan can be specified by giving values of w, l, and r.
If p < 0.5, then more weight is on minimizing H, and if p 0.5, then more
3.5 Model 2: An Outstanding Brownie Pan 35
Figure 3.7 HV (left) and HA (right) graphed against stretchiness and circleness
weight is put on maximizing N . We will first optimize the more important value
and then optimize the other. We will consider the two cases separately and describe
how the ultimate pan dimensions can be found.
N is emphasized (p 0.5)
The possible values of N are 1 through N ∗ ; a larger value of p puts more weight
on getting a larger value of N . We will consider an (unspecified) monotonically
increasing relaxation function f : [0.5, 1] → {1, · · · , N ∗ }, which will determine a
minimum target value of N depending on p. Let S1 be the set of triples (w, l, r)
such that f (p) pans specified by the triple will fit into the oven. Let (w, l, r) be the
triple in S1 that minimizes H (whether it be HA or HV ). This triple will describe
our ultimate pan.
H is emphasized (p < 0.5)
For this case, we will need a relaxation function g : [0, 0.5] → R, which will give a
maximum target value of H depending on p. Let S2 be the set of triples (w, l, r)
such that H will be less than g(p). Let (w, l, r) be the triple in S2 that maximizes
N . This triple will describe our ultimate pan.
3.5 Model 2: An Outstanding Brownie Pan
The outstanding-winner team at Shanghai Jiaotong University in China[18] de-

signed a brownie pan to combine space-efficiency with even heating. They used the
standard deviation of the temperature to determine the uniformity of the heating.
To determine how many pans could fit in the oven, they assumed that the oven
was large enough that any “edge effects” would be negligible and the only factor
to consider would be the fraction of space left in the over when the pans are placed
next to each other. We will work through their solution.
We will make similar simplifying assumptions as follows:
1. The only factor that the pan plays in the baking of the brownies is determining
the shape.
2. The pan material does not affect the heat being transferred to the brownies.
3. The only thing that heats the brownie is the heat from the air which is assumed
to have constant temperature.
3.5.1 Analyzing a Circular Pan
To begin with, we will make a simple model that we can solve analytically. For
this model, the pan will hold the brownie in the shape of a flat circular cylinder
of radius ρ0 and height L. We will set up a coordinate system so that the origin is
the center of the bottom of the pan (see Fig. 3.8). We will naturally use cylindrical
coordinates (ρ, ϕ, z) for position and t for time. Denote the temperature of the
brownie at a given point and time by u(ρ, ϕ, z, t). The heat equation for the batter
is
ut − a2 Δ3 u = 0,
where a2 is the thermal diffusivity of the batter.
Figure 3.8 Coordinates on a cylindrical pan
Let u0 be the fixed temperature of the air in the oven and u1 the initial
temperature of the batter (assumed to be the same throughout the batter). The air
temperature gives us the boundary conditions for u and the beginning temperature
of the batter gives us the initial condition; altogether we have
u(ρ = ρ0 ) = u0 ,
u(z = 0) = u0 ,
u(z = L) = u0 ,
u(t = 0) = u1 . (3.5)
This can be solved using separation of variables, resulting in

⎧ ⎡ ⎤ ⎫
∞
∞ ⎨ (0) 2 ⎬ (0) 2 2
⎣ xn p π
2 2
⎦ xn p π
u = u0 + Anp exp − 2
+ 2 a t J0 sin ,
⎩ ρ 0 L ⎭ ρ 0 L2
n=1 p=1
(3.6)
(0)
where J0 is the zero order Bessel function and xn is the nth positive zero of J0 .
The coefficients Anp need to satisfy
∞
∞
p2 π2
Anp sin = u1 − u0 .
n=1 p=1
L2
While we have managed to find a closed form solution (of sorts) for the tem-
perature of the pan, this approach will not extend to other more complicated pan
shapes.
3.5.2 Examining Two-dimensional Shape

To create a model that can help us find a good pan shape, we will reduce the
problem to a two-dimensional problem by only considering the cross sections of the
pan. Given the time constraints of the contest, this is an acceptable simplification
as long as it is explicitly addressed. The batter will then be lying on a region in the
plane. Letting u represent the temperature of the batter again, except that now
the position lies on a plane, the heat equation becomes
ut − a2 Δ2 u = 0,
where a2 = 1.48 × 10−7 m2 /s[21] . Again letting u0 be the oven temperature and u1
be the initial temperature of the batter, the initial condition is
u(t = 0) = u1
and the boundary condition is

u = u0
when the point is on the boundary of the region.

A common cooking temperature is 200 degrees Celsius, so if we measure
temperature in degrees Kelvin, we can use u0 = 473. Similarly, letting the bat-
ter start off at room temperature, we can use u1 = 293. Once a boundary is
given the problem is completely specified, although for most boundaries of in-
terest we will not be able to solve it analytically. We can use PDETools from
MATLAB to get numeric approximations to the heat distributions for various
shapes. Figs. 3.9 and 3.10 show graphs of the results for a circular and square
pan, respectively. The fact that the graphs show the temperature is uniform on
the edge of the circular pan and the temperature has peaks at the corners of the
square pan, results that we already know to be true, supports the validity of our
model.
Figure 3.9 The heat distribution on a circular pan
Figure 3.10 The heat distribution on a square pan
To measure the uniformity of the temperature on the sides of the brownies,

we will use the standard deviation of the boundary temperature at the end of
cooking. We will assume that our brownies are in the oven for 2,000 seconds. Since
we are getting a numerical solution to our temperature problem, we will find the
temperatures at evenly spaced points on the boundary and find the sample standard
deviation of these values. If we do that for various shapes (all with the same area
of 1/8 square meter), we might end up with the values given in Table 3.1 (standard
deviation measured in degrees Kelvin). As expected, the circle provides the most
uniform heating.
Table 3.1 The standard deviations of the temperature distributions of

different shapes
Shape Standard deviation (K)
Circle (radius 0.2 m) 30.85
Ellipse (semi-axes of 0.4 m and 0.1 m) 67.84
Square (0.3545 m by 0.3545 m) 43.43
Rectangle (0.2 m by 0.6285 m) 61.99
Equilateral triangle (side 0.5387 m) 40.41
3.5.3 Finding the Best Shape
Since we now have a reasonable model (i.e., a model that we can work with
and provides known answers), we can begin to work on finding a pan shape to
maximize the number of pans that can fit into an area and minimize the temper-
ature variation. We will again consider rounded rectangles (see Fig. 3.11). We are
supposed to solve the optimization problem for pans with area A (in our numeric
calculations, we will use A = 0.04 square meters), so from now on we will assume
that our pans have that area.
Figure 3.11 A rounded rectangle
Let S be the area of the enclosing rectangle and r be the radius of the circular
arcs. The value α = A/S tells us the fraction of the rectangle that is covered by
the pan, and if the rectangles can be used to cover the oven, we can use α in
place of N to measure how efficiently the pans cover the oven trays. If we place
these rectangles in the oven, there will likely be extra space around the rectangle
which is too thin to put in more rectangles, but we will assume the oven is large
enough that these edge effects are negligible. (Ignoring these edge effects, in fact,
Area of oven
gives us N = α, and so N and α would be proportional.) In this
A
case, the shape of the rectangle will not affect the efficiency of packing the rounded
rectangles, and table 3.1 indicates that a square would be the best rectangle to use.
The value of r will then determine the shape of our pan, and so α depends only on
r. In fact,
α = A/S = A/(A + 4r2 − πr2 ).
We will use σ to measure the uniformity of temperature, and we can use

PDETools to find σ for various values of r. If we plot the results we will get Fig.
3.12. It appears as if σ is approximately linear, and we can best fit the values to
the linear function
σ = −73.598 3r + 64.935 7.
Figure 3.12 The graph of σ
In order to ensure that the efficiency of packing the oven with pans and the
uniformity of the temperature have equal weights, we will normalize α and σ as
follows:
α − αmin
α̃ = ,
αmax − αmin
σ − σmin
σ̃ = ,
σmax − σmin
so they both take on values between 0 and 1. Since α and σ are functions of r, so
are α̃ and σ̃. Our goal now is to maximize
F (r, p) = p · α̃ + (1 − p)σ̃.
The graphs of F with respect to r (for various values of p) are given in Fig.
3.13. For p < 0.4, the bulk of the weight of F is on the uniformity of heat, and
since the graph of F is then increasing, the maximum will occur when r is as large
as possible; namely when the pan is a circle. For other values of p, the optimal
Exercises 41
values of r are given in Table 3.2 (note that when r = 0.1128, the pan is a circle).
In particular, the best shape for a brownie pan depends on how much emphasis the
baker puts on even heating versus filling the oven. However, notice in Fig. 3.13 that
for all p, the graph of F passes through the same point; namely when r = 0.0668
m, the value of F is always 0.592, which is a reasonably high value for F . So a
pan whose base is a rounded rectangle with that value of r will satisfy customers
regardless of which aspect of the pan they emphasize.
Figure 3.13 The graphs of F (·, p)
Exercises
3.1 In Section 3.4.1, the initial solution of using pans with sizes W × (A/W )
allows N ∗ = W L/A pans to fit into the oven. This solution was modified so that
the pans have sizes (W/k) × (kA/W ) for positive integers k. Show that at least
N ∗ − (k − 1) of these pans will fit in the oven.
3.2 Derive Equation 3.4, as described in Section 3.4.2.
3.3 Derive the update equation for cylindrical coordinates that is similar to Equa-
tion 3.4 for rectangular coordinates.
3.4 In Section 3.5, the effects of evaporating moisture and crust formation were
ignored in order to create a simpler model. How reasonable was it to ignore these
factors? If these effects were included, how could it affect the optimal pan shape?
3.5 Given the heat equation
ut − a2 Δ3 u = 0
and the conditions given by Equations (3.5), use separation of variables to derive
the solution (3.6). (Hint: Write u(ρ, ϕ, z, t) = v(ρ, ϕ, z)T (t) where v(ρ, ϕ, z) =
R(ρ)Φ(ϕ)Z(z). The heat equation then becomes the ordinary differential equations
T + k 2 a2 T = 0,
Δv + k 2 v = 0,
Φ + λΦ = 0,
Z + ν 2 Z = 0,
R + R /ρ + (k 2 − ν 2 − λ/ρ2 )R = 0,
for appropriate λ and k.)

3.6 The analytic model from Subsection 3.5.1 was solved, but the solution was
not used to determine how uniform the temperature was on the pan and it did not
help find the optimal shape of the pan. Why was it a good idea to include this
section?
3.7 In Section 3.5.2, the problem is simplified by focusing on the cross-sections of
the pan. As a result, the heat that the batter would get from the top and bottom
are ignored. How do you think this affects the solution? Does the thickness of the
brownie pans matter?
3.8 Your paper should be abundantly illustrated. Use a computer program (such
as PDETools) to solve the heat equation and derive graphs similar to Figs 3.9 and
3.10.
3.9 Discussed in Section 3.5, the square is the best shape of a brownie pan. How
much will non-square rectangles change the results?
Chapter 4 Water, Water,
Everywhere
Problem B of the MCM is a problem that can be attacked using discrete mathe-
matics. For 2013, the problem involved the distribution of water.
The exact statement of Problem B is given below:
Fresh water is the limiting constraint for development in much of the world. Build
a mathematical model for determining an effective, feasible, and cost-efficient
water strategy for 2013 to meet the projected water needs of [pick one country
from the list below] in 2025, and identify the best water strategy. In particular,
your mathematical model must address storage and movement; desalinization;
and conservation. If possible, use your model to discuss the economic, physical,
and environmental implications of your strategy. Provide a non-technical position
paper to governmental leadership outlining your approach, its feasibility and
costs, and why it is the “best water strategy choice.”
Countries: United States, China, Russia, Egypt, or Saudi Arabia
You are asked to create a model for determining a water strategy for one of several
countries, and then determine the best strategy of water distribution for the year
of 2025. You are asked to address at a minimum: transferring water, desalinization
of sea water, and conservation. Most Chinese teams chose to discuss the best water
strategy for China.
To begin the problem, you need to clarify what is being asked.
44 Chapter 4 Water, Water, Everywhere
What is a “water strategy”?
In this case, it was taken to mean a method of ensuring that all areas of the
country have sufficient fresh water for their needs. Since places which need water
(households, farms and industries, for example) are not necessarily close to fresh
water sources, transporting water from regions with plenty of water to regions
with insufficient water would be a big part of the strategy and one that most
teams focused on.
What does “best” mean?
Many teams began with finding the cheapest way moving water around appro-
priately. Particularly since you are asked to justify your position to government
leaders, “least expensive” is a good notion of “best” to begin with. The other
issues (environmental effects, etc.) should not be completely ignored, however.
Most papers concentrated on water transportation. It is not feasible to consider
every water pipe in the country, and so the water strategies did not involve local
transportation and instead divided the country into several regions and simply
considered transporting the water from on region to another.
4.2.1 Searching the Literature

A literature search should start by looking for a problem similar to the given
problem. The current problem is very similar to a problem from the 2009 HiMCM
(the high school version of the MCM):
Problem: Water, Water Everywhere

Fresh water is the limiting constraint for development in much of the
United States. Devise an effective, feasible, and cost-efficient national
water strategy for 2010 to meet the projected needs of the United States
in 2025. In particular, address storage and movement, desalinization, and
conservation as some of the possible components of your strategy. Con-
sider economic, physical, cultural, and environmental effects. Provide a
position paper for the United States Congress outlining your approach,
its costs, and why it is the best choice for the nation.
If you could find the outstanding papers from that contest, they might make a
good starting point. While the college contest would expect more sophisticated
and thorough papers, the high school papers could be a good source of starting
models.
Solving the problem involves knowing the water supply and water needs of
the regions of a country in 2025. It is not unreasonable to assume that the water
supply will be the same in 2025 as it is now, and this is information you would be
expected to look up. The water demands will almost certainly change, however, due
to changes in population as well as different uses of water (more efficient industrial
use, for example). To project the water needs, a reasonable strategy would be to
project the populations and the per capita water needs for the various regions in
2025. This can be done by projecting the populations and water consumption of the
regions given that information from the past few years, which is more information
that would require a literature search.
4.2.2 Creating a Model

A reasonable first (simple) model for this problem would cover the transfer of
water from the regions with an excess of water to those with a deficit. The cost of
transferring the water would involve, at a minimum, the one time cost of building
any necessary pipelines and the continuing cost of moving the water through the
pipes. Some teams also considered the cost of removing any contaminants that the
water may have picked up during the transport. The cost of the pipelines is simply
the length of the pipeline times the cost per unit length. To find the lengths,
a reasonable thing for you to do would be replaced each region by a point (for
example, the center of population or the geographical center) and use the distance
between these points as the distance between the regions. The cost of moving the
water through the pipes would be the amount of water being moved times the
cost per unit of water per unit distance. You would need to use the supply and
demands of the regions to determine how much water would have to be imported
and exported from each region.
To create a more complicated and more realistic model, you would need to
take into account water that can be added to the coastal region water supplies by
desalinizing seawater. The amount of water that can be added to the supply this
way is essentially unlimited, but it adds extra cost to the model per unit water
volume added.
You should also add conservation to your model. That could include making
more efficient use of water, which would decrease the water needs of a region
without adding costs; and recycling water, which would increase the water supply
but add extra cost.
4.2.3 Finding Solutions

Once the model is created, you need to determine the best strategy.
For the first (simple) model, the regions can be considered as nodes on a
graph and the pipelines as edges, each node has either a total amount of water
that can be exported or a total amount of water that needs to be imported and
each edge has a variable cost depending on how much water needs to be piped
through them. If desalinization and conservation are taken into account, then the
water supplies of the region could change, and the nodes could have a variable water
supply and in addition, a variable cost. If the “best” solution is the cheapest way of
transporting the water, then this problem is a nontrivial graph theory problem; a
good approximation to an optimal solution can be found using standard techniques.
If your notion of “best” contains more than just the cost, then your solutions
might change. One team used low environmental impact as a possible notion of
“best”, and they measured environmental impact by the total length of the pipeline.
A solution in this case would use the same model, but just minimize the length of
the edges.
4.2.4 Stability Test

Your model, and hence your solution, will likely depend on various quantities that
you need to approximate; for example, the water supply, the water demand, and
the various costs involved in your model. You need to make sure your solution
does not change if any of these quantities are changed; a good solution should be
stable under unexpected population changes, etc. If you used a stochastic method,
such as simulated annealing, to come up with your solution, you should rerun your
algorithm several times to make sure your “best” solution does not change much.
4.3 Judging

A good paper needs a clear goal, and so the judges look to make sure that the teams
explicitly state what they mean by “best”. What the judges looked for include the
following:
• the least cost,

• the most reliable,
• the least environmental impact,
4.3 Judging 47
• the highest quality.
A paper that constructs a model without clearly stating what it is for will not get
a high score from the judges.
Developing a water strategy for 2025 requires knowing the demand for water
in 2025. A discussion of transporting the water would require knowing where in the
country the water needs to go, and so the judges will check if the paper investigates
the regional demand for water as opposed to the demand for the country as a whole.
As a first approximation a team could predict the water needs based solely on the
predicted population of a region, but the judges would be more impressed if a paper
used more information to predict the demand. Similarly, the judges will make sure
that the water supply is predicted regionally rather than for the country as a whole.
While the price of transporting water will depend on the length of the pipeline,
the judges will make sure that papers use more than that to determine the cost.
The difference in altitudes between the water supply and the destination will play
a role in the cost, and the cost of the infrastructure and water loss will also have an
effect. Judges will be particularly impressed with papers that consider less obvious
and less direct costs, such as the cost associated with maintaining a worn-out in-
frastructure. Similarly, the costs associated with desalinization should also consider
the infrastructure.
Going above and beyond will impress the judges; for example, a team which
considers how their strategy will hold up past 2025 will be noticed by the judges.
The position paper needs to be convincing to a politician. It should explicitly
address, in nontechnical language, the goal of the strategy and explain how the
recommendation will meet it. The judges will look to see if the argument in the
position paper is well supported; it should include references.

The judges selected three Chinese teams and two US teams to receive the highest
honor as Outstanding Winners, listed below in alphabetical order of school names.
• Beijing University of Posts and Telecommunications, China.

Zhongxin Guo, Fan Wu and Beidan Wang.
Title: Five Models for China’s Water Scarcities.[23]
Advisor: Zuguo He.
• Colorado College, USA. (Frank R. Giordano Award Recipient)
Aradhya Sood, Yukiko Iwasaki and Namgyal Angmo.
Title: Water, Water Everywhere: Meeting the Demands of Saudi Arabia’s

Water Needs Today.[24]
Advisor: Andrea Bruder.
• Nanjing University, China. (INFORMS Award Recipient)
Wei Chen, Weizhi Liu and Cenying Yang.
Title: Make Wise Use of Every Drop.[25]
Advisor: Hui Qu.
• Tsinghua University, China. (SIAM Award Recipient)
Pengfei Gao, Boshuo He and Tianxin Zou.
Title: Quenching China’s Thirst in 2025: A Min-Cost-Max-Flow Network
Model.[26]
Advisor: Hao Wu.
• University of Colorado, Boulder, USA. (MAA Award Recipient)
Yue-Ya Hsu, Gregory S. Mcquie and David Thomas.
Title: Sustainable Water Management for Saudi Arabia in 2025 and Beyond.[27]
Advisor: Anne M. Dougherty.
4.4 Model 1: A Highly Regarded Water Strategy
One of the papers[28] , while not selected as an Outstanding Winner, was highly
regarded by the Chinese judges in the triage. This team divided China mainland
into seven regions, each composed of several provinces. They then built a series
of models, each taking into account more factors, and used linear programming to
develop their suggested water strategy for 2025.
The factors they took into account are
• water diversion and storage,

• desalinization,
• water conservation and recycling.
4.4.1 Water Supply and Demand

We will divide China into seven administrative divisions, each made up of several
provinces (see Fig. 4.1 and Table 4.1).
4.4 Model 1: A Highly Regarded Water Strategy 49
Table 4.1 The seven regions

Region 1 Region 2 Region 3 Region 4 Region 5 Region 6 Region 7
North Northeast East Central South West Northwest
Beijing Liaoning Shanhai Henan Guangdong Chongqing Shaanxi
Tianjin Jilin Jiangsu Hubei Guangxi Sichuan Gansu
Hebei Heilongjiang Zhejiang Hunan Hainan Guizhou Qinghai
Shanxi Anhui Yunnan Ningxia
Neimenggu Fujian Xizang Xinjiang
Jiangxi
Shandong
To determine how water-sufficient each region would be in 2025, we can con-

sider the degree of water demand for each region, which is the ratio of water demand
in 2025 to the water supply in 2012. We may consider a value close to 1/2 to be a
reasonable supply/demand balance; larger values would indicate a water deficiency
and smaller values a water surplus.
We need to know the water supply for each region in 2012. This information
is publicly available.
We also need to predict the water demand for each region in 2025; we can do
this by estimating the population and the per capita demand for the provinces and
then taking the product of these two values. Population growth is usually modeled
with an exponential function, when resources are plentiful, or a logistic function
when resources are limited. We can take the populations of all the provinces from
2000 to 2011 and then use the exponential growth model for provinces with plen-
tiful resources and the logistic growth model for provinces that are already well
developed. (In the paper, the legitimacy of this prediction method was verified by
showing that the prediction for the population of China mainland in 2025 closely
agrees with a prediction made by the UN.)
To predict the per capita water demand in 2025 for each province, we can begin
with the per capita water demand for the provinces from 2003 to 2010 and use a
combination of linear and logistic regression. The water demand for a province
will then be the product of the population and the per capita demand, and the
demand for a region would be the sum of the demands for the provinces which make
up the region. We should use a statistics program to crunch all of these numbers;
in the paper, the R statistics program was used. Table 4.2[28] contains the results,
where the columns represent, respectively, the provinces, the projected populations
for 2025, the water consumption per capita (in cubic meters) projected for 2025,
the projected water demand for 2025, and finally total supply for 2012. The last
column represents the degree of water demand for the regions.
Table 4.2 The degree of water demand for the regions

Region Province K (2025) W c/m3 D/108 m3 TOT/108 m3 d
Beijing 3 149.813 96.585 30.422 450 26.8
Tianjin 1 918.063 123.633 23.713 512 15.4
1 Hebei 7 945.012 226.973 180.330 083 157.2
Shanxi 3 979.139 186.114 74.057 228 124.3
Neimenggu 2 608.287 743.351 193.887 197 419.0
539.657 264 742.7 0.726 615 409
Liaoning 4 628.391 334.257 154.707 117 294.8
2 Jilin 2 804.441 520.466 145.961 479 315.9
Heilongjiang 8 594.770 1 094.769 940.928 776 629.5
1 041.522 355 1 240.2 0.839 801 931
Shanhai 3 896.291 450.173 175.400 618 20.7
Jiangsu 8 778.016 720.548 632.497 924 492.4
Zhejiang 6 702.647 255.697 171.384 740 745.0
3 Anhui 57.548 689.942 3.970 502 602.3
Fujian 4 042.964 605.288 244.715 598 774.9
Jiangxi 4 836.172 554.704 268.264 299 1 037.9
Shandong 10 499.480 224.170 235.366 528 347.6
1 940.945 129 4 020.8 0.482 726 106
Henan 8 876.614 247.789 219.952 731 328.0
4 Hubei 5 840.568 652.916 381.340 263 757.5
Hunan 6 919.186 509.571 352.581 445 1 126.9
1 017.108 423 2 212.4 0.459 730 800
Guangdong 13 639.110 222.595 303.599 224 1 471.3
5 Guangxi 5.358 649.959 0.348 245 1 350.0
Hainan 1 005.950 408.760 41.119 242 484.1
625.726 041 3 305.4 0.189 304 181
Chongqing 3 043.112 318.244 96.845 214 514.6
Sichuan 8 181.976 329.806 269.846 396 2 239.5
6 Guizhou 1 171.123 362.370 42.437 937 626.0
Yunnan 4 844.235 319.211 154.633 116 1 480.2
Xizang 372.085 1 372.526 51.069 689 4 402.7
951.836 183 9 263.0 0.102 756 794
Shaanxi 3 854.735 227.913 87.854 422 604.4
Gansu 2 573.87 467.741 120.390 479 242.2
7 Qinghai 600.058 554.652 33.282 336 733.1
Ningxia 714.053 1 064.383 76.002 577 8.8
Xinjiang 2 664.348 2 199.282 585.965 260 885.7
939.543 678 2 474.2 0.379 736 350
Notice that northwest China (i.e., region 7) has a degree of water demand
d = 0.38, which is neither too high nor too low. This means that northwest China
does not need to import water, and for geological reasons it is not a good region for
exporting water. Thus, we will leave that region out of our discussion of transferring
water. The remaining six regions, two of them have a degree of water demand well
above 1/2 (region 1 with d = 0.7266 and region 2 with d = 0.8398.) We will only be
considering these six regions and having regions 1 and 2 import water from regions
3, 4, 5 and 6.
4.4.2 Modeling Water Transfer
To begin with, we can create a model that only depends on water transfer and we
can minimize the total cost. As a rough approximation, we can take the cost for
a pipeline from one region to another to be proportional to the length of the pipe
times the amount of water transported, and the total cost to be the sum of the
individual costs. Specifically, the cost function is
⎛ ⎞
2
6
CO = w1 ⎝ Aij Lij ⎠ , (4.1)
j=1 i=3
where w1 is the unit cost of water diversion (the cost per kilometer per hundred
million cubic meters), Aij is the amount of water transferred from region i to region
j, and Lij is the distance from region i to region j. We can find the distances by
representing each region by a specific point in the region and using Google maps
(and get the results in Table 4.3) and use w1 = 10 000 yuan per 100 000 000 cubic
meters per kilometer. To find the cheapest water strategy, we want to find the
amounts of water to be transferred between the regions (i.e., the values of Aij )
that will minimize the cost denoted by CO.
Table 4.3 The distances between the regions (in km)

Region 1 (North) Region 2 (Northeast)
Region 3 (East) 903.7 1960.0
Region 4 (Central) 1120.0 2133.9
Region 5 (South) 1887.7 2836.7
Region 6 (West) 2034.3 5118.3
The constraints on each region are that the ratio of the demand for that region
to the water resources for the region (after the water is transferred between regions)
is less than some desired degree of water demand for that region; specifically,
Di
< di , (4.2)
NETi
where Di is the water demand for region i, NETi is the amount of water resources
for region i after the water is transferred. If TOTi is the total water resources
6
for the region in 2011, then NETi = TOTi + Aji for regions 1 and 2, and
j=3

2
NETi = TOTi − Aij for the regions which export the water. The necessary
j=1
conditions that no region exports all of its water

2
TOTi − Aij > 0 for i = 3, 4, 5, 6, (4.3)
j=1
and the amount of water to be transported is non-negative
Aij 0 for all i, j
are added.
These are linear equations and inequalities, and so this is a linear programming
problem. If we wish to lower the degree of water demands for regions 1 and 2 to
d = 0.6 and raise the degrees for regions 3 through 6 to d = 0.5, we can use a
computer program and get A31 = 13891 million cubic meters, A41 = 1781.883
million cubic meters, A42 = 16036.52 million cubic meters, A52 = 33530.48 million
cubic meters, and no other water being transferred.
We would need to check to make sure that small changes in the variables
produce small changes in our solution. The authors of the paper did that.
4.4.3 Water Transfer and Desalinization
We can make our model more sophisticated by adding desalinization to our model.
We will assume that regions 1, 2, 3 and 5 (the regions with coasts) now have
desalinization plants. For these regions, let Bi denote the amount that the region
gets from desalinization.
For this model, the cost function is
⎛ ⎞

2
6
CO = w1 ⎝ Aij Lij ⎠ + w2 Bi ,
j=1 i=3 i=1,2,3,5
where w2 is the unit cost of water desalinization (the cost per hundred million cubic
meters), which, after some research, we can take to be 10 thousand yuan per 100
million cubic meters.

The constraints on each region are that the ratio of the demand for that region
to the water resources for the region (after the water is transferred between regions
and now the desalinated water is included) is less than some desired degree of water
demand for that region; specifically,
Di
< di ,
NETi
where now NETi , the amount of water resources for region i after the water is
transferred and the desalinated water is added, is

6
NETi = TOTi + Aji + Bi
j=3
for regions 1 and 2,

2
NETi = TOTi − Aij + Bi
j=1
for the coastal regions which export water, and

2
NETi = TOTi − Aij
j=1
for the remaining two regions. The necessary conditions that no region exports all
of its water now become

2
TOTi − Aij + Bi > 0 for i = 3, 5,
j=1
and

2
TOTi − Aij > 0 for i = 4, 6,
j=1
and the amounts of water transported and desalinated are non-negative, that is,
Aij > 0 for all i, j,

Bi > 0 for all i.
Again, these are linear equations and inequalities, and again we can specify
degree of water demands for the regions and use a computer to minimize the cost.
4.5 Model 2: An Outstanding Water Strategy 55
In this case, the only water transferred is from Region 3 to Region 1 (13 891 million
cubic meters) and from Region 4 to Region 1 (1 781.883 million cubic meters). In
addition, Region 2 will need to desalinate 49 567 million cubic meters of water.
4.4.4 Including Water Conservation

We can then add conservation of water to our model. Water can be conserved in
a number of ways; less water will be used if the price of water is raised and if
improved irrigation methods are developed, more water will be available to use if
water is recycled. For conservation of water, our model will take the decrease in
water demand into account by multiplying the demand for each region by a number
λi between 0 and 1 (a different number for each region) and the increase in the
amount of available water by multiplying the net water supply for each region by
a number i greater than 1. The only effective change in the model is that the
conditions
Di
< di
NETi
are then replaced by
λi Di
< di .
i NETi
Once we determine what the values of the λ and are, we can again solve the linear
programming problem with a computer.
As an example, if we examine what will happen if less water is used in region
2 by setting λ2 = 0.95 and let everything else be unchanged (i.e., λi = 1 for
i = 1, 3, 4, 5, 6 and i = 1 for i = 1, 2, 3, 4, 5, 6). In this case, the total cost will show
a minor decrease and the amount of desalination that Region 2 would need to do
would decrease by about 20%.
4.5 Model 2: An Outstanding Water Strategy
The Outstanding-Winner team at Tsinghua University in China[26] also developed

a water strategy for China mainland. They considered each of the 30 contiguous
Chinese provinces as a separate unit (represented by a single point within the
province), and then created a network graph with 33 nodes; one for each province,
one to represent overall water supply, one to represent desalinated ocean water, and
one to represent overall water demand. After using regression analysis to estimate
the water supply and demand for each province in 2025, they viewed finding the
best water strategy as a min-cost-max-flow problem and used simulated annealing
to come up with a solution. They used the same model, but first deal with finding
the minimum cost; they then considered the environmental effects of the water
transport (“indirect costs”) and considered finding the minimum combined distance
of the pipelines.
4.5.1 Water Supply and Demand

To follow their model we will assume that the water resources will remain stable
for the near future and use the average amounts over the past few years given
by the National Bureau of Statistics of China as the resources available for each
province in 2025. Since not all of the resources can ever be used, we will use 30%
as a reasonable rate and 40% as the maximum rate.
To estimate the total water demand for each province, we will consider the
amount of water needed for agriculture, industry and households separately, since
they will likely have different growth rates, and then we can extrapolate the amount
that each province will need in 2025.
Table 4.4 is the water supply and demand values for the provinces[26].
Table 4.4 Water supply and demand for the provinces (×108 m3 )
Province Water supply (Sj ) Water demand (Dj )
Beijing 9.269 65.727
Tianjin 4.750 66.941
Hebei 54.135 245.072
Shanxi 38.848 124.140
Inner Mongolia 160.367 313.827
Liaoning 115.703 221.109
Jilin 156.434 179.065
Heilongjiang 273.793 707.434
Shanghai 12.355 176.209
Jiangsu 164.496 653.675
Zhejiang 368.831 304.748
Anhui 295.121 332.343
Fujian 443.322 255.121
Jiangxi 578.803 252.012
Shandong 128.399 319.070
Hubei 173.053 340.841
Hunan 391.604 359.371
Guangdong 676.361 394.774
Guangxi 694.642 504.610
Hainan 712.266 346.429
Continued
Province Water supply (Sj ) Water demand (Dj )

Chongqing 210.404 121.502
Sichuan 952.611 298.618
Guizhou 374.449 232.874
Yunnan 769.636 182.686
Tibet 1767.216 59.514
Shaanxi 178.558 126.121
Gansu 84.255 194.094
Qinghai 277.354 53.843
Ningxia 4.004 93.939
Xinjiang 367.684 684.183
4.5.2 Water Distribution Costs
The total cost of distributing water depends on
• the cost of creating the pipelines,

• the cost of transporting the water,
• the cost of removing pollutants from the transported water.
To determine the cost (per kilometer) of creating the pipelines, we can use the
projected per kilometer cost of Chinese South-to-North water diversion project.
This gives us an estimated cost of 111 724 138 RMB per kilometer. (Note that
this is an extremely large cost. The judges reading the paper will expect any data
and results to be reasonable and conform to common sense, and it looks like this
number might be unreasonably high. The authors of the paper acknowledged this
and stated that they did further research to verify its accuracy.)
To determine the cost of transporting the water, we can choose a point (prefer-
ably near the main water source close to the center of each province), find the
coordinates (latitude, longitude and altitude) of these points (which can be done
with Google Earth) and then find the distance between these points on a sphere.
When the water is transported through the pipeline, it will meet with resis-
tance which is typically measured in an equivalent height the water would have
to fall to have enough energy to overcome the resistance, called the head loss.
The head loss can be combined with the actual difference in altitude between two
provinces to determine an equivalent difference in altitude from one province to
another; namely, the altitude difference that would provide the amount of energy
to move the water between the provinces. A positive equivalent altitude difference
would require that much extra energy to transport the water from one province to
another. If the equivalent altitude is negative then no extra energy is required to
move the water.
In the case where extra energy is needed to transport the water, we can assume
that there will be a pump station providing the energy. The efficiency of pump
stations have been studied and shown that they will take about 70% of the input
energy (such as electricity) and transfer it to the water.
We would want to consider other issues that might affect the cost. Evaporation
can be shown to have negligible effect, and the cost of removing pollutants that
the water picks up during the travel will be jointly proportional to the amount of
water and the distance moved. We can estimate the constant of proportionality by
looking at the costs involved in an existing water project.
Since part of the water strategy may involve desalinization, we would have
to consider the associated costs. The authors of the paper briefly noted that there
were several methods of desalinating water, such as seawater reverse osmosis, mul-
tiple effect distillation and multi-stage flash. Since reverse osmosis has the smallest
marginal cost, that is the only method we will consider, but it is important that
the authors noted that they looked at the various methods.
Finally, some of the costs involved in a water strategy, such as transport and
pollutant removal, will be recurring. Assuming the program has a useful life of 50
years, we can estimate the present values of the yearly costs.
4.5.3 Modeling Water Transfer and Desalinization

Our water strategy will be to determine how much water should be transported
from one province to another in 2025 and how much water should be provided by
desalinization. We will use a directed graph to model the movement of water. The
graph will have a vertex for each of the 30 contiguous provinces ({v0 , · · · , v29 }), as
well as a vertex (s) representing the overall water supply, a vertex (t) to represent
the overall water demand, and a vertex to (vd ) represent desalinization water. The
edges of the graph are
• (s, vi ), where vi represents a province, the capacity of this edge is the water
supply of the province, and the cost is 0;
• (vi , t), where vi represents a province, the capacity of this edge is the water
demand of the province, and the cost is 0;
• (s, vd ), The capacity of this edge is infinite and the cost is the cost of water
desalinization per cubic meter;
• (vd , vi ), where vi represents a coastal province, the capacity of this edge is
infinite and the cost is 0;
• (vi , vj ), where vi and vj represent provinces, the capacity of this edge is infinite
and the cost is the average cost of moving a cubic meter of water from the
province represented by vi to the province represented by vj .
An illustration of such a graph, simplified to have only four cities, is given in

Fig. 4.2. The best strategy would be to find the amount of water to be transferred
from one province to the other that would maximize the flow from s to t at a
minimal cost.
Figure 4.2 A graph illustrating our model
4.5.4 Using the Model to Determine a Strategy

Given the graph, determining the optimal amount of water that should be trans-
ported between the provinces is an unreasonably large task. We can use simulated
annealing to minimize the total cost of the system (using the total cost of the
project as the “energy” function), which is expected to return a good approxima-
tion to the minimal cost. To compute the total cost at a state of the simulated
annealing algorithm, we can use the Shortest Path Faster Algorithm.
4.5.5 The Best Strategies

We can get various solutions by changing assumptions and emphases.
First solution(s)
To begin with, we will assume that China will be using its water resources at
the maximum rate of 40%. In this case, the overall water supply will be greater
than the water demand, and there would be no need for the relatively expensive
water desalinization. The amount of water that should be transferred between the
provinces that will minimize the overall cost is given by Table 4.5. On the whole
water will be transferred from the south to the north (due to the excess water in
the southern provinces) and from the west to the east (due to the lower costs in
transporting the water away from the relatively high provinces of western China).
In case there is a breakthrough in desalinization costs in the next few years,

we can use the model to investigate the best water strategy at lower desalinization
costs. Unless desalinization costs drop to about 1/4 of their current costs, it will
not be part of an optimal strategy.
If China will only use its water resources at a more environmentally-friendly
30%, then the total water supply will be less than the total demand, and so de-
salinization will be necessary. The strategy provided by the model in this case will
require more pipelines. The overall cost ends up being about four times the cost of
4.6 Model 3: Meeting the Demands of Saudi Arabia’s Water Needs 61
the previous situation; partially because the cost of the extra pipelines but mostly
because of the cost of the desalinization.
Second solution
The water pipelines can cause environmental damage, and so the team used their
model to develop a strategy which will minimize the total length of the pipelines.
This strategy requires desalinizing nearly as much water as is transferred.
4.6 Model 3: Meeting the Demands of Saudi

Arabia’s Water Needs
The Outstanding-Winner team at Colorado College in the USA[24] was also the
Frank R. Giordano Award Recipient. They developed a water strategy for Saudi
Arabia, a country with limited fresh water supply and largely surrounded by ocean.
While countries such as China have the option to use desalinization to add to their
water supply, Saudia Arabia needs to use desalinization to meet its water demands.
Currently, 70% of the drinking water in Saudi Arabia comes from desalinization.
They considered three aspects of a water strategy: desalinization, distribution
costs, and the treatment of wastewater. Rather than optimize these aspects simul-
taneously, this team found the best strategy for each part separately.
4.6.1 Modeling Desalinization
Like the previous team did in Subsection 4.5.2, this team briefly mentioned different
ways to desalinate water and quickly concluded to consider reverse osmosis, given
its lower costs.
We will use a Cobb-Douglas function to model the amount of water produced
by desalinization. If the amount of a good water produced P depends on two factors
F1 and F2 , then the Cobb-Douglas model gives the relation
P = αF1β1 F2β2
for constants α, β1 , and β2 . This model will have constant returns to scale (e.g.,
doubling both factors will double the production) if β1 + β2 = 1; more generally,
the constants β1 and β2 are “economy of scale” factors. We will use this model
throughout this section. For desalinization, a previous study[29] used the function
W = F 0.6 H 0.4 (4.4)
to model the amount of desalinated water produced, where H is the amount of

heat used in evaporation and F represents other factors, such as the cost of pumps
and labor. We will use this model for desalinization in Saudi Arabia.
We would like to desalinate as much water as possible, but there will be limited
resources. The total cost of desalinization will be
PH H + PF F,
where PH is the price per unit for heat (for which we will use the cost of the
electricity needed) and PF is the cost per unit of the other factors mentioned. The
most recent estimates for these costs are PH = $0.21/m3 and PF = $0.46/m3.
Saudi Arabia budgets about 6.4 billion dollars per year for water related projects.
We will assume that about 40% of this is directed towards desalinization giving us
2.56 billion dollars to work with. This gives us
0.21H + 0.46F 2.56 × 109 . (4.5)
To find the maximum amount of desalinated water, we want to maximize

W from Equation 4.4 with the condition given by Inequality 4.5. While this is a
straightforward calculus problem, there is no requirement that we do the calcu-
lations ourselves; the authors solved it using Mathematica. The maximum water
that can be produced is 3.9 × 109 m3 /year (with associated factors H = 4.9 × 109
kWh/year and F = 3.3 × 109 units/year).
To check the sensitivity of this model, the authors of the paper determined
the maximum amount of water with slight changes in the percent of the water
budget used for desalinization and the costs of the electricity and other factors.
(The number of times that W had to be maximized is a good reason why we would
want to have a computer to do the work for us.) As expected, small changes in
these values only resulted in small changes in the water production.
4.6.2 A Water Distribution Strategy

We next want to discuss a strategy for distributing water to where it is needed.
Here, our strategy will involve creating the necessary pipelines at the minimum
cost.
We will assume that the reservoirs of water and the destination of the water
is fixed and that the locations of the pipelines used to transport fresh water is
also fixed. (The authors considered a couple of different strategies for laying the
pipelines, namely, the “branched” and “looped” systems as shown in Fig. 4.3, but
their subsequent discussion does not depend on the strategy used.) We will also
assume that the water demand at the destinations is known.
Figure 4.3 The branched and looped methods of water distribution
The cost of laying the pipeline will depend on the length L and the diameter d
of the pipe. The length will be fixed; we will need to determine the best diameter.
For this fixed cost, we will use
F C = αLdβ
for constants α and β.
The cost of actually transporting the water will depend on the water demand
at each destination (C for consumption), the diameter of the pipes and the water
pressure inside the pipes (h for “hydraulic head”). We might expect, using a Cobb-
Douglas function, that the cost of transporting the water from a reservoir to the
destination would be proportional to
CP dθ hφ
for some θ and φ. While we might know the consumption at each destination,
we can also consider that demand might not be uniform throughout the day; if l
represents a possible “load”, then the cost to transport to a destination would be

l
γ CP dθ hφ .
l
The total variable cost of transporting the water will then be

l
VC =γ CP dθ hφ .
P l
The total cost of transporting the water will then be

l
T C = F C + V C = αLdβ + γ CP dθ hφ . (4.6)
P l
The only variables are d and h. The constraints on these variables are determined
by the flow rate
Q = Area × Velocity
πd2
= · 2gh
4 √
= 3.5d2 h,
where g is the gravitational constant. We will want the flow rate to equal the total
demand of water per unit time:
√
3.5d2 h = Q. (4.7)
We need to estimate the parameters L, α, β, γ, CP , θ, φ, and Q. Since there

is little information on the costs for the Saudi Arabian water system, the authors
used information about the USA water system[30] to get the estimates as follows:
α = 827,
β = 0.6,
γ = 763,
θ = 0.8,
φ = 0.4385,
Q = 218.
The parameters L and CP depend on the actual layout of the pipeline, which is
left unspecified.
Our strategy should be to minimize the total cost given in Equation 4.6 given
the condition Equation 4.7. In order to get a specific result, we will use a simplified
description of Saudi Arabia’s water demands. We will not divide Saudi Arabia by
region and just consider the agriculture, industry, and household uses as three des-
tinations. In Saudi Arabia, agriculture accounts for 88% of the demand, industry
accounts for 3% and household use for 8% of the demand. The authors used con-
sumptions of 876.41 m3 /s, 29.81 m3 /s, and 89.74 m3 /s respectively and pipelines of
lengths 490 m, 485 m, and 500 m. They then used Mathematica to get a minimum
cost, which results from a diameter of d = 5.37 m. They checked the sensitivity of
their result by varying the values of the length of the pipeline and the consumption
that they used.
4.6.3 Treating Waste Water
Finally, we will consider a strategy for the treatment of waste water. Again using a
Cobb-Douglas function to model the amount of water that can be treated, we get
the quantity
Q = F aEb, (4.8)
where E is the amount of electricity used and F represents other factors. We will
assume constant return to scale, so a + b = 1. The authors clearly state that there
is not enough data to determine a and b, but mention how they would find these
values if they had the data. To have something to work with, we will use a = 0.4
and b = 0.6. The total cost will be
PF F + PE E,
where PE is the price per unit for electricity and PF is the cost per unit of the
other factors. We will use PF = $0.038/m3 and PE = $0.46/m3. We will also
assume that 30% of the Saudi Arabian budget for water projects is allocated for
water treatment, so the total cost can be up to $2.56 billion. Our model is now to
find the maximum value of
Q = F 0.4 E 0.6
with the condition

0.038F + 0.46E 2.56 × 109 .
The authors used Mathematica to solve this problem, getting

F = 3.339 × 109 units,

E = 2.65 × 1010 kWh,
Q = 7.65 × 109 m3 .
For sensitivity analysis, they resolved this problem for different costs, ranging from
10% of the budget to 100% of the budget.
4.6.4 Comments
This model is very similar to their desalinization model. The authors clearly were
rushed to finish this section of their paper and the writing is sparse, but since they
already did something similar and because of their innovative economic model, the
overall paper gave them the Outstanding Winner standing.
Exercises
4.1 In the model created in Section 4.4.2, the changes in altitude between the
regions is ignored. Determine the relative altitudes of the regions and make a con-
jecture about how that would affect a water strategy.
4.2 Use a computer program to solve the linear programming problem given by
Equation 4.1 with constraints given in Inequalities 4.2 and 4.3, where d1 = d2 = 0.6
and d3 = d4 = d5 = d6 = 0.5 . (The authors of the paper used LINGO.)
4.3 Look up the water supply for one or two provinces over the last few years.
Use this information to predict the water supply for 2025. Compare your results
with the results in Table 4.4.
4.4 The models in Subsections 4.4.3 and 4.5.3 both involve transferring water
between provinces (or collections of provinces) as well as desalinization. The first
model concludes that the only water transferred is from Region 3 to Region 1
(13891 million cubic meters) and from Region 4 to Region 1 (1781.883 million
cubic meters). The second model concludes that the amount transferred should be
given by Table 4.5. Compare these results. Which do you think is more accurate?
4.5 The models in Subsections 4.4.3 and 4.5.3 both involve transferring water
between provinces (or collections of provinces) as well as desalinization. The first
model finds desalinization to be a useful tool and the second does not. Which one
is right? Justify your answer.
4.6 To find a “best” water strategy, why was it acceptable to deal with separately
Exercises 67
for Saudi Arabia. Would this have been acceptable for China?
4.7 What are some weaknesses of the models presented here? Given time and
resources, what would you do to fix these weaknesses?
Chapter 5 Network Modeling of
Earth’s Health I
ICM is an interdisciplinary modeling contest. It encourages combinations of stan-

dard mathematical techniques and non-mathematical methodologies in other dis-
ciplines to solve real-world problems. ICM judges consist of pure mathematicians,
applied mathematicians, statisticians, computer scientists, and experts in the dis-
ciplines where the contest problems are drawn from. For example, in ICM 2013
there was a biologist and earth environmental scientist serving as final judges, and
in ICM 2014 there was a behavioral scientist and a geologist serving as final judges.
Problem C this year is a network modeling problem of Earth’s health, which
is a complex network science problem. Network science is an emerging subject that
blends structures, processes, data, and applications from mathematics, computer
science, operations research, sociology, information science, and several other fields.
Network science and social network analysis was the topical area for the past two
ICM contests and will continue to be the topical area for ICM 2014[31].
Given below is the verbatim description of Problem C.
Background: Society is interested in developing and using models to forecast

the biological and environmental health conditions of our planet. Many scientific
studies have concluded that there is growing stress on Earth’s environmental and
biological systems, but there are very few global models to test those claims.
The UN-backed Millennium Ecosystem Assessment Synthesis Report found that
nearly two-thirds of Earth’s life-supporting ecosystems—including clean water,
pure air, and stable climate—are being degraded by unsustainable use. Humans
are blamed for much of this damage. Soaring demands for food, fresh water, fuel,
and timber have contributed to dramatic environmental changes; from deforesta-
5.1 Problem Description 69
tion to air, land, and water pollution. Despite the considerable research being con-
ducted on local habitats and regional factors, current models do not adequately
inform decision makers how their provincial polices may impact the overall health
of the planet. Many models ignore complex global factors and are unable to deter-
mine the long-range impacts of potential policies. While scientists realize that the
complex relationships and cross-effects in myriad environmental and biological
systems impact Earth’s biosphere, current models often ignore these relationships
or limit the systems’ connections. The system complexities manifest in multiple
interactions, feedback loops, emergent behaviors, and impending state changes
or tipping points. The recent Nature article written by 22 internationally known
scientists entitled “Approaching a state shift in Earth’s biosphere” outlines many
of the issues associated with the need for scientific models and the importance
of predicting potential state changes of the planetary health systems. The arti-
cle provides two specific quantitative modeling challenges in their call for better
predictive models:
1. To improve bio-forecasting through global models that embrace the com-
plexity of Earth’s interrelated systems and include the effects of local conditions
on the global system and vice versa.
2. To identify factors that could produce unhealthy global state shifts and
to show how to use effective ecosystem management to prevent or limit these
impending state changes.
The resulting research question is whether we can build global models using local
or regional components of the Earth’s health that predict potential state changes
and help decision makers design effective policies based on their potential impact
on Earth’s health. Although many warning signs are appearing, no one knows if
Planet Earth is truly nearing a global tipping point or if such an extreme state
is inevitable.
The Nature article and many others point out that there are several impor-
tant elements at work in the Earth’s ecosystem (e.g., local factors, global im-
pacts, multi-dimensional factors and relationships, varying time and spatial
scales). There are also many other factors that can be included in a predictive
model—human population, resource and habitat stress, habitat transformation,
energy consumption, climate change, land use patterns, pollution, atmospheric
chemistry, ocean chemistry, bio diversity, and political patterns such as social un-
rest and economic instability. Paleontologists have studied and modeled ecosys-
tem behavior and response during previous cataclysmic state shifts and thus
70 Chapter 5 Network Modeling of Earth’s Health I
historic-based qualitative and quantitative information can provide background

for future predictive models. However, it should be noted that human effects have
increased significantly in our current biosphere situation.
Requirements: You are members of the International Coalition of Modelers
(ICM) which will soon be hosting a workshop entitled “Networks and Health of
Planet Earth” and your research leader has asked you to perform modeling and
analysis in advance of the workshop. He requires your team to do the following:
Requirement 1: Build a dynamic global network model of some aspect of
Earth’s health (you develop the measure) by identifying local elements of this
condition (network nodes) and appropriately connecting them (network links) to
track relationship and attribute effects. Since the dynamic nature of these effects
is important, this network model must include a dynamic time element that
allows the model to predict future states of this health measure. For example,
your nodes could be nations, continents, oceans, habitats, or any combination of
these or other elements which together constitute a global model. Your links could
represent nodal or environmental influences, or the flow or propagation of physical
elements (such as pollution) over time. Your health measure could be any element
of Earth’s condition to include demographic, biological, environmental, social,
political, physical, and/or chemical conditions. Be sure to define all the elements
of your model and explain the scientific bases for your modeling decisions about
network measures, nodal entities, and link properties. Determine a methodology
to set any parameters and explain how you could test your model if sufficient
data were available. What kinds of data could be used to validate or verify the
efficacy of your model? (Note: If you do not have the necessary data to determine
parameters or perform verification, do not throw out the model. Your supervisor
realizes that, at this stage, good creative ideas and theories are as important as
verified data-based models.) Make sure you include the human element in your
model and explain where human behavior and government policies could affect
the results of your model.
Requirement 2: Run your model to see how it predicts future Earth health.
You may need to estimate parameters that you would normally determine from
data. (Remember, this is just to test and understand the elements of your model,
not to use it for prediction or decision making.) What kinds of factors will
your model produce? Could it predict state change or tipping points in Earth’s
condition? Could it provide warning about global consequences of changing local
conditions? Could it inform decision makers on important policies? Do you take
into account the human elements in your measures and network properties?
Requirement 3: One of the powerful elements of using network modeling is the
ability to analyze the network structure. Can network properties help identify
critical nodes or relationships in your model? If so, perform such analysis. How
sensitive is your model to missing links or changing relationships? Does your
model use feedback loops or take into account uncertainties? What are the data
collection issues? Does your model react to various government policies and could
it thus help inform planning?
Requirement 4: Write a 20-page report (summary sheet does not count in
the 20 pages) that explains your model and its potential. Be sure to detail the
strengths and weaknesses of the model. Your supervisor will use your report as
a major theme in the upcoming workshop and, if it is appropriate and insightful
to planetary health modeling, will ask you to present at the upcoming workshop.
Good luck in your network modeling work!
Potentially helpful references include:
• Anthony D. Barnosky, Elizabeth A. Hadly, Jordi Bascompte, Eric

L. Berlow, James H. Brown, Mikael Fortelius, Wayne M. Getz,
John Harte, Alan Hastings, Pablo A. Marquet, Neo D. Mar-
tinez, Arne Mooers, Peter Roopnarine, Geerat Vermeij, John W.
Williams, Rosemary Gillespie, Justin Kitzes, Charles Marshall, Nicholas
Matzke, David P. Mindell, Eloy Revilla, Adam B. Smith. “Approaching
a state shift in Earth’s biosphere,”. Nature, 2012; 486 (7401): 52 DOI:
10.1038/nature11018.
• Donella Meadows, Jorgen Randers, and Dennis Meadows. Limits to Growth:
The 30-year update, 2004.
• Robert Watson and A.Hamid Zakri. UN Millennium Ecosystem Assessment
Synthesis Report, United Nations Report, 2005.
• University of California at Berkeley. “Evidence of impending tipping point
for Earth.” Science Daily, 6 Jun. 2012. Web. 22 Oct. 2012.
The goal of the problem is to build a viable network model for studying the health
of the earth. Your model should be able to make predictions of condition changes
and compute a tipping point, which is the point from what is considered acceptable
in your model to unacceptable, and an unacceptable condition cannot be reversed

to acceptable unless extra measures outside the model are taken. Solving the prob-
lem requires creative thinking and innovative designs, as well as solid knowledge
in multiple areas, including discrete math, continuous math, computer science,
statistics, to name just a few.
To begin tackling the problem, you should first study the problem descrip-
tion (including all the requirements), dissect the problem to understand the exact
meaning of each sentence, and form a holistic view of the problem. This is par-
ticularly important for a problem with a long description. An effective method to
carry out this task is to rephrase the problem using your own words. Whenever
you have doubts about the meanings of a sentence, you should discuss it with your
teammates.
After obtaining a clear understanding of the problem, you should try to en-
vision what an acceptable solution would be like and what a reasonable approach
should be taken. You will use this vision to guide your exploration for a solution. In
particular, you should try to envision what nodes and links in your network model
should look like–these are your building blocks–and think about how to model the
health dynamics of the earth and how to predict a tipping point.
During the course of modeling, you will modify and justify your initial vision
a few times until you are satisfied with your approach. You should then generate a
list of assumptions and parameters and form a war plan. For example, your initial
strategy may include the following tasks:
1. Read the suggested references.

2. Write down the scientific facts and figures.
3. Use the facts and reasonable assumptions to form nodes and links for your
network model.
4. Search for data sets on Earth’s environment from the Internet and use them
to validate your model.
5. Think about how to model dynamic changes of the Earth’s environment for
deriving a tipping point.
After forming an initial strategy, you should work out an arrangement with your
teammates to best utilize the skills and talents of each member.
To figure out how to form nodes and links for your network model, it is neces-
sary to survey the relevant scientific research results through reading the suggested
materials posted in the problem description, as well as from other references. Us-
ing environmental factors to indicate the health of the planet Earth is an obvious
choice. Another choice would be economic data, for they are affected by envi-
ronmental factors. The modeling process may be broken down into several steps
described in the following subsections.
5.2.1 Select Measures

You must first determine your building blocks for constructing your network model.
Your building blocks depend on basic elements that affect Earth’s health. You may
obtain these elements from reading the references posted in the problem, or from
other references, and find out how to measure them. Remember to cite your sources.
These elements may be grouped into the following categories:
1. Pollutants. Pollution may happen in the air, under the water, and on land.
Air pollution may be measured by carbon emissions, CO2 level, or air quality;
water pollution may be measured by water quality or heavy metal pollution;
and land pollution may be measured by land use, forest coverage, or soil
quality, to name just a few.
2. Biological elements. A biological element may be measured by antibiotic health,
bio-diversity and habitat, food webs, or species extinctions.
3. Societal elements. A societal element may be measured by policies, economic
data, or financial data.
While it would be nice to include as many basic elements as possible in your design,
given the limited time you are given to complete your modeling, doing so and doing
it well would be difficult. Thus, it would be wise to focus only on a smaller number
of elements that you feel comfortable to work with. Next, you will need to figure out
how to measure the elements you selected using scalar values. More importantly,
you would need to obtain data sets for the elements you chose so that you can
validate your model.
The judges would look for innovation and creativity, not necessary a perfect
solution. For this problem, no perfect solution is known at this point. Some of the
world’s best minds are still trying to figure out a reasonably good solution. So any
reasonable approach would be worth a try.
5.2.2 Devise Network Models

Now it is time to incorporate the elements you selected to form a network model.
To do so you would need to make a few assumptions. Some are based on common
beliefs and some for simplicity. It is common sense to assume that humans are the
main cause for the degradation of the earth’s ecosystem. It is also reasonable to
assume that once an environmental measure goes below a critical threshold value
for humans to live in, it is impossible for it to return to its previous state without
extra force. Because data are observed and collected at discrete moments, the values
of the measures are discrete data, even though the variables representing some of
these elements are continuous. More specific assumptions may be made during the
course of modeling, and remember to state your assumptions explicitly and justify.
You may follow the following approaches to forming a network model:
1. Form a conceptual network to connect the elements you selected. For example,
elements that affect the ecosystem of the earth may be modeled as a Bayesian
Belief Network or as pitted competitors in the setting of predators and preys,
with humans as the predator and Earth the prey. Examples of competitors
include the following pairs:
(a) Human destroyers versus the planet ecosystem.

(b) Human technology benefits versus the destructive forces of the technology.
(c) Pollutants caused by humans versus the resilience of the Earth.
2. Form a geographic network that connects physical zones (regions) at various

levels of details: cities, states, countries, continents, and oceans. A zone in
the network is referred to as a zone node, or a region node. This separation
approach would allow you to focus on building your network at each level.
Within each zone you may include a subnetwork of elements you selected. The
simplest kind of this approach would be a two-level network with a top-level
network for interactions between geographical zones and sub-level network for
interactions between elements inside a zone.
After forming a network, you must describe how nodes will interact. Depend-
ing on your mathematical knowledge, you may choose, for example, (1) differential
equation models, (2) neural networks, or (3) discrete dynamical systems. These
methods offer predictive capability. You may also use the explicit network struc-
tures to determine local and global properties. Or you may choose to ignore any
network structure and perform classical data mining and element classification and
discrimination to identify properties.
You may then use real data sets to figure out the values of parameters you
used in your models using fitting and machine learning techniques, run simulations,
solve differential equations, and perform data analysis to predict future trends of
the elements. You will also find a threshold value to determine the tipping point of
5.3 Judging Criteria 75
your model. (Note: some models may not have a tipping point.)
5.2.3 Validate Models and Perform Sensitivity Analysis

Do not forget to validate your model and do not forget to perform sensitivity
analysis. For this purpose, you would need to obtain suitable data sets, which
can often be found from public sources on the Internet. Again, remember to cite
the sources of your data. Validation and sensitivity analysis would often go hand
in hand. Performing validation means to compare the results you obtained from
running your models against the real data sets. Performing sensitivity analysis
would often mean to simulate your model under slightly altered data sets and see
whether the results remain stable. Sensitivity analysis is often the weakest parts in
submissions. Some teams were confused what a sensitivity analysis is about.
5.3 Judging Criteria
When forming your thoughts and strategies, you should keep in mind, in addition to
the modeling components described in Section 5.2, what else the judges would like
to look at. Typically this would mean clear and concise writing, a good executive
summary, and sufficient explanations of the assumptions used in your models with
appropriate figures and charts. Most Chinese teams (outstanding winners included)
were having serious problems in writing. For an elaboration of these criteria the
reader is referred to [7].

The judging criteria for this year’s problem are listed below[31] .
Executive summary
Judges cannot stress enough the importance of writing a good executive sum-
mary. A well-versed summary will increase your chance a great deal to advance
to the category your solution deserves. Judges would check if your executive sum-
mary contained brief and informative descriptions of your modeling approach and
bottom-line results. Detailed explanations of what you presented in the executive
summary should then be provided in the main sections of your paper. An outstand-
ing paper must contain a well-written executive summary with a well-connected
and concise description of the approach you used, the results you obtained, and
any recommendations you would suggest.
Modeling
Judges would check whether you explicitly discussed why key assumptions were
made and how these assumptions affected the development of your model. You
should present motivations and rationales of your mathematics and prose. In other
words, you should avoid presenting equations and parameter values without expla-
nation. Using or misusing arbitrary parameters without any explanation or analy-
sis was a major defect found in submissions time and time again. Establishing and
explaining parameter values in models are at least as significant as making and
validating assumptions.
You were asked to assess if your models could predict state change or tipping
points in the earth’s condition so that warnings could be provided. This was proba-
bly the most challenging aspect. The judges recognized the difficulty in performing
this analysis and rewarded papers that presented models with strong analysis of
tipping points.
Sciences
Modeling Earth’s health requires facts and figures from scientific research, including
biological, environmental, and political science research, to name just a few. Judges
would be pleased to see a survey of the facts and figures used in your modeling, as
well as a good explanation of relevant scientific terminology you used.
Data/Validity/Sensitivity
Judges would check whether you used real data sets to validate your model and test
its accuracy and robustness. In particular, judges would like to see whether you
provided sensitivity analysis to determine if small changes of your assumptions or
data, or the presence of small errors, would affect the accuracy of your model. This
aspect is especially important for highly-structured and data-rich models such as
networks. Some network structures are robust and flexible while others are fragile
and sensitive to data. The judges were particularly interested in seeing how you
used the model to inform and guide potential policy decisions. Teams that did this
well quickly rose to the top of the judges’ evaluations.
Strengths/Weaknesses
Judges would like to know whether you really understand the models you created.
To demonstrate your understanding, you should discuss the strengths and weak-
nesses of your model, including the limitations or constraints of your assumptions
5.3 Judging Criteria 77
or the implications of your methodology. You should also suggest ways for improve-
ment. Networks are complex structures and so the strengths and weaknesses may be
hidden from direct views. Teams that did this well received favorable evaluations.
Communication/Visuals/Charts
Judges would like to read well-written and clearly-explained solutions. Using dia-
grams and tables appropriately helped a great deal. Readers were better informed
through well-designed arrays of powerful charts and graphs that explained both
models and results.
Recommendations
Judges would like to see a paper to include a paragraph that suggests, based on
the models and results presented, policies and regulations on human behaviors to
help preserve the ecosystem of the planet earth. This was an important indicator
in identifying strong papers.

There exist many different and interesting approaches to modeling the earth’s
health conditions. This is nice, for it provides a rich platform for creative teams to
design innovative and diverse solutions. Each reasonable approach would have the
full potential to reach the highest category. Papers that did not make to the final
judging often suffered from the following two defects:
1. Mathematical models were not connected well to the measurement they se-
lected.
2. Explanations of the modeling structure were either not provided or not clear.
In general, papers that were poorly written could not be understood, and they did
not progress to the final rounds of judging. You may find more information how to
write MCM/ICM papers in [7].
The judges selected four Chinese teams and one US team to receive the highest
honor as Outstanding Winners, listed below in alphabetical order of school names:
• Beijing University of Posts and Telecommunications, China (INFORMS Prize

Recipient).
Xu Yao, Yingnan Xiao and Chao Wang.
Title: Two-tier Communication Network Model of Global Health.[32]
Advisor: Xinchao Zhao.
• Northwestern Polytechnical University, China.

Yixiao Shu, Yuhui Wang and Chao Chen.
Title: Two-layered Coupled Network Model of Earth’s Health.[33]
Advisor: Lulu Pan.
• Peking University, China.
Ruofei Zhao, Xinqi Zhao and Minghe Hu.
Title: Measuring Earth’s Health by CO2 : a Technology Diffusion Network
Approach.[34]
Advisor: Shulin Zhou.
• Rensselaer Polytechnic Institute, USA.
Diogo Moitinho de Almeida, Eric Shapiro and Amanda Knight.
Title: Saving the Green with the Greens.[35]
Advisor: Mark Holmes.
• Zhejiang University, China.
Changyou Zhu, Weicheng Bai and Wenqing Yang.
Title: You pollute, I pollute![36]
Advisor: Yuan Ying.
These papers present five innovative models, which can be grouped into two
categories according to the types of math they used. The first category is differential
equations and the second is discrete math, including computer science. We will
present two models in the first category in this chapter and three models in discrete
math in the next chapter.
5.4 Model 1: A Two-level Network Model Using

Differential Equations
The Outstanding-Winner team at the Northwestern Polytechnical University[33]
constructed a two-level network model. They used the dynamic Lotka–Volterra
equations to model interactions between humans and other elements at the sub-
level network, and use reaction-diffusion equations to model interactions between
zone nodes at the top-level network. The Lotka–Volterra equations are also known
as the predator–prey equations.
5.4 Model 1: A Two-level Network Model Using Differential Equations 79
5.4.1 Network Model

The network consists of two levels. The top-level network, also referred to as the
global network, consists of nine zone nodes. Each zone node represents a geographic
region by dividing the countries into nine regions based on the global geography.
These regions are America and Cental America, Big Eastern Asia, Canada and
Greenland, Northern Africa and Middle East, Northern Europe and Russia, Ocea-
nia, South America, Sub-Saharan Africa, and Western and Southern Europe (see
Fig. 5.1 (a)). Each pair of zone nodes are directly connected (see Fig. 5.1 (b)).
Links between zone nodes represent interactions between physical elements inside
the zones.
Figure 5.1 Link segments labeled with the same circled number represent the same link
The sub-level network, also referred to as the elementary network, consists of

seven nodes, also referred to as the elementary nodes, for each zone node in the
top-level network, representing the following seven physical elements:
1. human population density,

2. air quality,
3. water quality,
4. biodiversity and habitat (biodiversity, in short),
5. forest coverage,
6. climate change, and
7. energy consumption.
The data set of the 2012 Environmental Performance Index (EPI)[37] provides
values for these elements. The values for these elements (except human population
density) were regularized by the Yale Center for Environmental Law and Policy
(YCELP)[38] and the Columbia University Center for International Earth Science
Information Network (CIESIN)[39] .
It is reasonable to assume that, based on common knowledge, an elementary
node inside each zone will directly interact with all the other elementary nodes
(including itself for self influence), and so the seven elementary nodes of the sub-
level network within each zone node form a complete graph with self loops. Across
the zones, we assume that, for simplicity, only elements of the same type will
interact with each other.
5.4.2 Sub-level Integrations as Predator and Preys

It is common practice to use the Lotka-Volterra equations to model predator-prey
interactions in a biological system of two (or more) species[40] . Let S1 and S2 denote,
respectively, Species 1 and Species 2. Let u1 (t) and u2 (t) denote, respectively, the
measurement of S1 and S2 at time t. To model the dynamic change of these two
species, we also consider self influence of the same species. Let a1 and a2 denote,
respectively, the natural growth rate of S1 and S2 , and b1 and b2 , respectively, the
influence factor of S1 to the natural growth rates of S1 and S2 . Similarly, let c1
and c2 denote, respectively, the influence factor of S2 to the natural growth rates
of S1 and S2 . The Lotka-Volterra equations are non-linear, first-order differential
equations defined as follows:
u1 (t) = u1 (t) (a1 + b1 u1 (t) + c1 u2 (t)) ,

u2 (t) = u2 (t) (a2 + b2 u1 (t) + c2 u2 (t)) .
Next, we extend these equations to model interactions between the seven ele-
mentary nodes inside a zone node. To do so we label the zone nodes as 1, 2, · · · , 9.
Within each zone node, we label the nodes for human population density, air qual-
ity, biodiversity, forest coverage, climate change, and energy consumption as nodes
1, 2, · · · , 7, respectively. Let uik (t) denote the kth elementary node within the ith
zone node, where i = 1, 2, · · · , 9 and k = 1, 2, · · · , 7. Let aik denote the change
rate of elementary node k within the ith zone node, and bijk the influence factor
of elementary node j to the change rate of elementary node k within the ith zone
node, where j = 1, 2, · · · , 7. The Lotka-Volterra equations can be written as
⎛ ⎞
duik (t)
7
= uik (t) ⎝aik + bijk uij (t)⎠ , (5.1)
dt j=1
i = 1, 2, · · · , 9,
k = 1, 2, · · · , 7.
5.4.3 Reaction Diffusions in the Top-level Network
Let z1 and z2 be two zone nodes. Let n1 ∈ z1 and n2 ∈ z2 be elementary nodes

of the same kind. We assume that n1 and n2 interacts as a flow, and the flow
moves from the higher-density node to the lower-density node. The rules provided
by YCELP and CIESIN support this assumption for all the elements except human
population. For simplicity, we also assume that human population interacts as a
flow, even though it may sometimes move from a lower density to a higher one. We
therefore model the top-level network as a flow network.
For each elementary node, we may view its values over the surface of the
globe as a scalar field in a discrete vector space. For simplicity, we will consider
only 2-dimensional space. (Note: Working on the 2-dimensional space may be
oversimplified, but it is easier to handle, and the judges recognized it.) Thus,
we may use a discrete version of the 2-dimensional reaction-diffusion equation
(R-D equation)[41] to model interactions between nodes at the top-level network.
In particular, let f (ρ, θ) denote a scalar field over the polar coordinate system for a
given variable (e.g., the air quality). Its gradient, denoted by ∇[f (ρ, θ)], represent-
ing the maximum rate of change of f per unit at the given point, can be described
as follows:
f 1 f
∇[f (ρ, θ)] = eρ + eθ ,
ρ ρ θ
where eρ is the unit vector on the radial coordinate and eθ the unit vector on the
tangential coordinate.
Completing the partial derivative transfers the scalar field into a polar space V .
For a point source flow, it was shown that streamlines are straight lines emanating
from a central point O[42] , as depicted in Fig. 5.2. It is straightforward to see
that the components in the radial and tangential directions are f / ρ and f / θ,
respectively, with f / θ = 0 (this is because on any point in a straight line, θ is a
constant).
Figure 5.2 A point source flow
It follows from the R-D equations that the divergence of every point in V ,
denoted by div V , can be calculated as follows:

2
f 1 f 1 f f
div V = div(∇f ) = div eρ = ρ· = + 2.
ρ ρ ρ ρ ρ ρ ρ
2
Note that f / ρ2 is insignificant to the final value, and so we can safely
ignore it to simplify calculation. We have
1 f
div V ≈ .
ρ ρ
This implies that the divergence of the gradient of the scalar field is inversely
proportional to the radial distance. In other words, the propagation of elements
between discrete points is inversely proportional to the distance of the points.
Let Δ denote the change of element k propagating from zone node j to zone
node i, where k ∈ {1, 2, · · · , 7} and i, j ∈ {1, 2, · · · , 9}. Then

i uik (t) − ujk (t)
Δ = βk ,
rij
where βki denotes the propagation factor of node i and rij the distance from node j
to node i. Because all neighboring nodes of node i will contribute to the propagation
of element k, we have

duik (t) i uik (t) − ujk (t)
= βk . (5.2)
dt rij
j=i
5.4.4 The Final Equation

The top-level network and the sub-level network interact simultaneously, and so
we combine Equation 5.1 and Equation 5.2 to get

⎛ ⎞
duik (t) 7 uik (t) − ujk (t)
i ⎝ i
= uk (t) ak + i i ⎠
bjk uj (t) + i
βk , (5.3)
dt j=1
rij
j=i
i = 1, 2, · · · , 9,
k = 1, 2, · · · , 7.
The values of the coefficients aik and bijk can be determined using the Ordinary
Least Square method on the EPI data sets over ten years[37].
5.4.5 Tipping Points

After the values of the coefficients are determined, we solve Equation 5.3 using a
differential equation solver (e.g., MATLAB or Mathematica). We can then predict
the trend of each element and compute a tipping point. The values of the seven
elements are normalized into the same scale from 0 to 100.
Prediction of future trends
Let A-Zone denote the zone node of America and Cental America. Fig. 5.3 depicts
the prediction of the seven elements in A-Zone for the next 50 years. We can see
that all elements but the human population density are increasing. In particular,
the forest coverage, water quality, and biodiversity are increasing more rapidly, and
the human population density is slightly decreasing. This indicates that, under the
current policies and practice, the ecosystem in A-Zone will remain healthy for the
next 50 years.
Figure 5.3 Trend in A-Zone for the next 50 years

Tipping points
One way to identify a tipping point is to try different initial values for a given
element. For example, to seek a tipping point for the element of air quality, we
alter its initial value from high to low, while keeping the initial values of the other
six elements intact. We then compute the trend of the air quality for the next 50
years to identify whether an irreversible change will occur at some point. When the
initial value makes the values of the air quality continue to decline and recovery
does not seem possible, this value is the tipping point for the element of air quality.
To put things into perspective, it means that when the air quality reaches this
value, from that time on, the air quality will keep on decreasing. For A-Zone, only
air quality and biodiversity have tipping points, depicted in Fig. 5.4, which are
Figure 5.4 Tipping points for A-Zone

38.76 and 59.86, respectively. To avoid these tipping points from happening, the
governments and decision makers in this zone should take special measures to
protect the air quality and biodiversity.
5.4.6 Sensitivity Analysis

As mentioned above, human activities are often the main cause that affect the
earth’s ecosystem. For this reason we carry out sensitivity analysis by focusing on
the influence of human population density to the other six elements. For demon-
stration purpose, we use forest coverage and energy consumption as examples. Let
PE and PF denote the influence factors of human population density to the change
rates of energy consumption and forest coverage, respectively, which are shown in
Fig. 5.5 for A-Zone.
We can see that the original PE increased slowly and smoothly, indicating
that it is a stable parameter. Reducing PE to 0.9 times the original PE seems
to cause subtle fluctuation and generate much worse effect than the original PE.
Increasing PE to 1.1 times the original PE generates an almost constant line—a
positive effect. We can also see that PF has the similar effect. For example, with
0.9 times the original PF, in the early stage, PF is weaken; but after that PF will
generate much worse effects than the original PF. On the other hand, increasing
PF will improve the health condition; for example, with 1.1 times the original PF
will generate a turning point for good around 2030.
Figure 5.5 Sensitivity analysis of influence factors of the human population density to
forest coverage and energy consumption in A-Zone. The solid line represents the original
trend, and the other lines represent adjusted trend where the dashed line matches the
reduced influence and the dotted line matches the increased influence
5.4.7 Network Structures
Based on the results obtained from Equation 5.3 and the EPI data set, we can
determine the influence effects between elements within a zone. There is a positive
influence and a negative influence. For each influence factor, if its value is less than
one-tenth of the average absolute value of all the influence factors, then we ignore
it for its influence would be weak. We can therefore obtain an interaction network
of all elements in a given zone. Fig. 5.6 shows the interaction network for A-Zone,
where the thickness of links represents the magnitude of the influence level, the
solid links represent positive influence, the dashed links represent negative influ-
ence, and the triangle represents self-influence.
The two-level network consists of a total of 63 nodes. We study its typology
using the following common notion of local centrality to find a central point, which
is the summation of the indegree and outdegree of a node. Using its influence factor
as a weight for an outgoing link, we define the notion of influence degree to capture
the influence of a node to other nodes, which is equal to the average absolute value
of the outgoing weighs of a node. (Note that outdegrees may have positive influence
and negative influence, and hence the absolute value.) Table 5.1 depicts the local
+
+
Figure 5.6 The interaction network for A-Zone
centrality and the influence degree of the seven nodes in A-Zone.
Table 5.1 Local centrality and influence degree of nodes in A-Zone

Element Local centrality Influence degree
Human population 9 7.93
Air quality 3 3.35
Water quality 8 6.08
Forest coverage 10 4.65
Biodiversity 6 2.56
Climate change 7 5.14
Energy consumption 5 1.48
Table 5.1 shows that the forest coverage is the most central point of the net-
work, while the human population density comes in second. With respect to the
influence levels, however, the human population density is the most influential while
the water quality is the second.
5.4.8 Adding Disturbances to Parameters

We may add a random variable called disturbance to every parameter to capture
uncertainties in the network flow model. We assume that the it is distributed ac-
cording to the normal distribution around the parameter. This gives rise to the
following equation:
⎡ ⎤
7
duik (t)
= uik (t) ⎣ aik + âik + bijk + b̂ijk uij (t)⎦ + (5.4)
dt j=1

ui (t) − uj (t)
i i k k
βk + β̂k ,
rij
j=i
i = 1, 2, · · · , 9,
k = 1, 2, · · · , 7,
where âik , b̂ijk , and β̂ki are disturbances. Fig. 5.7 depicts the trends of all elements
with disturbances for A-Zone.
Figure 5.7 The trends of all elements in A-Zone with disturbances
5.4.9 Comments
This two-level network model is focused on the interaction dynamics between hu-
mans and a number of typical environmental elements. It handles the interac-
tions between nodes in the top-level network as a flow network using the standard
reaction-diffusion equations, and the interactions between nodes in the sub-level
network as the predator-prey relation using the standard Lotka-Volterra equations.
Treating the interactions between the elements of human population as a flow
that flows from a higher-density node to a lower-density node is a major weakness
5.5 Model 2: A Network Model on Ocean Pollution Diffusion 89
in the model, for human populations do not always migrate in this manner, and so
it would be more desirable to come up with a better model for human migration.
For a better model on human migration the reader is referred to Section 6.1.5. It
would also be nice to provide a table of values of all the parameters in the final
equation presented in Section 5.4.4.
5.5 Model 2: A Network Model on Ocean Pollution

Diffusion
The Outstanding-Winner team at Zhejiang University used ocean pollution as a
marker for the Earth’s health, and subsequently modeled the flow of pollution
through eight regions of the global coastal water system. The team took a staged
approach to modeling their system and used differential equations to model pollu-
tion diffusion.
5.5.1 An Ocean Pollution Network

Ocean pollution is mainly caused by harmful chemicals dumping into the ocean
by humans during industrialization, diffused by ocean currents. Fig. 5.8 (a) shows
the human impact on the marine ecosystem[43] and Fig. 5.8 (b) shows the ocean
currents.
The coastal areas are affected most by ocean pollution, and they are also the
major sources of pollutants. Based on Fig. 5.8 and ignoring minor currents, we
divide the earth into eight regions for modeling ocean pollution diffusion as follows
(see Fig. 5.9 (a)):
1. The Asian-Pacific coastal area, including the Japan Sea, East China Sea, and
South China Sea.
2. The West America coastal area, including the Bering Sea.
3. The East North-America coastal area, including the Caribbean, and the Gulf
of Mexico.
4. The East South-America coastal area.
5. The West Europe coastal area, including the Mediterranean and the North
Sea.
6. The Indian Ocean area, including the gulf and the Red Sea.
7. The Antarctic coastal area.
8. The Arctic Ocean area.
Figure 5.8 (a) Human’s overall impact on the marine ecosystem. (b) Ocean currents
Figure 5.9 (a) Ocean division of eight nodes. (b) Ocean network
These nodes are connected with weighted directional links according to the
directions of ocean currents (see Fig. 5.9 (b)). For each directed link in the network,
the sink node is affected by pollutant emissions from the source node, and the
weight represents the distance between the linked nodes.
5.5.2 Model 2.1: Pollution Diffusion

The following assumptions are made for simplicity.
1. Pollutants appear only on the surface of the ocean, and the ocean surface is in
the 2-dimensional space. In other words, we ignore the curvature of the earth.
2. Pollutants travel between nodes in one dimension.
3. At any given moment, each node is affected by pollutants generated by itself
and by nodes with only one link away.
4. The temperature of ocean water is stable with a constant attenuation coeffi-
cient for pollutants.
5. The velocities of different ocean currents are the same.
6. The mutual effects of several nodes is the summation of the effect of each
individual node.
7. The conditions of the ocean are unlikely to incur dramatic changes for a long
period of time.
8. The pollution decay rate is proportional to the amount of pollutants and
inversely proportional to the distance between two nodes.
For a given node, let C(x, t) denote the amount of pollutants from a node at
position x and time t, where C(0, t) is the amount of pollutants generated by the
node itself at time t. We will later use Ci (x, t) to denote C(x, t) for node i. Ocean is
capable of self-purification, which will weaken pollutants during the transmission.

This effect can be modeled using a simplified Princeton Ocean Model (POM)[44,45] .
The original POM for water quality is a 3-dimensional partial differential equation.
Based on Assumption 1 that pollutants only stay on the 2-dimensional ocean sur-
face and Assumption 2 that pollutants travel in one dimension, we can simply use
a 1-dimensional model to simulate the propagation of pollutants by the following
partial differential equation[46] :
2
C C C
+u = DAB · + W (t) − k · C(0, t), (5.5)
t x x2
where u is the velocity of the ocean currents (e.g., see [47]), DAB the molecular
diffusion coefficient (see, e.g., [48]), W (t) the released amount of pollutants at time
t, k the attenuation coefficient, and −kC(0, t) the degradation amount at time t.
Let Pi (t) denote the amount of pollutants of node i at time t. Then based on
Assumption 3 we have

Pi = Ci (0, t) + Cj (xji , t),
j
where the summation is over nodes j that are directionally connected to i and xji
is the distance between node j and node i (i.e., the position from node j to node i
on a direct line).
The value of Ci (0, t) for a particular year t can be calculated by summing up
the amount of pollutants in each country contained in node i and in coastal water
contained in node i. The amount of pollutants in each country can be obtained from
public sources, while the amount of pollutants in coastal water can be obtained by
sampling ocean water (this is not done regularly in the current practice). To collect
water samples we would need to setup a number of observation stations along the
coastline, as evenly as possible for a better coverage.
Equation 5.5 can be solved numerically using the Finite Difference method
(e.g. the Four Point Implicit Difference method). To do so, we assume that pol-
lutants are released at discrete time of observation. We will later obtain a refined
solution using integration.
5.5.3 Prediction of Ocean Pollution Trends

Pollutant emissions follow certain patterns that are related to the degrees of the
countries’s development. For example, in Western Europe that consists of developed
countries, ocean pollution has grown slowly, while in Asia, where most countries
are developing countries, ocean pollution has gone up rapidly. Thus, it would be
interesting to figure out particular emission patterns and see how our model reacts
to such patterns. Unfortunately, data like this is hard to come by and so we will use
the following five common growth functions to approximate these patterns and use
them to analyze how ocean pollution would change over time. Results are shown
in Fig. 5.10.
1. Constant function: g(t) = C, where C is a constant.

2. Linear function: g(t) = at + b.
3. Quadratic function: g(t) = at2 + bt + c.
4. Exponential function: g(t) = Aeσt + B.
5. Logarithmic function: g(t) = A lnσ t + B.
Figure 5.10 Ocean pollution with different pollutant emission patterns
It follows from Fig. 5.10 that, while the ocean’s self-purification ability does
slowdown the growing rate of ocean water pollution, the trends in the aggregation of
pollutants in the coastal water areas are similar to the emission trends of pollutants.
This can be characterized by attenuation coefficient. If the pollutant emission is
stable, then the ocean pollution level remains a constant; if more pollutants are
discharged into the ocean, then the ocean will be polluted at a higher level. This
result agrees with our intuition.
5.5.4 Model 2.2: A Transmission Control System

As shown in Fig. 5.9, node 1 through node 6 represent the coastal areas of the
continents. The majority of human activities occur in these areas, and so these six
nodes are the major sources of pollutants. Node 7 represents the Arctic area and
node 8 the Antarctic area. Little human activities take place in these two nodes,
and so they are not the source of pollutants. These two nodes, however, are not iso-
lated from sources of pollutants, and so they will be affected, although at a slower
pace because of the long geographic distance from sources of pollutants, the slow
velocity of ocean currents, and the ocean’s self-purification effect. Nevertheless, a
small amount of pollutants, with the cumulative effect over time, could still cause
significant damage to the Polar Regions. Thus, we would want to study how nodes
7 and 8 would be affected. In other words, we would like to model how pollutants
would transmit to these two nodes from other nodes. We will use the ocean pol-
lution readings of the two Polar Regions to indicate the health condition of the
earth.
Model design
Pollutant transmissions can be measured using a control-theory method. Under

the assumption that pollutant transmissions are one directional and that ocean
conditions are unlikely to change dramatically for a long time, we assume that our
system is linear-time-invariant system (LTI)[49] . In this LTI system we associate six
inputs u1 , u2 , · · · , u6 to, respectively, node 1, node 2, · · · , node 6, and two outputs
y1 and y2 to node 7 and node 8, respectively (see Fig. 5.11).
Figure 5.11 Block diagram of the control system for pollutant transmissions
For i ∈ [1, 6], variable ui denotes the input to node i, which is a flow of pollu-
tants. Let xi denote the state variable of node i, representing the current pollution
condition of the node. Block gj (j = 1, 2, · · · , 9) represents the impulse response of

pollutant propagation (as signals) for every link in the network. According to the
LTI theory, if the input function is u(t) and the impulse response is g(t), then the
output is the convolution of u(t) and g(t), denoted by u(t) ∗ g(t), which is equal
to +∞
g(τ ) · u(t − τ )dτ.
−∞
Applying the Laplace Transform convolution to scalar multiplications and obtain

numerical results for outputs y1 and y2 after inputs are specified. This can be done,
e.g., by writing a MATLAB program.
5.5.5 Prediction of Pollution Trends

We divide the input functions into step functions and ramp functions. A step func-
tion releases pollutants at a constant rate while a ramp function releases pollutants
at a rate that follows a linear function. Prediction of pollution trends for node 7
and node 8 are given in two modes: the growing mode and the stable mode. The
growing-mode prediction shows what makes pollution go up and the stable-mode
prediction shows what makes pollution stable. We use the following input functions
to carry out simulations.
Input 1: u1 (t) = 2t.

Input 2: u2 (t) = 6.
Input 3: u3 (t) = 8.
Input 4: u4 (t) = 1.5t.
Input 5: u5 (t) = 8.
Input 6: u6 (t) = 2t.
Fig. 5.12 (a) depicts the pollution trend for the Antarctic area predicted by
the model with the given inputs, where pollution begins to increase in about three
time units and continues to increase linearly. It will reach a tipping point when time
goes by. However, if people control pollutant emissions within a certain constant
level, then pollution will become stable (see Fig. 5.12 (b)) and avoid reaching the
tipping point.
Fig. 5.13 (a) depicts the pollution trend for the Arctic area predicted by the
model with the given inputs, which is in a stable mode, for node 8 is affected only
by nodes 3 and 5, whose inputs are constant functions. If one of the input functions
Figure 5.12 (a) Growing mode for node 7 under the given input functions. (b) Stable
mode for node 7 if all ramping functions are changed to step functions of constants
is changed to a linear ramping function, then we will see an increase of pollution

(see Fig. 5.13 (b)).
Figure 5.13 (a) Stable mode for node 8 under the given input functions. (b) Growing
mode for node 8 if one of the input functions to nodes 3 and 5 is changed to a linear
ramping function

To analyze sensitivity, we set the input function to 0 for each of the six nodes one
at a time. Fig. 5.14 are results with respect to output y1 for node 7, from which
we can see that node 4 is the most sensitive to the variation of the input function
among all other nodes. The sensitivity ranking is in the following order from high
to low: nodes 4, 3, 5, 6, 1, 2.
Figure 5.14 Sensitivity for output y1
Fig. 5.15 are results with respect to output y2 for node 8, from which we can
see that node 5 is the most sensitive to the variation of the input function.
Figure 5.15 Sensitivity for output y2
We then generate noise at random (see Fig. 5.16), inject the same sequence
of noise to the most sensitive node and the least sensitive node. The results for
output y1 and y2 are shown in Fig. 5.17, from which we can see that the most
sensitive nodes act more like noise, while the least sensitive node for output y1
do not change much. Moreover, output y2 is more sensitive to noise than output
y1 .
Figure 5.16 The noise to be injected to inputs
Figure 5.17 Outputs with random noise added to the most sensitive node and the lease
sensitive node
5.5.7 Network Structures and Critical Nodes

The model constructed is a very small 8-node directional weighted network (see
Fig 5.18), where the size of each node is proportional to the amount of pollutants
it releases, and each link is weighted by the decay rates of pollutants related to the
distance.
In what follows we assume that node A connected to node B with a path has
influence on node B indirectly. We would like to find out which node does more
harm to the earth’s health. For this purpose, we define a notion of influencing
centrality and a notion of influenced centrality. . We divide nodes into two groups
Figure 5.18 A visualization of the network model
based on link directions. The first group consists of nodes with outgoing links, and
the second group consists of nodes with only incoming links. That is, the first group
is the set of nodes {1, 2, 3, 4, 5, 6} and the second group is {7, 8}. Take the Arctic
node (node 8) as an example. The node of West Europe transmits pollutants to
node 8 directly and the node of East North-America transmits pollutants indirectly
though West Europe. If a path has two or more links, the decay rate of the path
should be the product of the decay rates of each link.
The direct influencing centrality of node i, denoted by DIC(i), is the sum-

mation of the weights of its outgoing links. That is, DIC(i) = wij . Let pij be
i→j
a path from node i to node j of length greater than 1 (that is, all directions of
the links are the same that lead to j). Let w(pij ) denote the summation of weights
of all links on pij . Then the indirect influencing centrality of node i, denoted by
IIC(i), is the summation of w(pij ) for all different paths pij on the network. That
is,

IIC(i) = w(pij ).
pij
The influencing centrality of node i is the summation of its direct influencing

centrality and its indirect influencing centrality. We can similarly define influ-
enced centrality by reversing the direction of links and paths. Table 5.2 shows the
results.
Table 5.2 Influencing/Influenced centrality

Node Influencing Centrality Influenced Centrality Rank
Asian-Pacific 15 3
West America 8.82 5
East North-America 40.64 1
East South-America 4.5 6
West Europe 23.50 2
Indian Ocean 10 4
Antarctic Ocean 24.3 1
Arctic Ocean 17.4 2
These results agree with the common views. For example, they indicate that
North America is the most influencing region and the Antarctic is subject to pol-
lution more than the Arctic.
5.5.8 Comments
The model used ocean pollution as a unique marker of the earth’s health. Unlike
in the previous model (see Section 5.4) where multiple elements were used, this
approach does not need to consider interactions between multiple elements, which
allows the team to focus on pollutant transmissions by ocean currents. What stood
out the most was the simulations and network analysis, showing that the changes in
pollution inputs and the structure of the network would have real implications on
ocean pollution. The predictions would be more convincing if the input functions
were based on real data sets.
Exercises
5.1 Draw a diagram for the 63-node two-level network model constructed in Sec-
tion 5.4.1, including all the links for interactions between nodes. Explain why using
a conceptual two-level network modeling is an applaudable approach.
5.2 Modeling interactions between human populations as a flow that moves from
a higher-density node to a lower-density node is not flawless, for people may tend
to move from rural areas of low population density to a city with high population
density. Can you think about a better mechanism to model human interactions?
This would need a migration model for humans.
Exercises 101
5.3 The globe is a sphere. Thus, the distance between two points on the surface
of the earth is no longer Euclidean distance. Modify the models presented in this
chapter to take care of this problem.
5.4 Use the Ordinary Least Square method on the EPI data sets for 10 years[37]
to determine the values of the coefficients aik and bijk in Equation 5.3.
5.5 The sensitivity analysis given in Section 5.4.6 was rather limited, which might
be due to the time constraint. Suppose that you have more time available, how
would you improve the sensitivity analysis?
5.6 Incorporate your new model for human migration in Exercise 5.2 into the
two-level network model and do the following:
1. Derive new differential equations for modeling diffusion.

2. Validate your model using real data sets.
3. Compute future trends.
4. Predict tipping points.
5. Carry out sensitivity analysis.
5.7 Assumption 3 in Section 5.5.2 states that pollutants carried over from node
A to node B will remain in node B. In other words, pollutants only travel one link
and nodes receiving pollutants from other nodes will not pass them on. This is
obviously not a reasonable assumption. Explain why this assumption is needed in
the construction of Model 2.1.
5.8 Section 5.5.2 listed eight assumptions for constructing Model 2.1.
1. Try to justify these assumptions according to your understanding of ocean

science, and think about how you may revise the assumptions to make them
more reasonable.
2. Design a new Model 2.1 based on the revised assumptions you made.
5.9 Read about the Princeton Ocean Model (POM) from the references[44,45] or
other sources.
1. Describe POM in details.

2. Justify the 1-dimensional diffusion Equation 5.5 (a version of POM).
5.10 Describe how to solve Equation 5.5 numerically using the Four Point Implicit
Difference method.
5.11 Can you think of a way to obtain data sets of pollutant emission patterns?
5.12 Does it make sense to model pollutant transmission from one node to the
other node by ocean currents as an LTI system? Justify your answer.
5.13 Describe the LTI system in Fig. 5.11 under the Laplace Transform and write
a MATLAB program to compute numerical values for outputs y1 and y2 under the
input functions given in Section 5.5.5. Note: you will need to find attenuation co-
efficients for pollutants under a particular ocean current. You may also want to use
an approximation formula (e.g., Padé’s Approximation[50]) to compute exponential
functions.
5.14 Describe the pros and cons between the two-level network model presented
in Section 5.4 and the one-level network model presented in Section 5.5. Which
model would provide more accurate advise to the policy makers?
Chapter 6 Network Modeling of
Earth’s Health II
This chapter presents three different models for Earth’s health using discrete math
and computer science methods. These models do not involve differential equations
in any form. Instead, they use discrete methods to capture interactions between
nodes. These methods include communication networks, technology diffusion, and
machine learning.
6.1 Model 1: A Two-level Network over Forrester’s

World Model
The Outstanding-Winner team at the Beijing University of Posts and Telecommu-
nications[32] constructed a two-level network model, where they followed the pro-
tocol design paradigm in communication networks to model interactions between
top-level nodes, and used machine learning techniques to model the interactions
between sub-level nodes.
6.1.1 The Network

The network consists of two levels. The top-level network consists of nodes based
on geographic division. For example, we may use a node to represent a state or an
ocean. Fig. 6.1 depicts a network portion at the state level in the US.
The sub-level network consists of five nodes for each geographic zone, repre-
senting the following five elements based on the Forrester’s World Model[51] : envi-
ronment, resources, human population, industry, and agriculture. Fig. 6.2 depicts
the relations between these elements.
These five elements are measured as follows.
Environment
The environment is measured by the following three metrics:

104 Chapter 6 Network Modeling of Earth’s Health II
Figure 6.1 A network portion at the state level
Figure 6.2 Forrester’s World Model
1. Species diversity. An area with higher species diversity is generally considered

healthier than areas with lower species diversity. We uses Simpson’s Diversity
Index (SDI) to measure the species diversity[52]. Let SDI denote Simpson’s
diversity index, Ni the number of species i, and N the total number of species.
Then
1
SDI = 1 − (Ni (Ni − 1)).
N (N − 1) i
2. Air quality. We use the Air Quality Index (AQI) to measure air quality, which
6.1 Model 1: A Two-level Network over Forrester’s World Model 105
may be defined differently by different countries. For example, in the Peo-

ple’s Republic of China, AQI is divided into the following six categories:
(1) excellent, (2) good, (3) lightly polluted, (4) moderately polluted, (5) heav-
ily polluted, and (6) severely polluted. If AQI is less than 50 then it indicates
excellent air quality. If AQI is greater than 300 then it indicates severely pol-
luted air. The AQI is determined by the presence of five pollutants in the
air. They are ground-level ozone, particulate matter, carbon monoxide, sulfur
dioxide, and nitrogen dioxide, labeled by integers from 1 to 5, respectively.
Let Pi denote the AQI for pollutant i, then
C − Cl
Pi = (Ph − Pl ) + Pl ,
Ch − Cl
where C denotes the pollutant concentration, Cl the concentration breakpoint

lower than C, Ch the concentration breakpoint higher than C, Pl the index
breakpoint corresponding to Cl , and Ph the index breakpoint corresponding
to Ch . Then
AQI = max{Pi | i ∈ [1, 5]}.
3. Landform Change Ratio (LCR): The LCR is equal to the total area divided
by the area of landform changed, indicating the proportion of the land used
by humans (e.g. urban development versus farmland).
Resources
We will only consider the resource of land, forest, coal, and oil, labeled by integers
from 1 to 4, respectively. For a given resource type i, let ri denote the ratio of the
resource consumed over the available resource before consumption. We measure
resource using a resource index (ReI) defined by

4
ReI = ri .
i=1
Human population
The human population of a given zone is the number of people living in that zone.
Industry
We use the Industrial Production Index (IPI) defined by the US Federal Reserve
Board to measure industrial production, which is an economic indicator of the real
production output of manufacturing, mining, and utilities.
Agriculture
We use the Agricultural Production Index (API) defined by the Food and Agri-
culture Organization of the United Nations to measure agricultural production,
which is the sum of price-weighted quantities of different agricultural commodities
produced after deductions of quantities used as seed and feed.
6.1.2 Measuring Environmental Health

We define the following environmental health index (EHI) to measure environment:
EHI = ωs · NSDI + ωa · NAQI + ωl · NCLR,
where NSDI, NAQI, and NCLR denote, respectively, the normalized values of SDI,
AQI (the maximum value before normalization is 500), and CLR; and ωs , ωa , and
ωl denote, respectively, the weights of NSDI, NAQI, and NCLR.
We use the Analytical Hierarchy Process (AHP) to determine these weights.
First, we construct a pairwise comparison matrix to represent the relative impor-
tance of SDI, AQI, and CLR. Based on the study of Barnosky et. al.[53] , the earth
has experienced a worldwide shifts in species ranges, phenology and abundances,
which are concordant with ongoing climate change and habitat transformation.
Thus, it is reasonable to assign weights in decreasing order to species diversity,
air quality, and landform change. We label SDI, AQI, and CLR as 1, 2, and 3,
respectively. This gives a motivation to the following pairwise comparison matrix,
where the value at the ith row and jth column denotes the relative importance of
element i to element j (Note that the choice of values here is somewhat arbitrary):
⎡ ⎤
1 5/3 5/2
⎢ ⎥
A=⎢
⎣ 3/5 1 3/2 ⎥
⎦. (6.1)
2/5 2/3 1
The largest eigenvalue and the largest eigenvector (after normalization) of A are
λ = 3 and ω = [0.5, 0.3, 0.2].
We then check the consistency of the pairwise comparison matrix using the
consistency ratio (CR) (see, e.g., [54]), which can be calculated by
λ−r
CI = ,
r−1
CI
CR = ,
RI
where r is rank of A, CI the index of consistency, and RI the index of random

consistency. If CR is below 0.1, then the degree of inconsistency in this matrix is
acceptable. For this particular A, we have λ = n and so CR = 0. This implies that
it is reasonable to set ωs = 0.5, ωa = 0.3, and ωl = 0.2.
6.1.3 Sub-level Interactions as Bipartite Bayesian Belief Net-

works
A Bayesian belief network (BBN) models probabilistic dependencies of objects on
a connected graph, where each node represents a variable and the links represent
the causal relationships between connected nodes by these links. Each node may
assume one of a number of possible states or values. For a given node, the certainty
of (i.e., the belief in) each state in all possible states is determined by the certainty
of possible states of each node directly connected to it and its relationship with
each of these nodes. The certainty of each state of a node is updated whenever the
certainty of each state of any directly connected node changes. For a tutorial of
BBN the reader is referred to [55].
Construction of a bipartite BBN
We normalize the measures of elements in a zone node using their upper-bound

values and represent each element with five discrete states (values) of equal decre-
ments from the maximum value (see Table 6.1). Note that the values so chosen
are artificial without the support of real data sets, and we do so simply to make
computation manageable. Let P denote the current human population in a given
zone.
Table 6.1 Values of the five elements in each zone

Element Environment Resources Human Industry Agriculture
population
Maximum value 3P 1.0 4.0 100.0 100.0
Decrement 0.6P 0.2 0.8 20.0 20.0
For each zone node, we use a directed bipartite graph (V1 , V2 ; E) with one-
way links to model the states between the five elements (see Fig. 6.3), where each
node X in set V1 , called the X-set, is a 5-dimension vector with each component
representing a state of an element, and the elements are listed in the same order as
in Table 6.1. Each node in set V2 , called the Y -set, represents the changes of the
corresponding components, denoted by ΔX. Let S denote a subset of the X-set.
For each state node X ∈ S, the output of the BBN is a set of probabilities of ΔX
defined by
{P (ΔX | preconditionX ) : X ∈ S}, (6.2)
Figure 6.3 A bipartite BBN
where preconditionX denotes the set of vector states closely related to X.
Training the BBN model
To determine the values of probabilities in Formula 6.2, we will use the Forrester’s
World Model (see Fig. 6.2) to model the relations between the five elements and
use the method of gradient descent (see, e.g., [56]) based on real data sets.
6.1.4 Tipping Point for Environmental Health

According to Barnosky et. al.[53] , an extinction of at least 75% of Earth’s species
would cause a major biotic change on Earth. Thus, we set the tipping point of
the species diversity to be 0.25. In the People’s Republic of China, the empirical
tipping point of AQI is 300. An empirical landscape-scale study showed that the
critical threshold of landform change may lie between 50% and 90%[57] , and we use
an estimated value 0.68 as the threshold of LCR.
Let EHt denote the tipping point of the environment health. Then
EHt = 0.5 × 0.25 + 0.3 × (1 − 0.68) + 0.2 × (1 − 0.6)

= 0.301.
6.1.5 Top-level Interactions as Communication Networks

Interactions between top-level nodes may be viewed, in a way, as data exchange
between elements, which is similar to network communications between nodes. We
first introduce the terminologies of communication networks we will use later. For
an introduction to network communications, the reader is referred to standard
textbooks (e.g., see [58]).
• Protocol: A protocol is a set of rules that each party involved in a communi-

cation must follow for the communication to take place.
• Protocol data unit (PDU): Data is grouped into blocks for transmission. The
PDU is the size of the basic unit of data for transmission. At the physical level
a data block is transmitted as a sequence of signals.
• Protocol variables: A protocol may use variables, which are protocol variables.
• Protocol status: A protocol status includes waiting (for data) and transmitting
(data).
• Protocol action: A protocol action is the action a protocol entity performs to
transfer a protocol status.
• Service primitive: The user of a communication receives services using service
primitive. The protocols receive instruction through service primitive.
• Protocol event: A protocol event is a trigger of status transfer, including PDU
from peer protocol entities, service primitive from protocol users, and internal
timer signals.
A protocol status is transferred to another status when certain protocol event

occurs and actions are taken. We may view the old status as the precondition of
the new, while the new status as the postcondition of the old.
We view each zone node as a node in a communication network. The five
elements inside each zone node will change states (values) over time, and so the
elements may be viewed as protocol variables and the states as protocol status.
The following factors may cause element states to change:
• Internal interactions between elements inside a zone node;

• Outside impact from other zone nodes; or
• Policy changes that affect the network.
We may view internal interactions as internal timer signals, outside impact as

PDU, and policy changes as service primitive. Using the terminologies of precondi-
tion and postcondition, we assume that the changes of states within a zone node,
due to the actions taken in the current round of interactions, will not take effect
immediately. Instead, they will take place in the next round of interactions.
The links of the top-level network represent the geographical adjacency of the
nodes, modeled as signal channels for data exchange, which can be either population
relocation or air pollutant dispersion.
Population relocation
Within a zone node, if the environment degrades to the point not suitable for
humans to live in, we will call this node inactive. Otherwise, the node is called
active. People would move from a node that is about to become inactive to an
adjacent active node. Let Pn denote the population of zone node n before it becomes
inactive, m a neighboring zone node, EHi the health index of of zone node i. Let
ΔPnm denote the population relocating from zone n to zone m. Assuming people
will move to active adjacent nodes uniformly at random, and will not move further
away from the adjacent node, then we have
EHm
ΔPnm = Pn ,
EHi
i
where i is an active adjacent node to n.

The worst-case scenario is when none of zone n’s adjacent nodes is active. In
this case the population in node n will not survive. (Note that people may choose
to move further to non-adjacent nodes. For simplicity, we assume that they must
first move to an adjacent node that is active.) On the other hand, while humans
will not move into an inactive zone, it makes sense to assume that air pollutants
move between inactive nodes and active nodes.
We use Lowry’s migration method[59] to model how people move to adjacent
nodes, and subsequently move across the entire network. Let Mij denote the
population moving from zone j to zone i and Dij the distance between the centers
of these two zones. Then

EHi Pi · Pj
Mij = K · ,
EHj Dij
where K is a constant.
During a single iteration, population migration would take place at each zone
across the entire network. Let Pi denote the original population of node i before
the iteration takes place, and S the set of all adjacent nodes of node i. Then after
the iteration, the population Pi of node i can be calculated as follows:

Pi = Pi + Mij − Mji .
j∈S j∈S
Air pollutant dispersion
We use the Gaussian plume equation (see, e.g., [60]) to model air pollutant disper-
sion over the communication network. Suppose that the air pollution signal can
only transfer through a signal channel, which implies that if the zone nodes are not
adjacent, then one cannot directly affect the other.
Let C(x, y, z) denote the concentration level of pollutants on a certain zone
node, which could be the pollution source or an adjacent receiver. Let Q denote the
concentration level of air pollutants at the source, x the distance between the center
of the source and the center of the receiver in the downwind direction (Note that the
receiver could be the source itself), y the distance in the crosswind direction, and
z the vertical distance. Let μ denote the mean wind speed, and σy and σz denote
the standard deviation of the lateral and vertical concentration distribution. Then
„ «
y2 2
Q − 12 2
σy
z
+σ 2
C(z, y, z) = e z .
2πμσy σz
Note that each node can be both the pollution source and the receiver. Let
T Ci denote the total concentration of zone node i after the iteration, Si the set
of neighboring nodes of node i and node i itself, and Cij the concentration of
pollutants on node i with pollutant dispersion from adjacent node j, with node i
being the receiver and node j the source. Then

T Ci = Cij .
j∈Si
6.1.6 Global Health and the Tipping Point

We now describe how to model the health of the top-level network. Note that the
topology of a top-level network is similar to a 2-dimensional lattice and so we will
use a 2-dimensional lattice network to approximate the top-level network. We will
use the following concepts:
• Average degree: the average of degrees of all nodes.

• Giant component: a connected component of a given random graph that
contains a constant fraction of the nodes in the entire graph.
• Network Resilience: the ability of the network to provide and maintain an
acceptable level of service in the presence of faults and excessive service de-
mands.
• Site percolation: Each cell of the lattice becomes active with probability p,
which also means that each cell becomes inactive with probability (1 − p) (see
Fig. 6.4).
Figure 6.4 A 2-dimensional lattice network
It is known that in a lattice network, as the probability p of the site per-

colation reaches a threshold value, a giant component would suddenly emerge[61].
Link removal is the opposite process: When p drops below the threshold value,
the network becomes disconnected. The threshold values of the square, triangular,
honeycomb, and kagome are 0.5928716, 0.5, 0.765069, and 0.6546537. We use an
average value of 0.628 as the threshold value of a general lattice network, which
also indicates that once 37.2 percent of all the nodes have been removed, the lattice
network becomes disconnected, which implies the tipping point of Earth’s health.
To summarize, let GH denote the global health of the top-level network, Na
the number of active nodes, and Ni the number of inactive nodes. Then
Na
GH = .
Na + Ni
All the nodes are active if GH = 1, and the tipping point is approximately equal
to 0.628.
6.1.7 Prediction of Future Trends

The model can be used to predict future trends under different government policies.
The following examples are based on data from Meadows et. al.’s report[62] on a
developing zone adjacent to three areas of similar geographic, developmental, and
climate traits. These data are used as the initial data at time 0.
Scenario 1
Decision makers place the highest priority on industrial development and ignore
environment protection and resource consumptions. Under this policy, the future
trends are calculated and shown in Fig. 6.5 (a), which shows that the area’s in-
dustry is expected to develop rapidly in the first 50 years, but the resource and
environmental indexes will also decrease rapidly at the same time, and the envi-
ronment index is expected to reach its tipping point at a value slightly above 0.3
in about Year 65.
Scenario 2
Decision makers adopt a compromised strategy to keep the industrial development

under control, but still do not pay sufficient attention to environment protection
or resource reservation. Under this policy, the future trends are calculated and
shown in Fig. 6.5 (b), which shows that the environment index will not go below
the tipping point for 100 years. But still, the indexes of environment and resources
are decreasing at a constant rate, which are expected to reach their tipping points
soon after the 100 year.
Scenario 3
Decision makers decide to adopt sustainable policies to protect the environment and
reduce resource consumptions. Under this policy, the future trends are calculated
and shown in Fig. 6.5 (c), which shows that the industry, agriculture, and popu-
lation in the area will grow at a steady pase, while the environment and resources
will not be affected as much as the previous two scenarios.
6.1.8 Comments
This model used a Bayesian Belief Network to model interactions of elements within
a zone node for studying environment degradation with respect to the inherent
Figure 6.5 (a) Trends under Scenario 1 with an extremely negative policy on environment
protection and resource consumptions. (b) Trends under Scenario 2 with a moderate
policy on environment protection and resource consumptions. (c) Trends under Scenario
3 with a strongly positive policy on environment protection and resource consumptions
trait, and a communication network to model interactions between zone nodes as

signal transmissions. The model allows theoretical analysis for tipping points at the
sub-level network as well as at the top-level network using percolation theory on
lattice networks. The team also discussed in detail how policies and models could
6.2 Model 2: A Technology-diffusion Network Model 115
impact each other. Model validation was not performed using real data sets, neither
was sensitivity analysis. The team did not hide the deficiency. Instead, they stated
this difficulty and explained what data sets they would need and where they may
find them to validate their models and analyze sensitivity.
6.2 Model 2:A Technology-diffusion Network Model
The outstanding-winner team at the Peking University constructed a sequence of

three sub models to predict carbon dioxide (CO2 ) level, which is an important
indicator of the health condition of the earth. For example, elevation of CO2 will
increase temperature and decrease CO2 absorption by the ocean. On the other
hand, CO2 absorption by plants will increase at first, but will decrease eventually.
When the CO2 level is too high for humans to live in, the tipping point occurs.
The first sub model, called Technology Diffusion, is the most important of
the three, where the nodes represent individual countries measured by their CO2
emission-reduction technologies, and the links represent diffusions of technologies
between countries. The second sub model, called CO2 Regression, uses data sets
of economic growth and structural changes in energy consumption from the last
20 years to predict CO2 emissions. The third sub model, called CO2 Absorption,
describes the carbon cycle in terms of the nature’s ability to absorb CO2 .
6.2.1 Technology Diffusion Model

The carbon dioxide emission-reduction technology (ERT) is an important mech-
anism to keep the ecosystem of the earth healthy. The network of technology
diffusion consists of nodes representing individual countries. We assume that the
model starts from the beginning without any ERT in any country. However, each
country has a Technology Index (TI) that reveals its ability (speed) to produce
such technology. The links represent technology diffusions between countries.
Technology Index
Technology may be measured by the current technology levels of the country

and the amount of human resources and money invested in the research and de-
velopment of technologies. The technology level may be estimated by the coun-
try’s aggregated citations of ISI-journal articles in all fields. Table 6.2 shows an
example.
Table 6.2 Top-10 Technology Index

Country Technology Index
USA 20.53
Japan 19.28
China 16.11
Germany 15.80
UK 12.50
Russia 12.27
Brazil 10.94
Canada 10.86
India 10.59
Australia 8.69
Similarity
Technology is easier to diffuse between “similar” countries. Let FA and FB denote

the Economic Freedom Index (EFI) of countries A and B, respectively, and dAB
the geographic distance between the capital cities of them. Let SAB the denote the
similarity of A and B. Then
FA FB
SAB = α √ ,
dAB
where α is a balancing constant. The determination of EFI, published by the Her-
itage Foundation, depends on law, government, regulatory efficiency, and the open-
ness of the markets. Technology diffusion between countries often rely on trade and
direct foreign investment.
Accumulated Technology Index
The Accumulated Technology Index (ATI) is a function of time. Let AT IA (t) denote
the ATI of country A at time t. We make the following assumptions for simplicity:
• The unit of time is year, with 2013 being year 0, and AT IA (0) = 0.
• ATI depends only on technology produced by A and the inspiration it receives
from other countries.
• The inspiration that country B passes to country A in year t is measured by
AT IB (t) · SAB .
• The newly developed technology by country A in a year is measured by T IA A,
which is a positive constant.
Under these assumptions, we have

AT IA (t) = AT IA (t − 1) + T IA A + AT IB (t) · SAB ,
B
for every other country B.

This indicates that for every country A, AT IA (t) grows exponentially in t.
However, not all technology can be applied to reducing carbon emissions and the
marginal return of technology often decreases, and reduction rate (RR) needs to
be included, which is defined below:
RRA (t) = β ln (1 + AT IA (t)) , (6.3)
where β is a constant independent of A and t, but depends on the total number of

countries.
6.2.2 CO2 Regression Model

We use a heteroskedasticity-robust linear regression model to predict CO2 emission
trends. The use of this model was supported by the World-Bank data of CO2
emissions for most countries[63] . Let YA denote the amount of CO2 generated by
country A. Then YA is a function of a number of elements that contribute to CO2
emissions based on time, which was already shown in the data points. The linear
regression simply fits in these data for each country, which can be calculated by
YA = αA + βA t + eA ,
where eA represents the error. For example, after fitting in the data points for
India, US, and Germany, we came up one regress for India with e2India = 0.9271
(see Fig. 6.6 (a)), two regressions for US with e2US = 0.4785 from 1992 to 2010 and
0.8667 from 1992 to 2008 (see Fig. 6.6 (b)), and two regressions for Germany with
e2Germany = 0.6997 from 1992 to 2010 and 0.6802 from 1992 to 2008 (see Fig. 6.6
(c)).
6.2.3 CO2 Absorbtion Model

A NASA study[64] indicated that the temperature on Earth ultimately depends on
the atmospheric level of CO2 . While there is no precise formula to describe the
relationship of CO2 in terms of temperature, the data in the last 25 years seem to
indicate a linear relation (see Fig. 6.7).
Figure 6.6 Sample linear regressions
Figure 6.7 Relation between CO2 levels and temperatures
Let T (t) denote the global mean temperature and C(t) the atmospheric CO2
concentration at time t. Then, for the data in last 25 years, it is reasonable to use
the following linear function to define the relationship of T (t) in terms of C(t):
T (t) = η0 + η1 C(t),
where η0 and η1 are constants.

For simplicity, we assume that CO2 is distributed evenly in the atmosphere,
and the concentration of atmospheric CO2 , denoted by CCO2 , is proportional to its
mass. That is,

CCO2 (t) = μCO2 · mCO2 (t)
for a constant μCO2 .

Let E(t) denote the CO2 level at time t and A(t) the CO2 absorbtion amount
at time t. Then
mCO2 (t) = mCO2 (t − 1) + E(t) − A(t).
For simplicity, we assume that E(t) depends only on fossil fuels. We note that
A(t) can be divided into ocean absorbtion Aocean and land absorbtion Aland . That
A(t) = Aocean (t) + Aland (t). According to [65], we have
Aocean (t) = αocean − βocean · T (t − 1),

# $
2
Aland (t) = αland + γland · C(t − 1) − βland (T (t − 1) − 13.5) · Areaforest (t − 1),
where Areaforest (t) denotes the forest coverage at time t, αocean , αland , βocean, βland ,
and γland are coefficients depending on years. According to the World Bank’s data-
bank, for the year of 2008, we have
T η0 η1 αocean βocean αland γland βland Areaforest

14.5◦ C 10.65 0.01 15 000 500 1 000 30 3 000 1
6.2.4 Model Validation

The model of technology diffusion can be verified using historical data of an ar-
bitrary type of technology (e.g., waste disposal) using simulation. The model of
CO2 regression can be verified using simulation. Simulation results showed an a
47% increase of CO2 emissions over the next 25 years, which is in line with the
estimation of 50% increase of CO2 emissions estimated by the US Congress Office
of Technology Assessment. The model of CO2 absorbtion can be verified using the
World Bank data, confirming a reasonably good fit with an error rate of 2.94%.
6.2.5 Global Health Prediction

We select the following 15 countries and predict their CO2 levels: China, US, India,
Russia, Japan, Germany, Iran, Canada, UK, Saudi Arabia, South Africa, Brazil,
Mexico, Australia, and Indonesia. These countries are ranked among the top 20
nations in annual CO2 emissions, roughly evenly distributed across the continents
(see Fig. 6.8), and with a good mix of developed and developing countries.
Figure 6.8 The 15 selected countries
Data were collected from the following sources:
• Technology Diffusion Model: The Heritage Foundation’s 2013 Index of Eco-

nomic Freedom[66] .
• CO2 Regression Model: The US Energy Information Administration’s Inter-
national Energy Statistics[67] .
• CO2 Absorption Model: The World Bank’s databank[63] .
The results are listed below:
• Without emission-reduction technology in place, the tipping point is expected

to occur in 2051.
• With emission-reduction technology in place, the tipping point will not occur
for at least another 60 years.
• With emission-reduction technology in place, the maximum atmosphere CO2
concentration will reach 465 in 2062.
6.2.6 Structure Analysis

It is clear from the model that nodes with higher Tech Index are more important,
for they develop and apply new technologies faster. Table 6.3 lists the top-five
countries.
Each node has 14 links. Nodes with higher summation values of similarity are
more important to measure its ability to inspire other countries. Table 6.4 lists the
top-five countries.
We analyze the importance of a link by removing it from the network and

calculating the increase of CO2 in year 2023. Links with higher increases are more
important. Table 6.5 lists the top-five links.
Table 6.5 Top-five links of importance

Link CO2 increase
(US, Canada) 0.15
(Germany, UK) 0.08
(US, Germany) 0.05
(US, UK) 0.05
(US, Mexico) 0.04
These results coincide with the following intuitions:
• Countries with more advanced technology are more important.

• Countries with higher economic freedom indexes and better locations are more
important.
• Links between freer and more developed countries and shorter geographically
are more important.
6.2.7 Comments
The model is focused on technology diffusion on CO2 emission-reduction technolo-
gies between nations, with adequate explanations to the assumptions they made.
However, the team did not provide sensitivity analysis.
6.3 Model 3: A Data-driven Network Model 123
6.3 Model 3: A Data-driven Network Model
The Outstanding-Winner team at the Rensselaer Polytechnic Institute[35] con-

structed a data-driven network model to predict environmental outcomes from
human actions using machine learning techniques. In particular, they treated hu-
man action as an independent variable, instead of interactions between environ-
mental elements. Inaccuracies of the predicted data will be measured by the root
mean square error metrics, where error is obtained by first squaring the data, then
calculating the arithmetic mean of the difference between the actual and predicted
data, and finally taking the square root. This measure ensures that small errors in
different directions do not cancel each other.
6.3.1 Environmental Effects by Earth Damage Score

The annual Earth Damage Score (EDS) of a country is the economic loss by that
country induced by natural disasters and CO2 damage using data obtained from
the World Bank’s Databank[63] and Maplecroft’s global risk analysis[68]. Thus, EDS
represents compounded environmental effects in a format that is easily digestible
and relevant for environmental predictions. Fig. 6.9 depicts the economic loss for
China and the US compared to 1994.
Figure 6.9 EDS for China and the US
6.3.2 Human Actions

Listed below are economic variables used in the model. These variables can be
influenced directly by policies and environmental elements. The values of these
variables are available from the list of world development indicators in the World
Bank’s Databank[63] . We will use the first time derivative of EDS to run simulations
and obtain predictions, and the second derivative to represent policy changes in
each country.
Population growth
Population growth inherently increase agricultural and industrial growth, which

adds more bodies to shelter, transport, and contributes more waste[69] .
Agricultural growth
Modern farming causes deforestation, pesticides releasing, soil degradation, and

water pollution from runoff. Farming also consumes tremendous energy and is a
large contributing factor to the emission of CO2 and other toxic chemicals into the
atmosphere[70] .
Industrial growth
Industrial development is a double-edge saw to the environment. It is necessary

to reduce (and eliminate eventually) fossil fuel consumption and move toward sus-
tainable source of energy. On the other hand, it also continues to contribute to the
addition of toxic chemicals to the environment and CO2 emissions[71] .
GDP growth
There exist correlations between economic growth and the production rate of pol-
lutants by a country[71] .
Literacy rate
Literacy rate is a strong indicator of poverty within a country, which is correlated

to other factors such as agriculture and industrial growth[72] . It combines a number
of factors that are subject to human influence into an easily obtained number.
6.3.3 Assumptions
1. Environmental damage is influenced by human actions. Our model measures
the effect of human action on economic loss due to environmental factors. This
is widely perceived to be true.
2. Human action can be controlled, at least in part, by legal policy. This is
perceived to be true, and necessary for predicting the possible outcomes of
policy change.
3. Countries near each other share similar environmental effects. This is perceived
to be true, particularly for CO2 emissions.
4. Policy changes are likely to propagate across diplomatic links. This assumption
is useful to capture the ripple effect on the spreading of ideology between
nations.
5. Each country’s environmental invariants behave approximately as constant
throughout the time frame of the analysis. Our model does not take environ-
mental variables into account, and thus the model captures national invariants
such as size, latitude and longitude by solving for a constant.
6. The EDS is a good proxy for the true economic loss due to environmental
damage. Solving for a country’s true economic loss due to the environmental
factors is difficult (for example, it is hard to truly know how something such
as biodiversity loss will place a toll on the nation’s economy). Thus, this
assumption is necessary.
6.3.4 The Network Model
We develop a sequence of sub models, starting with a simple one, that lead to the
final network model.
Tikhonov regularization
Tikhonov regularization is a machine learning algorithm in the form of linear regres-

sion, including a regularization matrix to prevent overfitting of the data[73] . This
algorithm is especially well suited to problems with limited data and a relatively
small Vapnik-Chervonenkis dimension[74] .
Let matrix A denote the the economic variables, vector b the changes in EDS,
and the identity matrix the regularization matrix Γ . In addition, we add one binary
variable for each country as an additional feature to find optimal constant values
for each country, allowing the algorithm to take environmental and geographical
invariants into consideration. Let w be the vector of weights for each variable, with
the weights of each country binary variable equal to that country’s constant value.
Let ŵ be the set of weights that minimizes the objective function, which will be
used for prediction. We have the following minimization problem:
min
Aw − b
2 +
Γ w
2
w
ŵ = (AT A + Γ T Γ )−1 AT b
Geographic network
Tikhonov regularization treats each country as an independent entity, which cannot

handle more sophisticated parameters such as geographical proximity. We improve
the model by creating a global network using a weighted modification of the k-
nearest neighbor algorithm[75] . Each node in our network represents a country,
with a size proportional to the number of countries it shares border with (see Fig.
6.10). This measure is called degree centrality, which corresponds to the influence
of a country over its neighbors. Its color and relative position within the graph
correspond to the modularity class to which it belongs, a commonly used measure
which intuitively gives an indication of the influential community to which a country
belongs.
Figure 6.10 Geographic network
Two countries are connected by a link if they share common border or are
geographically close to each other. Label nodes by integers as 1, 2, · · · , n, where n
is the number of countries in the graph, and denote by d(i, j) the distance between
nodes i and j. We first use the Tikhonov regularization to compute the economic
damage score for each node i, denoted by EDSi,tr . We then compute a new eco-
nomic damage score based on the geographic network, denoted by EDSi,gn , which
is computed as follows:
%
EDSj,tr 1
EDSi,gn = .
d(i, j) + 1 d(i, j) + 1
d(i,j)2 d(i,j)2
These new values offers a 25% decrease in root mean square error for predicting
EDS.
Hybrid network
Policies of one country can easily influence its allies of similar political structures
and ideologies[76]. This leads to modeling another dynamic of diplomatic relation-
ships between countries. By designing an adjacency matrix with weights corre-
sponding to the perceived strength of the diplomatic relations and applying the
same algorithm we used for the geographic network model, we obtain a model that
is capable of simulating the ripple effect, demonstrating social influence[77] . We de-
fine an adjacency matrix by weighting each link’s strength by the number of times
each pair of countries were in common political and economic intergovernmental
organizations. This matrix was then normalized so that the maximum value of a
link’s weight could not exceed one (see Fig. 6.11). This value seems a good measure
Figure 6.11 Geographic network

of the diplomatic relationship between nations. We then apply an unsupervised

learning algorithm, k-means clustering with three means, to find relationships be-
tween any pair of countries with similar behavior. These similarities are also rep-
resented as edges in the modified adjacency matrix.
We apply a linear combination of the link weights in the two previous networks
and find that the minimum error was obtained when the scaling parameter for the
diplomatic network was nine times that of the geographic network, which results
in a 23% decrease in root mean square error for predicting EDS compared to the
geographic network model.
6.3.5 Measuring Centrality with PageRank

Because nodes with links of lower weight are less influential than those with higher
weight in the hybrid network, we cannot use degree centrality to measure impor-
tance of a node. Another common measure is betweenness centrality, which mea-
sures how many times a node acts as a bridge. However, it is not a good measure
either because there are a number of countries (e.g., Kyrgyzstan) that interact with
several distinct groups yet have little influential power.
Another choice is closeness centrality, which is essentially a measure of average
path length between a given node and any other node in the network. Again, this
is a bad choice because although it gives some sense of how quickly policy changes
could propagate through the network, it still neglects the weights of each connection
and thus does not provide a very good idea of a country’s influence.
Instead, we use the PageRank link analysis algorithm to determine our met-
ric of influential power[78] . This turns out to be a reasonable measure of relative
importance and it is represented in our graph by the size of each node. Color and
location again represent modularity class.
The biggest influences
As another method of identifying key nodes in our graph, we will solve for which
countries could make the largest overall effect on average EDS. To calculate this
for each country, we solve for the direction of the second derivative of economic
variables that would minimize average EDS. Because the economic variables could
be influenced by policies, we will compare the difference between implementing a
predicted optimal policy versus not implementing any policy. Not implementing
any policy is equivalent to setting the second derivative of the economic variables
to zero.
Our model predicts that the countries that could make the largest difference
in the world’s EDS are China, the United States, and to a lesser extent India.
The difference they make is several orders of magnitude larger than the maximum
possible difference made by the rest of the countries. Our model predicts that im-
plementing optimal policy in any two of the top three countries would be sufficient
to stabilize the EDS score for many generations, and implementing optimal pol-
icy in all three of China, the USA, and India, could bring the entire world’s EDS
down in slightly less than five years. Furthermore, this analysis gives us a human
interpretable direction of the optimal second derivative of our economic variables.
Based on the data, optimal policy greatly emphasizes a decrease in the growth
rate of the country’s population. This knowledge can be used to inform the policy
makers of the most effective measures to reduce EDS.

A certain amount of confidence in environmental models is necessary for policy
making decisions, and the currently policy makers do not have a sufficient level of
confidence to make the best decisions[79] . Because of this necessity, it is especially
important to achieve as much confidence in our predictions as possible, and we will
performe multiple forms of sensitivity analysis.
Cross validation
We first use 10-fold cross validation, a common standard for measuring the ro-
bustness of algorithms[80]. The method involves randomly partitioning the data
into tenths, and for each tenth, the remaining nine are used to predict its EDS.
The method shows that the cross-validation error is only slightly higher than the
training set error, demonstrating the robustness of the model (see Fig. 6.12 (a)).
Gaussian noise
As another form of testing the stability of our predictions, we then inject Gaussian
noise with increasing standard deviation to our input variables to verify that our
predictions are stable with perturbations in our data. We see that for low amounts
of noise, we achieve remarkably comparable accuracy to that of the model with
correct data, and for standard deviations of the noise less than 0.8, we obtain
comparable accuracy to the model, particularly for the environmental variables
(see Fig. 6.12 (b)).
Exercises 131
Figure 6.12 Sensitivity analysis
Prediction
Finally, as a third form of verifying model accuracy, we partition the last three
years of our data into a test set, and train the model, including clustering and
hyper- parameter optimization, on the remaining data. We predict the EDS of our
test set data given its economic variables[81] . This result is within 5% of the true
EDS, indicating that our model provides excellent accuracy for medium intervals
of time (less than 15 years) (see Fig. 6.12 (c)).
6.3.7 Comments
This is a data-driven network model based on a range of World Bank data on
economic losses affected by environmental factors, including population change,
agricultural growth, GDP growth, and literacy rates. The model starts with a linear
regression on the chosen economic indicators and then proceeds to a geographic
network. In particular, the team used simulations to understand how policy changes
in some of the prominent nodes would impacts economic losses.
Exercises
6.1 While the comparison matrix 6.1 defined in Section 6.1.2 is chosen somewhat
arbitrarily, try to argue that it is a reasonable choice. Can you come up with a
more convincing comparison matrix?
6.2 Using environmental indexes (such as SDI and AQI) to measure environment
elements has the advantage of knowing what values are unsuitable for humans
to live in. Under what conditions will these values be considered tipping points?
Justify your answer.
6.3 Think about how to incorporate the population relocation modeling intro-
duced in Section 6.1.5 to modeling the top-level network for reaction diffusion in
Section 5.4.3.
6.4 What are the major differences between pollutant dispersion in the air and
pollutant transmission in the ocean? Describe how you may follow the same mod-
eling approach described in Section 5.5 using air pollution instead of ocean water
pollution? Argue that using air pollutant dispersion may be easier for you can
obtain real data sets from public sources on air pollution.
6.5 In addition to the three scenarios given in Section 6.1.7, can you think of a
way to maintain economic growth while keeping human population stable?
6.6 Work out the details of gradient descent and compute the probability values
in Formula 6.2.
6.7 Other than the reasons provided in Section 6.1.5, can you think of different
reasons why modeling interactions between top-level nodes as network communi-
cations make sense? Is there any reason why you think network communications
may not be suitable to model such interactions?
6.8 Does it make sense to view a geographical network as a 2-dimensional lattice?
Justify your answer.
6.9 How to measure accurately a country’s technology index is an interesting
problem. Using the country’s aggregated citations of ISI-journal articles to indicate
its technology level is a workable approach but obviously insufficient. Can you think
of a few more approaches to measuring a country’s technology level?
6.10 Justify why Formula 6.3 makes sense for modeling technology reduction
rate.
6.11 Search for more information about Tikhonov regularization from the Inter-
net and explain why it is a good mechanism to predict economic loss for individual
countries.
6.12 What are the advantages of the hybrid model defined in Section 6.3.4?
Ⅱ
2013 年美国大学生
数学建模竞赛解题指南
1
(MCM) (ICM)

(COMAP)
, ,

,

MCM Ben Fusaro [1] : “

,

,

”
, 1927
(Putnam) 6 , , 3 ,
12

( ),

, ,

,
,

,

MCM 1984 , GNU
Ben Fusaro

,

,
,
, ,
,
,
,
Sol Garfunkel (COMAP ) Ben Fusaro FIPSE (the Department of
Education’s Fund for the Improvement of Postsecondary Education) ,
[2] :
,
, ,
, :
•

•
,
•
1 135
•
, ,

,

,

1984 , MCM ,
1985 ,
200 2013 , MCM/
ICM 6 000 , ,
90%
MCM
, A B A
, B ICM
, C ,
,

MCM/ICM :
(Unsuccessful Participants):

,
(Successful Participants):

,

(Honorable Mention): 25%
(Meritorious): 10%
(Finalist): 1%
(Outstanding Winner):
,
1%
(:

COMAP ,
,

)
, Fusaro Frank R. Giordano , Ben Fusaro
,
MCM ;
20
MCM Frank R. Giordano ,

MCM

! “”
, ,

,
, ,
, ,
!, "
, ,

,

136 1
2013 , "
,

, ,

, ,

,

,
10
,
, ,

A B C

, A B
, C
(
COMAP )
,
,

#

, , ,

, , ,

,

,

,

,

"
MCM/ICM ,
, $ 2013 MCM

ICM
, “ ” “ ” “ ” , “”
“ ”,
$ 2013 , “
”

2

,

, :
•
,

•

2.1
,
, %
,

, "
“”
MCM Ben Fusaro : “ ,
”
,
&

,

, ,
, , ( , )
,

,
,

,
,

,

,

,

,
,

,

, :
1.
,

, ' ,
,
,
138 2
MCM/ICM ,

2.

,

GNU Octave MATLAB, ,

,
( , Maxima Mathematica
,
'

UMAP , , '

3.

, , ,

,
,

),
2.2
, 4 ,

,

,
,
,

, ,

,

,

1.

, "

, , ,
, ), %
,
,
2010 MCM A “”, “)

‘
’”, ): “ ,

‘’,

, !”

), )

)
!,
),
),
,
)
,

2.3 139
)

,
,
) )
,

2.

2009

" “ ”, &
“” ( “

” ,

,

)
, “” " ,
“”
" ), (
)
2004 A "
, " “” (

#) , “”
3.

,

, ! ( ,
))
, ,

, 2005 “"”
: “
#", "
” , "

, # :
" ,

, ,

,
,
:
• ?
• !?
• ?
• "?
2.3

,
,
140 2

", #
UMAP

,

,

“” ,

,

!
“
"” ,

,
$
( #
, ,

)

,

“”

,

,

,

“"” ,
,
, ,
$"
,
, %

,
$
“
:

”[3] ,

(
&)
,
,

2011 A “ ” “
, ” , “

‘’ ”,
(,

, , A (%

A) , "
$ ,
%, ,
$
2.4
“” ,
"!,

2.5 141
,

,
“:

”[3]
, ,
,
" ,
( )

, :

, "

* ,
“: "
”[4] , " " ,
",
", #
, " '
,

, +

,

,
,

,

#

,

,

“
” “”
+ ,
K.
S. Cline
“MCM ”[5] :

, ,

,
,

:

, ——

2.5
,

,

!,
, ,

!, ,
,
142 2

", “: " ”
" , , ,
"
, ",
", (yield signs)

, "&, , '
%, %

,
,
#
(
)
“
:

” ,

,
#
, :

, ,
,
,

2.6
,
%
,

, #,
“ , ” [6] ,

+ &
, #
10% , & # 10%, (,
,
,
) , &

, ,
#,
#
,
#

,
,
# 2011 ##

(
), $

,

, #

,
,
143
2.7
MCM/ICM

,

,
!
!
MCM/ICM , [7]

,

, ,
,
:
•
• ,
•
•
•
• "
•
•

, , ,
#

", ,
,

,
,
,

,
2/3 ,

, ,

#(
MCM/ICM UMAP CD,
,
,

", "

3
MCM A ( ) ,

2013 A
3.1
A :
,
,
(
,
) ,

,
,
, ,

,

,

:
(1)
, W/L
(2) A
(3)

, :
(1) (N )
(2) (H)
(3) (1) (2) , (1) (2) p
(1 − p),
W/L p

MCM
,
1 ∼ 2 ,
3.2 145
3.2
, ,

,
,
, ,

(
3.2.1

,

1.

, %,
' &
-,

(, ( ,
# (
3.1) ,
&-#,
(

,
,
.$
;

, ,,

“”,
, “ ”
,
(
3.1
,

, “ ”
“ ” , ) ,
,

2. N

, , ,
,

$
,
146 3
3. H
H ,

&
, ,
H ,
pN + (1 − p)H
'
:
Tf − Tmax
(1) −1
Tf − Tmin
Tmax , Tmin , Tf
( Tmax Tmin)[8]
(2) # [9]
(3) C/P , , P , C
[10] :

" ( C ) C/P 1,
C/P 0,

,

3.2.2
,

,

,

% (!),

,
%

, (

UMAP ) ,

1994 ,
&

,

,

,
,

,

3.2 147

“”, (, “”),

% ,
, “!”,
&
,

,

, ,

, !
, #
,
, “!
%”[11]
“
)
”[12] , &
!, ,

“!”, !,
! (!
) “
”,
, “ ”,
,
,
,

, ',

,
( )

,

!:
• ;
• /;
• /;
• &
;
• #
, !,
,
)
$ , %
3.2.3
!,
, :
• ;
• /;
• /;
• &
*;
• ##;
148 3
• !#;
• !,

()
&
( ();
•

, ,

+

,

, # ,
, "
, !
( ),

/,
,
+ ,
;

#,

,
, ,

!, &

!! (!!
),
! ( ) , ,
( )
/&
, " ,
, ,

,
3.2.4
,
(N )

(H)
, (

),

, ,
$ p
(1 − p), ; W/L p # #
(
H ) ,
N H

pN + (1 − p)H ,

pN + (1 − p)H , H
3.3 149

, H N , N
H
, pN + (1 − p)H
,
#
MCM/ICM ,
, ,

3.2.5
,,
( )

, & ,

#

3.3
5 :
• ;
• ;
• ;
•
;
•

3.3.1
", ,
!,

[13]
,

, ,
,

!

,
,

, # 0 1 p
p
# , p

, ) p

150 3

, " ,
,
3.3.2

6 , 4 , 2 ,
" :
• (Bethel University), ( MAA )
: Michael Tetzlaff, Jake Smith, Anthony J. Burand, Jr.
: Cooking Up the Optimal Baking Strategy[14]
0: Nathan M. Gossett
• ),
: Kunrui Wang, Jing Xu, Wei Zheng
: Optimize the Shape for a Brownie Pan[15]
0: Ling Yang
• '(,
: Meng Wu, Chong Jin, Bowen Liu
: Applied Mathematicians Bake the Best Brownies: The Ultimate
Pan Design[16]
0: Xufeng Liu
• %*, ( INFORMS )
: Yankan Song, Ke Xu, Fan Yi
: The Design of the Ultimate Brownie Pan[17]
0: Hengxu Zhang
• ,
: Libin Wen, Jingyuan Wu, Cong Wang
: Solving the Brownie Pan Problem[18]
0: Yuehui Zhang
• " (University of Colorado, Boulder), (
SIAM )
: Christopher V. Aicher, Tracy Babb, Fiona Pigott
: The Best Rounded Rectangle for Ultimate Brownies[19]
0: Anne M. Dougherty
([20] Ben Fusaro
3.4 1: 151
3.4 1:
"[19] “”

, "
$,
(
) , )
,
,
$
3.4.1

W × L
A “” ,

1.
%
(
, #
,

, :
,
*
, N ∗ = W L/A ,
W , A/W , A,
N ∗ (1
A, 3.2)

,
47 cm ( Kenmore ), 760 cm2 (
), 47 cm × 16.2 cm
15.2 cm, +
3.2 N ∗ (1)

, +,
152 3
, ,
+, ! ,
(, &# ( 3.3) ,
k , k, 1/k,

, !
3.3 N ∗ (2)
2.

A
(" r = A/π ),

,
, ( 3.4) #
$,
3.4
3.
, , 4

3 : w, l r, w l “”
, r " ( 3.5)
A = πr2 + 4(lw + lr + wr).

3.4 1: 153

2(w + r) 2(l + r)
, WBR = (4 − π)r2
3.5
( 3.2),

1
,
,
,
, $
# ( W = 2(w + r)), (2(l + r))
, (N ∗ ) $
# " r, 1 W L − N ∗ A
1 N ∗ WBR ,

W L − AN ∗
r= .
N ∗ (4 − π)
L
r > , L, N ∗
2N ∗
% r

W L − AN ∗ L
r = min , .
N ∗ (4 − π) 2N ∗
2(w + r) = W
2(l + r)N ∗ = L, l w
3.4.2

1.
,

154 3
, ,

,
, !

T , T (x, y, z, t)
(x, y, z) t
T
T
= ∇ · (α∇T ),
t
α
α , :
T
= α∇2 T.
t

,
!
#, α ,

!

(1.5 × 10−7 m2 /s)
!)
(1 × 10−7 m2 /s) ,
α
!, !
&-, ,
,
# , Tb ;
Ti
Ω ),
T
= ∇ · (α∇T ), (3.1)
t

T (x, y, z, t) = Tb , (x, y, z) ∈ Ω,
T (x, y, z, 0) = Ti , (x, y, z) ∈ Ω.
2.
, ,

,
, W L H ( ,
W L , ),

α , Ω
0 x W,
0 y L,
0 z H.
3.4 1: 155
u(x, y, z, t) = T (x, y, z, t) − Tb , , (3.1) #

u
= α∇2 u
t
u(x, y, z, t) = 0, (x, y, z) ∈ Ω
u(x, y, z, 0) = Ti − Tb , (x, y, z) ∈ Ω.
# ,

∞ ∞ ∞ mπx nπy
lπz −λ2m,n,l αt
u(x, y, z, t) = Am,n,l sin sin sin e ,
m=1 n=1 l=1
W L H
mπ 2 nπ 2
2
lπ
λ2l,m,n = + + .
W L H
u(x, y, z, 0) = Ti − Tb , (x, y, z) ∈ Ω,

&
,
⎧
⎨ (Ti − Tb ) −64 , m, n, l ,
Am,n,l = mnlπ3
⎩
0, .

3.
,

&

(
A
')

, +

T
= ∇ · (α∇T )
t

T
= (αx Tx + αTxx ) + (αy Ty + αTyy ) + (αz Tz + αTzz ). (3.2)
t
', #, #

T 1 1
= αTr + αr Tr + αTrr + 2 (αθ Tθ + αTθθ ) + (αz Tz + αTzz ). (3.3)
t r r
156 3

, Δt, x y z
Δx Δy
Δz , T T (x, y, z, t), T (t+) T (x, y, z, t +
Δt), T (x+) T (x + Δx, y, z, t), T (x−) T (x − Δx, y, z, t),
T (y+) T (y−) T (z+) T (z−)
T T (t+) − T
≈
t Δt

3 :
T T (x+) − T (x−)
≈ ,
x Δx
2
T T (x+) − 2T + T (x−)
≈ .
x2 Δx2
y, z α (3.2) ,

T , !:

1 1 1
T (t+) = T · 1 − 2Δtα · + + +
Δx2 Δy 2 Δz 2

Δt α(x+) − α(x−)
T (x+) α + +
Δx2 4

Δt α(x+) − α(x−)
T (x−) α − +
Δx2 4

Δt α(y+) − α(y−)
T (y+) 2 α + +
Δy 4

Δt α(y+) − α(y−)
T (y−) 2 α − +
Δy 4

Δt α(z+) − α(z−)
T (z+) 2 α + +
Δz 4

Δt α(z+) − α(z−)
T (z−) 2 α − . (3.4)
Δz 4
', #!

!
: (x, y, z) ∈ Ω , T (x, y, z, t) = Tb (:
T
= −c(Tb − T ), c ,, n
n
2

)
,
# ,
:
3.4 1: 157
,
; , !
)
,

αwater , T < 100,
α(T ) =
αcake , T 100.

! , (
w l r ,
Tb , Ti ,
! ,
Δt Δx Δy
Δz
,

Ti , (x, y, z) ∈ Ω,
T (x, y, z, 0) =
Tb , (x, y, z) ∈ Ω.
t, ' t+ = t + Δt :
• (3.4) T ;
• # T ;
• # T ;
• (x, y, z) ∈ Ω ,
T (x, y, z, t+) = Tb , T ;
• α
,, ,
4.
$
, · :
A '

$ ( Δx = Δy), Δt Δx Δz :

2 1 1
αΔt + .
Δx2 Δz 2 2
,

(9 inch○
1
×11 inch) ,
, :
(1) : !
(
(2) : 3 , -! ,
!

(3)
(α): α , α #! #,
#! ; α , α # (

○
1 1 inch=2.54 cm.
158 3
) ,
#
,

(: ,
,
,

*, #
,
)
5.

α , ,
r = 0
α
20 inch 30 inch 3 inch, ) 3.6
, , ,

3.6 () ( )
6.
, ,

,
!!

,

T ,
,

!
t = 0, Δt, 2Δt, · · · , , T (t),
t

T (t)
T
t
, H :
• : 2

HV = T (x, y, z, t) − T
(x,y,z) t
3.4 1: 159
• *:

HA = T (x, y, z, t) − T

(x,y,z) t
,

,
γ1 :
l+r
γ1 = ,
w+r
γ2 , :
r
γ2 = .
l+w

,
; ,

,
3.7

3.7 HV () HA ( )

(HV : 108 ; HA : 104 )
3.4.3
N ,
H
,
, $
p pN (1 − p)H
, w l
r
p < 0.5 ,
' H $; p 0.5,
' N
$ ,

,
1. N (p 0.5)
N 1 N ∗ , p , N
(
) p !& f : [0.5, 1] → {1, · · · , N ∗ },
p
160 3
N S1 f (p)

(w, l, r) ), (w, l, r) ∈ S1 H ( HA HV ) ',

2. H (p < 0.5)
, p H & g : [0, 0.5] →

R S2 H g(p)
(w, l, r) ),
(w, l, r) ∈ S2 N ',

3.5 2:
[18] ,

,
, “ ”

, “1 ” $

:
(1) ,

(2) !!
(3) ,
3.5.1
"
,
" ρ0 L *
#(
( 3.8), # (ρ, ϕ, z) , t

t u(ρ, ϕ, z, t)

ut − a2 Δ3 u = 0,
a2

u0 (), u1 (
3.5 2: 161
3.8

) u0 u , u1 u ,
u(ρ = ρ0 ) = u0 ,
u(z = 0) = u0 , (3.5)
u(z = L) = u0 ,
u(t = 0) = u1 .
# ,
⎧ ⎡ ⎤ ⎫
∞
∞ ⎨ (0) 2 ⎬ (0)
x p π
2 2
x p2 π2
Anp exp −⎣ + 2 ⎦a t J0
n 2 n
u = u0 + sin , (3.6)
⎩ ρ0 L ⎭ ρ0 L2
n=1 p=1
J0 Bessel &, x(0)n J0 n , Anp
: 2 2
∞ ∞
p π
Anp sin = u1 − u0 .
n=1 p=1
L2
+, ,

3.5.2
, ,
))
, ,

,
, u
,

ut − a2 Δ2 u = 0,
a2 = 1.48 × 10−7 m2 /s,

[21]
u0
162 3
, u1 ,
u(t = 0) = u1
,
u = u0
200 ◦ C, * , u0 = 473 ,

4 20 ◦ C, u1 = 293 , ,

,
MATLAB PDE %, 3.9
3.10 ,
, , ,

3.9
3.10
!!

! 2000 s ,

,
( 1/8 m2 ), 3.1 ( *
) ,
3.5 2: 163
3.1

/K
(" 0.2 m) 30.85
() 0.4 m 0.1 m) 67.84
( 0.3545 m) 43.43
(0.2 m × 0.6285 m) 61.99
( 0.5387 m) 40.41
3.5.3
(
), , ,
# , 1/4
( 3.11) , A (
A = 0.04 m2 )
A
3.11
S , r ", α = A/S

)), α N
)))),

, +

, ( ,

N = α, N α )
,
A
, 3.1 ,
(
), " r , α

r ,
α = A/S = A/(A + 4r2 − πr2 ).
σ , PDE % r

σ, 3.12
, σ r &,
164 3
:
σ = −73.5983r + 64.9357.
3.12 σ
$ $, α

σ
α − αmin
α̃ = ,
αmax − αmin
σ − σmin
σ̃ = ,
σmax − σmin

0 1 α σ r &, α̃ σ̃
r &, &':
F (r, p) = p · α̃ + (1 − p)σ̃.
& F r (
p ) # 3.13 p < 0.4 ,
& F , F r !,
r
,
, F p 0.4 , r
3.2 (, r = 0.1128 , ) ,
! ,
3.13 ,
p , F , r = 0.0668 , F
0.592, F , " r = 0.0668

,

165
3.13 & F #
3.2 r
p r p r p r
0.0 0.1128 0.4 0.0942 0.8 0.0111
0.1 0.1128 0.5 0.0490 0.9 0.0049
0.2 0.1128 0.6 0.0308 1.0 0
0.3 0.1128 0.7 0.0193

3.1
3.4.1 , W × (A/W ) ,
N ∗ = W L/A (W/k) × (kA/W ), k ,
: # N ∗ − (k − 1)
3.2 3.4.2 ! (3.4)
3.3
# (3.4) !
3.4
3.5 , , &

) ,

3.5
ut − a2 Δ3 u = 0

(3.5), # (3.6)
(: u(ρ, ϕ, z, t) = v(ρ, ϕ, z)T (t), v(ρ, ϕ, z) = R(p)Φ(ϕ)Z(z),
166 3
#
T + k 2 a2 T = 0,
Δv + k 2 v = 0,
Φ + λΦ = 0,
Z + ν 2 Z = 0,
R + R /ρ + (k 2 − ν 2 − λ/ρ2 )R = 0,
λ k ,)
3.6
3.5.1 $
,
,

3.7
3.5.2 , )), ,
,
(

3.8
( PDE %)

, 3.9 3.10
3.9
3.5 ,
,
#
4 , ,
MCM B ,

2013
B
4.1
B :

, 2013
,
( ) 2025 ,

,

,

, , “

”
:

4.2
,

2025 #:

,

1. “ ”

, $

(*
)
, "# ,

168 4 , ,
2.
+

,
,

)
&
,
, ,
&
,
4.2.1

2013 B 2009
(HiMCM) :
: ,

2010 ,
2025
,
,

,

,

, HiMCM

, 2025 &
2025
,
,
#, #

() , 2025
, "#
*,

4.2.2
,, ( 1 ) +,
& " #

, (+) ),
,

4.3 169
,

,
, $

,

,
, *
, %, ,

4.2.3
),
,
,

,
, !#, #

,

#,
# ,

, ,
,
#

,

, ,
4.2.4

, ,

# #;

# #

, & ,
#
4.3
4.3.1
, &
“”

:
•
•
•
• %
170 4 , ,

,

2025 2025

, , ,
, ,
,
,
+ ,

,
,

,

*
-
' ,

,
2025
,
+
,
, )
,

4.3.2

""*:
• '(,
: Zhongxin Guo, Fan Wu, Beidan Wang
: Five Models for China’s Water Scarcities[23]
0: Zuguo He
• , (Colorado College), ( Frank R. Giordano )
: Aradhya Sood, Yukiko Iwasaki, Namgyal Angmo
: Water, Water Everywhere: Meeting the Demands of Saudi Arabia’s
Water Needs Today[24]
0: Andrea Bruder
• (, ( INFORMS )
: Wei Chen, Weizhi Liu, Cenying Yang
: Make Wise Use of Every Drop[25]
0: Hui Qu
•
, ( SIAM )
: Pengfei Gao, Boshuo He, Tianxin Zou
4.4 1: 171
: Quenching China’s Thirst in 2025: A Min-Cost-Max-Flow Network

Model[26]
0: Hao Wu
• " (University of Colorado, Boulder), (
MAA )
: Yue-Ya Hsu, Gregory S. Mcquie, David Thomas
: Sustainable Water Management for Saudi Arabia in 2025 and
Beyond[27]
0: Anne M. Dougherty
4.4 1:
[28] ,

7 , 5
, ", 2025
:
%
4.4.1

7 , 5 ( 4.1
4.1)
4.1 7
1 2 3 4 5 6 7
(') (*') (*) () () () (')
'( , * '
. +! , ' "
,' ", -, -
% ! .
. - -
,
%*
2025 ,

, 2025 2012
1/2 ,

; 1/2, "#;
1/2,
2012

2025 5 , /
172 4 , ,
4.4 1: 173
5 ,

&; , & 2000—
2011 5 , 5 ,
5
(:
, 2025
)
2025 5 , #.
2003—2010 5 5 5
,
5
, R
4.2 [28]
4.2

5 2025 2025 2025 2012
/m3 /108 m3 /108 m3
'( 3 149.813 96.585 30.422 450 26.8
. 1 918.063 123.633 23.713 512 15.4
1 ,' 7 945.012 226.973 180.330 083 157.2
% 3 979.139 186.114 74.057 228 124.3
. 2 608.287 743.351 193.887 197 419.0
539.657 264 742.7 0.726 615 409
4 628.391 334.257 154.707 117 294.8
2 +! 2 804.441 520.466 145.961 479 315.9
", 8 594.770 1 094.769 940.928 776 629.5
1 041.522 355 1 240.2 0.839 801 931
3 896.291 450.173 175.400 618 20.7
, 8 778.016 720.548 632.497 924 492.4
-, 6 702.647 255.697 171.384 740 745.0
3 ! 57.548 689.942 3.970 502 602.3
4 042.964 605.288 244.715 598 774.9
, 4 836.172 554.704 268.264 299 1 037.9
%* 10 499.480 224.170 235.366 528 347.6
1 940.945 129 4 020.8 0.482 726 106
, 8 876.614 247.789 219.952 731 328.0
4 ' 5 840.568 652.916 381.340 263 757.5
6 919.186 509.571 352.581 445 1 126.9
1 017.108 423 2 212.4 0.459 730 800
* 13 639.110 222.595 303.599 224 1 471.3
5 5.358 649.959 0.348 245 1 350.0
1 005.950 408.760 41.119 242 484.1
625.726 041 3 305.4 0.189 304 181
3 043.112 318.244 96.845 214 514.6
8 181.976 329.806 269.846 396 2 239.5
6 - 1 171.123 362.370 42.437 937 626.0
. 4 844.235 319.211 154.633 116 1 480.2
- 372.085 1 372.526 51.069 689 4 402.7
951.836 183 9 263.0 0.102 756 794
174 4 , ,

5 2025 2025 2025 2012
/m3 /108 m3 /108 m3
' 3 854.735 227.913 87.854 422 604.4
" 2 573.87 467.741 120.390 479 242.2
7 600.058 3 554.652 33.282 336 733.1
714.052 9 1064.383 76.002 577 8.8
- 2 664.348 2199.282 585.965 260 885.7
939.543 678 2474.2 0.379 736 350
7 ' 0.38, ,,

',
, ,

6 , ( d )
1/2 ( 1 d = 0.7266, 2 d = 0.8398),
6 , 3 4 5 6 1 2
4.4.2

, , # ,

,
&
⎛ ⎞
2
6
CO = w1 ⎝ Aij Lij ⎠ , (4.1)
j=1 i=3
w1 ( ), Aij i

j , Lij i j , Google
(() , ( 4.3) w1 = 1 !)
5,
( Aij ), CO
4.3 (: km)

1 (') 2 (*')
3 (*) 903.7 1 960.0
4 () 1 120.0 2 133.9
5 () 1 887.7 2 836.7
6 () 2 034.3 5 118.3
, ()

,

Di
< di , (4.2)
NETi
4.4 1: 175
Di i , NETi i

6
TOTi 2011 , i = 1, 2
NETi = TOTi + Aji
j=3

2

,
NETi = TOTi − Aij
j=1
,

2
TOTi − Aij > 0, i = 3, 4, 5, 6, (4.3)
j=1
:
Aij 0,
i, j.
,

1 2 d = 0.6, 3 6
d = 0.5, A31 = 1.3891 × 1010 m3 , A41 = 1.782 × 109 m3 , A42 =
1.6037 × 1010 m3 , A52 = 3.353 × 1010 m3 ,

#
#,
#
,
4.4.3

1 2 3

5 ($ )
Bi

, &
⎛ ⎞

2
6
CO = w1 ⎝ Aij Lij ⎠ + w2 Bi ,
j=1 i=3 i=1,2,3,5
w2 () , , 1 !

)
, ()
,

Di
< di ,
NETi
NETi i ,
176 4 , ,

6
NETi = TOTi + Aji + Bi , i = 1, 2.
j=3
$

2
NETi = TOTi − Aij + Bi ,
j=1

2
NETi = TOTi − Aij ,
j=1
, ,

2
TOTi − Aij + Bi > 0, i = 3, 5
j=1

2
TOTi − Aij > 0, i = 4, 6.
j=1
:
Aij > 0,
i, j,
Bi > 0,
i.
, ,

, 3 1
(1.3891 × 1010 m3 )
4 1 (1.782 × 109 m3 ) , 2
4.9567 × 1010 m3
4.4.4
, (
:
, &#; 0#,
%
, &#, λi
(
),
0 1 ;

, 1 i
#
Di
< di
NETi

4.5 2: 177
λi Di
< di .
i NETi
) λ , $
, , 2 λ2 = 0.95,
# ( λi = 1, i = 1, 3, 4, 5, 6; i = 1, i = 1, 2, 3, 4, 5, 6) ,
*, 2 &# 20%
4.5 2:

[26] 30
5
30 (5), 33
, 30 30 5 , 3

5
2025
, ,

, " ;
( ), ,
4.5.1

/, - $,
(# ) 2025 5

, 30%, 40%
/5 ,
,

, $ 2025 5
4.4 5
4.4

(: 108 m3 )

(Sj )

(Dj )
'( 9.269 65.727
. 4.750 66.941
,' 54.135 245.072
% 38.848 124.140
. 160.367 313.827
115.703 221.109
+! 156.434 179.065
", 273.793 707.434
12.355 176.209
, 164.496 653.675
-, 368.831 304.748
! 295.121 332.343
443.322 255.121
, 578.803 252.012
178 4 , ,

(Sj )

(Dj )
%* 128.399 319.070
' 173.053 340.841
391.604 359.371
* 676.361 394.774
694.642 504.610
712.266 346.429
210.404 121.502
952.611 298.618
- 374.449 232.874
. 769.636 182.686
- 1 767.216 59.514
' 178.558 126.121
" 84.255 194.094
277.354 53.843
4.004 93.939
- 367.684 684.183
4.5.2

:

% (), '
, /% 111 724 138 )( 6
' ,

)
,
( ,5
), (1$ ) ()7
), ,

,
, ,
$ %*,
!*!*5 5

5 ;
, 5
5
5
; ,

& ,
." ,
.",
() 70%

&
,
(
)
&,
4.5 2: 179

/
&
, 5
8",

, &

50 , /
4.5.3

2025 5
5

&,

5 ({v0 , · · · , v29 }), s , t , vd
:
• (s, vi ), vi 5 , 5 ,
0
• (vi , t), vi 5 , 5 ,
0
• (s, vd ), , 1 m3
• (vd , vi ), vi $5 , , 0
• (vi , vj ), vi vj 5 , , vi
1 m3 vj

,
4 9, 4.2
4.2
180 4 , ,
5
5 ,

4.5.4
, 5
, &# ( &),
, -
"
4.5.5
#
,
1.
, 40% ,

, 5
4.5 , ' ( )
* ( 5 #)

*,
* *
1/4,

, 30%, #
, , ,
4.6 3: 181
4 (,

,
2.

% ,
-
4.6 3:

,[24] (
Frank R. Giordano
)

,

,
70%
, :

,
4.6.1

, : ,
8"
, Cobb–Douglas & P
F1 F2 , Cobb–Douglas &
P = αF1β1 F2β2 ,
α, β1
β2 β1 + β2 = 1, (,
(,
() , β1 β2 (

, [29] &
:
W = F 0.6 H 0.4 , (4.4)
H &
*, F .

,

PH H + PF F,
182 4 , ,
PH ( ), PF

/ PH 0.21 )/m3 , PF 0.46 )/m3

# 64 ) 40% #, 25.6 )
#,
0.21H + 0.46F 2.56 × 109 . (4.5)
,
! (4.5) , (4.4) W
, $% , Mathemat-
ica , 3.9 × 109 m3 ( H * 4.9×
109 kWh , F * 3.3 × 109 )
, #
, ( W ,
$ ) , #

#
4.6.2

, %

(! ( -
, 4.3 ,
)
4.3 -

4.6 3: 183
L " d ,

" , :
F C = αLdβ ,
α β
(C *)
"
) (h )!) Cobb–Douglas &,

CP dθ hφ
, θ φ *,
, l 2,

γ (CP dθ hφ )l .
l
#

VC =γ (CP dθ hφ )l .
P l
,

T C = F C + V C = αLdβ + γ (CP dθ hφ )l . (4.6)
P l

d h #,
!
Q = ×
πd2
= · 2gh
4 √
= 3.5d2 h,
g :
√
3.5d2 h = Q. (4.7)
184 4 , ,
/ L α β γ CP θ φ
Q
#, [30] /

:
α = 827,
β = 0.6,
γ = 763,
θ = 0.8,
φ = 0.4385,
Q = 218.
L CP % ,
,
! (4.7) , (4.6)
% ,

,
%

,
88%, 3%, % 9%
3 876.41 m3 /s 29.81 m3 /s
89.74 m3 /s,
490 m 485 m
500 m Mathematica ,
" d = 5.37 m #
*

4.6.3
, Cobb–Douglas &
,
Q = F aEb, (4.8)
E , F a + b = 1, #

a b,

, a = 0.4 b = 0.6,
PF F + PE E,
PE , PF PF 0.038

)/m3 , PE 0.46 )/m3 ,
40% #,
185

25.6 )
,
!
0.038F + 0.46E 2.56 × 109
, &
Q = F 0.4 E 0.6

Mathematica , :
F = 3.339 × 109 ,

E = 2.65 × 1010 kWh,
Q = 7.65 × 109 m3 .

, # 10% " # 100%, ,

4.6.4
, /
!,

# (
,

4.1
4.4.2 ,
, 0%
4.2
(4.1)
! (4.2) (4.2) ,
d1 = d2 = 0.6, d3 = d4 = d5 = d6 = 0.5 (: LINGO )
4.3 5 2025
4.4
4.4
4.4.3 4.5.3 , 5
3 1 1.3891 × 1010 m3 , 4
1 1.782 × 109 m3 ; 5
4.5 , ?
4.5
4.4.3 4.5.3 , 5

,
?
186 4 , ,
4.6

,
? , ?
4.7 $
? ,
?
5 I
ICM ,
3

ICM

, 2013 ICM
, 2014 ICM

2013 C ,

,
&":

* (

:
ICM ,
ICM 2014
[31]
5.1
C :
:

,

,

:

, ,

, ,

,

,

,

,

188 5 I

22

,

,

:
(1)

, ,

(2)

,

,

,

,

,

,

(

)

,

,
,

,

,

: ICM ,
“

” ,
,
:
1:
(
), ,
()
()
,

,

5.1 189
,

,
( )

,

,

,

? (:
,
, ,
)

,

2: ,

, ,

(: !
,

)
"
?

?

?

?

?
3: #

,
? ,
?

? $ ?

?
4: 20 ( )
,
,
,

%,

!

:
• Anthony D. Barnosky, Elizabeth A. Hadly, Jordi Bascompte, Eric
L. Berlow, James H. Brown, Mikael Fortelius, Wayne M. Getz,
John Harte, Alan Hastings, Pablo A. Marquet, Neo D. Mar-
tinez, Arne Mooers, Peter Roopnarine, Geerat Vermeij, John
190 5 I
W. Williams, Rosemary Gillespie, Justin Kitzes, Charles Mar-

shall, Nicholas Matzke, David P. Mindell, Eloy Revilla, Adam B.
Smith. “Approaching a state shift in Earth’s biosphere”. Nature,
2012; 486 (7401): 52 DOI: 10.1038/nature11018.
• Donella Meadows, Jorgen Randers, and Dennis Meadows. Limits
to Growth: The 30-year update, 2004.
• Robert Watson and A. Hamid Zakri. UN Millennium Ecosystem
Assessment Synthesis Report, United Nations Report, 2005.
• University of California at Berkeley. “Evidence of impending
tipping point for Earth”. Science Daily, 6 Jun. 2012. Web. 22
Oct. 2012.
5.2

,

#
,

%

, ", .
,
, ,
,

,

,

,

, ,

),

, $
,

,

:
(1)
(2)
(3)
(4)
)

(5) #
5.2 191

, ,

, ,

( , (

5.2.1
"

,

:
(1)

&CO2 %;
%&-
;

)!*%
(2)

*
(3) :
:
+( &(

,
,

, ,
,
)
,

, )
),
,

,

,

,

5.2.2

,

,
,
)
/ , ,

),

, #

% ,

:
192 5 I
(1)

,
, ,
%
., , %, . :
1 ;
○
2
○ ;
○
3 %

(2) ( ) 9/

:

,

, ,

, ;, %

##%
, % "0)+-
' ) ,
, '&

*, (:

)
5.2.3
"# ,
), ) ,

' )
),

#

,
1
", ,

5.3

, 5.2 ,

,
1
,
,

5.3 193
, [7]
5.3.1
2013 C
[31]
1.
2

- $

)
,

2.

, 3 )
,

), %
"
/ )#
# ,
2, %

,
3.
, /

, +

,
),

4.

' ,

&
,

#, %
), &

,

, +

194 5 I
5.

' ,

", # !
#
,
"

,

6.
0 ,
,

7.

,
$,

+ ,

5.3.2

,

$%

" :
(1)

(2) )
"*
,
MCM/ICM , [7]

5 , 4 , 1 ,

" :
• '(,, ( INFORMS )
: Xu Yao, Yingnan Xiao, Chao Wang
: Two-tier Communication Network Model of Global Health[32]
0: Xinchao Zhao
• ',
: Yixiao Shu, Yuhui Wang, Chao Chen
: Two-layered Coupled Network Model of Earth’s Health[33]
0: Lulu Pan
5.4 1: 195
• '(,
: Ruofei Zhao, Xinqi Zhao, Minghe Hu
: Measuring Earth’s Health by CO2 : a Technology Diffusion Network
Approach[34]
0: Shulin Zhou
• ' , (Rensselaer Polytechnic Institute),
: Diogo Moitinho de Almeida, Eric Shapiro, Amanda Knight
: Saving the Green with the Greens[35]
0: Mark Holmes
• -,,
: Changyou Zhu, Weicheng Bai, Wenqing Yang
: You pollute, I pollute![36]
0: Yuan Ying
5
,
: ,

,
6 3

5.4 1:
'[33] , Lotka–Volterra

,

Lotka–Volterra
—

5.4.1

, , 9 ,
9 ,
/,/(!"' * '!
/+4 /,
/ ( 5.1 (a)),
( 5.1 (b))

(
)
7 (

)
,
7 :
(1)
(2) %
196 5 I
5.1 (a) 9 (b) (%

)
(3) %
(4) ()
(5) )!
(6) 5#
(7) *
2012 ! (EPI) [37] 7#
, !& )+ (YCELP)[38] %',
(CIESIN)[39]

,
<

, 7

%, ,

5.4 1: 197
5.4.2

Lotka–Volterra

[40] S1 S2 , 1 , 2, u1 (t) u2 (t)

t , #,
,
<*
, a1 a2 S1 S2 , b1 b2 S1
S1 S2 , c1 c2 S2 S1 S2
Lotka–Volterra , :
u1 (t) = u1 (t)(a1 + b1 u1 (t) + c1 u2 (t)),

u2 (t) = u2 (t)(a2 + b2 u1 (t) + c2 u2 (t)).
,
7 ,
, 9 1, 2, · · · , 9
,
%%)!5#
* 1, 2, · · · , 7 uik (t) i k ,
i = 1, 2, · · · , 9 k = 1, 2, · · · , 7; aik i k
#; bijk j i k
# , j = 1, 2, · · · , 7 Lotka–Volterra

⎛ ⎞
i
duk (t) 7
= uik (t) ⎝aik + bijk uij (t)⎠ , (5.1)
dt j=1
i = 1, 2, · · · , 9,
k = 1, 2, · · · , 7.
5.4.3
z1 z2 , n1 ∈ z1 n2 ∈ z2
n1 n2 , YCELP
CIESIN - , ,
,

,

,

, (:
, , ) ,

[41]
, f (ρ, θ)
198 5 I
# (%)
', ∇[f (ρ, θ)],
f # :
f 1 f
∇[f (ρ, θ)] = eρ + eθ ,
ρ ρ θ
eρ '), eθ
, '
, O
[42] , 5.2 $'
)$ f / ρ f / θ
θ
, f / θ = 0
5.2

, V , div V , :

2
f 1 f 1 f f
div V = div(∇f ) = div eρ = ρ· = + 2.
ρ ρ ρ ρ ρ ρ ρ
2
f / ρ2 , ,
1 f
div V ≈ .
ρ ρ
", , )
,
Δ k j % i #, k ∈{1, 2, · · ·, 7}
i, j ∈ {1, 2, · · · , 9}

uik (t) − ujk (t)
Δ= βki ,
rij
βki i % , rij j i ,

i
*) k %
6, %
5.4 1: 199

duik (t) i uik (t) − ujk (t)
= βk . (5.2)
dt rij
j=i
5.4.4

, (5.1)
(5.2),
⎛ ⎞
duik (t)
7 uik (t) − ujk (t)
= uik (t) ⎝aik + bijk uij (t)⎠ + βki , (5.3)
dt j=1
rij
j=i
i = 1, 2, · · · , 9,
k = 1, 2, · · · , 7.
6 EPI )[37] , aik bijk

5.4.5
, % ( MATLAB)
(5.3) , 7

$
0 100
1.

A / 5.3 A 7 50

,
3 , )
!% + , =
&# ,
5.3 50 A

200 5 I
+ , A
50
$
2.
, %

, , 6 #,
%
50 ,

#
, %/*#,
%, % #
A ,
%
, 38.76 59.86 ( 5.4) "
, +
$$%

5.4 A
5.4 1: 201
5.4.6
, , ,
6 )! * ,
PE PF )!#, #
) 5.5 (A )
5.5 A *)!

, , - & ,

202 5 I

, PE %+
,
PE
&# 0.9 (
, PE
1.1 ( #, '
, PF

, PF 0.9 (, PF & , + #;
PF , , PF 1.1 (, PF

2030
1
5.4.7
(5.3) EPI ),
,

*
6, ,

,
5.6 A
,
', - ',
+
5.6 A

5.4 1: 203
63
#
',
$, ,
,
$*
(

, *)
5.1 & A #
7
5.1 A

9 7.93
% 3 3.35
% 8 6.08
)! 10 4.65
6 2.56
5# 7 5.14
* 5 1.48
5.1 )! ,

, %, %
5.4.8

# (
),
#:
⎡ ⎤
7
duik (t)
= uik (t) ⎣ aik + âik + bijk + b̂ijk uij (t)⎦ + (5.4)
dt j=1

ui (t) − uj (t)
i i k k
βk + β̂k ,
rij
j=i
i = 1, 2, · · · , 9,
k = 1, 2, · · · , 7,
âik , b̂ijk β̂ki

5.7 A

*
204 5 I
5.7 A

*
5.4.9
- %

,

; ,
,
Lotka–Volterra

, ,
,
, 6.1.5

, 5.4.4 ,

5.5 2:

-,
, 8

,

5.5.1

7%/,

5.8 (a) [43] , 5.8 (b)

$
2,

! 5.8,
, 8
(
5.5 2: 205
5.8 (a) (b)
5.9 (a)):
(1) ,$, . *
(2) /,
(3) '*, %#
(4) *$
(5) $, '
(6) $, 8#
(7) '/$
(8) '4$

$ (
5.9 (b))
,
,

$ ,
206 5 I
5.9 (a) 8 (b)
5.5.2 2.1:
, :
(1)
&
,
,

(2)
"
(3)
,

(4) 9,
&
(5)
(6)

(7)

#
(8)
&
, ,
, C(x, t)
t x
, C(0, t)

t <
, Ci (x, t) i

5.5 2: 207
%
,

!0 (POM) [44,45] % !
0 (1),

(2),

" ,
*, ) [46] :

2
C C C
+u = DAB · + W (t) − k · C(0, t), (5.5)
t x x2
u ( [47]), DAB

( [48]), W (t)
t )
, k &, −kC(0, t) t
*
Pi (t) i
t
, (3),

Pi = Ci (0, t) + Cj (xji , t),
j
& , i j, xji j i ,

( ,)
t, Ci (0, t) i $

, $
# (

) )% ,

$",

(") (5.5)
,

5.5.3

%,

,

,
%+,

,/,

+ , ,,

: , 5
&,
#)
5.10
(1) &: g(t) = C, C
(2) &: g(t) = at + b
(3) &: g(t) = at2 + bt + c
(4) &: g(t) = Aeσt + B
(5) &: g(t) = A lnσ t + B
208 5 I
5.10

5.10 ,

#,
$
*
$, &

9,
$;
,

5.5.4 2.2:
5.9 , 1 6 $ , )

, 6
! 7 '/
, 8 '4 , #
,
,
',
3
,

%+,

+

, , #
%'
, 7 8
,
'

,

03#, #
(LTI)[49]
, 1 6 u1 , u2 , · · · , u6 ,
7 8 y1 y2 ( 5.11)
5.5 2: 209
5.11

i ∈ [1, 6], # ui i ,

xi
i #,
gj (j = 1, 2, · · · , 9)

% (
) ,$ LTI ,
& u(t) ,$ g(t), u(t) g(t) 4, u(t) ∗ g(t),
+∞
g(τ ) · u(t − τ )dτ.
−∞
#4 , y1

y2 , MATLAB
5.5.5

& &.&&9 )
, .&)
& 7 8

2 : !
, !
&
:
1: u1 (t) = 2t
2: u2 (t) = 6
3: u3 (t) = 8
4: u4 (t) = 1.5t
5: u5 (t) = 8
6: u6 (t) = 2t
5.12 (a) &'
: 3
, ,

,
210 5 I

,

3
( 5.12 (b) )
5.12 (a) & 7 (b) .&# & 7

5.13 (a) &&'4
,
, 8 3 5 , & &
&# .&,

( 5.13 (b))
5.13 (a) & 8 (b) 3 5 &#

.& 8
5.5.6
, 6 & 0 5.14 &
7
y1 , 4 &# 6
4 3 5 6 1 2
5.15 & 8
y2 , 5 &#

, 4> ( 5.16), 4>
, y1 y2 5.17 4>,
y1 # y2 y1 4>

5.5 2: 211
5.14 y1
5.15 y2
212 5 I
5.16 4>
5.17 4>
5.5.7
8
$ ( 5.18),

, $ ,

&

,
" A B, A B
,

, :
%
,
%
, )
{1, 2, 3, 4, 5, 6}, ) {7, 8} 8 ('') ,
8, '*
8
", &&
5.5 2: 213
5.18
DIC(i) i
,
$

, DIC(i) = wij pij i j 1 " (
i→j

), w(pij ) " pij
$
, IIC(i),
" pij w(pij ) ,

IIC(i) = w(pij ).
pij
i

"
, 5.2
5.2 /

, 15 3
8.82 5
'* 40.64 1
* 4.5 6
23.50 2
10 4

' 24.3 1
'4 17.4 2
214 5 I
, '*

, ''4

5.5.8
$

( 5.4 ), ,
)

,

#%
& , %

5.1 5.4.1 63
)
3
5.2
,

/,
9 -?

5.3 , , ,,
,
5.4 10 EPI )[37] (5.3)
aik bijk
5.5 -, 5.4.6
,
?
5.6
5.2 ,

:
(1)
,
(2)
(3) *
(4)
(5)
5.7 5.5.2 3 :

,

,

-

2.1
, ?
5.8 5.5.2 2.1, 8
215
(1) , ' ,

(2) 2.1
5.9 [44, 45]
, !0 (POM)
(1) !0
(2) POM
(5.5)
5.10 " (5.5)
5.11
)?
5.12
LTI %

,

%
-?
5.13 5.11 # LTI &
5.5.5 , MATLAB y1 y2 :

& , ( Padé

[50] ) &
5.14 5.4 5.5 5
+?
6 II
$ 3 ,
,

6.1 1: Forrester

'(%[32] ,

,

6.1.1

,
, 6.1 &
5 , Forrester [51] , 5

6.2 & 5

1.
3
(1) ,
, , ,# Simp-
son (SDI) ,[52] SDI Simpson , Ni
, i , N ,
1
SDI = 1 − (Ni (Ni − 1)).
N (N − 1) i
6.1 1: Forrester 217
6.1
6.2 Forrester
(2) %
% (AQI) %,
, % 6 :

% 50, % ;

% 300,
% 5

5
: %"&.
218 6 II
1, 1 5 Pi i
%,

C − Cl
Pi = (Ph − Pl ) + Pl ,
Ch − Cl
C
, Cl C , Ch C
, Pl Ph Cl Ch
AQI = max{Pi | i ∈ [1, 5]}.
(3) #
# (LCR) # ,
( 9
()
2.
)!)? 4 , 1 4

i, ri *, * *
(ReI)

4
ReI = ri .
i=1
3.
/
4.

(IPI) ,
%

(
5.

(API) ,

$
6.1.2
(EHI) , :
EHI = ωs · NSDI + ωa · NAQI + ωl · NCLR,

NSDI NAQI
NCLR SDI AQI ( 500)
CLR ; ωs ωa

ωl NSDI NAQI
NCLR $
(AHP) $
", %0 SDI AQI
CLR
Barnosky [53] , 5

##, #
5# #
, % #!&$
1 2 3 SDI AQI CLR, 0:
⎡ ⎤
1 5/3 5/2
⎢ ⎥
A=⎢
⎣ 3/5 1 3/2 ⎥
⎦ (6.1)
2/5 2/3 1
i j ) i ) j (

) 0 A λ = 3, ()
ω = [0.5, 0.3, 0.2]
, (CR) 0[54] , :
λ−r
CI = ,
r−1
CI
CR = ,
RI
r 0 A 1, CI , RI CR

0.1, 0 0 A,
λ = n,
CR = 0 $ ωs = 0.5 ωa = 0.3
ωl = 0.2

6.1.3

(BBN) ,
#,
,

,

#
#, [55]
1.
$ !,

220 6 II
, !& ! 5 ( 6.1) , !

,
, 6 P

6.1

5

3P 1.0 4.0 100.0 100.0
0.6P 0.2 0.8 20.0 20.0
, %

(V1 , V2 ; E) 5
( 6.3), V1 X ),
5
, !, 6.1
V2 Y ), X ) #, ΔX
6.3
S X ) ) X ∈ S, BBN ΔX

)
{P (ΔX | preconditionX ) : X ∈ S}, (6.2)
preconditionX X )
2.
(6.2) , Forrester ( 6.2)

* ( [56])
' ) 5
6.1.4
" Barnosky [53] , * 75% ,
%
# , 0.25
, % 300

, ,
#
50% 90% [57] ,
0.68 #

EHt ,
EHt = 0.5 × 0.25 + 0.3 × (1 − 0.68) + 0.2 × (1 − 0.6)

= 0.301.
6.1.5

,
$
,
%! ( [58])
• :

%
• ) (PDU): PDU )

,
• #: #, #
• : ( )
( )
• : #

• : ,

• : )

!

, *

,

5

##, #,

#:
• ;
•
;
• +#
222 6 II
,

PDU, +#
,

#

*

,

1.

, ,/,
,
!*
Pn
n
#, * m, EHi i
, ΔPnm n m
!*
, !*
EHm
ΔPnm = Pn ,
EHi
i
i n
!*
n
!* ,
(
!*, ,

!* )
,
,

Lowry [59] *,
Mij j i , Dij
,

EHi Pi · Pj
Mij = K · ,
EHj Dij
K

%,
Pi
i , S i
!*)%,
i Pi :

Pi = Pi + Mij − Mji .
j∈S j∈S
2.
/ ( [60])


, !*

C(x, y, z)

,

Q

, x

, (

<), y

&,, z ", μ
, σy σz
)"
„ «
y2 2
Q − 12 2
σy
z
+σ 2
C(z, y, z) = e z .
2πμσy σz

,

T Ci
i

, Si i
!*), Cij
i j

, i
,
j

T Ci = Cij .
j∈Si
6.1.6
$

'
,
:
•
:

• ' : ,
9
• 2:
%#&, $

• 8/: ) p,
, )
(1 − p) ( 6.4)

, p 8/, '
[61]
: p *, $$%

( 0.5928716 0.5 0.765069
0.6546537
0.628 , 37.2% , $,

GH
, Na
, Ni
Na
GH = .
Na + Ni
GH = 1,

, 0.628
224 6 II
6.4
6.1.7

+ + ,
Meadows [62] *5#
, 0
1. 1

$$ *
+
, 6.5 (a) ,
50 +
, 03*,
65 *
, 0.3
2. 2

1+
, $$

+, 6.5 (b)
,
100
3
100 ,
9 *,
!

3. 3

+, $$
+&,
6.5 (c) , ,
6.5 (a) 1 : $$ *' (b) 2 :

$$ * (c) 3 : $$ *'
+
, ,
3
6.1.8
$ ,

, ,

226 6 II
,

8/

"#

' ) ,
"-, ,

, ) ,
)
6.2 2:
'( 3
& (CO2 ) , , CO2
#!, * CO2
, 2 CO2
&# CO2 #,? #?

,
,
&&
!,

CO2 , 20 (
*# CO2 CO2 ,
& &%
6.2.1
&& (ERT) $

,

%
&&,
(TI),

1.

&#

ISI
/ 6.2 %
2.

“”
FA FB A
B ( (EFI), dAB " ,, SAB A
B
FA FB
SAB = α √ ,
dAB
6.2 2: 227
6.2 10

20.53
. 19.28
16.11
15.80
! 12.50
! 12.27
10.94
10.86
10.59
, 8.69
α
EFI &
(Heritage Foundation) ,
)+
9

#
3.
(ATI) #& ATIA (t) t A

ATI , :
• , 2013 0 , ATIA (0) = 0
• ATI A -
• B
t A - ATIB (t) · SAB
• A
TIA A,

, B,

ATIA (t) = ATIA (t − 1) + TIA A + ATIB (t) · SAB .
B
, A, ATIA (t) t

,

&#&,
!&, !& (RR) :
RRA (t) = β ln(1 + ATIA (t)), (6.3)
β A t ,

6.2.2 CO2

– CO2
$ CO2 )[63] YA A CO2 ,
228 6 II
CO2
&,

(
, :
YA = αA + βA t + eA ,
eA ,
,
e2India = 0.9271 ( 6.6 (a));
: 1992 —
2010 e2US = 0.4785, 1992 — 2008 0.8667 ( 6.6 (b));

: 1992 — 2010 e2Germany = 0.6997, 1992 — 2008 0.6802 (
6.6 (c))
6.2 2: 229
6.6
6.2.3 CO2

;;# (NASA) [64] ,
CO2
CO2 , 25

( 6.7)
6.7 CO2
T (t)
, C(t) t CO2 ,
25 , & T (t) C(t) :
T (t) = η0 + η1 C(t),
230 6 II
η0 η1
, CO2 , CO2 CCO2 ,
% mCO2 ,
CCO2 (t) = μCO2 · mCO2 (t),
μCO2
E(t) t CO2 , A(t) t CO2 ,
mCO2 (t) = mCO2 (t − 1) + E(t) − A(t).
, E(t) 0 A(t)

, Aocean (t)
Aland (t) [65],
Aocean (t) = αocean − βocean · T (t − 1),

# $
2
Aland (t) = αland + γland · C(t − 1) − βland (T (t − 1) − 13.5) · Areaforest (t − 1),
Areaforest (t) t )!, αocean αland βocean βland
γland
2008 $ ,
T η0 η1 αocean βocean αland γland βland Areaforest

14.5◦ C 10.65 0.01 15 000 500 1 000 30 3 000 1
6.2.4

(") '
CO2 ' ' :

25 , CO2 47%,
/4/
50% CO2 CO2 $
, 2.94%
6.2.5

15
CO2 : !
. $!
%
,
",
CO2 / 20 ,

/ ( 6.8),

6.2 2: 231
6.8
15
:
•
: &
(Heritage Foundation) 2013 (
[66]
• CO2 : @ [67]
• CO2 : $ [63]
5:
• &,
2051 #
• &, #
60
#
• &, CO2
2062 465
6.2.6
, %
,

6.3 5
6.3 5
. !
20.53 19.28 15.80 12.50 10.94

14 -
6.4 5
232 6 II
6.4 5
! .
1.19 1.15 1.14 1.14 0.90
,
2023 CO2 ,
CO2
6.5
5
6.5 5
CO2

(, ) 0.15
(, !) 0.08
(, ) 0.05
(, !) 0.05
(, %) 0.04
:
•
• (
•

6.2.7
$+6 CO2
&,
) ,

6.3 3:
' ,[35] ,

,
#,
,

3 ' $

4
6.3.1

$ [63]
Maplecroft [68] ,
(EDS) 27
& (*
6.3 3: 233
, EDS , 6.9

1994
6.9 1994
6.3.2
(#, #+
#$
)[63]
EDS , , EDS
+#
1.

# , %,
%'[69]
2.
)!&%&# *
"
,

*',
CO2
5 [70]
3.

11&# ( ) ?0
*,

,
/
5
CO2 [71]
4. GDP
(

[71]
234 6 II
5.
# ,

[72]
, #

6.3.3
(1) % %(
* '
(2) # )++#% ,

(3) !* , CO2
(4) +#%
%
(5) #

#, #, &
(6) EDS %(*'
%(* (, '
2*(
%*) ,
6.3.4
,
,
1. Tikhonov
Tikhonov , "

0[73] ,
Vapnik-Chervonenkis
[74]
0 A (#, b EDS #, Γ 0
0 , )#,
#, w #$,
)#$9 ŵ &$
),

:
min
Aw − b
2 +
Γ w
2
w
−1
ŵ = AT A + Γ T Γ AT b
6.3 3: 235
2.
Tikhonov
), ,
!* , $ k * [75]
, !*
( 6.10)
6.10
, *

0! ((
) -, -%

!*,,

1, 2, · · · , n , n , d (i, j) i j
, Tikhonov ( , EDSi,tr
236 6 II
, ( , EDSi,gn ,

:
%
EDSj,tr 1
EDSi,gn = .
d(i, j) + 1 d(i, j) + 1
d(i,j)2 d(i,j)2

EDS , ( &# 25%

3.
+%
+[76]
*0, $
, ,
:
[77]
*0,
+(<
0, $ 1 ( 6.11)

, 6
, 3 k−+, %

'*0 $
6.11
$

,
6.3 3: 237
9 (, ,

EDS &# 23%
6.3.5 PageRank

, $ $
,

,
,
,
(++), %,

,

"

+#%

,
$,

PageRank [78]

,
0!-

,
,
EDS , (#

EDS (#+,
+
(#
, EDS
=

,
3 EDS ,
, EDS *
5 , (#)
, +'*

& EDS
6.3.6
+
%
,

[79] ,
,
,
238 6 II
1.
10 1 &

[80] ,
10 , 1 ,
9 EDS
),
&
( 6.12 (a))
6.3 3: 239
6.12
2.

, #"2%
4>
, #4>
, , 0.8
4>, #
$ ( 6.12 (b))
3.
3

),
1 , +,
(#
,
) EDS [81] ' EDS 5%,

' ( 15 ) , %
(
6.12 (c))
6.3.7
$ , $
(* , #GDP

( ,

, !+ #(
*
240 6 II

6.1
6.1.2 0 (6.1) %
,

0?
6.2 ( SDI AQI)
SDI AQI , ,/

?
6.3 6.1.5 5.4.3

6.4

% ? 5.5

,

,

'
6.5 6.1.7 3 , (
$?
6.6 *, (6.2)
6.7 6.1.5 , )

?
,
?
6.8

?
6.9

ISI )
,

-?
6.10 (6.3) &
6.11
Tikhonov , )

(*
6.12 6.3.4 ?

A diffusivity, 33
Accumulated Technology Index (ATI), 116 direct influencing centrality, 99
Agricultural Production Index (API), 106 directed bipartitie graph, 108
Air Quality Index (AQI), 105 disturbance, 88
Analytical Hierarchy Process (AH), 106
average degree, 112 E
Earth Damage Score (EDS), 123
B Economic Freedom Index (EFI), 116
baking pan, 16 eigenvalue, 106
Bayesian Belief Network (BBN), 107 eigenvector, 106
betweenness centrality, 128 emission reduction, 121
brownie, 16 emission-reduction technology (ERT), 115
brownie pan, 16, 17 environmental health index (EHI), 106
exponential growth model, 49
C
circleness, 34 F
classification of papers, 3, 5 Finalist, 4
Cobb-Douglas function, 61, 65 finite difference approximation, 31
COMAP, 1, 5 FIPSE, 3, 134
communication protocol, 109 Forrester’s World Model, 103, 108
comparison matrix, 107 Frank R. Giordano Award, 4, 61
computer skill, 7 Fusaro Award, 4
consistency ratio (CR), 107 Fusaro, Ben, 2, 6, 134, 137
cross validation, 129
cylindrical coordinates, 36 G
Garfunkel Sol, 2, 134
D Gaussian noise, 129
degree of water demand, 49 Gaussian plume equation, 111
desalinization, 47, 53, 58 genetic algorithms, 22
242
giant component, 112 Meritorious, 3

gradient, 81 migration method, 110
gradient descent, 108
N
H network protocol, 109
half-pipe, 10 network resilience, 112
head loss, 57 numeric approximation, 29
heat distribution, 17, 33 numeric solutions, 30
heat equation, 21, 28, 30, 37
heteroskedasticity-robust linear regression, O
117 Outstanding Winner, 4
homogeneous boundary conditions, 29
Honorable Mention, 3 P
packing circles, 26
I packing rectangles, 25
impulse response, 95 packing rounded rectangles, 27
indirect influencing centrality, 99 PageRank, 128
Industrial Production Index (IPI), 106 physics baking, 19
influence degree, 86 pollutant transmission, 94
influence factor, 86 pollution diffusion, 91
influenced centrality, 98 propagation equation, 83
influencing centrality, 98 protocol action, 109
protocol data unit (PDU), 109
J protocol event, 109
judging, 4, 5, 13 protocol primitive, 109
protocol status, 109
L Putnam Mathematical Competition, 2
linear programming, 53, 55
linear regression, 49 R
linear-time-invariant system (LTI), 94 ramp function, 95
literacy rate, 124 reaction-diffusion equation, 81
local centrality, 86 reduction rate (RR), 117
logistic regression, 49 regression analysis, 55
Lotka-Volterra equations, 80, 81 relaxation function, 35
resource index (ReI), 105
M
mathematical model for baking, 20 S
mathematics skill, 7 scalar field, 81
243
self purification, 93 Tollbooth problem, 9

sensitivity, 13 traffic circle, 12
sensitivity analysis, 5 triage, 4
Simpson’s Diversity Index (SDI), 104
site percolation, 112 U
snowboarder, 13 Unsuccessful Participant, 3
stability, 13
stability test, 23, 32 V
step function, 95 variance, 18
stretchiness, 34 von Neumann analysis, 32
Successful Participant, 3 voting, 8
sweet spot, 8
W
T water strategy, 43, 44
team, 6 water transportation, 44
technology diffusion, 115 World Bank’s databank, 121
Technology Index (TI), 115 writing skill, 7
The fingerprint problem, 10
The Heritage Foundation, 121 Y
the World Bank’s Databank, 124 yield sign, 12
thermal diffusivity, 28
Tikhonov regularization, 125 Z
tipping point, 69, 71, 83–85, 108, 112, 113, zero order Bessel function, 37
121 zone node, 74

B
(reaction-diffusion equation),
(Unsuccessful Participant), 135 197
(Bayesian Belief Network, (variance), 146

BBN), 219 · (von Neumann analysis),
(scalar field), 197 157

(voting), 138 (protocol primitive), 221
C G
Cobb-Douglas &, 181, 184
(disturbance), 203
(Analytical Hierarchy Process, / (Gaussian plume equation),
AH), 219 222
(triage), 135, 136 (Industrial Production Index,
IPI), 218
D
(degree of water demand), 171 H
(Earth Damage Score, EDS), " (traffic circle), 142
233 (Successful Participant), 135
(zone node), 192 (desalinization), 181
!& (reduction rate, RR), 227 ! (mathematical model for
baking), 147
E & (snowboarder), 142
&& (emission-reduction tech- (half-pipe), 140
nology, ERT), 226 (environmental health index,
EHI), 218
F (regression analysis), 177
Forrester , 216, 220
Frank R. Giordano , 135, 181 J
Fusaro , 135 (Meritorious), 135
245
(Accumulated Technology In- (Putnam Mathematical

dex, ATI), 227 Competition), 134
(Technology Index, TI), 226
(computer skill), 138 Q

(indirect influencing cen- (homogeneous boundary con-

trality), 213 ditions), 155
& (step function), 209 (brownie), 144
( (Economic Freedom Index, (brownie pan), 144, 145
EFI), 226
# (local centrality), 203 R
(packing rectangles), 151
(yield sign), 142
' (giant component), 223
(heat equation), 148, 154, 155,
161
K
(thermal diffusivity), 154, 158
(baking pan), 144 (heat distribution), 145, 158
% (Air Quality Index, AQI), 217
S
L Simpson (SDI), 216
Lotka-Volterra , 197 " (tollbooth problem), 139
(tipping point), 188
(influenced centrality), 212
Bessel & (zero order Bessel func- ) (protocol data unit, PDU), 221
tion), 161 (mathematics skill), 137
(classification of papers), 135, 136 (numeric approximation), 154
#. (logistic regression), 173 (numeric solutions), 155
(water transportation), 168
M (water strategy), 167
,$ (impulse response), 209 & (relaxation function), 159
(sensitivity), 142, 157
(sensitivity analysis), 136 T
Tikhonov , 234
N (gradient), 198
(Agricultural Production In- * (gradient descent), 220
dex, API), 218 (Finalist), 135
(team), 137
P
(average degree), 223 W
(stretchiness), 159 2 (network resilience), 223

(judging), 135, 142 (network protocol), 221
246
"8/ (site percolation), 223 (consistency ratio, CR), 219

(stability), 142, 149, 157 (genetic algorithms), 149
(influencing centrality), 212 – (heteroskedasticity-

! (physics baking), 147 robust linear regression), 227
(influence degree), 203
X
(finite difference ap-
(linear programming), 175 proximation), 155
(linear regression), 173 (circleness), 159
# (protocol variables), 221 (packing rounded rectan-
(protocol event), 221 gles), 153
(protocol action), 221 (packing circles), 152
(protocol status), 221
.& (ramp function), 209 Z
(writing skill), 138
(direct influencing central-

ity), 213
Y " (fingerprinting), 140
(Honorable Mention), 135 (resource index, ReI), 218
(Outstanding Winner), 135 (sweet spot), 138

[1] Vanisko M, Anspach P. Judge’s Commentary: The Fusaro Award for the Geographic
Profiling Problem. UMAP Journal, 2010, 31(2): 153-156.
[2] Fusaro B A. Background and History of the MCM. UMAP Models: Tools for Teach-
ing, 1994: 1-6.
[3] Mou Y, Diao P, Quabili R. The Sweet Spot: A Wave Model of Baseball Bats. UMAP
Journal, 2010, 31(2): 105-122.
[4] Chang C, Fan Z, Sun Y. One Ring to Rule Them All: The Optimization of Traffic
Circles. UMAP Journal, 2009, 30(3): 247-260.
[5] Cline K S. Secrets of the Mathematical Contest in Modeling. http://www.carroll.
edu/kcline/mcm.pdf, 2009.
[6] Gong E, Wang X Y and Li R S. A half-blood half-pipe, a perfect performance.
UMAP Journal, 2011, 32(2): 109-122.
[7] Belanger J, Wang J. Write Right for the American Mathematical Contest in Mod-
eling. Beijing: Higher Education Press, 2013.
[8] Peking University, China. 2013 MCM Problem A entry, control number 21596.
[9] Northwestern Polytechnical University, China. 2013 MCM Problem A entry, control
number 17389.
[10] Tongji University, China. 2013 MCM Problem A entry, control number 17602.
[11] Sablani S S, Marcotte M, Baik O D, et al. Modeling of Simultaneous Heat and Water
Transport in the Baking Process. Lebensmittel-Wissenschaft und-Technologie, 1998,
31(3): 201-209.
[12] Wang L J and Sun D W. Modeling three conventional cooling processes of cooked
meat by finite element method. International Journal of Refrigeration, 2002, 25(1).
100–110.
[13] Black K. Judges’ Commentary: The Ultimate Brownie Pan Papers. UMAP Journal,
2013, 34(2-3): 141-149.
[14] Tetzlaff M, Smith J and Burand A J. Cooking up the optimal baking algorithm.
2013 MCM Problem A entry.
[15] Wang K R, Xu J and Zheng W. Optimize the Shape for a Brownie Pan. 2013 MCM
Problem A entry.
[16] Wu M, Jin C and Liu B W. Applied mathematicians bake the best brownies: the
248
ultimate pan design. UMAP Journal, 2013, 34(2-3): 115-140.

[17] Song Y K, Xu K and Yi F. The Design of the Ultimate Brownie Pan. 2013 MCM
Problem A entry.
[18] Wen L B, Wu J Y and Wang C. Solving the Brownie Pan Problem. 2013 MCM
Problem A entry.
[19] Aicher C V, Babb T and Pigott F. The Best Rounded Rectangle for Ultimate Brown-
ies. 2013 MCM Problem A entry.
[20] Tongi University, China. 2013 MCM Problem A entry, control number 17617.
[21] Hellman D R. Handbook of Food Engineering. 2nd ed. Boca Raton: CRC Press,
2007.
[22] Olwell D H. Judges’ Commentary: The Outstanding National Water Strategy Pa-
pers. UMAP Journal, 2013, 34(2-3): 189-195.
[23] Guo Z X, Wu F and Wang B D. Five Models for China’s Water Scarcities. 2013
MCM Problem B entry.
[24] Sood A, Iwasaki Y, Angmo N. Water, Water Everywhere: Meeting the Demands of
Saudi Arabia’s Water Needs Today. 2013 MCM Problem B entry.
[25] Chen W, Liu W Z and Yang C Y. Make Wise Use of Every Drop. 2013 MCM Problem
B entry.
[26] Gao P F, He B S, Zou T X. Quenching China’s Thirst in 2025: A Min-Cost-Max-
Flow Network Model. 2013 MCM Problem B entry.
[27] Hsu Y Y, Mcquie G S and Thomas D. Sustainable water management for Saudi
Arabia in 2025 and beyond. UMAP Journal, 2013, 34(2-3): 163-188.
[28] Sun Yat-Sen University, China. 2013 MCM Problem B entry, control number 17691.
[29] Zagourasa N G, Caourisa Y G, Kantos E T. Production and cost functions of water
lowtemperature solar desalination. Applied Economics, 1989, 21: 1177-1189.
[30] Field R, Heaney J P, Pitt R. Innovative urban wet-weather flow manage-
ment systems. General books. http://unix.eng.ua.edu/rpitt/Publications/
BooksandReports/Innovative/achap10.pdf.
[31] Arney C, Haye K, Wang J. Judges’ commentary for ICM 2013: modeling of Earth’s
health. UMAP Journal, 2013, 34(2-3): 229-244.
[32] Yao X, Xiao Y N and Wang C. Two-tier Communication Network Model of Global
Health. 2013 ICM entry.
[33] Shu Y X, Wang Y H, Chen C. Two-layered Coupled Network Model of Earth’s
Health. 2013 ICM entry.
[34] Zhao R F, Zhao X Q and Hu M H. Measuring Earth’s Health by CO2 : a Technology
Diffusion Network Approach. 2013 ICM entry.
[35] Almeida D M, Shapiro E, Knight A. Saving the green with the greens. UMAP Jour-
nal, 2013, 34(2-3): 211-228.
[36] Zhu C Y, Bai W C, Yang W Q. You pollute, I pollute! 2013 ICM entry.
249
[37] Environmental Performance Index and Pilot Trend Environmental Perfor-

mance Index, 2012, Release (2000-2010). sedac.ciesin.columbia.edu/data/set/
epi-environmental-performance-index-pilot-trend-2012.
[38] Yale Center for Environmental Law and Policy (YCELP). http://envirocenter.
yale.edu.
[39] Columbia University Center for International Earth Science Information Network
(CIESIN). http://www.ciesin.org.
[40] Lotka A J. Elements of Physical Biology. Baltimore: Williams and Wilkins, 1925.
[41] Britton N F. Reaction-Diffusion Equations and Their Applications to Biology. Salt
Lake City: Academic Press, 1986.
[42] Anderson J D. Fundamentals of Aerodynamics. New York: McGraw-Hill, 2001.
[43] Halpern B S, Walbridge S, Selkoe K A, et al. A global map of human impact on
marine ecosystems. Science, 2008, 319(5865): 948–952.
[44] The Princeton Ocean Model. http://www.aos.princeton.edu.
[45] Mellor G L. Users Guide for A Three-Dimensional, Primitive Equation, Numerical
Ocean Model. June 2004.
[46] Hundsdorfer W. Numerical Solution of Advection-Diffusion-Reaction Equations. Lec-
ture Notes, Thomas Stieltjes Institute, 2000.
[47] Speed of Ocean Current. http://hypertextbook.com/facts/2002/
EugeneStatnikov.shtml.
[48] Molecular Diffusion Coefficients in Sea Water. http://visumod.freeshell.org/
thermo/difcoef.html.
[49] D’Azzo J J, Houpis O H. Linear Control System Analysis and Design Convention
and Modern. 4th ed. Beijing: Tsinghua University Press, 2000.
[50] Baker G A, Graes-Morris P. Padé Approximants. Cambridge: Cambridge University
Press, 1996.
[51] Hritonenko N, Yatsenko Y. Mathematical Modeling in Economics, Ecology, and the
Environment. Dordrecht: Kluwer Academic Publishers, 2006.
[52] Simpson E H. Measurement of diversity. Nature, 1949, 163: 688.
[53] Barnosky A D, Hadly E A, Bascompte J, et al. Approaching a state shift in Earth’s
biosphere. Nature, 2012, 486: 52–58.
[54] Saaty T L. The Analytic Hierarchy Process. New York: McGraw Hill, 1980.
[55] Krieg M L. A Tutorial on Bayesian Belief Networks. DSTO Electronics and Surveil-
lance Research Laboratory, Australia, 2001. http://www.dsto.defence.gov.au/
publications/2424/DSTO-TN-0403.pdf.
[56] Russell S, Norvig P. Artificial Intelligence: A Modern Approach. 2nd ed. Upper
Saddle River: Prentice Hall, 2003.
[57] Jackson S T, Betancourt J L, Booth R K, et al. Ecology and the ratchet of events:
climate variability, niche dimensions, and species distributions. In Proceedings of the
250
US National Academy of Sciences. 2009, 106: 19685–19692.

[58] Stefan B. Objected-oriented Network Protocols. Upper Saddle River: Addison-
Wesley, 2000.
[59] Lowry I S. Migration and Metropolitan Growth: Two Analytical Models. San Fran-
cisco: Chandler Publishing, 1966.
[60] Turner D B. Workbook of Atmospheric Dispersion Estimates: An Introduction to
Dispersion Modeling. 2nd ed. Massachusetts: CRC Press, 1994.
[61] Rosowsky A. An analytical method to compute an approximate value of the site
percolation threshold. The European Physical Journal B –Condensed Matter and
Complex Systems, 2000, 15: 77-86.
[62] Meadows D, Randers J, Meadows D. Limits to Growth: The 30-year Update.
Vomont: Chelsea Green Publishing, 2004.
[63] World Development Indiators. World Databank. http://www.worldbank.org.
[64] NASA: http://www.nasa.gov/topics/earth/features/co2-temperature.html.
[65] Malhi Y, Meir P, Brown S. Forests, carbon and global climate. Philosophical Trans-
actions: Mathematical, Physical and Engineering Sciences, 2002, 360(1797): 1567–
1591.
[66] The Heritage Foundation. 2013 Index of Economic Freedom. http://www.heritage.
org/index/.
[67] US Energy Information Administration: International Energy Statistics. http://
www.eia.gov.
[68] Maplecroft. USA and China top global risk ranking for economic loss due to nat-
ural disasters linked to climate change. 2009. http://businessassurance.com/
downloads/2009/03/natural-disaster-and-economic-loss1.pdf.
[69] Goodland R. The case that the world has reached limits: more precisely that cur-
rent throughput growth in the global economy cannot be sustained. Population and
Environment, 1992, 13(3): 167–182.
[70] Trautmann N M. Modern Agriculture: Its Effects on the Environment. Cornell
Cooperative Extension. 2012. http://psep.cce.cornell.edu/facts-slides-self/
facts/mod-ag-grw85.aspx.
[71] Grubb M, Muller B, Butter L. The relationship between carbon dioxide emissions
and economic growth. Oxbridge study on CO2 -GDP relationships, 2004.
[72] Ahluwalia M S. Inequality, poverty and development. Journal of Development Eco-
nomics, 1976, 3(4): 307–342.
[73] Hoerl A E, Kennard R W. Ridge regression: biased estimation for nonorthogonal
problems. Technometrics, 1970, 12(1): 55–67.
[74] Vapnik V V. The Nature of Statistical Learning Theory. Heidelberg: Springer-Verlag,
2000.
[75] Coomans D, Massart D L. Alternativec k-nearest neighbour rules in supervised pat-
251
tern recognition: Part 1. k-Nearest neighbour classification by using alternative vot-

ing rules. Analytica Chimica Acta, 1982, 136: 15–27.
[76] Hartigan J A, Wong M A. Algorithm AS 136: A k-means clustering algorithm.
Applied Statistics, 1979, 28(1): 100–108.
[77] Cialdini R B. Influence: Science and Practice. Vol. 4. Boston: Allyn and Bacon,
2001.
[78] Page L, Brin S, Motwani R, et al. The PageRank citation ranking: bringing order
to the web. Standford InfoLab, http://ilpubs.stanford.edu:8090/422/.
[79] Pindyck R S. Uncertainty in environmental economics. Review of Environmental
Economics and Policy, 2007, 1(1): 45–65.
[80] Krogh A, Vedelsby J. Neural network ensembles, cross validation, and active learn-
ing//Tesauro G, Tourezky D S, Leen T K. eds. Advances in Neural Information
Processing Systems 7. Massachusetts: MIT Press, 1995.
[81] Schapire R E, Singer Y. Improved boosting algorithms using confidence-rated pre-
dictions. Machine Learning, 1999, 37(3): 297–336.

美国大学生数学建模竞赛题解析与研究 第5辑

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

美国大学生数学建模竞赛题解析与研究 第5辑

Uploaded by

Copyright:

Available Formats

内容提要

Ⅰ. ①美… Ⅱ. ①王… Ⅲ. ①数学模型 - 竞赛题 - 研

中国版本图书馆 CIP 数据核字（2014）第 039903 号

策划编辑 刘 英 责任编辑 刘 英 封面设计 李卫青 版式设计 杜微言

the Mathematical Contest in Modeling, MCM

While it is hard for me to believe, the Mathematical Contest in Modeling (MCM)

Sol Garfunkel, PhD

   

Undergraduate students who receive instruction and experiences in mathematical

time to be a skilled modeler since methodology to provide visualization and ﬁnd

Chris Arney, PhD

the Mathematical Contest in Modeling, MCM/

2012 MCM/ICM 5 026 

2013 MCM/ICM 1985

  

   3  4

 5  6  C  

  5  6 

  

;  

Forward by Sol Garfunkel

Forward by Chris Arney

I Mathematical Modeling for MCM/ICM 2013

Chapter 2 General Methodology · · · · · · · · · · · · · · · · · · · · 6

Chapter 3 The Ultimate Brownie Pan · · · · · · · · · · · · · · · · 16

3.2 How to Approach the Problem · · · · · · · · · · · · · · · · · · · · 17

Chapter 4 Water, Water, Everywhere · · · · · · · · · · · · · · · · · 43

4.5 Model 2: An Outstanding Water Strategy · · · · · · · · · · · · · · 55

Chapter 5 Network Modeling of Earth’s Health I · · · · · · · · · · 68

5.5.2 Model 2.1: Pollution Diﬀusion · · · · · · · · · · · · · · · · 91

Chapter 6 Network Modeling of Earth’s Health II · · · · · · · · · 103

6.3.6 Sensitivity Analysis · · · · · · · · · · · · · · · · · · · · · · 129

2 · · · · · · · · · · · · · · · · · · · · · · · · · · · 137

3 · · · · · · · · · · · · · · · · · · · · · · · 144

3.5 2:  · · · · · · · · · · · · · · · · · · 160

4 , , · · · · · · · · · · · · · · · · · · · · · · · · · · 167

5  I · · · · · · · · · · · · · · · · · · · · · · · · 187

6  II · · · · · · · · · · · · · · · · · · · · · · · 216

6.1.3     · · · · · 219

The Mathematical Contest in Modeling (MCM) and the Interdisciplinary Contest

ment of Postsecondary Education (FIPSE). This resulted in a three year grant.

The purpose of this competition is to involve students and faculty in

Finalist About one percent of the papers are Finalists.

2.1 Before the Contest

2.2 Reading the Problem

• what you will be modeling;

2.3 Literature Search

2.4 Creating Models

2.5 Finding Solutions

2.6 Stability and Sensitivity

2.7 Writing up the Solution

• A restatement of the problem, as you interpret it.

3.1 Problem Description

The exact statement of Problem A is given below:

3.2 How to Approach the Problem

3.2.1 Deciding What Needs to Be Done

Figure 3.1 Standard brownie pans

to consider diﬀerent shapes. The problem statement suggests “rectangular to circu-

What is N , the number of pans which can ﬁt in the oven?

How do you measure H, the distribution of heat for the pan?

Some measurements that were used are listed below:

3.2.2 Searching the Literature

• Convection heat from the air

3.2.3 Creating a Model

美国大学生数学建模竞赛题解析与研究第5辑

美国大学生数学建模竞赛题解析与研究第5辑

　　Ⅰ. ①美… Ⅱ. ①王… Ⅲ. ①数学模型 - 竞赛题 - 研

　　中国版本图书馆 CIP 数据核字（2014）第 039903 号

策划编辑刘英　　　　责任编辑　刘英　　　封面设计　李卫青　　　　版式设计　杜微言

2012 MCM/ICM 5 026

2013 MCM/ICM 1985

3 4

5 6 C

5 6

;

3.5 2: · · · · · · · · · · · · · · · · · · 160

5 I · · · · · · · · · · · · · · · · · · · · · · · · 187

6 II · · · · · · · · · · · · · · · · · · · · · · · 216

6.1.3 · · · · · 219