Geometric Programming

for Design and Cost Optimization

(with illustrative case study problems and solutions)
Second Edition
Copyright © 2011 by Morgan & Claypool

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in
any form or by any means—electronic, mechanical, photocopy, recording, or any other except for brief quotations in
printed reviews, without the prior permission of the publisher.

Geometric Programming for Design and Cost Optimization (with illustrative case study problems and solutions)
- Second Edition
Robert C. Creese
www.morganclaypool.com

ISBN: 9781608456109 paperback

ISBN: 9781608456116 ebook

DOI 10.2200/S00314ED1V01Y201011ENG012

A Publication in the Morgan & Claypool Publishers series

SYNTHESIS LECTURES ON ENGINEERING

Lecture #12
Series ISSN
Synthesis Lectures on Engineering
Print 1939-5221 Electronic 1939-523X
 Synthesis Lectures on
Engineering
Geometric Programming for Design and Cost Optimization (with illustrative case study
problems and solutions) - Second Edition
Robert C. Creese
2010

Survive and Thrive: A Guide for Untenured Faculty

Wendy C. Crone

August 2010
Geometric Programming for Design and Cost Optimization (with Illustrative Case Study Problems
and Solutions)
Robert C. Creese
2009

Style and Ethics of Communication in Science and Engineering

Jay D. Humphrey and Jeffrey W. Holmes
2008

Introduction to Engineering: A Starter’s Guide with Hands-On Analog Multimedia

Explorations
Lina J. Karam and Naji Mounsef
2008

Introduction to Engineering: A Starter’s Guide with Hands-On Digital Multimedia and

Robotics Explorations
Lina J. Karam and Naji Mounsef
2008

CAD/CAM of Sculptured Surfaces on Multi-Axis NC Machine: The DG/K-Based

Approach
Stephen P. Radzevich
2008

Tensor Properties of Solids, Part Two: Transport Properties of Solids

Richard F. Tinder
2007
iv
Tensor Properties of Solids, Part One: Equilibrium Tensor Properties of Solids
Richard F. Tinder
2007

Essentials of Applied Mathematics for Scientists and Engineers

Robert G. Watts
2007

Project Management for Engineering Design

Charles Lessard and Joseph Lessard
2007

Relativistic Flight Mechanics and Space Travel

Richard F. Tinder
2006
Geometric Programming for
Design and Cost Optimization
(with illustrative case study problems and solutions)
Second Edition

Robert C. Creese
West Virginia University

SYNTHESIS LECTURES ON ENGINEERING #12

M
&C Morgan & cLaypool publishers
ABSTRACT
Geometric programming is used for design and cost optimization, the development of generalized
design relationships, cost ratios for specific problems, and profit maximization. The early pioneers of
the process - Zener, Duffin, Peterson, Beightler, Wilde, and Phillips – played important roles in the
development of geometric programming. There are three major areas: 1) Introduction, History, and
Theoretical Fundamentals, 2) Applications with Zero Degrees of Difficulty, and 3) Applications with
Positive Degrees of Difficulty. The primal-dual relationships are used to illustrate how to determine
the primal variables from the dual solution and how to determine additional dual equations when
the degrees of difficulty are positive. A new technique for determining additional equations for
the dual, Dimensional Analysis, is demonstrated. The various solution techniques of the constrained
derivative approach, the condensation of terms, and dimensional analysis are illustrated with example
problems.The goal of this work is to have readers develop more case studies to further the application
of this exciting tool.

KEYWORDS
posynominials, primal, dual, cost optimization, design optimization, generalized de-
sign relationships, cost ratios, profit maximization, constrained derivatives, dimensional
analysis, condensation of terms
 vii

Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

PART I Introduction, History, and Theoretical

Fundamentals of Geometric Programming . . . . . . . .1
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Optimization and Geometric Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Geometric Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Brief History of Geometric Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.1 Pioneers of Geometric Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Theoretical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1 Primal and Dual Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

PART II Geometric Programming Applications with

Zero Degrees of Difficulty . . . . . . . . . . . . . . . . . . . . .11
4 The Optimal Box Design Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1 The Optimal Box Design Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
viii

5 Trash Can Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Problem Statement and General Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.3 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

6 The Open Cargo Shipping Box Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

6.1 Problem Statement and General Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.2 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

7 Metal Casting Cylindrical Riser Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.2 Problem Formulation and General Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.4 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

8 Inventory Model Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

8.1 Problem Statement and General Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
8.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
8.3 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

9 Process Furnace Design Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

9.1 Problem Statement and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
9.2 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

10 Gas Transmission Pipeline Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

10.1 Problem Statement and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
10.2 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
 ix
11 Profit Maximization Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
11.1 Profit Maximization and Geometric Programming . . . . . . . . . . . . . . . . . . . . . . . . . 47
11.2 Profit Maximization using the Cobb-Douglas Production Function . . . . . . . . . . . 47
11.3 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

12 Material Removal/Metal Cutting Economics Case Study . . . . . . . . . . . . . . . . . . . . 51

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
12.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
12.3 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

PART III Geometric Programming Applications with

Positive Degrees of Difficulty . . . . . . . . . . . . . . . . . 57
13 Journal Bearing Design Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
13.2 Primal and Dual Formulation of Journal Bearing Design . . . . . . . . . . . . . . . . . . . . 59
13.3 Dimensional Analysis Technique for Additional Equation . . . . . . . . . . . . . . . . . . . 63
13.4 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

14 Metal Casting Hemispherical Top Cylindrical Side Riser Case Study . . . . . . . . . . 67

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
14.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
14.3 Dimensional Analysis Technique for Additional Two Equations . . . . . . . . . . . . . . 72
14.4 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

15 Liquefied Petroleum Gas (LPG) Cylinders Case Study . . . . . . . . . . . . . . . . . . . . . . . 77

15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
15.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
15.3 Dimensional Analysis Technique for Additional Equation . . . . . . . . . . . . . . . . . . . 82
15.4 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
x
16 Material Removal/Metal Cutting Economics with Two Constraints . . . . . . . . . . . 85
16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
16.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
16.3 Problem Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
16.4 Example Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
16.5 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

17 The Open Cargo Shipping Box with Skids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
17.2 Primal-Dual Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
17.3 Constrained Derivative Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
17.4 Dimensional Analysis Approach for Additional Equation . . . . . . . . . . . . . . . . . . . . 98
17.5 Condensation of Terms Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
17.6 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

18 Profit Maximization Considering Decreasing Cost Functions of Inventory

Policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .103
18.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
18.2 Model Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
18.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
18.4 Transformed Dual Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
18.5 Evaluative Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

19 Summary and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

19.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
19.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
19.3 Development of New Design Relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

20 Thesis and Dissertations on Geometric Programming . . . . . . . . . . . . . . . . . . . . . . 119

Author’s Biography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
 Preface
The purpose of this text is to introduce manufacturing engineers, design engineers, manufac-
turing technologists, cost engineers, project managers, industrial consultants and finance managers
to the topic of geometric programming. I was fascinated by the topic when first introduced to it
at a National Science Foundation (NSF) Short Course in Austin, Texas in 1967. The topic was
only for a day or so of a three week course, but I recognized its potential in the application to riser
design in the metal casting industry during the presentation. I was fortunate to have two of the
pioneers in Geometric Programming make the presentations, Doug Wilde of Stanford University
and Chuck Beightler of the University of Texas, and had them autograph their book “Foundations
of Optimization” for me which I fondly cherish even now.
I finally wrote my first publication using geometric programming in 1972 and have written
several journal papers using geometric programming on metal cutting and metal casting riser design
problems, but was able to teach a complete course on the topic.Thus, before I retire, I decided to write
a brief book on the topic illustrating the basic approach to solving various problems to encourage
others to pursue the topic in more depth. Its ability to lead to design and cost relationships in an
integrated manner makes this tool essential for engineers, product developers and project managers
be more cost competitive in this global market place.
This book is dedicated to the pioneers of geometric programming such as Clarence Zener,
Richard Duffin, Elmor Peterson, Chuck Beightler, Doug Wilde, Don Phillips, and several others
for developing this topic. This work is also dedicated to my family members, Natalie and Jennifer;
Rob, Denie, Robby and Sammy, and Chal and Joyce.
I also want to recognize those who have assisted me in reviewing and editing of this work
and they are Dr. M. Adithan, Senior Professor in Mechanical Engineering and Dean of Faculty at
VIT University, Vellore, Tamil Nadu, India and Dr. Deepak Gupta, Assistant Professor at Southeast
Missouri State University, USA.
I used the first edition of the book to teach a course on geometric programming at the MS level.
The students, Yi Fang, Mohita Yalamanchi, Srikanth Manukonda, Shri Harsha Chintala, Kartik
Ramamoorthy, and Joshua Billups developed various examples, both original and from the literature
which are included in this new edition and they taught me a lot about geometric programming. One
of the important items presented during the course was the dimensional analysis technique for the
xii PREFACE
obtaining of additional equations when the degrees of difficulty was positive, and no reference to
this technique could be found in the literature.

December 2010 Dr. Robert C. Creese, CCE

Industrial & Management Systems Engineering
West Virginia University, Morgantown, West Virginia, USA
 PART I

Introduction, History, and Theoretical

Fundamentals of Geometric
Programming
 3

CHAPTER 1

Introduction
1.1 OPTIMIZATION AND GEOMETRIC PROGRAMMING
1.1.1 OPTIMIZATION
Optimization can be defined as the process of determining the best or most effective result utilizing a
quantitative measurement system. The measurement unit most commonly used in financial analysis,
engineering economics, cost engineering or cost estimating tends to be currency such as US Dollars,
Euros, Rupees, Yen, Won, Pounds Sterling, Kroner, Kronor, Pesos or specific country currency. The
optimization may occur in terms of net cash flows, profits, costs, benefit/cost ratio etc. Other mea-
surement units may be used, such as units of production or production time, and optimization may
occur in terms of maximizing production units, minimizing production time, maximizing profits,
or minimizing cost. Design optimization determines the best design that meets the desired design
constraints at the desired objective, which typically is the minimum cost. Two of the most impor-
tant criteria for a successful product are to meet all the functional design requirements and to be
economically competitive.
There are numerous techniques of optimization methods such as linear programming, dynamic
programming, geometric programming, queuing theory, statistical analysis, risk analysis, Monte
Carlo simulation, numerous search techniques, etc. Geometric programming is one of the better tools
that can be used to achieve the design requirements and minimal cost objective. The development
of the concept of geometric programming started in 1961. Geometric programming can be used not
only to provide a specific solution to a problem, but it also can in many instances give a general solution
with specific design relationships. These design relationships, based upon the design constants, can
then be used for the optimal solution without having to resolve the original problem. A second
concept is that the dual solution gives a constant ratio between the terms of the objective function.
These fascinating characteristics appear to be unique to geometric programming.

1.1.2 GEOMETRIC PROGRAMMING

Geometric programming is a mathematical technique for optimizing positive polynomials, which
are called posynominials. This technique has many similarities to linear programming, but has
advantages in that:

1. a non-linear objective function can be used;

2. the constraints can be non-linear; and

4 1. INTRODUCTION
3. the optimal cost value can be determined with the dual without first determining the specific
values of the primal variables.

Geometric programming can lead to generalized design solutions and specific relationships
between variables.Thus, a cost relationship can be determined in generalized terms when the degrees
of difficulty are low, such as zero or one.This major disadvantage is that the mathematical formulation
is much more complex than linear programming and complex problems are very difficult to solve.
It is called geometric programming because it is based upon the arithmetic-geometric inequality
where the arithmetic mean is always greater than or equal to the geometric mean. That is:

(X1 + X2 + . . . Xn )/n ≥ (X1 ∗ X2 ∗ . . . ∗ Xn )(1/n) . (1.1)

Geometric programming was first presented over 50 years but has not received adequate
attention similar to that which linear programming has obtained over its history of less than 70 years.
Some of the early historical highlights and achievements of geometric programming developments
are presented in the next chapter.

1.2 EVALUATIVE QUESTIONS

1. What is the most common unit of measurement used for optimization?

2. The following series of costs ($) were collected: 2, 4, 6, 8, and 10

(a) What is the arithmetic mean of the series of costs?

(b) What is the geometric mean of the series of costs?

3. The following series of costs (€) were collected: 20, 50, 100, 500, and 600

(a) What is the arithmetic mean of the series of costs?

(b) What is the geometric mean of the series of costs?

4. The following series of costs (Rupees) were collected: 5, 7, 8, 12, 16, and 18

(a) What is the arithmetic mean of the series of costs?

(b) What is the geometric mean of the series of costs?

5. What is the year recognized as the beginning of geometric programming?

REFERENCES
[1] R.J. Duffin, E.L. Peterson and C. Zener, Geometric Programming, John Wiley and Sons, New
York, 1967.
 5

CHAPTER 2

Brief History of Geometric

Programming
2.1 PIONEERS OF GEOMETRIC PROGRAMMING

Clarence Zener, Director of Science at Westinghouse Electric in Pittsburgh, Pennsylvania, USA,

is credited as being the father of Geometric Programming. In 1961 he published a paper in the
Proceedings of the National Academy of Science on “A mathematical aid in optimizing engineering
designs” [6] which is considered as the first paper on geometric programming. Clarence Zener is
better known in electrical engineering for the Zener diode. He later teamed with Richard J. Duffin
and Elmor L. Peterson of the Carnegie Institute of Technology (now Carnegie-Mellon University,
USA) to write the first book on geometric programming, named “Geometric Programming” in
1967 [1]. A report by Professor Douglas Wilde and graduate student Ury Passey on “Generalized
Polynomial Optimization” was published in August 1966. Professor Douglas Wilde of Stanford
University and Professor Charles Beightler of the University of Texas included a chapter on Geo-
metric Programming in their text “Foundations of Optimization” [2]. I attended an Optimization
Short Course at the University of Texas in August 1967 and that is when I first became interested
in geometric programming. I realized at that time that geometric programming could be used for
the metal casting riser design problem, and I published a paper on it in 1971 [7].
Other early books by these leaders were “Engineering Design by Geometric Programming” by
Clarence Zener in 1971 [3], “Applied Geometric Programming” by C.S. Beightler and D.T. Phillips
in 1976, and the second edition of “Foundations of Optimization” by C.S. Beightler, D.T. Phillips,
and D. Wilde in 1979 [4]. Many of the initial applications were in the area of transformer design as
Clarence Zener worked for Westinghouse Electric and in the area of Chemical Engineering which
was the area emphasized by Beightler and Wilde. It is also important that several graduate students
played an important role in the development of geometric programming, namely Elmor Peterson at
Carnegie Institute of Technology and Ury Passy and Mordecai Avriel at Stanford University.
Geometric programming has attracted a fair amount of interest and a list of the various thesis
and dissertations published that have either mentioned or focused on geometric programming are
listed in an appendix. Web sites on geometric programming [5] have appeared with additional
interesting applications.
6 2. BRIEF HISTORY OF GEOMETRIC PROGRAMMING
2.2 EVALUATIVE QUESTIONS
1. Who is recognized as the father of geometric programming?

2. When was the first book published on geometric programming and what was the title of the
book?

3. Which three Universities played an important role in the development of geometric program-
ming?

REFERENCES
[1] R.J. Duffin, E.L. Peterson and C. Zener, Geometric Programming, John Wiley and Sons, New
York, 1967. 5

[2] D.J. Wilde and C.S. Beighler, Foundations of Optimization, Prentice-Hall, Englewood Cliffs,
New Jersey, 1967. 5

[3] C. Zener, Engineering Design by Geometric Programming, John Wiley and Sons, New York,
1971. 5

[4] C.S. Beightler, D.T. Phillips and D.J. Wilde, Foundations of Optimization, 2nd Edition, Prentice
Hall, Englewood Cliffs, New Jersey, 1979. 5

[5] http://www.mpri.lsu.edu/textbook/Chapter3.htm (Chapter 3 Geometric Program-

ming) (site visited 5–20-09) 5

[6] C. Zener, “A Mathematical Aid in Optimizing Engineering Design”, Proceedings of the National
Academy of Science, Vol. 47, 1961, p. 537. 5

[7] R.C. Creese, “Optimal Riser Design by Geometric Programming”, AFS Cast Metals Research
Journal, Vol. 7, 1971, pp. 118–121. 5
 7

CHAPTER 3

Theoretical Considerations
3.1 PRIMAL AND DUAL FORMULATION
Geometric programming requires that the expressions used are posynomials, and it is necessary to
distinguish between functions, monomials and posynomials [1]. Posynomial is meant to indicate
a combination of “positive” and “polynomial” and implies a “positive polynomial.” Examples of
functions, which are monomials, are:

5x, 0.25, 4x 2 , 2x 1.5 y −.15 , 160, 65x −15 t 10 z2 .

Examples of posynomials are, which are monomials or sums of monomials, are:

5 + xy, (x + 2Y Z)2 , x + 2y + 3z + t x/y + z35x 1.5 + 72Y 3 .

Examples of expressions which are not posynomials are:

−1.5 (negative sign)

(2 + 2yz)3.2 (fractional power of multiple term which cannot be expanded)

x − 2y + 3z (negative sign)

x + sin(x) (sine expression can be negative).

The coefficients of the constants must be positive, but the coefficients of the exponents can
be negative.
The mathematics of geometric programming are rather complex, however the basic equations
are presented and followed by an illustrative example. The theory of geometric programming is
presented in more detail in some of the references [1, 2, 3, 4, 5] listed at the end of the chapter. The
primal problem is complex, but the dual version is much simpler to solve. The dual is the version
typically solved, but the relationships between the primal and dual are needed to determine the
specific values of the variables in the primal. The primal problem is formulated as:


Tm 
N
Ym (X) = σmt Cmt Xnamtn ; m = 0, 1, 2, . . . M , (3.1)
T =1 n=1
8 3. THEORETICAL CONSIDERATIONS
with σmt = ±1 and Cmt > 0
and Ym (X) ≤ σm for m = 1, . . . M for the constraints
where Cmt = positive constant coefficients in cost and constraint equations
and Ym (X) = primal objective function
and σmt = signum function used to indicate sign of term in the equation
(either +1 or −1).
The dual is the problem formulation that is typically solved to determine the dual variables
and value of the objective function. The dual objective function is expressed as:
 σ

M 
Tm
d(ω) = σ (Cmt ωm0 /ωmt )σmt ωmt m = 0, 1, . . . M and t = 1, 2, . . . Tm , (3.2)
m=0 t=1

where
σ = signum function (± 1)
Cmt = constant coefficient
ωm0 = dual variables from the linear inequality constraints
ωmt = dual variables of dual constraints, and
σmt = signum function for dual constraints,
and by definition:
ω00 = 1 . (3.3)

The dual is formulated from four conditions:

(1) a normality condition

Tm
σ0t ω0t = σ where σ = ±1 , (3.4)
T =1

where
σ0t = signum of objective function terms
ω0t = dual variables for objective function terms.
(2) N orthogonal conditions

M 
T
σmt amtn ωmt = 0 , (3.5)
m=0 t=1

where
σmt = signum of constraint term
amtn = exponent of design variable term
ωmt = dual variable of dual constraint.
(3) T non-negativity conditions (dual variables must be positive):

ωmt ≥ 0 m = 0, 1 . . . M and t = 1, 2 . . . . . . Tm . (3.6)

 3.1. PRIMAL AND DUAL FORMULATION 9
(4) M linear inequality constraints:


Tm
ωmo = σm σmt ωmt ≥ 0 . (3.7)
t=1

The dual variables, ωmt , are restricted to being positive, which is similar to the linear program-
ming concept of all variables being positive. If the number of independent equations and variables
in the dual are equal, the degrees of difficulty are zero. The degrees of difficulty is the difference
between the number of dual variables and the number of independent linear equations; and the
greater this degrees of difficulty, the more difficult the solution. The degrees of (D) can be expressed
as:
D = T − (N + 1) (3.8)
where
T = total number of terms (of primal)
N= number of orthogonality conditions plus normality condition
(which is equivalent to the number of primal variables).
Once the dual variables are found, the primal variables can be determined from the relation-
ships:

N
Cot Xnamtn = ωot σ Yo t = 1, . . . To , (3.9)
n=1
and

N
Cmt Xnamtn = ωmt /ωmo t = 1, . . . To and m = 1, . . . M . (3.10)
n=1
The theory may appear to be overwhelming with all the various terms, but various examples
will be presented in the following chapters to illustrate the application of the various equations.
There are two sections of examples, the first considering basic examples with zero degrees of
difficulty, and then the second section considering problems with more than zero degree of difficulty
and presenting various approaches to solving the problem. Problems with zero degrees of difficulty
typically will have a linear dual formulation with an equal number of equations and dual variables
and can be solved relatively easily.
When there are more than zero degrees of difficulty, the solution is much more difficult. Some
of the approaches to solve these problems are:
1) Finding additional equations so the number of variables and number of equations are equal,
but the additional equations are usually non-linear. The additional equations can be determined by
either: a) dimensional analysis of the primal dual relationships or b) by substitution techniques.
2) Express the dual in terms of only one dual variable by substitution and take the derivative
of the dual, set it to zero, and obtain the value of the unknown dual variable.
3) Condensation techniques by combining two or more of the primal terms to reduce the
number of dual variables.This technique gives an approximate optimal solution and will be illustrated.
10 3. THEORETICAL CONSIDERATIONS
4) Transformed Problem Approach. This technique is often used on maximization problems
to transform signomial problems into posynominial problems. It is also used to have the terms of
the transformed objective function as a function of a single variable to make the solution easier.

3.2 EVALUATIVE QUESTIONS

1. What version of the geometric problem formulation is solved for the objective function and
why?

2. What values can the signum function have?

3. How are the primal variables determined?

4. Which of the following terms are posynomials?

a)3.4 b)4x c)5xy d)4x 2.1 e)(x + 2)4
f )5x − 3 g)6x −2.4 h)e−3 i)3x + 4y + 5z−1.4 j)(x + 4)2.2
k)(x − y + 3)2 l)cot y m)x/2y n)5l −2 t 5

REFERENCES
[1] S. Boyd, S-J Kim, L. Vandengerghe, and A. Hassibi, “A Tutorial on Geometric Programming”,
Optimization and Engineering, 8(1), pp 67–127, Springer, Germany, 2007. 7

[2] R.J. Duffin, E.L. Peterson and C. Zener, Geometric Programming, John Wiley and Sons, New
York, 1967. 7

[3] D.J. Wilde and C.S. Beighler, Foundations of Optimization, Prentice-Hall, Englewood Cliffs,
New Jersey, 1967. 7

[4] C. Zener, Engineering Design by Geometric Programming, John Wiley and Sons, New York,
1971. 7

[5] C.S. Beightler, D.T. Phillips and D.J. Wilde, Foundations of Optimization, 2nd Edition, Prentice
Hall, Englewood Cliffs, New Jersey, 1979. 7
 PART II

Geometric Programming Applications

with Zero Degrees of Difficulty
 13

CHAPTER 4

The Optimal Box Design Case

Study
4.1 THE OPTIMAL BOX DESIGN PROBLEM

The optimal box design problem is a relatively easy problem which illustrates the procedure for solving
a geometric programming problem with zero degrees of difficulty. It also indicates the importance
of a general solution which is possible with geometric programming; that is formulas for the box
dimensions can be developed which will give the answers without needing to resolve the problem if
the costs or box volume changes.

EXAMPLE:
A box manufacturer wants to determine the optimal dimensions for making boxes to sell to customers.
The cost for production of the sides is C1 ($ 2/sq ft) and the cost for producing the top and bottom
is C2 ($ 3/sq ft) as more cardboard is used for the top and bottom of the boxes. The volume of the
box is to be set at a limit of “V ” (4 ft3 ) which can be varied for different customer specifications. If
the dimensions of the box are W for the width, H for the box height, and L for the box length, what
should the dimensions be based upon the cost values and box volume?
The problem is to minimize the box cost for a specific box volume. The primal objective
function is:

Minimize Cost(Y ) = C2 W L + C1 H (W + L) (4.1)

Subject to: W LH ≥ V . (4.2)

However, in geometric programming the inequalities must be written in the form of ≤ and
the right-hand side must be ±1. Thus, the primal constraint becomes:

Minimize Cost(Y ) = C1 H W + C1 H L + C2 W L (4.3)

Subject to − W H L/V ≤ −1 . (4.4)
14 4. THE OPTIMAL BOX DESIGN CASE STUDY
From the coefficients and signs, the signum values for the dual are:

σ01 =1
σ02 =1
σ03 =1
σ11 = −1
σ1 = −1

Thus, the dual formulation from Chapter 3 would be:

Objective Function (using Eqs. (3.4) and (4.3)) ω01 + ω02 + ω03 =1 (4.5)
L terms (using Eqs. (3.5), (4.3), and (4.4)) ω02 + ω03 − ω11 =0 (4.6)
H terms (using Eqs. (3.5), (4.3), and (4.4)) ω01 + ω02 − ω11 =0 (4.7)
W terms (using Eqs. (3.5), (4.3), and (4.4)) ω01 + ω03 − ω11 =0 (4.8)

The degrees of difficulty (D) are equal to:

D = T − (N + 1) = 4 − (3 + 1) = 0 ,

where
T = total number of terms of primal and
N = number of orthogonality conditions plus normality condition or the number
of primal variables (H , W , and L gives 3 primal variables).
Thus, one has the same number of variables as equations, so this can be solved by simultaneous
equations as these are linear equations.
Using Equations (4.5)–(4.8), the values for the dual variables are found to be:

ω01 = 1/3
ω02 = 1/3
ω03 = 1/3
ω11 = 2/3
and by definition
ω00 = 1

Using the linearity inequality equation expressed by Equation (3.7),


ω10 = ωmt = σm σmt ωmt = (−1) ∗ (−1 ∗ 2/3) = 2/3 > 0 where m = 1 and t = 1.
 4.1. THE OPTIMAL BOX DESIGN PROBLEM 15
The objective function can be found using Equation (3.2):
 σ

M 
Tm
d(ω) = σ σmt ωmt
(Cmt ωmo /ωmt ) (3.2)
 m=0 t=1
(1)∗(1/3)  (1)∗(1/3)
d(ω) = 1 (C1 ∗ 1)/(1/3) ∗ (C1 ∗ 1)/(1/3)
 (1)∗(1/3)  (−1)∗(2/3) 1
∗ (C2 ∗ 1)/(1/3) ∗ ((1/V ) ∗ 1)/(2/3)





= 1 (3C1 )1/3 ∗ (3C1 )1/3 ∗ (3C2 )1/3 ∗ (1/V )−2/3
2/3 1/3
= 3C1 C2 V 2/3 (4.9)

= 3 22/3 31/3 42/3 = $17.31 .

The solution has been determined without finding the values for L, W, or H . Also note that
the dual expression is expressed in constants and thus the answer can be found without having to
resolve the entire problem as one only needs to use the new constant values. To find the values of
L, W, and H , one must use Equations (3.9) and (3.10), which are repeated here.


N
Cot Xnamtn = ωot σ Yo t = 1, . . ., To (3.9)
n=1
and

N
Cmt Xnamtn = ωmt /ωmo t = 1, . . ., To and m = 1, . . ., M . (3.10)
n=1

Using Equation (3.9) the relationships are:

C1 H W = ω01 Y = Y/3
C1 H L = ω02 Y = Y/3
C2 W L = ω03 Y = Y/3 .

Combining the first two of these relationships one obtains

W =L. (4.10)

Combining the last two of these relationships one obtains

H = (C2 /C1 )L . (4.11)

Since V = H W L = (C2 /C1 )LLL = (C2 /C1 )L3 .

16 4. THE OPTIMAL BOX DESIGN CASE STUDY
Thus,

L = [V (C1 /C2 )]1/3 (4.12)

W = L = [V (C1 /C2 )]1/3 (4.13)
 1/3
H = (C2 /C1 )L = (C2 /C1 ) [V (C1 /C2 )]1/3 = V (C22 /C12 ) . (4.14)

The specific values for this particular problem would be:

L = [4(2/3)]1/3 = 1.386 ft
W = L = 1.386 ft
 1/3
H = 4(32 /22 ) = 2.080 ft .

The volume of the box, LW H = (1.386)(1.386)(2.080) = 4.0 ft3 , which was the minimum
volume required for the box. To verify the results, the parameters are used in the primal problem
(Equation (4.3)) to make certain the solution obtained is the same.

Cost (Y ) = C1 H W + C1 H L + C2 W L (4.3)
  1/3   1/3
Y0 = C1 V C22 /C12 [V (C1 /C2 )]1/3 + C1 V C22 /C12 [V (C1 /C2 )]1/3
+ C2 [V (C1 /C2 )]1/3 [V (C1 /C2 )]1/3
  
2/3 1/3 2/3 1/3 2/3 1/3
Y0 = C1 V 2/3 C2 + C1 V 2/3 C2 + C1 V 2/3 C2
2/3 1/3
Y0 = 3C1 C2 V 2/3 . (4.15)

The expressions for Equation (4.9) from the primal and Equation (4.15) from the dual are
equivalent. The geometric programming solution is in general terms, and thus can be used for any
values of C1 , C2 , and V . This ability to obtain general relationships makes the use of Geometric
Programming a very valuable tool for cost engineers.

4.2 EVALUATIVE QUESTIONS

1. A large box is to be made with the values of C1 = 4 Euros/m2 , C2 = 4 Euros/m2 , and V =
8 m3 . What is the cost (Euros) and the values of H, W, and L?

2. The cost of the top and bottom is increased to 6 Euros/m2 and what is the increase in the box
cost and the change in box dimensions?

3. The box is to be open (that is there is no top). Determine the expressions for H, W, and L for
an open box and determine the cost if C1 = 4 Euros/m2 , C2 = 6 Euros/m2 , and V = 8 m3 .
Compare the results with Problem 2 and discuss the differences in the formulas, dimensions,
and costs.
 17

CHAPTER 5

Trash Can Case Study

5.1 INTRODUCTION
Various case studies are used to illustrate the different applications of geometric programming as
well as to illustrate the different conditions that must be evaluated in solving the problems. The
second case study, the trash can case study, is easy to solve and has zero degrees of difficulty. It is
similar to the box problem, but involves a different shape. The solution is provided in detail, giving
the general solution for the problem in addition to the specific solution. These examples are provided
so that the readers can develop solutions to specific problems that they may have and to illustrate
the importance of the generalized solution.

5.2 PROBLEM STATEMENT AND GENERAL SOLUTION

5.2.1 EXAMPLE
Bjorn of Sweden has entered into the trash can manufacturing business and he is making cylindrical
trash cans and wants to minimize the material cost.The trash can is an open cylinder and designed to
have a specific volume. The objective will be to minimize the total material cost of the can. Figure 5.1
is a sketch of the trash can illustrating the design parameters used, the radius and the height of the
trash can. The bottom and sides can be of different costs as the bottom is typically made of a thicker
material. The primal objective function is:

Minimize: Cost(Y ) = C1 π r 2 + C2 2π rh (5.1)

Subject to: V = π r 2h , (5.2)

where:
r = radius of trash can bottom
h = height of trash can
V = volume of trash can
C1 = material constant cost of bottom material of trash can
C2 = material constant cost of side material of trash can.
The constraint must be written in the form of an inequality, so

V ≥ π r 2h . (5.3)

And it must be written in the less than equal form, so it becomes

−π r 2 h/V ≤ −1 . (5.4)
18 5. TRASH CAN CASE STUDY

2r

h

Figure 5.1: Trash can parameters r and h.

Thus, the primal problem is:

Minimize: Cost(Y ) = C1 π r 2 + C2 2π rh (5.5)
Subject to: − π r 2 h/V ≤ −1 . (5.6)
From the coefficients and signs, the signum values for the dual are:
σ01 = 1
σ02 = 1
σ11 = −1
σ1 = −1 .
The dual formulation is:
Objective Function ω01 + ω02 =1 (5.7)
r terms 2ω01 + ω02 − 2ω11 = 0 (5.8)
h terms ω02 − ω11 = 0 . (5.9)
Using theses equations, the values of the dual variables are found to be:
ω01 = 1/3
ω02 = 2/3
ω11 = 2/3
and by definition
ω00 = 1
 5.2. PROBLEM STATEMENT AND GENERAL SOLUTION 19
The degrees of difficulty are equal to:

D = T − (N + 1) = 3 − (2 + 1) = 0 . (5.10)

Using the linearity inequality equation,


ω10 = ωmt = σm σmt ωmt = (−1) ∗ (−1 ∗ 2/3) = 2/3 > 0 where m = 1 and t = 1 .

The objective function can be found using the dual expression:

 M T σ
  m
Y = d(ω) = σ (Cmt ωmo /ωmt )σmt ωmt (5.11)
m=0 t=1
   1
=1 {(π C1 ∗ 1/(1/3))}(1∗1/3) {π C2 ∗ 1/(2/3))}(1∗2/3) {(π/V ) ∗ ((2/3)/(2/3)}(−1∗2/3)

1/3 2/3
= 3π 1/3 C1 C2 V 2/3 . (5.12)

The values for the primal variables can be determined from the relationships between the
primal and dual as:

C1 π r 2 = ω01 Y = 1/3 ∗ Y . (5.13)

And

C2 2π rh= ω02 Y = 2/3 ∗ Y . (5.14)

Dividing these expressions and reducing terms one can obtain:

r = (C2 /C1 ) ∗ h . (5.15)

Setting

V = π r 2h . (5.16)

And using the last two equations one can obtain

h = ((V /π )(C12 /C22 ))1/3 . (5.17)

and

r = ((V /π )(C2 /C1 ))1/3 . (5.18)

Using (5.17) and (5.18) in Equation (5.5) for the primal, one obtains:

Y = C1 π r 2 + C2 2π rh (5.5)
1/3 2/3
= 3π 1/3
C1 C2 V 1/3 . (5.19)
20 5. TRASH CAN CASE STUDY
Note that Equations (5.19) and (5.12) are identical, which is what should happen, as the
primal and dual objective functions must be identical.
An important aspect about the dual variables is that they indicate the effect of the terms
upon the solution. The values of ω02 = 2/3 and ω01 = 1/3 indicates that the second term has twice
the impact as the first term in the primal. For example, if C1 = 9 $/sq ft, C2 = 16 $/sq ft, and
V = 4π = 12.57 cubic feet, then

h = ((V /π)(C12 /C22 ))1/3 = ((4π/π)(92 /162 ))1/3 = (4 ∗ 81/256)1/3 = 1.082 ft

and r = ((V /π)(C2 /C1 ))1/3 = ((4π/π)(16/9))1/3 = (4 ∗ 16/9)1/3 = 1.923 ft .

Note that:

V = π r 2 h = 3.1416 ∗ 1.082 ft ∗ (1.923 ft)2 = 12.57 ft3

And Y = C1 π r 2 + C2 2π rh = 9 ∗ 3.14 ∗ 1.9232 + 16 ∗ 2 ∗ 3.14 ∗ 1.923 ∗ 1.082
= $104.5 + $209.0
= $313.5 .

The contribution of the second term is twice that of the first term which is what is predicted
by the value of the dual variables. This occurs regardless of the values of the constants used and this
is an important concept for cost analysis.

5.3 EVALUATIVE QUESTIONS

1. A trash can is designed to hold 3 cubic meters of trash. Determine the cost and the design
parameters (radius and height) in meters for if the costs C1 and C2 are 20 Swedish Kroner per
square meter and 10 Swedish Kroner per square meter, respectively.

2. If the volume is doubled to 6 cubic meters, what are the new dimensions and cost?

3. If the trash can is to have a lid which will have the same diameter as the bottom of the trash
can, what are the cost and dimensions of the trash can with the lid?
 21

CHAPTER 6

The Open Cargo Shipping Box

Case Study
6.1 PROBLEM STATEMENT AND GENERAL SOLUTION

This is a classic geometric programming problem as it was the first illustrative problem presented in
the first book [1] on geometric programming. This problem presented here is expanded as not only
is the minimum total cost required, but also the dimensions of the box. The problem is: “Suppose
that 400 cubic yards (V ) of gravel must be ferried across a river. The gravel is to be shipped in
an open cargo box of length x1 , width x2 and height x3 . The sides and bottom of the box cost
$ 10 per square yard (A1 ) and the ends of the box cost $ 20 per square yard (A2 ). The cargo box will
have no salvage value and each round trip of the box on the ferry will cost 10 cents (A3 ).
a) What is the minimum total cost of transporting the 400 cubic yards of gravel?
b) What are the dimensions of the cargo box?
c) What is the number of ferry trips to transport the 400 cubic yards of gravel?”
Figure 6.1 illustrates the parameters of the open cargo shipping box.

x2

x3
x1
Bottom
Side

End

Figure 6.1: Open cargo shipping box. x1 = Length of the Box, x2 = Width of the Box, x3 = Height
of the Box.
22 6. THE OPEN CARGO SHIPPING BOX CASE STUDY
The first issue is to determine the various cost components to make the objective function.
The ferry transportation cost can be determined by:
T 1 = V ∗ A3 /(x1 ∗ x2 ∗ x3 ) = 400 ∗ 0.10/(x1 ∗ x2 ∗ x3 ) = 40/(x1 ∗ x2 ∗ x3 ) . (6.1)
The cost for the ends of the box (2 ends) is determined by:
T 2 = 2 ∗ (x2 ∗ x3 ) ∗ A2 = 2 ∗ (x2 ∗ x3 ) ∗ 20 = 40 ∗ (x2 ∗ x3 ) . (6.2)
The cost for the sides of the box (2 sides) is determined by:
T 3 = 2 ∗ (x1 ∗ x3 ) ∗ A1 = 2 ∗ (x1 ∗ x3 ) ∗ 10 = 20 ∗ (x1 ∗ x3 ) . (6.3)
The cost for the bottom of the box is determined by:
T 4 = (x1 ∗ x2 ) ∗ A1 = (x1 ∗ x2 ) ∗ 10 = 10 ∗ (x1 ∗ x2 ) . (6.4)
The objective function (Y ) is the sum of the four components and is:
Y = T1+T2+T3+T4 (6.5)
Y = 40/(x1 ∗ x2 ∗ x3 ) + 40 ∗ (x2 ∗ x3 ) + 20 ∗ (x1 ∗ x3 ) + 10 ∗ (x1 ∗ x2 ) . (6.6)
The primal objective function can be written in terms of generic constants for the cost variables
to obtain a generalized solution.
Y = C1 /(x1 ∗ x2 ∗ x3 ) + C2 ∗ (x2 ∗ x3 ) + C3 ∗ (x1 ∗ x3 ) + C4 ∗ (x1 ∗ x2 ) , (6.7)
where C1 = 40, C2 = 40, C3 = 20 and C4 = 10.
From the coefficients and signs, the signum values for the dual are:
σ01 =1
σ02 =1
σ03 =1
σ04 =1.
The dual formulation is:
Objective Function ω01 + ω02 + ω03 + ω04 =1 (6.8)
x1 terms −ω01 + ω03 + ω04 =0 (6.9)
x2 terms −ω01 + ω02 + ω04 =0 (6.10)
x3 terms −ω01 + ω02 + ω03 =0. (6.11)
Using these equations, the values of the dual variables are found to be:
ω01 = 2/5
ω02 = 1/5
ω03 = 1/5
ω04 = 1/5
 6.1. PROBLEM STATEMENT AND GENERAL SOLUTION 23
and by definition

ω00 = 1 .

Thus, the dual variables indicate that the first term of the primal expression is twice as im-
portant as the other three terms. The degrees of difficulty are equal to:

D = T − (N + 1) = 4 − (3 + 1) = 0 . (6.12)
The objective function can be found using the dual expression:
⎡ ⎤σ
 
M T m
Y = d(ω) = σ ⎣ (Cmt ωmo /ωmt )σmt ωmt ⎦ (6.13)
m=0 t=1

= 1[[{(C1 ∗ 1/(2/5))}(1∗2/5) ][{C2 ∗ 1/(1/5))}(1∗1/5) ][{C3 ∗ 1/(1/5)}(1∗1/5) ][{C4 ∗ 1/(1/5)}(1∗1/5) ]]1

= 1002/5 ∗ 2001/5 ∗ 1001/5 ∗ 501/5
= 1002/5 ∗ 10000001/5
= 100 2/5 ∗ 1003/5
= $100 .

Thus, the minimum cost for transporting the 400 cubic yards of gravel across the river is $ 100.
The values for the primal variables can be determined from the relationships between the primal
and dual as:

C1 /(x1 ∗ x2 ∗ x3 ) = ω01 Y = (2/5)Y (6.14)

C2 ∗ x2 ∗ x3 = ω02 Y = (1/5)Y (6.15)
C3 ∗ x1 ∗ x3 = ω03 Y = (1/5)Y (6.16)
C4 ∗ x1 ∗ x2 = ω04 Y = (1/5)Y . (6.17)

If one combines Equations (6.15) and (6.16) one can obtain the relationship:

x2 = x1 ∗ (C3 /C2 ) . (6.18)

If one combines Equations (6.16), (6.17) and (6.18), one can obtain the relationship:

x3 = x2 ∗ (C4 /C3 ) = x1 ∗ (C3 /C2 ) ∗ (C4 /C3 ) = x1 ∗ (C4 /C2 ) . (6.19)

If one combines Equations (6.14) and (6.15) one can obtain the relationship:

x1 ∗ x22 ∗ x32 = (1/2) ∗ (C1 /C2 ) . (6.20)

Using the values for x2 and x3 in Equation (6.20), one can obtain:

x1 = [(1/2) ∗ (C1 C23 /(C32 C42 ))]1/5 . (6.21)

24 6. THE OPEN CARGO SHIPPING BOX CASE STUDY
Similarly, one can solve for x2 and x3 and the equations would be:

x2 = [(1/2) ∗ (C1 C33 /(C22 C42 ))]1/5 (6.22)

and
x3 = [(1/2) ∗ (C1 C43 /(C22 C32 ))]1/5 . (6.23)
Now using the values of C1 = 40, C2 = 40, C3 = 20 and C4 = 10, the values of x1 , x2 , and
x3 can be determined using Equations (6.21), (6.22), and (6.23) as:

x1 = [(1/2) ∗ (40 ∗ 403 /(202 102 ))]1/5 = [32]1/5 = 2 yards

x2 = [(1/2) ∗ (40 ∗ 203 /(402 102 ))]1/5 = [1]1/5 = 1 yard
x3 = [(1/2) ∗ (40 ∗ 103 /(402 202 ))]1/5 = [0.03125]1/5 = 0.5 yard .

Thus, the box is 2 yards in length, 1 yard in width, and 0.5 yard in height. The total box volume is
the product of the three dimensions, which is 1 cubic yard.
The number of trips the ferry must make is 400 cubic yards/1 cubic yard/trip = 400 trips.
If one uses the primal variables in the primal equation, the values are:

Y = 40/(x1 ∗ x2 ∗ x3 ) + 40 ∗ (x2 ∗ x3 ) + 20 ∗ (x1 ∗ x3 ) + 10 ∗ (x1 ∗ x2 ) (6.6)

Y = 40/(2 ∗ 1 ∗ 1/2) + 40 ∗ (1 ∗ 1/2) + 20 ∗ (2 ∗ 1/2) + 10 ∗ (2 ∗ 1) (6.24)
Y = 40 + 20 + 20 + 20
= $100

Note that the primal and dual give the same result for the objective function. Note that the
components of the primal solution (40, 20, 20, 20) are in the same ratio as the dual variables(2/5,
1/5, 1/5, 1/5). This ratio will remain constant even as the values of the constants change and this is
important in the ability to determine which of the terms are dominant in the total cost. Thus, the
transportation cost is twice the cost of the box bottom and the box bottom is the same as the cost of
the box sides and the same as the cost of the box ends. This indicates the optimal design relationships
between the costs of the various box components and the transportation cost associated with the
design.

6.2 EVALUATIVE QUESTIONS

1. The ferry cost for a round trip is increased from $ 0.10 to $ 3.20. What is the new total cost, the
new box dimensions, and the number of ferry trips required to transport the 400 cubic yards
of gravel?

2. The ferry cost for a round trip is increased from $ 3.20 to $ 213.06. What is the new total cost,
the new box dimensions, and the number of ferry trips required to transport the 400 cubic yards
of gravel.
 REFERENCES 25
3. A cover must be added to the box and it is made having the same costs as the box bottom.
Determine the new total cost, the new box dimensions, and the number of ferry trips required
to transport the 400 cubic yards of gravel.

REFERENCES
[1] R.J. Duffin, E.L. Peterson and C. Zener, Geometric Programming, John Wiley and Sons, New
York, 1967. 21
 27

CHAPTER 7

Metal Casting Cylindrical Riser

Case Study
7.1 INTRODUCTION
The riser design problem in metal casting is always a concern for foundry engineers. The riser (also
called feeders in many parts of the world) is an amount of additional metal added to a metal casting
to move the thermal center of the casting and riser into the riser so there will be no solidification
shrinkage in the casting. The risers are typically shaped as cylinders as other shapes are difficult for
the molding process and this shape has been successfully used for decades. The riser also has other
design conditions such as to supply sufficient feed metal, but thermal design issues are typically the
primary concern. There are several papers on riser design using geometric programming for side
riser, top riser, insulated riser and many other riser design types.
For a riser to be effective, the riser must solidify after the casting in order to provide liquid
feed metal to the casting. The object is to have a riser of minimum volume to improve the yield of
the casting process which improves the economics of the process. The case study considered is for a
cylindrical side riser which consists of a cylinder of height H and diameter D. Figure 7.1 indicates
the relationship between the casting and the side riser and the parameters of the riser.

D


Riser H
 
Casting

Figure 7.1: The cylindrical riser.

The theoretical basis for riser design is Chvorinov’s Rule, which is:

t = K(V /SA)2 , (7.1)

28 7. METAL CASTING CYLINDRICAL RISER CASE STUDY
where

t = solidification time (minutes or seconds)

K = solidification constant for molding material (minutes/in2 or seconds/cm2 )
V = riser volume (in3 or cm3 )
SA = cooling surface area of the riser (in3 or cm3 ) .

The objective is to design the smallest riser such that

tR ≥ tC (7.2)

where

tR = solidification time of the riser

tC = solidification time of the casting .

This constraint (Equation 7.2) can be written as:

KR (VR /SAR ) ≥ KC (VC /SAC ) . (7.3)

The riser and the casting are assumed to be molded in the same material so the KR and KC
are equal and thus the equation can be written as:

(VR /SAR ) ≥ (VC /SAC ) . (7.4)

The casting has a specified volume and surface area, the right-hand side of the equation can be
expressed as a constant Y = (VC /SAC ), which is called the casting modulus (Mc ), and Equation (7.4)
becomes
(VR /SAR ) ≥ Y . (7.5)
The volume and surface of the cylindrical riser can be written as:

VR = π D 2 H /4 (7.6)
SAR = π DH + 2π D 2 /4 . (7.7)

The surface area expression neglects the connection area between the casting and the riser as
the effect is small. Thus, Equation (7.5) can be rewritten as:

(πD 2 H /4)/(π DH + 2π D 2 /4) = (DH )/(4H + 2D) ≥ Y . (7.8)

The constraint must be rewritten in the less than equal form with the right hand side being
less than or equal to one which becomes

4Y D −1 + 2Y H −1 ≤ 1 . (7.9)
 7.2. PROBLEM FORMULATION AND GENERAL SOLUTION 29
7.2 PROBLEM FORMULATION AND GENERAL SOLUTION
The primal form of the side cylindrical riser design problem can be stated as:

Minimize: V = π D 2 H /4 (7.10)
Subject to: 4Y D −1 + 2Y H −1 ≤ 1 . (7.11)

From the coefficients and signs, the signum values for the dual are:

σ01 =1
σ11 =1
σ12 =1
σ1 =1.

The dual problem formulation is:

Objective Function ω01 =1 (7.12)

D terms 2ω01 − ω11 =0 (7.13)
H terms ω01 − ω12 =0. (7.14)

Using Equations (7.12) to (7.14), the values of the dual variables were found to be:

ω01 = 1
ω11 = 2
ω12 = 1 .

The degrees of difficulty (D) are equal to:

D = T − (N + 1) = 3 − (2 + 1) = 0 . (7.15)

Using the linearity inequality equation,


ω10 = ωmt = σm σmt ωmt = (1) ∗ (1 ∗ 2 + 1 ∗ 1)
= 3 > 0 where m = 1 and t = 1 . (7.16)

The objective function can be found using the dual expression:

 σ

M 
Tm
V = d(ω) = σ (Cmt ωmo /ωmt )σmt ωmt (7.17)
m=0 t=1
= 1[[{(π/4 ∗ 1/1)}(1∗1) ][{(4Y ∗ 3/2)}(1∗2) ][{(2Y ∗ 3/1)}(1∗1) ]]1
= (π/4) ∗ (6Y )2 ∗ (6Y )
= (π/4) ∗ (6Y )3 . (7.18)
30 7. METAL CASTING CYLINDRICAL RISER CASE STUDY
The values for the primal variables can be determined from the relationships between the
primal and dual as:

4Y D −1 = ω11 /ω10 = 2/3 (7.19)

and 2Y H −1 = ω11 /ω10 = 1/3 . (7.20)

The equations for H and D can be determined as:

D = 6Y (7.21)
and H = 6Y . (7.22)

Using (7.21) and (7.22) in Equation (7.10) for the primal, one obtains:

V = (π/4) ∗ (6Y )3 . (7.23)

Equations (7.18) and (7.23) are identical, which is what should happen, as the primal and
dual objective functions must be identical. The values for the riser diameter and the riser height
are both six times the casting modulus. These relationships hold for the side cylindrical riser design
with negligible effects for the connecting area. This also indicates that the riser height and riser
diameter are equal for the side riser. Designs for other riser shapes and with insulating materials
using geometric programming are given in the references.

7.3 EXAMPLE
A rectangular plate casting with dimensions L = W = 10 cm and H = 4 cm is to be produced and
a cylindrical side riser is to be used. The optimal dimensions for the side riser can be obtained from
the casting modulus Y and Equations (7.21) and (7.22). The casting modulus is obtained by:

Y = (VC /SAC )
= (10 cm × 10 cm × 4 cm)/[2(10 cm × 10 cm) + 2(10 cm × 4 cm) + 2(10 cm × 4 cm)]
= 400 cm3 /360 cm2 = 1.111 cm .

Thus,

H = 6Y = 6 × 1.111 cm = 6.67 cm
D = 6Y = 6 × 1.111 cm = 6.67 cm .

The volume of the riser can be obtained from Equation (7.23) as:

V = (π/4)(6Y )3 = 233 cm3 .

Thus, once the modulus of the casting is determined, the riser height, diameter, and volume
can be determined using Equations (7.21)-(7.23).
 7.4. EVALUATIVE QUESTIONS 31
7.4 EVALUATIVE QUESTIONS
1. A side riser is to be designed for a metal casting which has a surface area of 340 cm2 and
a volume of 400 cm3 . The hot metal cost is 100 Rupees per kg and the metal density is
3.0 gm/cm3 .

a) What are the dimensions in centimeters for the side riser (H and D)?
b) What is the volume of the side riser (cm 3 )?
c) What is the metal cost of the side riser (Rupees)?
d) What is the metal cost of the casting (Rupees)?

2. Instead of a side riser a top riser is to be used; that is the riser is placed on the top surface of
the casting. The cooling surface area for the top riser is:

SAR = π DH + π D 2 /4 .

Show that for the top riser that D = 6Y and H = 3Y .

REFERENCES
[1] Creese, R. C., “Optimal Riser Design by Geometric Programming,” AFS Cast Metals Research
Journal, Vol. 7, 1971, pp. 118–121.

[2] Creese, R.C., “Dimensioning of Risers for Long Freezing Range Alloys by Geometric Pro-
gramming,” AFS Cast Metals Research Journal, Vol. 7, 1971, pp. 182–184.

[3] Creese, R.C., “Generalized Riser Design by Geometric Programming,” AFS Transactions,
Vol. 87, 1979, pp. 661–664.

[4] Creese, R.C., “An Evaluation of Cylindrical Riser Designs with Insulating Materials,” AFS
Transactions, Vol. 87, 1979, pp. 665–669.

[5] Creese, R.C., “Cylindrical Top Riser Design Relationships for Evaluating Insulating Materi-
als,” AFS Transactions, Vol. 89, 1981, pp. 345–348.
 33

CHAPTER 8

Inventory Model Case Study

8.1 PROBLEM STATEMENT AND GENERAL SOLUTION.
The basic inventory model is to minimize the sum of the unit set-up costs and the unit inventory
holding costs.The objective is to determine the optimal production quantity which will minimize the
total costs. The problem has been solved using the method of Lagrange Multipliers, but it can also
easily be solved using geometric programming which permits a general solution for the production
quantity in terms of the constant parameters.
The assumptions for the model are:

1. Replenishment of the order is instantaneous

2. No shortage is permitted

3. The order quantity is a batch

The model can be formulated in terms of annual costs as:

Total Cost = Total Unit Costs + Annual Inventory Carrying Cost

+ Annual Set-up Cost (8.1)
T C = DCu + CA + SD/Q . (8.2)

where
T C = Total Annual Cost ($)
D = Annual Demand (pieces/year)
Cu = Item Unit Cost ($/piece)
C = Inventory Carrying Cost ($/piece-yr)
A = Average Inventory (Pieces)
S = Set-up Cost ($/set-up)
Q = Order Quantity (Pieces/order) .
The average inventory for this model is given by:

A = Q/2 . (8.3)

Thus, the model can be formulated as:

T C = DCu + CQ/2 + SD/Q (8.4)

34 8. INVENTORY MODEL CASE STUDY
The model can be formulated in general terms as:
Minimize Cost(Y ) = C00 + C01 Q + C02 /Q (8.5)
where
C00 =DCu
C01 =C/2
C03 =SD .
Since C00 is a constant, the objective function can be rewritten for the variable cost as:
Minimize Variable Cost(Y ) = Yvc = C01 Q + C02 /Q . (8.6)
From the coefficients and signs, the signum values for the dual are:
σ01 = 1
σ02 = 1 .
Thus, the dual formulation would be:
Objective Function (using Equation 8.6) ω01 + ω02 = 1 (8.7)
Q terms (using Equation 8.6) ω01 − ω02 = 0 (8.8)
The degrees of difficulty are equal to:
D = T − (N + 1) = 2 − (1 + 1) = 0 .
Thus, one has the same number of variables as equations, so this can be solved by simultaneous
equations as these are linear equations.
Using Equations (8.7) and (8.8), the values for the dual variables are found to be:
ω01 = 1/2
ω02 = 1/2
and by definition
ω00 = 1 .
The objective function can be found using the dual expression:
 M T σ
  m
d(ω) = σ (Cmt ωmo /ωmt ) σmt ωmt
(8.9)
m=0 t=1
d(ω) = 1[{(C01 ∗ 1)/(1/2)}(1)∗(1/2) ∗ {(C02 ∗ 1)/(1/2)}(1)∗(1/2) ]1
= 1[{(2C01 )1/2 } ∗ {(2C02 )1/2 }]
1/2 1/2
= 2C01 C02
= 2(C/2 ∗ SD)1/2
 8.2. EXAMPLE 35
or
Yvc = (2CSD)1/2 . (8.10)
And
T C = DCu + Yvc
or
T C = DCu + (2CSD)1/2 . (8.11)
The solution has been determined without finding the value for Q. Also note that the dual
expression and total cost expressions are expressed in constants and thus the answer can be found
without having to resolve the entire problem as one only needs to use the new constant values.

8.2 EXAMPLE
The values for the parameters for the example problem are:
D = Annual Demand (pieces/year) = 100,000/yr
Cu = Item Unit Cost ($/piece) = $ 1.5/piece
C = Inventory Carrying Cost ($/piece-yr) = $0.20/piece-year
A = Average Inventory (Pieces) = Q/2
S = Set-up Cost ($/set-up) = $ 400/set-up
Q = Order Quantity (Pieces/order) =Q
Note the total cost can be found using Equation (8.11) as:
T C = 100, 000 ∗ 1.5 + (2 ∗ 0.20 ∗ 400 ∗ 100, 000)1/2 (8.12)
= 150, 000 + 4, 000
= $154, 000 . (8.13)
The value of Q can be determined from the primal-dual relationships which are:
(C/2)Q = ω01 Yvc = (1/2)(2CSD)1/2
or
Q = (2SD/C)1/2 . (8.14)
And for the example problem
Q = (2 ∗ 400 ∗ 100, 000/0.20)1/2
Q = 20, 000 units . (8.15)
The primal problem can now be evaluated using Equation (8.4) as:
T C = DCu + CQ/2 + SD/Q (8.4)
T C = 100, 000 ∗ 1.5 + 0.20 ∗ 20, 000/2 + 400 ∗ 100, 000/20, 000
= 150, 000 + 2, 000 + 2, 000
= $154, 000 (8.16)
36 8. INVENTORY MODEL CASE STUDY
The number of set-ups per year would be 100,000/20,000 or 5 set-ups per year.
Thus, the primal and dual give identical values for the solution of the problem as indicated
by Equations (8.13) and (8.16). The dual variables are both equal to 1/2 which implies that the
total annual inventory carrying cost and the total annual set-up costs are equal as illustrated by the
example.

8.3 EVALUATIVE QUESTIONS

1. The following data was collected on a new pump.
D = Annual Demand (pieces/year) = 200/yr
Cu = Item Unit Cost ($/piece) = $ 300/piece
C = Inventory Carrying Cost ($/piece-yr) = $ 20/piece-year
S = Set-up Cost ($/set-up) = $ 500/set-up
a. Determine the total cost for the 200 pumps during the year.
b. Determine the average total cost per pump.
c. What is the number of set-ups per year?
d. What is the total inventory carrying cost for the year?

2. The demand for the pumps increased dramatically to 3,000 because of the oil spill in the Gulf.
D = Annual Demand (pieces/year) = 3,000/yr
Cu = Item Unit Cost ($/piece) = $ 300/piece
C = Inventory Carrying Cost ($/piece-yr) = $ 20/piece-year
S = Set-up Cost ($/set-up) = $ 500/set-up
a. Determine the total cost for the 3,000 pumps during the year.
b. Determine the average total cost per pump.
c. What is the number of set-ups per year?
d. What is the total inventory carrying cost for the year?

3. Resolve Problem 2 if the order must be completed in one year, so recalculate the answers be
since the number of set-ups was not an integer?

REFERENCES
[1] H. Jung and C.M. Klein, “Optimal Inventory Policies Under Decreasing Cost Functions via
Geometric Programming”, European Journal of Operational Research, 132, 2001, pp. 628–642.
 37

CHAPTER 9

Process Furnace Design Case

Study
9.1 PROBLEM STATEMENT AND SOLUTION
An economic process model was developed [1, 2] for an industrial metallurgical application. The
annual cost for a furnace operation in which the slag-metal reaction is a critical factor of the process
was considered and a modified version of the problem is presented. The objective was to minimize
the annual cost and the primal equation representing the model was:
Y = C1 /(L2 ∗ D ∗ T 2 ) + C2 ∗ L ∗ D + C3 ∗ L ∗ D ∗ T 4 . (9.1)
The model was subject to the constraint that:
D≤L.
The constraint must be set in geometric programming for which would be:
(D/L) ≤ 1 (9.2)
where
D = Depth of the furnace (ft)
L = Characteristic Length of the furnace (ft)
T = Furnace Temperature (K) .
For the specific example problem, the values of the constants were:
C1 = 1013 ($ − ft2 − K2 )
C2 = 100($/ft2 )
C3 = 5 ∗ 10−11 (ft−2 − K−4 ) .
From the coefficients and signs, the signum values for the dual are:
σ01 =1
σ02 =1
σ03 =1
σ11 =1
σ1 =1.
38 9. PROCESS FURNACE DESIGN CASE STUDY
L

T=?

D

Figure 9.1: Process furnace.

The dual problem formulation is:

Objective Function ω01 + ω02 + ω03 =1 (9.3)

L terms −2ω01 + ω02 + ω03 − ω11 =0 (9.4)
D terms −ω01 + ω02 + ω03 + ω11 =0 (9.5)
T terms −2ω01 + 4ω03 =0. (9.6)

Using Equations (9.3) to (9.6), the values of the dual variables were found to be:

ω01 = 0.4
ω02 = 0.4
ω03 = 0.2
ω11 = −0.2 .

The dual variables cannot be negative and the negative value implies that the constraint is
not binding, that is it is a loose constraint. Thus, the problem must be reformulated without the
constraint and the dual variable is forced to zero, that is, ω11 = 0 and the equations resolved. This
means that the constraint D ≤ L will be loose, that is D will be less than L in the solution. The new
dual becomes:

Objective Function ω01 + ω02 + ω03 =1 (9.7)

L terms −2ω01 + ω02 + ω03 =0 (9.8)
D terms −ω01 + ω02 + ω03 =0 (9.9)
T terms −2ω01 + 4ω03 =0. (9.10)

Now the problem is that it has 4 equations to solve for three variables. If one examines
Equations (9.8) and (9.9), one observes that Equation (9.8) is dominant over Equation (9.9) and
thus Equation (9.9) will be removed from the dual formulation. The new dual formulation is:
 9.1. PROBLEM STATEMENT AND SOLUTION 39

Objective Function ω01 + ω02 + ω03 = 1 (9.11)

L terms −2ω01 + ω02 + ω03 = 0 (9.12)
T terms −2ω01 + 4ω03 = 0 . (9.13)

The new solution for the dual becomes:

ω01 = 1/3
ω02 = 1/2
ω03 = 1/6

and by definition

ω00 = 1.0 .

The dual variables indicate that the second term is the most important, followed by the first
term and then the third term. The degrees of difficulty are now equal to:

D = T − (N + 1) = 3 − (2 + 1) = 0 .

The objective function can be found using the dual expression:

 M T σ
  m
Y = d(ω) = σ (Cmt ωmo /ωmt ) σmt ωmt
(9.14)
m=0 t=1

= 1[[{(C1 ∗ 1/(1/3)1)}(1/3∗1) ][{(C2 ∗ 1/(1/2)}(1/2∗1) ][{(C3 /(1/6))}(1/6∗1) ]]1

= 1[[{(1 ∗ 1013 ∗ 1/(1/3)1)}(1/3∗1) ][{(100 ∗ 1/(1/2)}(1/2∗1) ][{(5 ∗ 10−11 /(1/6))}(1/6∗1) ]]1

= $11, 370 .

This can be expressed in a general form in terms of the constants as:

Y = (3C1 )1/3 (2C2 )1/2 (6C3 )1/6 . (9.15)

The values for the primal variables can be determined from the relationships between the
primal and dual which are:

C1 ∗ L−2 ∗ D −1 ∗ T −2 = ω01 Y (9.16)

C2 ∗ L ∗ D = ω02 Y (9.17)
and C3 ∗ L ∗ D ∗ T 4 = ω03 Y . (9.18)
40 9. PROCESS FURNACE DESIGN CASE STUDY
The fully general expressions are somewhat difficult, but the variables can be expressed in
terms of the constants and objective function as:

T = [(ω03 /ω02 ) ∗ (C2 /C3 )]1/4 (9.19)

L = [(C1 ∗ (C2 ∗ C3 )(1/2) )]/[ω01 ∗ (ω02 ∗ ω03 )(1/2) ∗ Y 2 ] (9.20)
(3/2) (1/2) (3/2) (1/2)
D = [(ω01 ∗ ω02 ∗ ω03 Y 3 )/(C1 ∗ C2 ∗ C3 ]. (9.21)

The expressions developed for the variables in terms of the constants in a reduced form were:

T = (C2 /3C3 )1/4 (9.22)

L = (3C1 )1/3 (2C2 )−1/2 (6C3 )1/6 (9.23)
D=1. (9.24)

Using the values of Y = 11370, C1 = 1013 , C2 = 100, C3 = 5 ∗ 10−11 , ω01 = 1/3, ω02 =
1/2 and ω03 = 1/6, the values for the variables are:

T = 903 K
L = 56.85 ft

and

D = 1.00 ft.

Using the values of the variables in the primal equation, the objective function is:

Y = C1 /(L2 ∗ D ∗ T 2 ) + C2 ∗ L ∗ D + C3 ∗ L ∗ D ∗ T 4
= 1013 /(56.852 ∗ 1 ∗ 9032 ) + 100 ∗ 56.85 ∗ 1 + 5 ∗ 10−11 ∗ 56.85 ∗ 1 ∗ (9034 )
= 3, 795 + 5, 685 + 1, 890
= $11, 370 .

The values of the objective function for the primal and dual are identical, which implies that
the values for the primal variables have been correctly obtained. The costs terms are in the same
ratio as the dual variables; the third term is the smallest, the first term is twice the third term and
the second term is three times the third term.
This problem was presented to indicate the difficulties in that when the constraint is loose, the
problem must be restated with the loose constraint removed and the new dual variables are resolved.
The constraint is loose as D = 1 ft is much lower than L = 56.85 ft. The other item of interest
was that equations dominated by other equations can prevent a solution and must be removed. The
removal of the dominated equation was necessary to obtain a solution and may be the cause of D
being unity.
 9.2. EVALUATIVE QUESTIONS 41
9.2 EVALUATIVE QUESTIONS
1. The problem constraint was given as D ≤ L, but the designer decided that was incorrect and
reversed the constraint to L ≤ D. Resolve the problem and determine the dual and primal
variables as well as the objective function.

2. Resolve the problem making the initial assumption that L = D and reformulate the primal
and dual problems and find the variables and objective function.

REFERENCES
[1] http://www.mpri.lsu/edu/textbook/Chapter3-b.htm (visited May 2009). 37

[2] Ray, W.H. and J Szekely, Process Optimization with Applications in Metallurgy and Chemical
Engineering, John Wiley and Sons, Inc., New York (1973). 37
 43

CHAPTER 10

Gas Transmission Pipeline Case

Study
10.1 PROBLEM STATEMENT AND SOLUTION
The energy crisis is with us today, and one of the problems is in the transmission of energy. A gas
transmission model was developed [1, 2] to minimize the total transmission cost of gas in a new gas
transmission pipeline. The problem is more difficult than the previous case studies as several of the
exponents are not integers. The primal expression for the cost developed was:

C = C1 ∗ L1/2 ∗ V /(F 0.387 ∗ D 2/3 ) + C2 ∗ D ∗ V + C3 /(L ∗ F ) + C4 ∗ F /L . (10.1)

Subject to:
(V /L) ≥ F .

The constraint must be restated in the geometric form as:

−(V /(LF )) ≤ −1 , (10.2)

where
L = Pipe length between compressors (feet)
D = Diameter of Pipe (in)
V = Volume Flow Rate (ft3 /sec)
F = Compressor Pressure Ratio Factor.
Figure 10.1 is a sketch of the problem indicating the variables and is not drawn to scale.

Compressor Compressor
L

F D V F

Figure 10.1: Gas transmission pipeline.

44 10. GAS TRANSMISSION PIPELINE CASE STUDY
For the specific problem, the values of the constants were:

C1 = 4.55 ∗ 105
C2 = 3.69 ∗ 104
C3 = 6.57 ∗ 105
C4 = 7.72 ∗ 105 .

From the coefficients and signs, the signum values for the dual are:

σ01 = 1
σ02 = 1
σ03 = 1
σ04 = 1
σ11 = −1
σ1 = −1 .

The dual problem formulation is:

Objective Function ω01 + ω02 + ω03 + ω04 =1 (10.3)

L terms 0.5ω01 − ω03 − ω04 + ω11 =0 (10.4)
F terms − 0.387ω01 − ω03 + ω04 + ω11 =0 (10.5)
V terms ω01 + ω02 − ω11 =0 (10.6)
D terms −0.667ω01 + ω02 =0. (10.7)

Using Equations (10.3) to (10.7), the values of the dual variables were found to be:

ω01 = 0.26087
ω02 = 0.17391
ω03 = 0.44952
ω04 = 0.11570
ω11 = 0.43478

and by definition
ω00 = 1 ,
and

ω10 = ωmt = σm σmt ωmt = (−1) ∗ (−1 ∗ 0.43478) = 0.43478 where m = 1 and t = 1 .
 10.1. PROBLEM STATEMENT AND SOLUTION 45
The objective function can be found using the dual expression:
 σ

M 
Tm
Y = d(ω) = σ (Cmt ωmo /ωmt )σmt ωmt (10.8)
m=0 t=1
= 1[[{(4.55 ∗ 105 ∗ 1/0.26087)}(1∗0.26087) ]∗[{(3.69 ∗ 104 ∗ 1/0.17391)}(1∗0.17391) ]∗
[{(6.57 ∗ 105 ∗ 1/0.44952)}(1∗0.44952) ]∗[{(7.72 ∗ 105 ∗ 1/0.11570)}(1∗0.11570) ]∗
[{(1 ∗ 0.43478/0.43478)}(−1∗0.43478) ]]1
= $1.3043 ∗ 106 / yr .

The degrees of difficulty are equal to:

D = T − (N + 1) = 5 − (4 + 1) = 0 . (10.9)

The values for the primal variables can be determined from the relationships between the
primal and dual which are:

C1 ∗ L1/2 ∗ V /(F 0.387 ∗ D 2/3 ) = ω01 Y (10.10)

C2 ∗ D ∗ V = ω02 Y (10.11)
C3 /(L ∗ F ) = ω03 Y (10.12)
C4 ∗ F /L = ω04 Y (10.13)
V /(F ∗ L) = ω11 /ω10 = 1 . (10.14)

The fully general expressions are somewhat difficult, but the variables can be expressed in
terms of the constants and objective function as:

F = [(C3 ω04 )/(C4 ω03 )]1/2 (10.15)

V = C3 /(ω03 ∗ Y ) (10.16)
L = [(C3 C4 )/(ω03 ω04 )]/Y (10.17)
D = [(ω02 ∗ ω03 )/(C2 ∗ C3)] ∗ Y 2 . (10.18)

Using the values of Y = 1.3043 ∗ 106 , C1 = 4.55 ∗ 105 , C2 = 3.69 ∗ 104 , C3 = 6.57 ∗ 105 ,
C4 = 7.72 ∗ 105 , ω01 = 0.26087, ω02 = 0.17391, ω03 = 0.44952, ω04 = 0.11570, and ω11 =
0.43478 one obtains:

F = [(C3 ∗ ω04 )/(C4 ∗ ω03 )]1/2 = [(6.57 ∗ 105 ∗ 0.11570)/(7.72 ∗ 105 ∗ 0.44952)]1/2 = 0.468
V = C3 /(ω03 ∗ Y ) = 6.57 ∗ 105 /(0.44952 ∗ 1.3043 ∗ 106 ) = 1.1205 ft3 /sec
L = [(C3 ∗ C4 )/(ω03 ∗ ω04 )]1/2 /Y = [(6.57 ∗ 105 ∗ 7.72 ∗ 105 )/(0.44952 ∗ 0.1157)]1/2 /1.3043 ∗ 106 = 2.3943 ft
D = [(ω02 ∗ ω03 )/(C2 ∗ C3 )] ∗ Y 2 = [(0.17391 ∗ 0.44952)(3.69 ∗ 104 6.57 ∗ 105 )] ∗ (1.3043 ∗ 106 )2 = 5.4857 in .
46 10. GAS TRANSMISSION PIPELINE CASE STUDY
The primal expression can now be solved using the primal variables and the contribution of
each of the terms can be observed.

C = C1 ∗ L1/2 ∗ V /(F 0.387 ∗ D 2/3 ) + C2 ∗ D ∗ V + C3 /(L ∗ F ) + C4 ∗ F /L

= 4.55 ∗ 105 ∗ 2.39431/2 ∗ 1.1205/(0.4680.387 ∗ 5.48572/3 ) + 3.69 ∗ 104 ∗ 5.4857 ∗ 1.1205+
6.57 ∗ 105 /(2.3943 ∗ 0.468) + 7.72 ∗ 105 ∗ 0.468/2.3943
= 3.4029 ∗ 105 + 2.26814 ∗ 105 + 5.8633 ∗ 105 + 1.5090 ∗ 105
= $1, 304, 300 .

The third term is slightly higher than the others, but all terms are of the same magnitude.
Since the constraint is binding, that is V = L ∗ F , and the results indicate that holds as:

1.1205 = 2.3943 ∗ 0.468 = 1.1205 .

The values of the dual variables were more complex for this problem than the previous prob-
lems, but the values of these dual variables still have the same relationship to the terms of the primal
cost function. The first dual variable, ω01 was 0.26087, and the relation between the first cost term
of the primal to the total cost is 3.4029 ∗ 105 /1.3043 ∗ 106 = 0.2609. The reader should show that
the other dual variables have the same relationships between the terms of the primal cost function
and the total primal cost.

10.2 EVALUATIVE QUESTIONS

1. Resolve the problem with the values of C1 = 6 ∗ 105 , C2 = 5 ∗ 104 , C3 = 7 ∗ 105 , and C4 =
8 ∗ 105 . Determine the effect upon the dual variables, the objective function, and the primal
variables. Also examine the percentage of each of the primal terms in the objective function
and in the original objective function.

2. The constraint is a binding constraint. If the constraint is removed, the objective function
should be lower. What problem(s) occurs when the constraint is removed that causes concern?

REFERENCES
[1] http://www.mpri.lsu.edu/textbook/Chapter3-b.htm (visited May 2009). 43

[2] Sherwood, T.K., A Course in Process Design, MIT Press, Cambridge, Mass. (1963). 43
 47

CHAPTER 11

Profit Maximization Case Study

11.1 PROFIT MAXIMIZATION AND GEOMETRIC
PROGRAMMING
The examples thus far have been minimization problems and the maximization problems are slightly
different. In minimization problems, the terms of the objective function tend to have positive signs
or all positive Signum functions. The constraints may have negative signum functions in a mini-
mization problem, but those in the objective function tend to be positive. In the profit maximization
problem, the revenues have positive coefficients and the costs have negative coefficients.The solution
formulation for the maximization problem is slightly different than the minimization problem and
a solution to an example is presented to illustrate the problem formulation and solution procedure.

11.2 PROFIT MAXIMIZATION USING THE COBB-DOUGLAS

PRODUCTION FUNCTION
The problem selected was presented by Liu [1] who adapted it from Keyzer and Wesenbeeck [2].
The objective is to maximize the profit, π , using a Cobb-Douglas production function and the
component costs. The problem given is

Maximize π = pAx10.1 x20.3 x30.2 − C01 x1 − C02 x2 − C03 x3 (11.1)

where
Parameter Term Value for Problem
p = market price = 20
A = scale of production(Cobb-Douglas function) = 40
C01 = Cost of Product 1 = 20
C02 = Cost of Product 2 = 24
C03 = Cost of Product 3 = 30
C00 = pA = 800
To solve the problem, one minimizes the negative of the profit function; that is the primal
objective function is:

Minimize Y = −C00 x10.1 x20.3 x30.2 + C01 x1 + C02 x2 + C03 x3 (11.2)

were Y = −π .
48 11. PROFIT MAXIMIZATION CASE STUDY
The dual objective function would be:

D(Y ) = −1[(C00 /ω01 )−ω ω ω ω −1

01 (C01 /ω02 )02 (C02 /ω03 )03 (C03 /ω04 )04 ] . (11.3)

For maximization of profits, the signum function σ00 = −1, whereas in the dual objective function
for the minimization of costs the signum function σ00 = 1, a positive value. From the coefficients
and signs, the signum values for the dual are:

σ01 = −1
σ02 = 1
σ03 = 1
σ04 = 1.

The dual formulation would be:

Objective Fctn − ω01 +ω02 + ω03 +ω04 = −1(max) (11.4)

x1 terms − 0.1ω01 +ω02 = 0 (11.5)
x2 terms − 0.3ω01 + ω03 = 0 (11.6)
x3 terms − 0.2ω01 +ω04 = 0 (11.7)

Solving Equations (11.4) through (11.7) for the dual variables one obtains:

ω01 = 2.50
ω02 = 0.25
ω03 = 0.75
ω04 = 0.50 .

The dual variables do not sum to unity which was the case for the cost minimization problems.
However, the difference between the profit dual variable and cost dual variables is equal to unity.
Also, the cost terms should be in the same ratio as their corresponding dual variables. The dual
objective function would result in

D(Y ) = −1[(800/2.5)−2.5 (20/0.25)0.25 (24/0.75)0.75 (30/0.50)0.50 ]−1 (11.8)

= −5, 877.12 .

The primal variables can be found from the primal-dual relationships and the values of the dual
objective function and dual variables similar to that in the cost minimization procedure.
Thus,

x1 = ω02 Y/C02 = 0.25 × 5877.12/20 = 73.46 (11.9)

x2 = ω03 Y/C03 = 0.75 × 5877.12/24 = 183.66 (11.10)
x3 = ω04 Y/C04 = 0.50 × 5877.12/30 = 97.95 . (11.11)
 11.3. EVALUATIVE QUESTIONS 49
If the values for x1 , x2 and x3 are used in the primal, one obtains:
Maximize π = pAx10.1 x20.3 x30.2 − C01 x1 − C02 x2 − C03 x3 (11.1)
= 800(73.46) (183.66) (97.95) − 20 × 73.46 − 24 × 183.66 − 30 × 97.95
0.1 0.3 0.2

= 14, 692.66 − 1469.2 − 4, 407.84 − 2, 938.5

π= 5, 877.12
Therefore, the values of the primal and dual objective functions are equal. The three cost terms of
1469.2, 4,407.84, and 2.938.5 are in the same ratio as the cost dual variables of 0.25, 0.75, and 0.50
or 1:3:2. The primal variables, x1 , x2 and x3 , are not in the same ratio as the dual variables.

11.3 EVALUATIVE QUESTIONS

1. What is the history of the Cobb-Douglas Production Function?
2. What is the primary difference between the dual objective functions for the minimum cost
and maximum profit models?
3. The profit function is:
Maximize π = pAx10.1 x20.3 x30.2 − C01 x1 − C02 x2 − C03 x30.80
And the values for the primal coefficients are:
pA = $500
C01 = $40
C02 = $30
C03 = $20
(a) Solve for the dual variables and the dual objective function.
(b) Solve for the primal variables and the primal objective function.
(c) Show that the cost terms of the primal are in the same ratio as the dual variables.
4. The following problem is from the web site http://www.mpri.lsu.edu/textbook/
Chapter3-b.htm
Maximize the function:
Y = 3x10.25 − 3x11.1 x20.6 − 115x2−1 x3−1 − 2x3 .
(a) Solve for the dual variables and the dual objective function.
(b) Solve for the primal variables and the primal objective function.
(c) Show that the cost terms (negative functions) of the primal are in the same ratio as the
dual variables.
(Hint – solve the dual in terms of fractions answers rather than decimal values.)
50 11. PROFIT MAXIMIZATION CASE STUDY
REFERENCES
[1] Shiang-Tai Liu,“A Geometric Programming Approach to Profit Maximization”, Applied Math-
ematics and Computation, 182, (2006), pp 1093–1097. 47

[2] M. Keyzer and L.Wesenbeck,“Equilibrium Selection in Games; the Mollifier Method”, Journal
of Mathematical Economics, 41, (2005), pp 285–301. 47
 51

CHAPTER 12

Material Removal/Metal
Cutting Economics Case Study
12.1 INTRODUCTION
Material removal economics, also known as metal cutting economics or machining economics, is an
example of a problem which has non-integer exponents, and this makes the problem challenging.
This problem has been presented previously [1, 2], but this version is slightly different from, and
easier than, those presented earlier. The material removal economics problem is based upon the
Taylor Tool Life Equation, which was developed by Frederick W. Taylor over 100 years ago in the
USA.There are several versions of the equation, and the form selected is one of the modified versions
which includes cutting speed and feed rate. The equation selected was:
T V 1/n f 1/m = C , (12.1)
where
T = tool life (minutes)
V = cutting speed (ft/min or m/min)
F = feed rate (inches/rev or mm/rev)
1/n = cutting speed exponent
1/m = feed rate exponent
C = Taylor’s Modified Tool Life Constant (ft/min or m/min).
The object is to minimize the total cost for machining, operator, tool cost and tool changing
cost.

12.2 PROBLEM FORMULATION

An expression for the machining cost, operator cost, tool cost and tool changing cost was devel-
oped [1] and the resulting expression was:
Cu = K00 + K01 f −1 V −1 + K02 f (1/m−1) V (1/n−1) , (12.2)
where
Cu = total unit cost
K00 = (Ro + Rm )tl
K01 = (Ro + Rm )B
K02 = [(Ro + Rm )tch + Ct ]QBC −1 ,
52 12. MATERIAL REMOVAL/METAL CUTTING ECONOMICS CASE STUDY
and
Ro = operator rate ($/min), derived from Operator’s Hourly Cost
Rm = machine rate ($/min), derived from Machine Hour Cost
tl = machine loading & unloading time (min)
tch = tool changing time (min)
B = cutting path surface factor of tool (in-ft, or mm-m)
Q = fraction of cutting path that tool is cutting material
Ct = tool cost ($/cutting edge), derived from Tool Insert Cost
C = Taylor’s Modified Tool Life Constant (min).
The first part of the objective function expression represents the loading and unloading costs,
the second part represents the cutting costs, and the third part represents the tool and tool changing
costs. The loading and unloading costs are not a function of the feed and cutting speed. Since K00
is a constant, the primal problem can be formulated as solving for the variable cost, Cu (var) as:

Cu (var) = K01 f −1 V −1 + K02 f (1/m−1) V 1/n−1) . (12.3)

Subject to a maximum feed constraint written as:

K11 f ≤ 1 , (12.4)

where
K11 = 1/fmax .
From the coefficients and signs, the signum values for the dual are:

σ01 =1
σ02 =1
σ11 =1
σ1 =1.

The dual problem formulation is:

Objective Function ω01 + ω02 =1 (12.5)

f terms −ω01 + (1/m − 1)ω02 + ω11 = 0 (12.6)
V terms −ω01 + (1/n − 1)ω02 =0. (12.7)

The degrees of difficulty are equal to:

D = T − (N + 1) = 3 − (2 + 1) = 0 . (12.8)

From the constraint equations, which have only one term, it is apparent that:

ω10 = ω11 . (12.9)

 12.2. PROBLEM FORMULATION 53
Since there are zero degrees of difficulty, the dual parameters can be solve directly. Thus, if
one adds Equations (12.5) and (12.7), one can solve for ω02 directly and obtain:
ω02 = n . (12.10)
Then, from Equation (12.5), one obtains;
ω01 = 1 − n . (12.11)
Finally, by using the values for ω01 and ω02 , one can determine ω11 as:
ω11 = 1 − n/m . (12.12)
The objective function can be evaluated using the dual expression:
 M T σ
  m
Cu (var) = d(ω) = σ (Cmt ωmo /ωmt ) σmt ωmt
(12.13)
m=0 t=1
Cu (var) = 1{[(K01 ω00 /ω01 )ω01 ][(K02 ω00 /ω02 )ω02 ][(K11 ω10 /ω11 )ω11 ]}1 (12.14)
= [(K01 1/(1 − n)) (1−n)
][(K02 1/n) ][(K11 (1 − n/m)/(1 − n/m))
n (1−n/m)
]
= [K01 /(1 − n)][(K02 /K01 )((1 − n)/n)] [(K11 )
n (1−n/m)
]
1−n n
= (K11 )1−n/m K01 K02 (1 − n)n−1 /nn . (12.15)
The primal variables, V and f , can be evaluated from the primal-dual relationships.
K01 f −1 V −1 = ω01 Cu (var) (12.16)
K02 f 1/m−1 V 1/n−1 = ω02 Cu (var) (12.17)
K11 f = 1 . (12.18)
From Equation (12.18), is seen that:
f = 1/K11 . (12.19)
If one divides Equation (12.17) by Equation (12.16), one obtains:
f 1/m V 1/n = (K01 /K02 )(n/(1 − n)) . (12.20)
Using Equation (12.19) in (12.20) and solving for V , one obtains
n/m
V = [(n/(1 − n)]n (K01 /K02 )n K11 . (12.21)
Now, if one uses the values of f and V from Equations (12.19) and (12.21) in the primal
Equation (12.3) and also using Equation (12.20), one obtains:
Cu (var) = K01 f −1 V −1 + K02 f (1/m−1) V (1/n−1) , (12.3)
= f −1 V −1 [K01 + K02 f 1/m V 1/n ]
−n/m
= K11 [(n/(1 − n))−n (K01 /K02 )−n K11 [K01 + K02 (n/(1 − n)(K01 /K02 )]
(n/(1 − n))−n (K01 /K02 )−n [K01
1−n/m
= K11 + K01 (n/(1 − n)]
1−n/m 1−n n
= K11 K01 K02 (1 − n)n−1 /nn . (12.22)
54 12. MATERIAL REMOVAL/METAL CUTTING ECONOMICS CASE STUDY
The variable unit cost expressions, Cu (var), are identical for both the primal and dual formu-
lations. The expressions for the primal variables and the variable unit cost are more complex than
the expressions obtained in the previous models because of the non-integer exponents.
Elaborate research work has been done with the material removal problems, and the disser-
tation by Pingfang Tsai [3] has solutions for problems with an additional variable, the depth of cut,
and additional constraints on horsepower of the motor driving the main spindle of the lathe and
depth of cut. With the additional constraints, there is the possibility of loose constraints, and a flow
chart has been developed for the different solutions depending upon which constraints are loose.
Chapter 16 presents a material removal problem with one degree of difficulty and two constraints.

12.3 EVALUATIVE QUESTIONS

1. A cylindrical bar, 6 inches long and 1 inch in diameter is to be finished turned on a lathe. The
maximum feed to be used to control the surface finish is 0.005 in/rev. Find the total cost to
machine the part, the variable cost to machine the part, the feed rate, the cutting speed, and
the tool life in minutes. Use both the primal and dual equations to determine the variable unit
cost. The data are:
Ro = 0.60 $/min
Rm = 0.40 $/min
Ct = $ 2.00/edge
tl = 1.5 min
tc h = 0.80 min
D = 1 inch
L = 6 inches
1/m = 1.25 (m = 0.80)
1/n = 4.00 (n = 0.25)
C = 5.0 x 108 min
Q = 1.0 (for turning).
Using these values, one can obtain:

K00 = 1.50
K01 = 1.57 in-ft
K02 = 8.8 x 10−9
(solution f = 0.005 in/rev, V = 459 ft/min, Cu (var) = 0.91, and T = 8.5 min).

2. A cylindrical bar, 150 mm long and 25 mm in diameter is to be finished turned on a lathe. The
maximum feed to be used to control the surface finish is 0.125 mm/rev. Find the total cost to
machine the part, the variable cost to machine the part, the feed rate, the cutting speed, and
the tool life in minutes. The data are:
 REFERENCES 55
Ro = 0.60 $/min
Rm = 0.40 $/min
Ct = $ 2.00/edge
tl = 1.5 min
tc h = 0.8 min
D = 25 mm
L = 150 mm
1/m = 1.25 (m = 0.80)
1/n = 4.00 (n = 0.25)
C = 2.46 x 108 min
Q = 1.0 (for turning).
Using these values, one can obtain:
K00 = 1.50
K01 = 11.78 mm-m
K02 = 1.34 x 10−7
(solution f = 0.125 mm/rev, V =140 m/min, Cu (var)=0.90, and T =8.6 min).

REFERENCES
[1] Robert C. Creese and Pingfang Tsai, “Generalized Solution for Constrained Metal Cutting
Economics Problem,” 1985 Annual International Industrial Conference Proceedings, Institute of
Industrial Engineers U.S.A, pp 113–117. 51

[2] Ermer, D.S., “Optimization of the Constrained Machining Economics Problem by Geometric
Programming,” Journal of Engineering for Industry, Transactions of the ASME, November
1971, pp 1067–1072. 51

[3] Tsai, Pingfang An Optimization Algorithm and Economic Analysis for a Constrained Machining
Model, PhD Dissertation, West Virginia University, Morgantown, WV. 54
 PART III

Geometric Programming Applications

with Positive Degrees of Difficulty
 59

CHAPTER 13

Journal Bearing Design Case

Study
13.1 INTRODUCTION
An interesting problem with one degree of difficulty is a journal bearing design problem presented
by Beightler, Lo, and Bylander [1]. The objective was to minimize the cost (P ), and the variables
were the half-length of the bearing (L) and the radius of the journal (R). The objective function and
the constants in the problem are those presented in the original paper, and the derivations of the
constants were not detailed. The solution is based upon deriving an additional equation whereas the
original problem was solved by reducing the degree of difficulty and determining upper and lower
bounds to the solution. The solution presented solves the problem, directly using the additional
equation and without needing to use search techniques.

13.2 PRIMAL AND DUAL FORMULATION OF JOURNAL

BEARING DESIGN
The generalized primal problem was:

Minimize P = C01 R 3 L−2 + C02 R −1 + C03 RL−3 (13.1)

Subject to: C11 ∗ R −1 ∗ L3 ≤ 1 , (13.2)

where
P = Cost ($)
R = radius of the journal (in)
L = half-length of the bearing (in)
C01 = 0.44 (for example problem)
C02 = 10 (for example problem)
C03 = 0.592 (for example problem)
and C11 = 8.62 (for example problem).
Figure 13.1 is a sketch illustrating the variables for the problem.
From the coefficients and signs, the signum values for the dual from Equations (13.1)
and (13.2) are:
60 13. JOURNAL BEARING DESIGN CASE STUDY

R
L

Figure 13.1: Journal bearing parameters of half-lengh and radius.

σ01 =1
σ02 =1
σ03 =1
σ11 =1
σ1 =1.

The dual problem formulation is:

Objective Function ω01 + ω02 + ω03 =1 (13.3)

R terms 3ω01 − ω02 + ω03 − ω11 = 0 (13.4)
L terms −2ω01 − 3ω03 + 3ω11 = 0 . (13.5)

From the constraint equation there is only one term, so:

ω10 = ω11 . (13.6)

This adds one additional equation but also one additional term, so the degrees of difficulty
are equal to:
D = T − (N + 1) = 4 − (2 + 1) = 1 ≥ 0 . (13.7)
The dual has more variables than equations, and thus another equation is needed to solve for
the dual variables. The relationships between the primal and dual variables will be used to determine
an additional equation, and the equation typically is non-linear. The approach presented [2] is
the “substitution approach” to obtain the additional equation needed. The “dimensional analysis
approach” will be presented later in this Chapter. The relationships between the primal and dual
which are:

C01 R 3 L−2 = ω01 P (13.8)

C02 R −1 = ω02 P (13.9)
C03 RL−3 = ω03 P (13.10)
C11 R −1 L3 = (ω11 /ω10 ) . (13.11)
 13.2. PRIMAL AND DUAL FORMULATION OF JOURNAL BEARING DESIGN 61
Since ω10 = ω11 , Equation (13.11) can be used to relate the primal variables, that is:

R = C11 L3 . (13.12)

Using Equation (13.12) in Equation (13.9), one obtains:

P = C02 /(ω02 R)
= [C02 /(C11 ω02 L3 )] . (13.13)

Using Equation (13.10) with Equations (13.12) and (13.13), one obtains after reducing terms:

L3 = (C03 R)/(ω03 P )
= [(C02 /(C03 C11
2
)) ∗ (ω03 /ω02 )] . (13.14)

Now using Equation (13.8) and the values for R and L, one obtains:

(C01 C11 L )/C02 = ω01 /ω02 .

4 10
(13.15)

Now using Equation (13.14) in Equation (13.15) and reducing it, one can obtain:
7/3 10/3 8/3 7/3 10/3
(C01 C02 )/(C03 C11 ) = (ω01 ω02 /ω03 ) , (13.16)

or the form of
3 7
(C01 C02 )/(C03 C11 ) = (ω01
10 8 3 7 10
ω02 /ω03 ). (13.17)

Now using Equations (13.3) to (13.5) to solve for the dual variables in terms of ω02 , one obtains:

ω01 = (3/7)ω02 (13.18)

ω03 = 1 − (10/7)∗ ω02 (13.19)
ω11 = 1 − (8/7)∗ ω02 . (13.20)

Using Equations (13.18) to (13.20) in Equation (13.17), one can obtain:

(3/7ω02 )3 (ω02 )7 /[1 − ((10/7) ∗ ω02 )] = (C01

3 7
C11 ) = A
10 8
C02 )/(C03
or (ω02 /(1 − ((10/7)ω02 ) = [A ∗ (7/3) ] 3 1/10
=B
or ω02 = (7B/(7 + 10B)) . (13.21)

Thus, the remaining dual variables can be solved for as:

ω01 = 3B/(7 + 10B) (13.22)

ω03 = 7/(7 + 10B) (13.23)
ω11 = (7 + 2B)/(7 + 10B) . (13.24)
62 13. JOURNAL BEARING DESIGN CASE STUDY
Using Equations (13.10) and (13.12)
P = (C03 /ω03 )RL−3
= (C03 /ω03 )(C11 L3 )L−3
= C11 C03 /ω03
= (C11 C03 )(1 + (10/7)B)
= (C11 C03 )(1 + (10/7)[(7/3)3 A]1/10 )
= (C11 C03 )(1 + (10/7)(7/3)3/10 [(C01
3 7 10 8 1/10
C02 )/(C03 C11 )]
= C11 C03 + C11 C03 (10/7)(7/3) 3/10
[(C01 C02 )/(C03
3 7 10 8 1/10
C11 )]
= C11 C03 + (10/7)[((7/3)(C01 )] 3/10
∗ (C02 ) 7/10
∗ (C11 )2/10 . (13.25)
Using Equation (13.9) to solve for R, one obtains:
R = C02 /(ω02 P )
= C02 /[(7B/(7 + 10B)) ∗ (C11 C03 )(1 + (10/7)B)
= C02 /[(7B/(7 + 10B)) ∗ (C11 C03 )(7 + (10B)7)
= (C02 /(C11 C03 ))/B
= (C02 /(C11 C03 ))/[(7/3)3 ∗ C01
3 7 10 8 1/10
C02 /(C03 C11 )]
−2/10
= [(3/7) ∗ (C02 /C01 )]3/10 C11 . (13.26)
Using Equation (13.10) to solve for L, one obtains:
L = [C03 R/(ω03 P )]1/3
−2/10
= [[C03 ∗ [(3/7) ∗ (C02 /C01 )]3/10 C11 ]/[ω03 ∗ C11 C03 /ω03 ]]1/3
−4/10
= [(3/7)(C02 /C01 )]1/10 ∗ C11 . (13.27)
The equations for P , R, and L are general equations but are rather complex equations com-
pared to the previous problems illustrated. The solution was based upon determining an additional
equation from the primal-dual relationships, which was highly non-linear and resulted in rather com-
plex expressions for the variables. The additional equation along with the dual variables was used in
the equations relating the primal and dual to determine the final expressions for the variables. This
frequently happens when the degrees of difficulty are greater than zero.
For this particular example problem where C01 = 0.44, C02 = 10, C03 = 0.592 and C11 =
8.62, the value for A and B are:
A = (C01
3 7
C11 ) = [(0.44)3 (10)7 ]/[(0.592)10 (8.62)8 ] = 5.285
10 8
C02 )/(C03 (13.28)
B = [A(7/3)3 ]1/10 = [5.285(7/3)3 ]1/10 = 1.523 . (13.29)
Now using the equations for the dual variables, Equations (13.21) to (13.24), one obtains
ω02 = 7B/(7 + 10B) = 0.480
ω01 = 3B/(7 + 10B) = 0.205
ω03 = 7/(7 + 10B) = 0.315
ω11 = (7 + 2B)/(7 + 10B) = 0.452 .
 13.3. DIMENSIONAL ANALYSIS TECHNIQUE FOR ADDITIONAL EQUATION 63
From Equation (13.25) the value of P can be found as:

P = C11 C03 + (10/7)[((7/3)(C01 )]3/10 ∗ (C02 )7/10 ∗ (C11 )2/10

= (8.62)(0.592) + (10/7)[((7/3)(0.44)]0.3 (10)0.7 (8.62)0.2
= 5.10 + 11.10
= $16.2 .

The primal variables can be determined from Equations (13.26) and (13.27) as:

−2/10
R = [(3/7)(C02 /C01 )]3/10 C11
= [(3/7)(10/0.44]3/10 8.62−2/10
= 1.29 in
−4/10
L = [(3/7)(C02 /C01 )]1/10 C11
= [(3/7)(10/0.44)]1/10 8.62−4/10
= 0.530 in .

Now, if the values of R and L are used in Equation (13.1) for evaluating the primal, one
obtains:

P = C01 R 3 L−2 + C02 R −1 + C03 RL−3

= 0.44(1.29)3 (0.53)−2 + 10(1.29)−1 + 0.592(1.29)(0.53)−3
= 3.363 + 7.752 + 5.130
= $16.2 .

As in the previous case studies, the value of the primal and dual objective functions are
equivalent.

13.3 DIMENSIONAL ANALYSIS TECHNIQUE FOR

ADDITIONAL EQUATION
It was difficult to determine the additional equation by repeated substitution and other methods
can be used. One method is the technique of dimensional analysis and this will be demonstrated
to obtain Equation (13.17). The dimensional analysis approach sets up the primal dual relations
of Equations (13.8)-(13.11) and setting the primal variables on one side and the dual variables,
constants, and objective function on the other side and giving the terms variable exponents as
illustrated in Equation (13.30)

(R 3 L−2 )A (R −1 )B (RL−3 )C (R −1 L3 )D = 1
= (ω01 P /C01 )A (ω02 P /C02 )B (ω03 P /C03 )C (1/C11 )D . (13.30)
64 13. JOURNAL BEARING DESIGN CASE STUDY
One must balance the exponents to remove the primal variables (R, L) and the dual objective
function (P ). This is done by:

R values 3A − B + C − D= 0 (13.31)
L Values − 2A − 3C + 3D= 0 (13.32)
P values A+B + C =0. (13.33)

There are four variables and three equations, so one will find three variables in terms of the
fourth variable. If one adds 3 times Equation (13.31) to Equation (13.32), one obtains:

A = 3/7B . (13.34)

Using Equations (13.34) and (13.33), one can determine that:

B = −7/10C , (13.35)

which results in:

A = −3/10C . (13.36)
Using Equations (13.31), (13.35), and (13.36), one obtains that

D = 8/10C . (13.37)

If one lets C = 10, the A = −3, B = −7 and D = 8. Using these values in Equation (13.30),
one can obtain Equation (13.17), that is:
3 7
(C01 C02 )/(C03 C11 ) = (ω01
10 8 3 7
ω02 10
/ω03 ). (13.17)

Another solution method that is often used is the constrained derivative approach. This
method has the dual equations rearranged in terms of one unknown dual variable and these are
substituted into the dual objective function. The objective function is set into logarithmic form
and differentiated with respect to the unknown dual variable, set to zero, and then solved for the
unknown dual variable. The solved dual variable is used in the dual equations to obtain the values
of the other dual variables. This technique is demonstrated in Chapters 16, 17, and 18.

13.4 EVALUATIVE QUESTIONS

1. Use the values of C01 = 0.54, C02 = 10, C03 = 0.65 and C11 = 13.00, determine the values
of A and B, of the dual variables and the value of the objective function. Also determine the
values of L and R, and use these to determine P .

2. Determine the sensitivity of the objective function and the primal variables of R and L by
changing one of the constants by 20% (such as C01 ).
 REFERENCES 65
3. Use dimensional analysis to develop the additional equation for the dual variables (ω values) in
terms of the constants (C values) from the following data where D and H are primal variables
and Y is the dual objective function.

C01 D 2 H = ω01 Y
C02 D 3 = ω02 Y
C11 D −1 = ω11 /ω10
C12 H −1 = ω12 /ω10
C13 DH −1 = ω13 /ω10
[ Answer (C02 /C01 )(C11 /C12 ) = (ω02 /ω01 )(ω11 /ω12 )] .

REFERENCES
[1] Beightler, C.S., Lo, Ta-Chen, and Bylander, H.G., “Optimal Design by Geometric Program-
ming,” ASME, Journal of Engineering for Industry, 1970, pp. 191–196. 59

[2] Creese, R.C., “A Primal-Dual Solution Procedure for Geometric Programming,” ASME,
Journal of Mechanical Design, 1971. (Also as Paper No. 79-DET-78.) 60
 67

CHAPTER 14

Metal Casting Hemispherical

Top Cylindrical Side Riser
Case Study
14.1 INTRODUCTION
The sphere is the shape which will give the longest solidification time, so a riser with a hemispheric
shaped top should be a more efficient riser than a cylindrical riser and thus be more economical.
However, the problem of designing this type of riser is more complex and has two degrees of difficulty.
The case study considered is a side riser with a hemispherical top and a cylindrical bottom for ease
of molding and to provide a better connection to the casting. The top hemispherical cap has the
same diameter(D) as the cylinder.

14.2 PROBLEM FORMULATION

The volume of the riser is the sum of the cylinder part and the hemisphere part and can be written
as:

V = cylindrical part + Hemispherical part

V = π D H /4
2
+ π D 3 /12 (14.1)
SA = π D /4 + π DH + π D 2 /2
2

= 3/4π D 2 + π DH . (14.2)

The constraint for riser design is Chvorinov’s Rule, which is

t = K(V /SA)2 , (14.3)

where
t = solidification time (minutes or seconds)
K = solidification constant for molding material (minutes/in2 or seconds/cm2 )
V = volume (in3 or cm3 )
SA = cooling surface area (in2 or cm2 ).
An illustration of the hemispherical top side riser is shown in Figure 14.1 where the radius of
the hemisphere is the same as the radius of the cylinder which is the diameter (D) divided by two.
68 14. METAL CASTING HEMISPHERICAL TOP CYLINDRICAL SIDE RISER
D

Riser H

Casting

Figure 14.1: Hemispherical top side riser design.

This results in the relation:

(V /SA) ≥ Mc = K , (14.4)
where
Mc = the modulus of the casting (a constant K for a particular casting).
Thus, using the cooling surface area (SA) and volume (V ) expressions, Equation (14.4) can
be rewritten as:

V /SA = (πD 2 H /4 + π D 3 /12)/(3/4π D 2 + π DH ) ≥K

= (πD 2 /12) ∗ (3H + D)/[(π D/4) ∗ (3D + 4H ) ≥K
= D ∗ (3H + D)/[3 ∗ (3D + 4H )] ≥K. (14.5)

Rearranging the equation in the less than equal form results in:

4KD −1 + 3KH −1 − (1/3)DH −1 ≤ 1 . (14.6)

Thus, the primal form of the problem can be stated as:

Min V = π D 2 H /4 + π D 3 /12 . (14.7)

Subject to:
4KD −1 + 3KH −1 − (1/3)DH −1 ≤ 1 . (14.8)
From the coefficients and signs, the signum values for the dual are:

σ01 = 1
σ02 = 1
σ11 = 1
σ12 = 1
σ13 = −1
σ1 = 1.
 14.2. PROBLEM FORMULATION 69
The dual problem formulation is:

Objective Function ω01 + ω02 =1 (14.9)

D terms 2ω01 + 3ω02 − ω11 − ω13 = 0 (14.10)
H terms ω01 − ω12 + ω13 = 0 . (14.11)

The degrees of difficulty are equal to:

D = T − (N + 1) = 5 − (2 + 1) = 2 . (14.12)

Using the linearity inequality equation,


ω10 = ωmt = σm σmt ωmt = (1) ∗ (1 ∗ ω11 + 1 ∗ ω12 + (−1) ∗ ω13 )
ω10 = ω11 + ω12 − ω13 . (14.13)

The dual variables cannot be determined directly as the degrees of difficulty are 2; that is,
there are two more variables than there are equations. The objective function can be found using the
dual expression:
 M T σ
  m
V = d(ω) = σ (Cmt ωmo /ωmt ) σmt ωmt
(14.14)
m=0 t=1
V = 1[[{(π/4) ∗ 1/ω01 )}(1 ∗ ω01 )] ∗ [{(π/12 ∗ ω02 )}(1∗ω02 ) ] ∗ [{(4K ∗ ω10 /ω11 )}(1∗ω10 ) ]∗
[{(3K ∗ ω10 /ω12 )}(1∗ω10 ) ] ∗ [{((1/3) ∗ ω10 /ω13 )}(−1∗ω13 ) ]]1 . (14.15)

The relationships between the primal and dual variables can be written as:

(π/4)KD 2 H = ω01 V (14.16)

(π/12)KD 3 = ω02 V (14.17)
(4K)D −1 = ω11 /ω10 (14.18)
(3K)H −1 = ω12 /ω10 (14.19)
(1/3)DH −1 = ω13 /ω10 . (14.20)

If one takes Equation (14.16) and divides by Equation (14.17), one obtains:

3H /D = ω01 /ω02 . (14.21)

If one takes the inverse of Equation 10.20, one obtains:

3H /D = ω10 /ω13 . (14.22)

Now comparing Equations (14.21) and (14.22), one can obtain an equation between the dual
variables as:
ω01 /ω02 = ω10 /ω13 . (14.23)
70 14. METAL CASTING HEMISPHERICAL TOP CYLINDRICAL SIDE RISER
If one takes Equation (14.18) and divides by Equation (14.19), one obtains:

(4/3)H /D = ω11 /ω12 . (14.24)

Now comparing Equations (14.21) and (14.24), one can obtain an additional equation between
the dual variables as:
ω01 /ω02 = (9/4)ω11 /ω12 . (14.25)
Now there are six equations with only the six dual variables, and they are Equations (14.9)
to (14.11), (14.13), (14.23), and (14.25). The procedure used was to solve for all of the variables in
terms of ω02 and then obtain the specific value of ω02 .
From Equation (14.9), one obtains:

ω01 = 1 − ω02 . (14.26)

If one adds Equations (14.10) and (14.11), one obtains:

3ω01 + 3ω02 − ω11 − ω12 = 0 .

Which can be reduced to:

ω11 + ω12 = 3 . (14.27)
Using Equations (14.13) and (14.27), one obtains:

ω10 = ω11 + ω12 − ω13

ω10 = 3 − ω13 . (14.28)

Now using Equations (14.28) and (14.23), one obtains:

ω01 /ω02 = ω10 /ω13 = (3 − ω13 )/ω13 = (1 − ω02 )/ω02 .

Solving for ω13 one obtains:

ω13 = 3ω02 . (14.29)
From Equations (14.28) and (14.29):

ω10 = 3 − ω13 = 3 − 3ω02

ω10 = 3(1 − ω02 ) . (14.30)

Using Equations (14.26) and (14.29) in Equation (14.11), one obtains:

ω12 = ω10 + ω13

= (1 − ω02 ) + 3ω02
= 1 + 2ω02 . (14.31)
 14.2. PROBLEM FORMULATION 71
Using Equations (14.10), (14.26), and (14.29), one has:

ω11 = 2ω01 + 3ω02 − ω13

= 2(1 − ω02 ) + 3ω02 − 3ω02
= 2(1 − ω02 ) . (14.32)

Now using Equation (14.25), one can solve for ω02 using the values for ω01 , ω11 , and ω12

ω01 /ω02 = (9/4)ω11 /ω12

(1 − ω02 )/ω02 = (9/4)2(1 − ω02 )/(1 + 2ω02 ) .

And solving for ω02 results in:

ω02 = 0.4 .
Therefore,
ω01 = 0.6
ω11 = 1.2
ω12 = 1.8
ω13 = 1.2
ω10 = 1.8 .
Now using the dual variables in Equation (14.15) to find the minimum volume, one obtains:

V = 1[[{(π/4 ∗ 1/ω01 )}(1∗ω01 ) ] ∗ [{(π/12 ∗ ω02 )}(1∗ω02 ) ] ∗ [{(4K ∗ ω10 /ω11 )}(1∗ω11 ) ]∗
[{(3K ∗ ω10 /ω12 )}(1∗ω12 ) ] ∗ [{((1/3) ∗ ω10 /ω13 )}(−1∗ω13 ) ]]1
= 1[[{(π/4) ∗ 1/0.6)}(1∗0.6) ] ∗ [{(π/12 ∗ 0.4)}(1∗0.4) ] ∗ [{(4K ∗ 1.8/1.2)}(1∗1.2) ]∗
[{(3K ∗ 1.8/1.8)}(1∗1.8) ] ∗ [{((1/3) ∗ 1.8/1.2)}(−1∗1.2) ]]1
= 1[[{((5/12)π)}(0.6) ] ∗ [{((5/24)π )}(0.4) ] ∗ [{(6K)}(1.2) ] ∗ [{(3K)}(1.8) ] ∗ [{(1/2)}(−1.2) ]]1
= 1[[{((2 ∗ 5/24)π)}(0.6) ] ∗ [{((5/24)π )}(0.4) ] ∗ [{(2 ∗ 3K)}(1.2) ] ∗ [{(3K)}(1.8) ] ∗ [{(2)}(1.2) ]]1
= 20.6 [(5/24)π ](0.6+0.4) ∗ 21.2 ∗ (3K)(1.2+1.8) 21.2
= (5π/24)1 ∗ 2(0.6+1.2+1.2) ∗ (3K)3
V = (5π/24) ∗ (6K)3 . (14.33)

The values for H and D can be found from Equations (14.18) and (14.19)

D = 4K ∗ (ω10 /ω11 )
= 4k ∗ (1.8/1.2)
D = 6K (14.34)

and

H = 3K ∗ (ω10 /ω12 )
= 3K ∗ (1.8/1.8)
H = 3K . (14.35)
72 14. METAL CASTING HEMISPHERICAL TOP CYLINDRICAL SIDE RISER
Now the primal can be evaluated using Equation (14.7) with the values for H and D from
Equations (14.34) and (14.35).

V = π D 2 H /4 + π D 3 /12
V = π(6K)2 (3K)/4 + π(6K)3 /12
= π 27K 3 + π 18K 3
= 45π K 3 (14.36)
= (5π/24) ∗ (6K)3 . (14.37)

The values for the primal and dual are equivalent, which is required for the solution. The
expression of Equation (14.37) is the preferred expression for foundry as 6K is the value for the
diameter for the simple cylindrical risers where K is the modulus of the casting. This example
illustrates that it is possible to solve problems with two degrees of difficulty in some instances,
but there are numerous mathematical operations that must be performed to obtain the additional
equations and solution. There have been several articles written concerning various riser design
shapes and the use of insulating materials to improve casting yield [1, 2, 3, 4, 5].

14.3 DIMENSIONAL ANALYSIS TECHNIQUE FOR

ADDITIONAL TWO EQUATIONS

Although it was relatively easy to determine the two additional equations, they also could have
been obtained by dimensional analysis. The dimensional analysis approach sets up the primal dual
relations of Equations (14.16)-(14.20) and setting the primal variables on one side and the dual
variables, constants, and objective function on the other side and giving the terms variable exponents
as illustrated in Equation (14.38)

(D 2 H )A (D 3 )B (D −1 )C (H −1 )D (DH −1 )E = 1
= (ω01 V /C01 )A (ω02 V /C02 )B (ω11 /ω10 C11 )C (ω12 /ω10 C12 )D (ω13 /ω10 C13 )E , (14.38)

where

C01 = (Kπ/4) (14.39)

C02 = (Kπ/12) (14.40)
C11 = 4K (14.41)
C12 = 3K (14.42)
C13 = 1/3 . (14.43)
 14.3. DIMENSIONAL ANALYSIS TECHNIQUE FOR ADDITIONAL TWO EQUATIONS 73
One must balance the exponents to remove the primal variables (DH ) and the dual objective
function (V ). An equation will also be written to consider ω10 . This is done by:

V values A + B =0 (14.44)
ω10 values +C + D+E =0 (14.45)
D values 2A + 3B−C +E =0 (14.46)
H values A − D−E =0. (14.47)

There are five variables and four equations, so one will find three variables in terms of the
fourth variable. From Equation (14.44), one notes that

A = −B . (14.48)

If Equations (14.46) and (14.47) are added and using the relation of (14.48), one obtains

C = −D . (14.49)

Using the relationship of (14.49) and (14.45), one notes that

E=0. (14.50)

If one takes Equation (14.46) and subtracts two times Equation (14.44) and using E = 0,
one obtains that
C=B. (14.51)
Thus, if one lets B = 1, then C = 1, D = −1, A = −1 and E = 0. Using these values
in (14.38) one obtains that

(ω02 /ω01 )(ω11 /ω12 ) = (C02 /C01 )(C11 /C12 )

= ((Kπ/12)/(Kπ/4))((4K)/(3K) = 4/9 . (14.52)

Upon rearranging terms, this is equivalent to Equation (14.25).

The second equation is obtained by allowing ω10 to be part of the new equation. Thus, taking
the Equations (14.44)-(14.47) and allow ω10 to enter the solution by making the RH S = 1 in
Equation (14.45), which is renumbered as (14.53).

V values A + B =0 (14.44)
ω10 values +C + D+E =1 (14.53)
D values 2A + 3B−C +E =0 (14.46)
H values A − D−E =0. (14.47)

There are five variables and four equations, so one will find three variables in terms of the
fourth variable. From Equation (14.44), one notes that

A = −B . (14.48)
74 14. METAL CASTING HEMISPHERICAL TOP CYLINDRICAL SIDE RISER
If Equations (14.46) and (14.47) are added and using the relation of (14.48), one obtains

C = −D . (14.49)

Using the relationship of (14.49) and (14.52), one notes that

E=1. (14.53)

Using Equations (14.47) and (14.53), one obtains that

A=D+1 (14.54)

and thus
B = −(D + 1) . (14.55)
Using Equation (14.46) and (14.48), one obtains that

C = B + E = −D . (14.56)

Thus, if one lets D = 1, then A = 2, B = −2, C = −1 and E = 1. Using these values

in (14.38) one obtains that

(ω01 V /C01 )2 (ω02 V /C02 )−2 (ω11 /ω10 C11 )−1 (ω12 /ω10 C12 )1 (ω13 /ω10 C13 )1 = 1 .

Reducing terms and putting the constants on the right-hand side

(ω01 /ω02 )2 (ω12 /ω11 )(ω13 /ω10 ) = (C12 C13 C01

2 2
)/(C02 C11 ) = (3K ∗ 1/3 ∗ (Kπ/4)2 /((Kπ/12)2 4K)
(ω01 /ω02 )2 (ω12 /ω11 )(ω13 /ω10 ) = 9/4 . (14.57)

But rearranging Equation (14.25), one observes that

(ω01 /ω02 )(ω12 /ω11 ) = 9/4 .

Cancelling these values from both sides, the result is:

(ω01 /ω02 )(ω13 /ω10 ) = 1 , (14.58)

which is equivalent to Equation (14.23).

ω01 /ω02 = ω10 /ω13 . (14.23)

Thus, the dimensional analysis approach gives the same results as the substitution approach
for obtaining the additional equations. The advantage of the dimensional analysis approach is that
it is more straight forward in obtaining the relationships.
 14.4. EVALUATIVE QUESTIONS 75
14.4 EVALUATIVE QUESTIONS
1. A side riser with a hemispheric top is to be designed for a casting which has a surface area
of 40 cm2 and a volume of 120 cm3 . The hot metal cost is 100 Rupees per kg and the metal
density is 3.0 gm/cm3 . Compare these results with Problem 1 in Section 6.2.

(a) What are the dimensions of the hemispherical side riser (H and D)?
(b) What is the volume of the hemispherical side riser (cm3 )?
(c) What is the metal cost of the hemispherical side riser (Rupees)?
(d) What is the metal cost of the casting (Rupees)?

2. Two castings of equal volume but of different dimensions are to be cast. If one is a 3 inch
cube and the other is a plate of 1 × 3 × 9 inches and a top riser is to be used, what are the
dimensions (H and D) of the risers for the two cases?

REFERENCES
[1] Creese, R. C., Optimal Riser Design by Geometric Programming,” AFS Cast Metals Research
Journal, Vol. 7, 1971, pp. 118–121. 72

[2] Creese, R.C., “Dimensioning of Risers for Long Freezing Range Alloys by Geometric Pro-
gramming,” AFS Cast Metals Research Journal, Vol. 7, 1971, pp. 182–184. 72

[3] Creese, R.C., “Generalized Riser Design by Geometric Programming,” AFS Transactions,
Vol. 87, 1979, pp. 661–664. 72

[4] Creese, R.C., “An Evaluation of Cylindrical Riser Designs with Insulating Materials,” AFS
Transactions, Vol. 87, 1979, pp. 665–669. 72

[5] Creese, R.C., “Cylindrical Top Riser Design Relationships for Evaluating Insulating Materi-
als,” AFS Transactions, Vol. 89, 1981, pp. 345–348. 72
 77

CHAPTER 15

Liquefied Petroleum Gas (LPG)

Cylinders Case Study
15.1 INTRODUCTION
This case study problem [1, 2] deals with the design of liquefied petroleum gas cylinders, more
commonly known as propane gas cylinders in the USA. This is a very interesting problem as it has
two general solutions as well as one degree of difficulty. The two general solutions occur depending
upon the relationship between the constants. What happens in this particular problem is that one
of the two constraints can be either binding or loose, depending upon the value of the constants.

15.2 PROBLEM FORMULATION

The problem was to minimize the drawing force (Z) to produce the tank by deep drawing and two
constraints were considered so the tank would have a minimum volume and the height/diameter
ratio would be less than one. The formulation of the primal problem was:

Minimize Z = K1 hd + K2 d 2 , (15.1)

subject to the two constraints:

π d 2 h/4 ≥ Vmin , (15.2)

or in the proper geometric programming form as

(4Vmin /π )d −2 h−1 ≤ 1 , (15.3)

and
h/d ≤ 1 , (15.4)

where

K1 = π P Y C/F , and (15.5)

K2 = ((C − E)π P Y/2F , (15.6)

where
78 15. LIQUEFIED PETROLEUM GAS (LPG) CYLINDERS CASE STUDY
Z = drawing force
P = internal gas pressure
Y = material yield strength
F = hoop stress
C = constant = 1.04
E = constant = 0.65 .
If all the constants are combined, the primal form can be written as:

Z = K1 hd + K2 d 2 , (15.1)

subject to:
K3 h−1 d −2 ≤ 1 , (15.7)
and
K4 hd −1 ≤ 1 , (15.8)
where
K3 = (4Vmin /π ) , (15.9)
and
K4 = 1 (This could be taken as the minimum d/ h ratio) . (15.10)
From the coefficients and signs, the signum values for the dual are:

σ01 =1
σ02 =1
σ11 =1
σ21 =1
σ1 =1
σ2 =1.

The dual problem formulation is:

Objective Function ω01 + ω02 =1 (15.11)

h terms ω01 − ω11 + ω21 = 0 (15.12)
d terms ω01 + 2ω02 − 2ω11 − ω21 = 0 . (15.13)

The degrees of difficulty are equal to:

D = T − (N + 1) = 4 − (2 + 1) = 1 . (15.14)

From the constraint equations which have only one term it is apparent that:

ω10 = ω11 , (15.15)

 15.2. PROBLEM FORMULATION 79
and
ω20 = ω21 . (15.16)
An additional equation is needed, and one must examine the primal dual relationships to find
the additional relationship, which are:

K1 hd = ω01 Z (15.17)
K2 d 2 = ω02 Z (15.18)
K3 h−1 d −2 = ω11 /ω10 = 1 (15.19)
K4 hd −1 = ω21 /ω20 = 1 . (15.20)

If one adds Equations (15.12) and (15.13), one can solve for ω11 by:

2ω01 + 2ω02 − 3ω11 =0 (15.21)

2(ω01 + ω02 ) − 3ω11 =0 (15.22)
2 − 3ω11 =0 (15.23)
ω11 = 2/3 . (15.24)

If one takes Equation (15.17) and divides it by Equations (15.18) and (15.20), one obtains:

K1 hd/(K2 d 2 ∗ K4 hd −1 ) = K1 /(K2 K4 ) = ω01 Z/(ω02 Z ∗ 1) = ω01 /ω02 . (15.25)

This can be solved for ω01 in terms of ω02 and the constants:

ω01 = ω02 (K1 /(K2 K4 )) . (15.26)

Now using Equations (15.26) and (15.11), one can solve for ω01 and ω02 and obtain:

ω01 = K1 /(K1 + K2 K4 ) (15.27)

ω02 = K2 K4 /(K1 + K2 K4 ) . (15.28)

Now using Equation (15.12) and substituting the values for ω01 and ω11 , one obtains:

ω21 = ω11 − ω01 = (2K2 K4 − K1 )/[3(K1 + K2 K4 )] . (15.29)

Now ω21 must be ≥ 0, so that implies that:

2K2 K4 − K1 ≥ 0 , (15.30)

or
K2 K4 /K1 ≥ 1/2 . (15.31)
Thus, there are two sets of solutions, depending upon whether Equation (15.31) holds; this
can be illustrated in Figure 15.1.
80 15. LIQUEFIED PETROLEUM GAS (LPG) CYLINDERS CASE STUDY
Now, the solutions can be found for the two cases. If the answer is “No,” then the objective
function can be evaluated using the dual expression:
 M T σ
  m
Z = d(ω) = σ (Cmt ωmo /ωmt )σmt ωmt , (15.32)
m=0 t=1

where by definition
ω00 = 1 . (15.33)

IF
(K2K4/K1)ш1/2

Yes No(looseconstraint)
(thatish=d) (thatish<d)
 
ʘ11=2/3    ʘ11=2/3
ʘ21=(2K2K4ͲK1)/[3(K1+K2K4)] ʘ21=0(can’tbenegative)
ʘ01=K1/(K1+K2K4)   ʘ01=ʘ11=2/3  
ʘ02=K2K4/(K1+K2K4) ʘ02=1/3

Figure 15.1: Values of Dual Variables based upon making ω21 ≥ 0.

Z = [K1 ∗ 1/(2/3)]1∗2/3 [K2 ∗ 1/(1/3)]1∗1/3 [K3 ∗ (2/3)/(2/3)]1∗2/3 , (15.34)

which can be reduced to:

2/3 1/3 2/3
Z = [3/22/3 ]K1 K2 K3 . (15.35)
Now, the primal variables can be determined from the primal dual relationships. If one uses
Equation (15.18) and solves for d 2 and then for d, one obtains:
2/3 1/3 2/3
d 2 = ω02 Z/K2 = (1/3) ∗ {[3/22/3 ]K1 K2 K3 }/K2 = (K1 K3 /2K2 )2/3 . (15.36)

Solving for d results in:

d = (K1 K3 /2K2 )1/3 . (15.37)
 15.2. PROBLEM FORMULATION 81
Similarly, if one uses Equation (15.17) to solve for h, one obtains:
2/3 1/3 2/3
h = ω01 Z/K1 d = {(2/3) ∗ {[3/22/3 ]K1 K2 K3 }}/[K1 ∗ (K1 K3 /2K2 )1/3 ] . (15.38)

Reducing terms, one obtains:

−2/3 2/3 1/3
h = 22/3 K1 K2 K3 . (15.39)

Now substituting the primal variables into the primal objective function, one has:

Z = K1 hd + K2 d 2 , (15.1)
−2/3 2/3 1/3
= K1 ∗ 2 K1 K2 K3 ∗ (K1 K3 /2K2 )1/3 + K2
2/3
∗ (K1 K3 /2K2 ) 2/3
2/3 1/3 2/3 2/3 1/3 2/3
= 2K1 K2 K3 /22/3 + K1 K2 K3 /22/3
2/3 1/3 2/3
= 3K1 K2 K3 /22/3
2/3 1/3 2/3
= [3/22/3 ] ∗ K1 K2 K3 . (15.40)

The equations for the primal and dual, Equations (15.35) and (15.40), give the same results.
Thus, one has a general solution for the objective function and the two primal variables when the
“No” route was taken.
The “Yes” route has the dual variables as functions of the constants and thus is the more
complex route. The objective function can be evaluated using the dual expression:
 M T σ
  m
Z = d(ω) = σ (Cmt ωmo /ωmt )σmt ωmt
, (15.32)
m=0 t=1

where by definition
ω00 = 1 . (15.33)
The dual variables are:

ω11 = 2/3 (15.41)

ω21 = (2K2 K4 − K1 )/[3(K1 + K2 K4 )] (15.42)
ω01 = K1 /(K1 + K2 K4 ) (15.43)
ω02 = K2 K4 /(K1 + K2 K4 ) . (15.44)

Using these dual variables and the signum values the dual objective function is:

Z = (K1 ω00 /ω01 )1∗ω01 (K2 ω00 /ω02 )1∗ω02 (K3 )1∗ω11 (K4 )1∗ω21
= (K1 + K2 K4 )ω01 [(K1 + K2 K4 )/K4 ]ω02 (K3 )2/3 (K4 )ω11 −ω01
= [(K1 + K2 K4 )/K4 ]ω01 [(K1 + K2 K4 )/K4 ]ω02 (K3 K4 )2/3
= [(K1 + K2 K4 )/K4 ](K3 K4 )2/3
−1/3 2/3
= (K3 )2/3 [(K1 K4 + K2 K4 )] . (15.45)
82 15. LIQUEFIED PETROLEUM GAS (LPG) CYLINDERS CASE STUDY
The primal variables can be obtained from the relationships between the primal and dual
variables. Using Equation (15.20), one can obtain a relation between h and d which is:
h = d/K4 . (15.46)
If one combines Equations (15.19) and (15.20) one obtains:
(K3 h−1 d −2 )(K4 hd −1 ) = 1 ∗ 1 ,
which yields:
K3 K4 d −3 = 1 ,
or
d = (K3 K4 )1/3 . (15.47)
Now from Equations (15.46) and (15.47) one obtains:
1/3 −2/3
h = d/K4 = K3 K4 . (15.48)
Now, using the primal equation for the objective function with the primal variables, one has:
Z = K1 hd + K2 d 2 , (15.1)
1/3 −2/3
= K1 (K3 K4 )(K3 K4 )1/3 + K2 (K3 K4 )2/3
2/3 −1/3
= K1 (K3 K4 ) + K2 (K3 K4 )2/3
2/3 −1/3 2/3
= K3 (K1 K4 + K2 K4 ) . (15.49)
Note that the general objective function is the same for both the primal and dual solutions.
This problem illustrates that it is possible to solve a problem with more than one degree of difficulty
and have two solutions, depending upon the specific values of the constants in the problem.

15.3 DIMENSIONAL ANALYSIS TECHNIQUE FOR

ADDITIONAL EQUATION
It was not very difficult to determine the additional equation by repeated substitution, but the
technique of dimensional analysis can be used to verify it and this will be demonstrated to obtain
Equation (15.26). The (15.17)– (15.20) and setting the primal variables on one side and the dual
variables, constants, and objective function on the other side and giving the terms variable exponents
as illustrated in Equation (15.50)
(hd)A (d 2 )B (h−1 d −2 )C (hd −1 )D = 1 = (ω01 Z/K1 )A (ω02 Z/K2 )B (1/K3 )C (1/K4 )D . (15.50)
One must balance the exponents to remove the primal variables (d, h) and the dual objective
function (Z). This is done by:
Z values A + B =0 (15.51)
h values A − C + D= 0 (15.52)
d values A + 2B − 2C − D= 0 . (15.53)
 15.4. EVALUATIVE QUESTIONS 83
There are four variables and three equations, so one will find three variables in terms of the
fourth variable. From Equation (15.51), one obtains

A = −B . (15.54)

Using Equations (15.54) and (15.52), one notes that

B =D−C . (15.55)

If one subtracts Equation (15.52) from Equation (15.53), one obtains that

B = D + C/2 . (15.56)

If one compares Equations (15.55) and (15.56), the only solution was for both equations to
be satisfied is if:
C=0. (15.57)
If one sets D = 1, then B = 1 and A = −1. Using these values in Equation (15.50) one
obtains

(hd)−1 (d 2 )1 (h−1 d −2 )0 (hd −1 )1 = 1 = (ω01 Z/K1 )−1 (ω02 Z/K2 )1 (1/K3 )0 (1/K4 )1 . (15.58)

This results in Equation (15.59) which is equivalent to Equation (15.26)

ω01 = ω02 (K1 /(K2 K4 ) . (15.59)

The solution to the problem is the same, but the obtaining of the additional equation is
often the difficult step in the process. The dimensional analysis technique is often easier than the
substitution method as it is more consistent and often faster in obtaining the necessary additional
equation(s).

15.4 EVALUATIVE QUESTIONS

1. A tank is to be designed with a minimum volume of 17,500,000 mm3 and the values for
parameters are:
P = 0.2535 kg/mm2
F = 32.33 kg/mm2
Y = 25 kg/mm2
C = 1.04
E = 0.65 .
Determine the amount of the drawing force (kg) and the height (mm) and diameter (mm) of
the tank.
84 15. LIQUEFIED PETROLEUM GAS (LPG) CYLINDERS CASE STUDY
2. A new procedure was developed for another older machine which changed the expression for
K2 . The new expression was:

K2 = (2C − E)π P Y/(2F ) . (15.50)

Using the same data as in Evaluative Question 1, determine the amount of the drawing
force (kg) and the height (mm) and diameter (mm) of the new tank.

3. A tank is to be designed with a minimum volume of 11,000 in3 and the values for parameters
are:
P = 360 lb/in2
F = 46,000 lb/in2
Y = 35,500 lb/in2
C = 1.04
E = 0.65 ω01 = ω02 (K1 /(K2 K4 )) .
Determine the amount of the drawing force (lb and tons) and the height (in) and diameter (in)
of the tank.

REFERENCES
[1] Ravivarna Pericherla, Design and Manufacture of Liquefied Petroleum Gas Cylinders, MS Thesis,
West Virginia University, 1992, Morgantown, WV, USA. 77

[2] B. Gopalakrishnan and R. Pericherla, “Design for Manufacturability of the Liquefied

Petroleum Gas Cylinder,” Proceedings of the International Industrial Engineering Research Con-
ference, IIE, Miami, FL, May 1997. 77
 85

CHAPTER 16

Material Removal/Metal
Cutting Economics with Two
Constraints
16.1 INTRODUCTION
The metal cutting economics case study in Chapter 12 had only a feed constraint, but the problem
presented here also includes the horsepower constraint. The additional constraint leads to a problem
with one degree of difficulty and to the possibility of multiple solutions similar to that in the Liquefied
Petroleum Gas Cylinder Case Study. Thus, the problem is more difficult to solve and needs to be
considered separately, but the equations to define the problem are similar to those of Chapter 12.
The theory is from the work by Tsai [1, 3] and the example data is from that of Ermer [2].

16.2 PROBLEM FORMULATION

The problem is based upon the modified form of the Taylor’s Tool Life Equation and two constraints,
the feed rate and horsepower constraints. The cost function represents the sum of the various costs
and is expressed as:

Cu = (Ro + Rm )th + (Ro + Rm )tc + Ct nt + (Ro + Rm )tch nt , (16.1)

where
Cu = Unit Cost, $/piece
Ro = Operator Rate, $/min
Rm = Machine Rate, $/min
Ct = Tool Cost (per cutting edge for insert tools), $
th = Handling Time (to insert and remove piece), min/piece
tc = Cutting Time, min/piece
tch = Tool Changing Time (to change tool), min
nt = Number of tool changes per piece.
The modified form of Taylor’s Tool life Equation is:

T V 1/n f 1/m = C (16.2)

where
86 16. MATERIAL REMOVAL/METAL CUTTING ECONOMICS
T = Tool Life, min
V = Cutting Speed, feet/min
1/n = Cutting Speed Exponent
f = Feed Rate, in/rev
1/m = Feed Rate Exponent
C = Taylor’s Modified Tool Life Constant (min).
The cutting time can be expressed as:
tc = Bf −1 V −1 (16.3)
where
B = Cutting Path Surface Factor (an input value), ft
f = Feed Rate, in/rev
V = Cutting Speed, feet/min.
The number of tool changes per piece can be found by:
nt = Qtc /T (16.4)
where
Q = Fraction of cutting time that the tool is worn
tc = Cutting Time, min/piece
T = Tool Life, min.
The value of Q is approximately 1.0 for operations such as turning, but for other metal cutting
operations it may be as low as 0.10, say for horizontal milling operations.
Utilizing Equations (16.2) and (16.3) in Equation (16.4), one obtains:
nt = QBC −1 f (1/m−1) V (1/n−1) . (16.5)
The total cost can be expressed as the sum of the cost components as:
Cu = Machine Cost + Operator Cost + Tool Cost + Tool Changing Cost
= Rm (tc + tl ) + Ro (tc + tl ) + Ct nt + (Ro + Rm )tch nt . (16.6)
After substituting for nt , tc , and T to obtain the cost expression in terms of V and f , one obtains:
Cu = (Ro + Rm )tl + (Ro + Rm )Bf −1 V −1
+ [(Ro + Rm )tch + Ct ]QBC −1 f (1/m−1) V (1/n−1) . (16.7)
The constants can be combined and the unit cost objective function can be restated as:
Cu = Koo + K01 f −1 V −1 + K02 f (1/m−1) V (1/n−1) (16.7)
where
Koo = (Ro + Rm )tl
K01 = (Ro + Rm )B
K02 = [(Ro + Rm )tch + Ct ]QBC − .
 16.3. PROBLEM SOLUTION 87
Since Koo is a constant, the objective function to be minimized is:

Y = K01 f −1 V −1 + K02 f (1/m−1) V (1/n−1) (16.8)

where
Y = Variable portion of the unit cost.
The two constraints must be developed into geometric programming format. The feed con-
straint is:
f ≤ fmax (16.9)
where
fmax = Maximum Feed Limit (in/rev).
The feed constraint is typically used to control the surface finish as the smaller the feed, the
better the surface finish.
The horsepower constraint is given as:

aV b f c ≤ Hp (16.10)

where
a = Horsepower Constraint Constant
b = Velocity Exponent for Horsepower Constraint
c = Feed Rate Exponent for Horsepower Constraint
H P = Horsepower Limit.
These constraints can be put into geometric programming form as:

K11 f ≤ 1 (16.11)
K21 V b f c ≤ 1 (16.12)

where

K11 = 1/fmax
K21 = a/Hp .

16.3 PROBLEM SOLUTION

The primal problem can be stated as:

Minimize Y = K01 f −1 V −1 + K02 f (1/m−1) V (1/n−1) . (16.13)

Subject to the constraints:

K11 f ≤ 1 (16.14)
K21 V b f c ≤ 1 . (16.15)
88 16. MATERIAL REMOVAL/METAL CUTTING ECONOMICS
From the coefficients and signs, the signum values for the dual are:

σ01 =1
σ02 =1
σ11 =1
σ21 =1
σ1 =1
σ2 =1.

The dual problem can be formulated as:

ω01 + ω02 =1 (16.16)

f terms − ω01 + (1/m − 1)ω02 + ω11 + c ω21 = 0 (16.17)
V terms − ω01 + (1/n − 1) ω02 + b ω21 = 0 . (16.18)

The degrees of difficulty (D) are equal to:

D = T − (N + 1) = 4 − (2 + 1) = 1 . (16.19)

From the constraint equations which have only one term it is apparent that:

ω10 = ω11 (16.20)

ω20 = ω21 . (16.21)

The dual objective function is:

Y = (K01 /ω01 )ω01 (K02 /ω02 )ω02 (K11 )ω11 (K21 )ω21 . (16.22)

Thus, we have 4 variables and 3 equations, or one degree of difficulty. To find an additional
equation one must get the number of variables reduced to one in the dual objective function, differ-
entiate the logarithm of the equation to obtain another equation by setting the derivative to zero. It
was decided to determine the dual variables in terms of ω02 .
From Equation (16.16) one obtains:

ω01 = 1 − ω02 . (16.23)

From Equation (16.18) one obtains:

bω21 = ω01 − (1/n − 1)ω02

= (1 − ω02 ) − (1/n − 1)ω02
= 1 − ω02 /n
ω21 = 1/b − ω02 /bn . (16.24)
 16.3. PROBLEM SOLUTION 89
From Equation (16.17) one obtains:
ω11 = ω01 − (1/m − 1)ω02 − cω21
= 1 − ω02 − (1/m − 1)ω02 − c(1/b − ω02 /bn)
= (1 − c/b) + ω02 (c/bn − 1/m)
ω11 = (1 − c/b) + ω02 Z (16.25)
where
Z = (c/bn − 1/m) . (16.26)
The dual can be written as:
Y = (K01 /(1 − ω02 ))(1−ω02 ) (K02 /ω02 )ω02 (K11 )((1−c/b)+ω02 Z) (K21 )(1/b−ω02 /bn) . (16.27)
The log of the dual is:
Log (Y ) = (1 − ω02 ) log(K01 /(1 − ω02 )) + ω02 log(K02 /ω02 )
+ (1 − c/b) log(K11 ) + ω02 Z log(K11 )
+ 1/b log(K21 ) − ω02 /bn log(K21 ) . (16.28)
Find ∂(log Y )/∂ω02 and set it to zero to find an additional equation.
∂(log Y )/∂ω02 = (1 − ω02 )[1/{K01 /(1 − ω02 )}][K01 (−1)(1 − ω02 )−2 (−1)]
+ [log(K01 /(1 − ω02 ))](−1)
−2
+ ω02 [1/{K02 /ω02 }][K02 (−1)ω02 (1)] + [log(K02 /ω02 )](1)
+ 0 + Z log(K11 ) + 0 − (1/bn) log(K21 )
∂(log Y )/∂ω02 = (−1) − [log(K01 /(1 − ω02 ))] + 1 + [log(K02 /ω02 )]
+ 0 + Z log(K11 ) + 0 − (1/bn) log(K21 )
∂(log Y )/∂ω02 = log[(K02 /K01 )((1 − ω02 )/(ω02 ))] + log(K11 )Z − log(K21 )1/bn
∂(log Y )/∂ω02 = log[(K02 /K01 )((1 − ω02 )/(ω02 ))] + log[(K11 )Z / log(K21 )1/bn ]
∂(log Y )/∂ω02 = log[(K02 /K01 )((1 − ω02 )/(ω02 ))(K11 )Z /(K21 )1/bn ] (16.29)
Setting the expression equal to zero and then use the anti-log, one goes through various steps
to obtain an expression for ω02 :
1 = (K02 /K01 )((1 − ω02 )/(ω02 ))(K11 )Z /(K21 )1/bn
(1 − ω02 )/(ω02 ) = (K01 /K02 )((K21 )1/bn /(K11 )Z )
1/ω02 − 1 = (K01 /K02 )((K21 )1/bn /(K11 )Z )
1/ω02 = (K01 /K02 )((K21 )1/bn /(K11 )Z ) + 1
1/bn
ω02 = (K02 K11
Z
)/(K01 K21 + K02 K11
Z
). (16.30)
Now the values for other three dual variables can be obtained in terms of ω02 as:
ω01 = −ω02 (16.31)
ω21 = 1/b − ω02 /bn (16.32)
ω11 = (1 − c/b) + ω02 Z . (16.33)
90 16. MATERIAL REMOVAL/METAL CUTTING ECONOMICS
Substituting the values of the dual variables into the dual objective function, the dual objective
function can be expressed in terms of ω02 and is:

Y = (K01 /(1 − ω02 ))(1−ω02 ) (K02 /ω02 )ω02 (K11 )((1−c/b)+ω02 Z) (K21 )(1/b−ω02 /bn) (16.34)

and the minimum cost would be:

Cu = Y + K00 (16.35)
or Cu = K00 + (K01 /(1 − ω02 ))(1−ω02 ) (K02 /ω02 )ω02
(K11 )((1−c/b)+ω02 Z) (K21 )(1/b−ω02 /bn) . (16.36)

By using the primal-dual relationships of the objective function, one can obtain the following
equation for the primal variables. Starting with

K01 f −1 V −1 = ω01 Y (16.37)

and
K02 f 1/m−1 V 1/n−1 = ω02 Y . (16.38)
One can obtain equations for the primal variables in terms of the dual variables, dual objective
function and constants. These equations are for the case when both constraints are binding.

f = (ω02 K01 /(ω01 K02 ))m (K01 /ω01 )(m(m−1)/(n−m))

(ω02 K02 )((mm)/(n−m) Y (m/(n−m)) (16.39)
V = (K01 /ω01 )(n(m−1)/(m−n)) (ω02 /K02 )((mn)/(m−n)) Y (n/(m−n)) . (16.40)

Upon examination of the equations for the dual variables, it is possible for either ω11 or ω21
to be negative, which means that the constraints are not binding. Thus, the dual variable must be set
to zero if the constraint is not binding. If ω11 is zero, then Equations (16.16), (16.17), and (16.18)
become:

ω01 + ω02 =1 (16.41)

f terms − ω01 + (1/m − 1)ω02 + c ω21 = 0 (16.42)
V terms − ω01 + (1/n − 1) ω02 + b ω21 = 0 . (16.43)

The degree of difficulty becomes zero with the variable ω11 removed, and the new values for
the dual variables become:

ω02 = (b − c)/(b/m − c/n) (16.44)

ω01 = 1 − ω02 (16.45)
ω11 =0 (loose constraint) (16.46)
ω21 = (1/m − 1/n)/(b/m − c/n) . (16.47)
 16.4. EXAMPLE PROBLEM 91
If ω21 is zero, then Equations (16.16), (16.17), and (16.18) become:
ω01 + ω02 =1 (16.48)
f terms − ω01 + (1/m − 1)ω02 + ω11 = 0 (16.49)
V terms − ω01 + (1/n − 1) ω02 =0. (16.50)
The degree of difficulty becomes zero with the variable ω21 removed, and the new values for
the dual variables become:
ω02 =n (16.51)
ω01 = 1 − ω02 (16.52)
ω11 = 1 − n/m (16.53)
ω21 =0. (16.54)
The equations for velocity and feed for this set of dual variables are:
(n/m)
V = (n/(1 − n))n (K01 /K02 )n K11 (16.55)
f = (K01 /(ω01 Y V ) (16.56)
Thus, there are three different solutions based upon the values of the constants in the problem.
However, the equations for the dual variables for the three possibilities have been obtained and the
only question is to determine which of the three conditional are applicable. The results for the
primal variables have been given for the case when both constraints are binding and when the feed
constraint is binding (horsepower is non-binding). The horsepower constraint is often not binding,
but the feed constraint is usually a binding constraint.

16.4 EXAMPLE PROBLEM

An example problem will be used to illustrate the application of the formulas and the magnitude of
the results obtained. The data used is from Ermer (2) and results obtained are in Table 16.1.
The model results indicate that the cutting speed is 290 ft/min and the feed rate is at the max
of 0.005 in/rev. The variable cost (Y ) is $1.05 and the total cost is $1.25.

16.5 EVALUATIVE QUESTIONS

1. In a problem with two constraints, how many solutions are possible and describe the conditions
under which the solutions would occur?
2. The data for the metal cutting problem has been modified and is in Table 16.2. Calculate the
values for the constants and constants and variables.
3. The data for the metal cutting problem has been modified and is in Table 16.3. Calculate the
values for the constants and constants and variables. However, for this problem consider the
horsepower to be non-binding.
92 16. MATERIAL REMOVAL/METAL CUTTING ECONOMICS

Table 16.1: Data for Metal Cutting Example Problem

Input Parameters Calculated Constants and Variables
Q 1 K00 0.20
R0 0.04 $/min K01 1.2566
Rm 0.06 $/min K02 1.80 ∗ 10−8
Ct 0.50 $/edge K11 200
th 2 min K21 0.358
tch 0.50 min Z 2.2684
D 6 inches ω01 0.8216
L 8 inches ω02 0.1784
m 0.862 ω11 0.5476
n 0.25 ω21 0.3146
C 3.84 ∗ 108 min Y $ 1.05
a 3.58 Cu $ 1.25
b 0.91 B 12.56 in-ft = π D (in) L (ft)
c 0.78 f 0.005 in/rev
Hp 10 V 290 ft/min
fmax 0.005 in/rev

REFERENCES
[1] Robert C. Creese and Pingfang Tsai, “Generalized Solution for Constrained Metal Cutting
Economics Problem”, 1985 Annual International Industrial Conference Proceedings, Institute of
Industrial Engineers U.S.A, 113–117. 85

[2] Ermer, D.S., “Optimization of the Constrained Machining Economics Problem by Geometric
Programming,” Journal of Engineering for Industry,Traisactions of the ASME, November 1971,
pp 1067–1072. 85

[3] Tsai, Pingfang An Optimization Algorithm and Economic Analysis for a Constrained Machining
Model, PhD Dissertation, West Virginia University, Morgantown, WV, 214pp. 85
 REFERENCES 93

Table 16.2: Data for Problem 2

Input Parameters Calculated Constants and Variables
Q 1 K00
R0 0.4 $/min K01
Rm 0.6 $/min K02
Ct 2.0 $/edge K11 200
th 1.5 min K21 0.358
tch 0.80 min Z
D 1 inches ω01
L 6 inches ω02
m 0.80 ω11
n 0.25 ω21
C 5.00 ∗ 108 min Y 1.14
a 3.58 Cu
b 0.91 B
c 0.78 f 0.005 in/rev
Hp 10 V 290 ft/min
fmax 0.005

Table 16.3: Data for Problem 3 (horsepower constraint removed)

Input Parameters Calculated Constants and Variables
Q 1 K00
R0 0.4 $/min K01
Rm 0.6 $/min K02
Ct 2.0 $ K11 200
th 1.5 min K21 0.358
tch 0.80 min Z
D 1 inches ω01
L 6 inches ω02
m 0.80 ω11
n 0.25 ω21
C 5.00 ∗ 108 min Y 0.91
a 3.58 Cu
b 0.91 B
c 0.78 f 0.005 in/rev
Hp NA V 460 ft/min
fmax 0.005
 95

CHAPTER 17

The Open Cargo Shipping Box

with Skids
17.1 INTRODUCTION
The open cargo shipping box was presented in Chapter 6 and is the classical problem in geometric
programming. It had zero degrees of difficulty and was solved relatively easily. The open cargo
shipping box problem is adjusted to add skid rails at $5/unit length and if two rails are used, the
additional cost would be $10/box length. This becomes a problem with one degree of difficulty and
the solution is more difficult, and the purpose of this chapter is to introduce different methods of
obtaining a solution.
The first method to be applied is the constrained derivative approach, that is obtaining four of
the dual variables in terms of the fifth variable, taking the derivative of the function and setting it to
zero and obtaining a fifth independent dual equation. The problem then becomes a problem with
zero degrees of difficulty and the solution is obtained.
The second method applied is the dimensional analysis approach, which takes the primal dual
equations and separating them into the primal variables on one side of the equation and the dual
variables and constants on the other side. Solving this for the exponents of the terms leads to the
required additional independent equation. The second method does not require the taking of the
derivatives of logarithmic functions which can be difficult. In some manners, the approach is similar
to the method of solving the dual equations. One advantage of the dimensional analysis method is
that students are often familiar with this technique in evaluating units engineering problems.
The third method is the condensation of terms approach. This method combines two terms
to reduce the degrees of difficulty by one. This method, however, does not guarantee an optimal
solution and the selection of which terms to combine is important. The terms selected should have
exponents that close in value and should be selected such that the equations in the new dual do not
result in dual variables becoming zero or that redundant equations result in the formation of the
dual. The terms are typically combined geometrically with equal weights.

17.2 PRIMAL-DUAL PROBLEM FORMULATION

The problem can be stated as: “400 cubic yards of gravel must be ferried across a shallow river. The
box is an open box with skids used because of the low water level. The box has length L, width W ,
and depth H . The sides of the box cost $10/yd2 and have a total area of 2(L + W )H and the bottom
96 17. THE OPEN CARGO SHIPPING BOX WITH SKIDS
of the box cost $20/yd2 and has an area of LW . There are 2 skids at a cost of $5/yd on the length of
the box and the total skid length is 2L. Each round trip of the box on the ferry will cost 10 cents
per cubic yard of gravel shipped.” The objective function becomes:

C = 40/LW H + 10LW + 20LH + 40H W + 10L . (17.1)

The previous cost solution was $100 and the box length was 2 units (yds), so the new cost
would be $100 + $20 = $120 if the same dimensions are used. However, one must determine if the
additional cost yields the same dimensions (L = 2 yds, W = 1 yd, H = 1/2 yd) and 400 trips were
required.
The primal objective function can be written in general terms of:

C = C01 /(LW H ) + C02 LW + C03 LH + C04 H W + C05 L . (17.2)

The degrees of difficulty is

5 − (3 + 1) = 1 . (17.3)
The dual formulation is:

Objective Function ω01 + ω02 + ω03 + ω04 + ω05 =1 (17.4)

L terms −ω01 + ω02 + ω03 + ω05 =0 (17.5)
W terms −ω01 + ω02 + ω04 =0 (17.6)
H terms −ω01 + ω03 + ω04 =0. (17.7)

Subtracting Equation (17.7) from Equation (17.6) one obtains that:

ω02 = ω03 . (17.8)

Considering Equation (17.4) and Equation (17.5), one obtains

ω04 = 1 − 2ω01 . (17.9)

Using Equation (17.6) with Equations (17.7) and (17.8), one obtains

ω02 = 3ω01 − 1 (17.10)

and using Equation (17.4) with Equations (17.8) thru (17.10), one obtains

ω05 = 2 − 5ω01 . (17.11)

The dual can be written as:

Dual (Y ) = {(C01 /ω01 )ω01 (C02 /ω02 )ω02 (C03 /ω03 )ω03 (C04 /ω04 )ω04 (C05 /ω05 )ω05 } (17.12)
Y = {(C01 /ω01 )ω01 (C02 /(3ω01 − 1))(3ω01 −1) (C03 /(3ω01 − 1))(3ω01 −1)
(C04 /(1 − 2ω01 ))(1−2ω01 ) (C05 /(2 − 5ω01 ))(2−5ω01 ) } . (17.13)
 17.3. CONSTRAINED DERIVATIVE APPROACH 97
17.3 CONSTRAINED DERIVATIVE APPROACH
The constrained derivate approach takes the derivative of the log Y with respect to ω01 , sets it to
zero and solve for ω01 .

Log Y = ω01 log(C01 /ω01 ) + (3ω01 − 1) log(C02 /(3ω01 − 1)) + (3ω01 − 1)

log(C03 /(3ω01 − 1)) + (1 − 2ω01 ) log(C04 /(1 − 2ω01 ))
+ (2 − 5ω01 ) log(C05 /(2 − 5ω01 )) (17.14)
∂(log Y )/∂(ω01 ) = 0 = −1 + log(C01 /ω01 ) + −3 + 3 log(C02 /(3ω01 − 1)) − 3
+ 3 log(C03 /(3ω01 − 1)) + 2 − 2 log(C04 /(1 − 2ω01 ))
+ 5 − 5 log(C05 /(2 − 5ω01 )) (17.15)
= log(C01 /ω01 ) + 3 log(C02 /(3ω01 − 1)) + 3 log(C03 /(3ω01 − 1))
− 2 log(C04 /(1 − 2ω01 )) − 5 log(C05 /(2 − 5ω01 )) (17.16)

The antilog is taken and the constants are put on one side and the dual variable equations on the
other to obtain:
3 3
C01 C02 2 5
C03 /(C04 C05 ) = Z = ω01 (3ω01 − 1)3 (3ω01 − 1)3 /(1 − 2ω01 )2 (2 − 5ω01 )5 (17.17)
Z = 40 × 103 × 203 /(402 × 105 ) = 2 (17.18)
ω01 (3ω01 − 1)3 (3ω01 − 1)3 /(1 − 2ω01 )2 (2 − 5ω01 )5 = 2 . (17.19)

Note that the terms of Equation (17.19) must be positive, so that indicates that ω01 must be
< 1, > 1/3, < 0.5, and < 0.4 which indicates it is between 1/3 and 0.4.
If one solves Equation (17.19) for ω01 (using search techniques), the value obtained was:

ω01 = 0.3776519 . (17.20)

Using Equations (17.8) to (17.11) for the other variables in terms of ω01 ; one obtains

ω01 = 0.378 (17.21)

ω02 = 0.133 (17.22)
ω03 = 0.133 (17.23)
ω04 = 0.245 (17.24)
ω05 = 0.111 . (17.25)

Using the values of C and ω in the Equation (17.12) for the dual, one obtains

Y = (40/0.378)0.378 (10/0.133)0.133 (20/0.133)0.133 (40/0.245)0.245

(10/0.111)0.111 = $115.72 . (17.26)

This compares with the cost of $100 for the initial problem without skids.
Using the primal-dual relationships and the objective function, the primal variables can be found as:

L = ω05 Y/10 = (.111)(115.72)/10 = 1.284 yd (versus 2) or ω05 Y/C05 (17.27)

98 17. THE OPEN CARGO SHIPPING BOX WITH SKIDS
Similarly,
H = ω03 Y/(20L) = ω03 Y/(20ω05 Y/10) = (ω03 /ω05 )(C05 /C03 )
= (0.133/0.111)(10/20) = 0.599 (versus 0.500) (17.28)
and
W = ω02 Y /(10L) = ω02 Y/(10ω05 Y/10) = (ω02 /ω05 )(C05 /C02 )
= (0.133/0.111)(10/10) = 1.198 (versus 1.000) (17.29)
Therefore, the primal becomes
C = C01 /(LW H ) + C02 LW + C03 LH + C04 H W + C05 L (17.30)
= C01 /(1.284 ∗ 0.599 ∗ 1.198) + C02 (1.284 ∗ 1.198) + C03 (1.284 ∗ 0.599)
+ C04 (0.599 ∗ 1.198) + C05 ∗ (1.284)
= 43.41 + 15.38 + 15.38 + 28.70 + 12.84
= 115.71 versus 115.72 . (17.31)
The volume of the box is 1.284 × 0.599 × 1.198 = 0.921 cubic yards versus the 1.0 cubic yard
volume in the original problem. The number of trips will increase from 400 in the original problem
to 434.3 or 435 trips. The shipping cost increases from $40 to $43.41, but the box cost (side, end
and bottom) totals $59.46 versus $60 and the primary increase is the cost of the skids, that is $12.84.

17.4 DIMENSIONAL ANALYSIS APPROACH FOR

ADDITIONAL EQUATION
The dimensional analysis approach starts with the primal and dual equations which are:
C = C01 /(LW H ) + C02 LW + C03 LH + C04 H W + C05 L (17.2)
and
Dual (Y ) = {(C01 /ω01 )ω01 (C02 /ω02 )ω02 (C03 /ω03 )ω03 (C04 /ω04 )ω04 (C05 /ω05 )ω05 } (17.12)
The primal dual relationships are:
L−1 W −1 H −1 = ω01 Y/C01 (17.32)
LW = ω02 Y/C02 (17.33)
LH = ω03 Y/C03 (17.34)
H W = ω04 Y/C04 (17.35)
L = ω03 Y/C03 (17.36)
These relationships can be combined to give:
(L−1 W −1 H −1 )A (LW )B (LH )C (H W )D (L)E = 1
= (ω01 Y/C01 )A (ω02 Y/C02 )B (ω03 Y/C03 )C (ω04 Y/C04 )D (ω05 Y/C05 )E (17.37)
 17.4. DIMENSIONAL ANALYSIS APPROACH FOR ADDITIONAL EQUATION 99
The equations from dimensional analysis are:
L terms −A+B +C +E =0 (17.38)
W terms −A+B +D =0 (17.39)
H terms −A +C+D =0 (17.40)
Y +A+B +C+D+E =0. (17.41)
From Equations (17.39) and (17.40), one obtains that:
B=C (17.42)
Subtracting Equation (17.38) from Equation (17.41), one obtains:
D = −2A (17.43)
Subtracting Equation (17.38) from Equation (17.39) results in
D =C+E (17.44)
From Equation (17.38) one observes that:
C+E =A−B (17.45)
Thus,
D = −2A = A − B, (17.46)
so
B = 3A (17.47)
C = 3A (17.48)
D = −2A (17.49)
and via Equation (17.38)
E = −5A . (17.50)
Thus, if A = 1, then B = 3, C = 3, D = −2, and E = −5.
Thus, from Equation (17.37)
1 = (ω01 Y/C01 )1 (ω02 Y /C02 )3 (ω03 Y/C03 )3 (ω04 Y/C04 )−2 (ω05 Y/C05 )−5 . (17.51)
This results in
(ω01 )(ω02 )3 (ω03 )3 (ω01 )−2 (ω01 )−5 = (C01 )(C02 )3 (C03 )3 (C04 )−2 (C05 )−5
= (40)(2)3 (10)3 (40)−1 (10)−5 (17.52)
or in terms of ω01
(ω01 )(3ω01 − 1)3 (3ω01 − 1)3 (1 − 2ω01 )−2 (2 − 5ω01 )5 = 2 . (17.53)
Equation (17.53) is the same as Equation (17.19) and the solution procedure from that point on
would be the same as that used for the constrained derivative approach.
100 17. THE OPEN CARGO SHIPPING BOX WITH SKIDS
17.5 CONDENSATION OF TERMS APPROACH
The technique of condensation involves the combining of terms to reduce the degrees of difficulty to
make the solution easier. D.J.Wilde presented this technique in his book “Globally Optimum Design”
in 1978 [1] and this example is presented in “Engineering Design – A Material and Processing
Approach” by George Dieter (1991) [2].
The primal problem was:

C = 40/LW H + 10LW + 20LH + 40H W + 10L (17.1)

or as
C = C01 /(LW H ) + C02 LW + C03 LH + C04 H W + C05 L (17.2)
The dual is

Dual (Y ) = {(C01 /ω01 )ω01 (C02 /ω02 )ω02 (C03 /ω03 )ω03 (C04 /ω04 )ω04 (C05 /ω05 )ω05 } , (17.12)

and there is one degree of difficulty.

The condensation technique combines two terms to reduce the degrees of difficulty. The
selection of the terms and the weighing of the terms is important. The terms selected should have
exponents that are not that different and should be selected such that equations in the new dual do
not result in dual variables resulting to be zero or that redundant equations result in the formulation
of the dual. The weights for combining the two terms will be considered to be equal. The formation
of the new term will be illustrated.
If the third and fifth terms are combined with equal weights, the result is

t3 + t5 ≥ (C03 LH /(1/2))1/2 (C05 L/(1/2))1/2 = 2(C03 C05 )1/2 LH 1/2 = C06 LH 1/2 (17.53)

where
C06 = 2(C03 C05 )1/2 . (17.54)
The new primal would be:

C = C01 /(LW H ) + C02 LW + C04 H W + C06 LH 1/2 . (17.55)

The new Dual would be

Dual (Y ) = {(C01 /ω01 )ω01 (C02 /ω02 )ω02 (C04 /ω04 )ω04 (C06 /ω06 )ω06 } . (17.56)

The new dual formulation is:

Objective Function ω01 + ω02 + ω04 + ω06 = 1 (17.57)

L terms − ω01 + ω02 + ω06 = 0 (17.58)
W terms − ω01 + ω02 + ω04 =0 (17.59)
H terms − ω01 + ω04 + ω06 /2= 0 . (17.60)
 17.5. CONDENSATION OF TERMS APPROACH 101
Equations (17.58) and (17.59) result in:

ω06 = ω04 (17.61)

Equation (17.60) and (17.61) result in:

ω01 = ω04 + ω06 /2 = (3/2)ω04 (17.62)

Using Equations (17.61) and (17.62) in Equation (17.58) results in:

ω02 = ω01 − ω06 = ω04 /2 (17.63)

Using Equations (17.61), (17.62), and (17.63) with Equation (17.59) results in:

ω04 = 0.25 (17.64)

ω06 = 0.25 (17.65)
ω01 = 0.375 (17.66)
ω02 = 0.125 . (17.67)

Using the values for the constants and the dual variables, the dual objective function becomes:

Dual (Y ) = {(40/0.375)0.375 (10/0.125)0.125 (40/0.25)0.25 (2(20 × 10)1/2 /0.25)0.25 }

= (5.761)(1.729)(3.557)(3.261) = 115.5 (versus 115.71) . (17.68)

The primal dual relationships yield

C01 /H W L = ω01 Y (17.69)

C02 LW = ω02 Y (17.70)
C04 H W = ω04 Y (17.71)
C06 LH 1/2 = ω06 Y . (17.72)

Combining Equations (17.70) and (17.71) result in:

H /L = (C02 /C04 )(ω04 /ω02 ) = (10/40)(0.25/0.125) = 1/2 . (17.73)

Combining Equations (17.70) and (17.72) result in:

(W/H 1/2 ) = (C06 /C02 )(ω02 /ω06 )

= (2(20 × 10)1/2 /(10) × (0.125/0.25)) = 21/2 = 1.414 . (17.74)

Combining Equations (17.69) and (17.70) result in

L2 W 2 H = (C01 /C02 )(ω02 /ω01 ) = (40/10)(0.125/0.375) = 4/3 (17.75)

Using Equation (17.73) and 17.74 with Equation (17.75) one obtains:

(2H )2 (2H )H = 4/3 (17.76)

102 17. THE OPEN CARGO SHIPPING BOX WITH SKIDS
and this results in:

H = (1/6)0.25 = 0.6389 yd (vs 0.599 yd in constrained derivative solution) (17.77)

L = 2H = 1.2779 yd (vs 1.284 yd in constrained derivative solution) (17.78)
W = (2H ) = 1.1304 yd (vs 1.198 yd in constrained derivative solution)
1/2
(17.79)

The primal objective function would be:

C = C01 /(LW H ) + C02 LW + C04 H W + C06 LH 1/2 (17.80)

= 40/(1.2779 × 1.1304 × 0.6389) + 10(1.2779 × 1.1304) + 40(0.6389 × 1.1304)
+ 20 × 21/2 (1.2779 × 0.63891/2 )
= 43.34 + 14.45 + 28.89 + 28.89
= 115.57 . (17.81)

The values for the primal and dual are in good agreement and in good agreement with the constrained
derivative approach and the dimensional analysis approach. The primal variables for the condensed
version are slightly different than the primal variables of the other solutions. This is because the
objective function is slightly different, but the values of the various objective functions are quite
close.

17.6 EVALUATIVE QUESTIONS

1. If the cost of the skid rails was $20 instead of $10, what is the effect on the total cost, the
number of trips, and the box dimensions?

2. Use the dimensional analysis approach on the metal cutting problem with two constraints to
derive Equation (16.30).

3. Use the condensation technique with second and fifth terms of the primal objective function
and compare the solutions of the original problem and when the condensation technique was
used on terms 3 and 5.

REFERENCES
[1] D. J. Wilde, Globally Optimum Design, Wiley-Interscience, New York, 1978, pp. 88–90. 100

[2] George E. Dieter, Engineering Design-A Materials and Processing Approach, 2nd Edition,
McGraw-Hill, New York, 1991, pp. 223–225. 100
 103

CHAPTER 18

Profit Maximization
Considering Decreasing Cost
Functions of Inventory Policy
18.1 INTRODUCTION
The classical inventory models consider the unit costs and unit prices to be fixed, but as the quantity
increases the economies of scale permit decreased in costs and prices. In the model presented by
Jung and Klein [1] in 2001, they considered cost per unit and price per unit to be power functions
of demand. The variables in the profit maximization model considered are the order quantity and
the product demand. The inventory model considered is the basic model with the assumptions that
(1) replenishment is instantaneous; (2) no shortage is allowed; and (3) the order quantity is a batch.

18.2 MODEL FORMULATION

The variables and parameters used in the model are listed in Table 18.1. The variables are demand
per unit time and order quantity. The parameter price is a function of the price scaling constant,
the demand, and the price elasticity constant. The parameter cost is a function of the cost scaling
constant, the demand, and the cost elasticity constant, also called the economy of scale factor.
The profit maximization model can be stated on a per unit time basis as:

Maximize Profit (π ) = Total Variable Revenue (R)

− [Total Variable Cost (T V C)
+ Total Set-up Cost (T SC)
+ Inventory Holding Cost (I H C)] . (18.1)

The revenue per unit can be represented as:

P = aD −α (18.2)

where

D = Demand per Unit Time

a = Scaling Constant for Price
α = Price Elasticity with respect to Demand .
104 18. PROFIT MAXIMIZATION

Table 18.1: Variables and parameters for inventory model

Symbol Used Description of Variable or Parameter
D Demand per Unit Time (decision variable)
Q Order Quantity (decision variable)
P Price per Unit ($/Unit)
C Cost per Unit ($/Unit)
A Set-up Cost ($/Batch)
i Inventory Holding Cost Rate (Percent/Unit Time)
a Scaling Constant for Price (Initial Price)
b Scaling Constant for Cost (Initial Cost)
α Price Elasticity with respect to Demand
β Cost Elasticity with respect to Demand
(Degree of Economies of Scale)
π Profit per unit time
R Total Revenue
TVC Total Variable Cost
T SC Total Set-up Cost

The total revenue (R) is the product of the demand and the revenue per unit:
R = D ∗ P = D ∗ aD −α = aD 1−α . (18.3)
The cost per unit (C) can be represented as:
C = bD −β (18.4)
where
D = Demand per Unit Time
b = Scaling Constant for Cost
β = Cost Elasticity with respect to Demand .
The total variable cost (T V C) is the product of the demand and the cost per unit:
T V C = D ∗ C = D ∗ bD −β = bD 1−β . (18.5)
The total set-up cost (T SC) is the product of the set-up cost times the number of set-ups
which can be expressed as:
T SC = A ∗ D/Q = ADQ−1 . (18.6)
The inventory holding cost (I H C) represents the product of the average inventory (Q/2),
the unit cost (bD −β ) and the inventory holding cost rate (i) and results in:
I H C = Q/2 ∗ bD −β ∗ i = (ib/2) ∗ QD −β . (18.7)
 18.2. MODEL FORMULATION 105
Inserting Equations (18.3), (18.5), (18.6) and (18.7) into Equation (18.1), the result is the
primal objective function:

Max (π) = aD 1−α − (bD 1−β + ADQ−1 + (ib/2)QD −β ) . (18.8)

The problem can be written in general terms, such as

Max (π) = C00 D 1−α − C01 D 1−β − C02 DQ−1 − C03 QD −β . (18.9)

To solve the problem one minimizes the negative of the profit function, that is, the primal
objective function becomes:

Min Y = −C00 D 1−α + C01 D 1−β + C02 DQ−1 + C03 QD −β (18.10)

where Y = −π .
The signum functions would be:

σ01 = −1
σ02 = σ03 = σ04 = 1 ,
and σ00 = −1 as this is a maximization problem.

The dual objective function is:

D(Y ) = −1[(C00 /ω01 )−ω01 (C01 /ω02 )ω02 (C02 /ω03 )ω03 (C03 /ω04 )ω04 ]−1 . (18.11)

The dual formulation would be:

Objective Function − ω01 + ω02 + ω03 + ω04 = −1 (18.12)

D Terms − (1 − α)ω01 + (1 − β)ω02 + ω03 − βω04 = 0 (18.13)
Q Terms − ω03 + ω04 = 0 . (18.14)

The degree of difficulty (D) is:

Degrees of Difficulty = Dual Variables(4) − (Primal Variables(2) + 1) = 1 .

Since the degrees of difficult is 1, one additional equation is needed to solve the problem. First
one must get the four dual variables in terms of one variable and then obtain an equation in terms
of that variable. If one examines Equation (18.14) one observes that:

ω03 = ω04 . (18.15)

Thus, one selects ω04 as the unknown variable and now must find ω01 and ω02 in terms of
ω04 . If one multiplies Equation (18.12) by (1 − α) and subtracts it from Equation (18.13) and using
Equation (18.15), one obtains:

ω02 [(1 − β) − (1 − α)] + ω04 [(1 − β) − 2(1 − α)] = 1 − α . (18.16)

106 18. PROFIT MAXIMIZATION
This results in:

ω02 = [(1 − α)/(α − β)] + [(1 + β − 2α)/(α − β)]ω04 . (18.17)

Now if one solves for ω01 via Equations (18.12) and (18.15) one obtains:

ω01 = 1 + ω02 + 2ω04 . (18.18)

Now using Equation (18.17) in Equation (18.18), one obtains the expression for ω01 in terms
of ω04 which is:
ω01 = [(1 − β)/(α − β)][1 + ω04 ] . (18.19)
To obtain the additional equation, the approach of dimensional analysis will be used. The
additional equations from the primal dual relationships are:

C00 D 1−α = ω01 Y (18.20)

C01 D 1−β = ω02 Y (18.21)
C02 DQ−1 = ω03 Y (18.22)
C03 D −β Q = ω04 Y . (18.23)

The dimensional analysis equation is:

(D 1−α )−A (D 1−β )B (DQ−1 )C (D −β Q)D = 1

= (ω01 Y/C00 )−A (ω02 Y/C01 )B (ω03 Y/C02 )C (ω04 Y/C03 )D . (18.24)

Since this is a maximization problem, the sign on the revenue term exponent is negative and
the three cost terms is positive. The equations would be:

D terms −A(1 − α) + B(1 − β) +C − βD =0 (18.25)

Q terms −C +D =0 (18.26)
Y Dual −A +B +C +D =0 (18.27)

From Equation (18.26) one observes that

C=D. (18.28)

If one multiplies Equation (18.27) by (1 − α) and subtracts Equation (18.25) and using Equa-
tion (18.28) one obtains:
B = [(1 + β − 2α)/(α − β)]D . (18.29)
Using Equation (18.27) one obtains that

A=B +C+D .
 18.3. EXAMPLE 107
Using the values for B and C in terms of D the result is:

A = [(1 − β)/(α − β)]D . (18.30)

If one selects the value of D = 1, the values are A = [(1 − β)/(α − β)], B = [(1 + β − 2α)/(α −
β)] and C = 1. Now Equation (18.24) becomes:

(ω01 Y/C00 )−[(1−β)/(α−β)] (ω02 Y/C01 )[(1+β−2α)/(α−β)]

(ω03 Y/C02 )(ω04 Y/C03 ) = 1 . (18.31)

Separating the constants and dual variables, the relationship obtained was:
−[(1−β)/(α−β)] [(1+β−2α)/(α−β)]
ω01 ω02 ω03 ω04
−[(1−β)/(α−β)] [(1+β−2α)/(α−β)]
= C00 C01 C02 C03 =K. (18.32)

The right-hand side of the equation consists of only constants, so the product is also a constant. If
one substitutes the dual variables in terms of ω04 , the expression becomes quite complex and would
need to be solved with specific values of the parameters. The equation would be:

[((1 − β)/(α − β))(1 + ω04 )]−[(1−β)/(α−β)] [(1 − α)/(α − β)

+ (1 + β − 2α)/(α − β)ω04 ][(1+β−2α)/(α−β)] ω04 ω04 = K . (18.33)

18.3 EXAMPLE
The input parameters for the example are in Table 18.2.

Table 18.2: Parameters for Illustrative Problem

Symbol Value Description of Variable or Parameter
A $ 10 Set-up Cost ($/Batch)
i 0.10 Inventory Holding Cost Rate (Decimal Percent/Unit Time)
a 200 Scaling Constant for Price (Initial Price)
b 20 Scaling Constant for Cost (Initial Cost)
α 0.5 Price Elasticity with respect to Demand
β 0.1 Cost Elasticity with respect to Demand
(Degree of Economies of Scale)
C00 200 C00 = a
C01 20 C01 = b
C02 10 C02 = A
C03 1 C03 = ib/2

Using the values of Table 18.2 in Equation (18.33), the result is:

[2.25(1 + ω04 )]−2.25 [1.25 + 0.25ω04 ]0.25 ω04

2
= 1.4058 × 10−4 (18.34)
108 18. PROFIT MAXIMIZATION
Solving (18.34) via search techniques for ω04 the result is:

ω04 = 0.0296 (18.35)

Using this value for ω04 , the values of the remaining dual variables can be found from Equa-
tion (18.15), (18.17), and (18.19) as:

ω03 = ω04 = 0.0296 (18.36)

ω02 = [(1 − α)/(α − β) + ((1 + β − 2α)/(α − β))ω04 ]
= [1.25 + 0.25 ∗ (0.0296] = 1.2574 (18.37)
ω01 = [((1 − β)/(α − β))(1 + ω04 )] = 2.25 ∗ (1 + 0.0296) = 2.3166 . (18.38)

The value of the dual objective function is found from Equation (18.11) as:

D(Y ) = −1[(C00 /ω01 )ω01 (C01 /ω02 )ω02 (C02 /ω03 )ω03 (C03 /ω04 )ω04 ]−1 (18.11)

= −[(200/2.3166)−2.3166 (20/1.2574)1.2574 (10/0.0296)0.0296 (1/0.0296)0.0296 ]−1

= −[0.0013982]−1
= −715.2 . (18.39)

The minus sign indicates that it is a profit and not a cost. It is also interesting to note from
the dual variables that the first cost term is the dominant cost term and that the second and third
terms have equal value. The difference between the revenue dual variable and the sum of the cost
dual variables is 1.0, which is the total of the dual variables when a cost only model is used.
The primal variables can be obtained from the dual variables using the primal-dual relation-
ships. Rearranging the primal-dual relationship of Equation (18.20), the demand D can be evaluated
as:

D = (ω01 Y/C00 )(1/(1−α)) (18.40)

= (2.3166 ∗ 715.18/200)(1/0.5)
= 68.6 . (18.41)

The value for the order quantity Q can be determined rearranging Equation (18.22) as:

Q = ((C02 D)/(ω03 Y )) (18.42)

= ((10 ∗ 68.6)/(0.0296 ∗ 715.18))
= 32.4 . (18.43)

The value of the primal objective function can now be found using Equation (18.10)

Min Y = −C00 D 1−α + C01 D 1−β + C02 DQ−1 + C03 QD −β . (18.10)

= −200(68.6)0.5 + 20(68.6)0.9 + 10 ∗ 68.6/32.4 + 1 ∗ 32.4(68.2)−0.1
= −1656.5 + 898.9 + 21.2 + 21.2
= −715.2 . (18.44)
 18.4. TRANSFORMED DUAL APPROACH 109
The values of the primal and dual are in agreement. This problem was also solved by Yi Fang [2]
via the transformed dual approach in which a constraint is added and the problem is solved via the
constrained derivative approach as the degree of difficulty is one. The results are identical, although
the solution approaches are quite different as the method presented obtains the additional equation
via the method of dimensional analysis and solves a problem with zero degrees of difficulty.

18.4 TRANSFORMED DUAL APPROACH

The objective function of the primal is given by Equation (18.8) and is repeated here:

Max (π) = aD 1−α − (bD 1−β + ADQ−1 + (ib/2)QD −β ) . (18.8)

Since the objective function was a signomial problem, Jung and Klein [1] used the transformed
dual method developed by Duffin et al. [3] to solve the problem and their solution is presented. The
problem transforms the primal objective function into a constraint with positive terms. The problem
was formulated as:
Max z (18.45)
subject to:
aD 1−α − bD 1−β − ADQ−1 − (ib/2)QD −β ≥ z (18.46)
The transformed primal problem is:
Min z−1 (18.47)
subject to:

a −1 D α−1 z + a −1 bD α−β + a −1 AD α Q−1 + a −1 (ib/2)QD α−β−1 ≤ 1 (18.48)

The transformed primal function is a constrained posynomial function.The degree of difficulty

remains the same as one variable and one terms was added, that is:

D = T − (N + 1) = 5 − (3 + 1) = 1

From the coefficients and signs, the signum functions for the dual are:

σ01 = 1
σ11 = σ12 = σ13 = σ14 = 1
σ1 = 1 .

The dual problem formulation is:

Objective Function ω01 =1 (18.49)

z terms − ω01 + ω11 =0 (18.50)
D terms + (α − 1)ω11 + (α − β)ω12 + αω13 + (α − β − 1)ω14 =0 (18.51)
Q terms − ω13 + ω14 =0 (18.52)
110 18. PROFIT MAXIMIZATION
Using Equations (18.49)–(18.52) the values of the dual variables ω01 and ω11 can be found and ω12
and ω13 can be found in terms of ω14 . The results are:
ω01 =1 (18.53)
ω11 =1 (18.54)
ω12 = [(1 − α + (1 + β − 2α)ω14 )/(α − β)] (18.55)
ω13 = ω14 (18.56)
and ω10 = ω11 + ω12 + ω13 + ω14
or ω10 = 1 + [(1 − α − (2α − β − 1)ω14 )/(α − β)] + 2ω14
= ((1 − β)(1 + ω14 ))/(α − β)) (18.57)
The dual objective function can be stated in terms of ω14 as:
D(Y ) =[(1/ω01 )ω01 (a −1 ω10 /ω11 )ω11 (a −1 bω10 /ω12 )ω12
(a −1 Aω10 /ω13 )ω13 ((a −1 ib/2)ω10 /ω14 )ω14 ]−1 (18.58)
which, upon substitution, becomes:
D(Y ) =[1]1 ∗ [a −1 ∗ ((1 − β)(1 + ω14 ))/(α − β)]1 ∗
[[(a −1 b) ∗ ((1 − β)(1 + ω14 ))/(α − β)]/{(1 − α + (1 + β − 2α)
ω14 /(α − β)}]{(1−α+(1+β−2α)ω14 /(α−β)} ]
∗ [(a −1 A ∗ ((1 − β)(1 + ω14 )/(ω14 (α − β))ω14 ]
∗ [(a −1 (ib/2) ∗ ((1 − β)(1 + ω14 )/(ω14 (α − β))ω14 ] (18.59)
The logarithm form of Equation (18.59) is:
Log[D(Y )] = log[(a −1 (1 + β)/(α − β)) ∗ (1 + ω14 )]
+ {(1 − α + (1 + β − 2α)ω14 /(α − β)}) log[[(a −1 b)
∗ ((1 − β)(1 + ω14 ))/(α − β)/{(1 − α + (1 + β − 2α)ω14 /(α − β))}]
+ ω14 log(a −1 A ∗ (1 − β)(1 + ω14 ))/(ω14 (α − β))
+ ω14 log(a −1 (ib/2) ∗ (1 − β)(1 + ω14 ))/(ω14 (α − β)) . (18.60)
Taking the derivative of Equation (18.60) and setting it to zero and reducing the terms results
in:
Log{[(a −1 b)(1 − β)(1 + ω14 )]/[1 − α + (1 + β − 2α)ω14 ]}(1+β−2α))/(α−β)
+ Log[a −1 A((1 − β)/(α − β)) ∗ ((1 + ω14 )/ω14 )]
+ Log[a −1 (ib/2)((1 − β)/(α − β)) ∗ ((1 + ω14 )/ω14 )] = 0 (18.61)
Taking the anti-log this becomes:
{[(a −1 b)(1 − β)(1 + ω14 )]/[1 − α + (1 + β − 2α)ω14 ]}(1+β−2α))/(α−β)
∗ [a −1 A((1 − β)/(α − β)) ∗ ((1 + ω14 )/ω14 )]
∗ [a −1 (ib/2)((1 − β)/(α − β)) ∗ ((1 + ω14 )/ω14 )] = 1 (18.62)
 18.4. TRANSFORMED DUAL APPROACH 111
Using the values for the example problem of α = 0.5, β = 0.1, a = 200, b = 20, A = 10, and
i = 0.10 to solve for ω14 , Equation (18.62) becomes;
{[(200−1 20)(1 − 0.1)(1 + ω14 )]/[1 − 0.5 + (1 + 0.1 − 2 ∗ 0.5) ∗ ω14 ]}[(1+0.1−2∗0.5)/(0.5−0.1)]
∗ [(200−1 ∗ 10 ∗ ((1 − 0.1)/(0.5 − 0.1)) ∗ ((1 + ω14 )/ω14 )]
∗ [(200−1 ∗ (0.1 ∗ 20/2) ∗ ((1 − 0.1)/(0.5 − 0.1)) ∗ ((1 + ω14 )/ω14 )] = 1 (18.63)
or
[0.09 ∗ (1 + ω14 )/[0.5 + 0.1 ∗ ω14 ]0.25
∗ [0.1125 ∗ (1 + ω14 )/(ω14 )] ∗ [0.01125 ∗ (1 + ω14 )(ω14 ) = 1 (18.64)
Solving for ω14 one obtains:
ω14 = 0.0296 (18.65)
This is the same result as obtained by the previously in Equation (18.35) for ω04 and thus the
solutions will be the same as that of the dimensional analysis approach used. The constraint dual
variables are the same as the objective dual variables as the constraint was equivalent to the earlier
objective function. The remaining dual variables are:
ω12 = 1.2574 (same as ω02 earlier,
but represents the same function in this problem) (18.65)
ω13 = 0.0296 (same as ω03 earlier,
but represents the same function in this problem) (18.66)
ω11 = 1.0 (18.67)
ω10 = 2.3167 (18.68)
The value of the dual objective function using Equation (18.58) is
D(Y ) = [(1/ω01 )ω01 (a −1 ω10 /ω11 )ω11 (a −1 bω10 /ω12 )ω12
(a −1 Aω10 /ω13 )ω13 ((a −1 ib/2)ω10 /ω14 )ω14 ]−1 (18.58)
= [(1/1)1 ∗ (200−1 ∗ 2.3174/1)1
∗ (200−1 ∗ 20 ∗ 2.3167/1.2574)1.2574
∗ (200−1 ∗ 10 ∗ 2.3167/0.0296)0.0296
∗ (200−1 ∗ (0.1 ∗ 20/2) ∗ 2.3167/0.0296)0.0296 ]−1
= [1.3978 ∗ 10−3 ]−1
= 715.4 . (18.69)
The dual solution is in agreement with the dual solution obtained using the dimensional
analysis approach. The solution for the primal variables can be obtained from the primal-dual
relationships. The equations used to solve for the dual variables were quite different, but gave the
same value for ω14 and its equivalent of ω04 of 0.0286. The approach using dimensional analysis
was somewhat easier as it did not require taking the derivative of the log of the dual and it had one
less dual variable than the transformed dual approach.
112 18. PROFIT MAXIMIZATION
18.5 EVALUATIVE QUESTIONS
1. Solve the example problem changing the price elasticity to determine the effects of D and Q
and compare the results with the original example problem.

Table 18.3: Parameters for Problem 1

Symbol Value Description of Variable or Parameter
A $ 10 Set-up Cost ($/Batch)
i 0.10 Inventory Holding Cost Rate
(Decimal Percent/Unit Time)
a 200 Scaling Constant for Price (Initial Price)
b 20 Scaling Constant for Cost (Initial Cost)
α 0.3 Price Elasticity with respect to Demand
β 0.1 Cost Elasticity with respect to Demand

2. Solve the example problem changing the set-up cost to determine the effects on D and Q and
compare the results.

Table 18.4: Parameters for Problem 2

Symbol Value Description of Variable or Parameter
A $ 80 Set-up Cost ($/Batch)
i 0.10 Inventory Holding Cost Rate
(Decimal Percent/Unit Time)
a 200 Scaling Constant for Price (Initial Price)
b 20 Scaling Constant for Cost (Initial Cost)
α 0.5 Price Elasticity with respect to Demand
β 0.1 Cost Elasticity with respect to Demand

3. Solve for the primal variables using the dual variables obtained in the transformed dual ap-
proach example problem.

4. Take the derivative of Equation (18.59) and show all the steps to obtain Equation (18.63)
including the terms that are canceled.

5. Discuss the advantages and disadvantages of the methods of dimensional analysis versus the
constrained derivative approach.

REFERENCES
[1] Jung, H., Klein, C. M., Optimal inventory policies under decreasing cost functions via geometric
programming, European Journal of Operational Research 132 (2001), 628–642. 103, 109
 REFERENCES 113
[2] Yi Fang, IENG 593 F, “Geometric Programming Research Paper,” April 15, 2010, 23 pages.
109

[3] Duffin, R. J., Peterson, E. L., Zener, C., Geometric Programming – Theory and Application,
Wiley, New York, 1976. 109
 115

CHAPTER 19

Summary and Future Directions

19.1 SUMMARY
The object of this text is to generate interest in geometric programming amongst manufacturing
engineers, design engineers, manufacturing technologists, cost engineers, project managers, industrial
consultants and finance managers by illustrating the procedure for solving certain industrial and
practical problems. The various case studies were selected to illustrate a variety of applications as
well as a set of different types of problems from diverse fields. Several additional problems were added
focusing on profit maximization and additional problems with degrees of difficulty. In addition, the
methods of dimensional analysis and the constrained derivative approach have been presented in
detail. Table 19.1 is a summary of the case studies presented in this text giving the type of problem,
degrees of difficulty, and other details.
The metal removal economics example also had variable exponents in the general solution.
The problems were worked in detail so general solutions could be obtained and also to show that
the dual and primal solutions were identical. The problems were selected to illustrate a variety of
types and also to show the use of the primal-dual relationships to determine the equations for the
primal variables. It is by showing the various types of applications in detailed examples that others
can follow the procedure and develop new applications.

19.2 FUTURE DIRECTIONS

The author is hopeful that others will communicate him additional examples to illustrate new appli-
cations that can be included in future editions. New applications will attract new practitioners to this
fascinating area of geometric programming. It is believed that the scope of geometric programming
will expand with new applications.
The author would like to include some software for different applications in geometric pro-
gramming in the future and would welcome contributions.

19.3 DEVELOPMENT OF NEW DESIGN RELATIONSHIPS

There are many different types of problems that can be solved by geometric programming and one
of the significant advantages of the method is that it is possible in many applications to develop
general design relationships. The general design relationships can save considerable time and effort
in instances where the constants are changed.
116 19. SUMMARY AND FUTURE DIRECTIONS
Although geometric programming was first presented nearly 50 years ago, the applications
have been rather sparse compared to that of linear programming. One goal is that as researchers
take advantage of the potential to develop design relationships that new applications will rapidly
occur. The development of new design relationships can significantly reduce the development time
and cost for new products and this is essential for companies to remain competitive in the global
economy.
117
19.3. DEVELOPMENT OF NEW DESIGN RELATIONSHIPS

Table 19.1: Summary of Case Study Problem

Degrees Number Number Number
Special
Chapter Case Study of of of Variable Description of
Characteristics
Difficulty Constraints Variables Solutions
4 The Optimal Box Design 0 1 3 Height, Width, Length 1
5 Trash Can 0 1 2 Height, Diameter 1
6 Open Cargo Shipping Box 0 1 3 Height, Width, Length 1 Classical Problem
7 Metal Casting Cylindrical Riser 0 1 2 Height, Diameter 1
8 Inventory Model 0 0 1 Lot Size 1
9 Process Furnace Design 0 1 3 Temperature, Length, 1 Dominant Equation
Height Negative Dual Variable
Dominant Equation
10 Gas Transmission Pipe Line 0 1 4 Length, Diameter, Flow 1 Four Variables
Length
Pressure Ratio Factor
11 Profit Maximization 0 0 3 1 Cobb-Douglas Profit Function
12 Material Removal/Metal Cutting 0 1 2 Feed Rate, Cutting Speed 1 Fractional Exponents
13 Journal Bearing Design 1 1 2 Journal Radius, 1 Dimensional Analysis Approach
Bearing Half-Length
14 Metal Casting Hemispherical Top 2 1 2 Height, Diameter 1 Dimensional Analysis Approach
Riser
15 Liquefied Petroleum(LPG) Cylinder 1 1 2 Height, Diameter 2 Multiple Solutions
Dimensional Analysis Approach
16 Metal Cutting Economics- 1 2 2 Feed Rate, Cutting Speed 3 Multiple Solutions
2 Constraints Derivative Approach
17 Open Cargo Shipping Box with Skids 1 0 3 Height, Width, Length 1 Derivative Approach
Dimensional Analysis Approach
Condensation of Terms Approach
18 Profit Maximization with Decreasing 1 0 2 Lot Size, Demand 1 Dimensional Analysis Approach
Cost Functions 1 1 3 Transformed Dual Approach
 119

CHAPTER 20

Thesis and Dissertations on

Geometric Programming
A search was made to find the thesis and dissertations on geometric programming. The listed
items had geometric programming listed in the abstract, but that does not necessarily imply that
the research focused on geometric programming. However, most the research work did focus on
geometric programming. The search was made using Dissertation Abstracts Online and they do
have the thesis/dissertation available for sale on their web site.
 120

Type
Author Title School Year
M.
1 Cheung, Wing Tai Geometric Programming and Signal Flow Graph Assisted Design of Interconnect and Analog Circuits University of Hong Kong 2008 Phil
2 Firouzabadi, Sina Jointly Optimal Placement and Power Allocation of Nodes of a Wireless Network University of Maryland 2007 MS
3 Gupta, Deepak Prakash Energy-Sensitive Machining Parameter Optimization Model West Virginia University 2005 MSIE
4 Kokatnur, Ameet Design and Tevelopment of a Target-Costing Model for Machining West Virginia University 2004 MSIE
5 Dev. and Opt. of a NL Multibumper Design for Spacecraft Protective Structures Using Simulated
Helba, Michael John Univ. of Alabama, Huntsville 1994 MS
Annealing
6 Liu, Guili The Hoelder Inequality and Its Application in Banach Algebra and Geometric Programming The University of Regina(CAN) 1992 MSc
7 Yang, Edward V. Sensitivity of Optimum Heat Exchanger Design by Geometric Programming Univ. of Massachusetts-Lowell 1976 MS
8 Meteer, James W. A Geometric Programming Solution to a Strata Allocation Problem University of Wyoming 1976 MS
9 Yeh, Lucia Lung-Chia Geometric Programming Solution of a Problem Involving Sampling from Overlapping Frames University of Wyoming 1974 MS
10 Shah, Shivji Virji Optimization of Finned Tube Condenser by Geometric Programming Univ. of Massachusetts-Lowell 1973 MS

11 Shah, Hemendrakumar R. Optimal Heat-Exchanger Design by Geometric Programming Univ. of Massachusetts-Lowell 1971 MS

Figure 20.1: Original data was obtained from Dissertation Abstracts Online.
20. THESIS AND DISSERTATIONS ON GEOMETRIC PROGRAMMING
121

Author Title School Year Type

1 Charafeddine, Mohamad Communication over n-user Interference Channel Stanford University 2009 Ph.D.
2 Kim, Jintae Multi-level Design Optimizations of Pipelined A/D Converter UCLA 2008 Ph.D.
Cross-Layer Design for Reliable and Efficient Data Transmission over Multiple Antenna Mobile Infostation
3 Liu, Hongbo Rutgers University 2008 Ph.D.
Networks
4 Tan, Chee Wei Nonconvex Power Control in Multiuser Communication Systems Princeton University 2008 Ph.D.
5 Joshi, Siddarth Large-Scale Geometric Programming for Devices and Cirsuits Stanford University 2008 Ph.D.
6 Roy, Sanghamitra Novel Modeling and Optimization Techniques for Nano-scale VLSI Designs Wisconsin University 2008 Ph.D.
7 Patil, Dinesh Design of Robust Energy-Efficient Digital Circuits Using Geometric Programming Stanford University 2008 Ph.D.
8 Seong, Kibeom Cross-Layer Resource Allocation for Multi-user Communication Systems Stanford University 2008 Ph.D.
9 Hsiung, Kan-Lin Geometric Programming Under Uncertainity withEngineering Applications Stanford University 2008 Ph.D.
10 Zheng, Gan Optimization in Linear Multiuser MIMO Systems University of Hong Kong 2007 Ph.D.
11 Zhang, Wei Bivariate Cubic L1 Splines and Applications North Carolina State 2007 Ph.D.
12 Zhou, Quming Reliability-driven Circuit Optimization and Design Rice University 2007 Ph.D.
13 Chinney, David Graeme Low Power Design Amtomation UC-Berkeley 2006 Ph.D.
14 Colleran, David M. Optimization of Phase-locked Loop Circuits via Geometric Programming Stanford University 2006 Ph.D.
15 Vanderhaegen, John Peter A Design Methodology for Analog Circuits Based on Global Optimization UC-Berkeley 2005 Ph.D.
16 Wang, Yong Theory and Algorithms for Shape-preserving Bivariate L1 Splines North Carolina State 2005 Ph.D.
17 Yun, Sunghee Convex Optimization for Digital Integrated Circuit Applications Stanford University 2005 Ph.D.
18 Xu, Yang Affordable Analog and Radio Frequency Integrated Circuits Design and Optimization Carnegie Mellon University 2004 Ph.D.
19 O'Neill, Daniel C. Jointly Optimal Network Performance: A Cross-layer Approach Stanford University 2004 Ph.D.
20 Chaing, Mung Solving Nonlinear Problems in Communication Systems Using Geometric Programming and Dualities Stanford University 2003 Ph.D.
21 Qin, Zhanhai Topological Circuit Reduction: Theory and Applications UC-San Diego 2003 Ph.D.
22 Cheng, Hao Tehory and Algorithms for Cubic L(1) Splines North Carolina State 2002 Ph.D.
23 Jacobs, Etienne Theodorus Power Dissapation and Timing in CMOS Circuits Technische Univ. Eindhoven 2001 Dr.
24 Jung, Hoon Optimal Inventory Policies for an Economic Order Quantity Models Under Various Cost Functions Univ. Missouri-Columbia 2001 Ph.D.
25 Hershenson, Maria del Mar CMOS Analog Circuit Design via Goemetric Programming Stanford University 2000 Ph.D.
26 Lou, Jinan Integrated Logical and Physical Optimizations for Deep Submicron Circuits Univ. Southern California 1999 Ph.D.
27 Xu, Li Na Optimization Methods for Computing Empirically Constrained Extremal Probability Distributions The University of Iowa 1996 Ph.D.
28 Zao, John Kar-kin Finite-Precision Representation and Data Abstraction for Three-Dimensional Euclidean Transformations Harvard University 1995 Ph.D.
29 Maranas, Costas D. Global Optimization in Computational Chemistry and Process Systems Engineering (Potential Energy) Princeton University 1995 Ph.D.
30 Ince, Erdem A Parallel Balanced Posynomial Unconstrained Geometric Programming Algorithm Colorado School of Mines 1995 Ph.D.
Figure 20.1: Thesis and Dissertations on Geometric Programming. Continues.
 122

Author Title School Year Type

31 Yang, Hsu-Hao Investigation of Path-Following Algorithms for Signomial Geometric Programming Problems The University of Iowa 1994 Ph.D.
32 Jackson, Jack Allen, Jr. A Mathematical Experiment in Dual Geometric Programming Colorado School of Mines 1994 Ph.D.
33 Mader, Douglas Paul An Improved Method for Sensitivity Analysis in Geometric Programming Colorado School of Mines 1994 Ph.D.
34 Cheng, Zhao-Yang A Least Squares Approach to Interior Point Methods with an Application to Geometric Programming Rensselaer Poly. Institute 1993 Ph.D.
35 Zhu, Jishan On the Path-Following Methods for Linearly constrained Convex Programming Problems The University of Iowa 1992 Ph.D.
Issues and Methodologies in Auto. Proc. Planning for a Three-Dimensional Part Representation to Recomend
36 Narang, Ramesh Vashoomal The University of Iowa 1992 Ph.D.
Machining Parameters
37 Martino, Thomas John A Tandem Joint Inspection and Queueing Model Employing Posynomial Geometric Programming University of Rhode Island 1991 Ph.D.
An Algorithm for Solving a Class of Nonlinear, Unconstrained, Multi-Variable, Signomial Optimization Problems
38 Katz, Joseph Harold Colorado School of Mines 1991 Ph.D.
(GP)
39 Dull, Owen S. A G P Based Method for Determining the Required Constraints in a Class of Chemical Blending Problems Colorado School of Mines 1991 Ph.D.
40 Viriththamulla, Gamage I. Mathematical Programming Models and Heuristics for Standard Modular Design Problem The University of Arizona 1991 Ph.D.
41 Knowles, James Alexander A Geometric Programming Approach to the Solution of Multistage Countercurrent Heat Exchanger Systems Colorado School of Mines 1990 Ph.D.
42 Rama, Dasaratha V. Planning Models for Communications Satellites Offering Customer Premises Services The University of Iowa 1990 Ph.D.
43 No, Hoon Accelerated Convergent Methods for a Class of Nonlinear Programming Problems The University of Iowa 1990 Ph.D.
44 Wessels, Gysbert Johannes A Geometric Programming Algorithm for Solving a Class of Nonlinear, Signomial Optimization Problems Colorado School of Mines 1989 Ph.D.
45 Bailey, Steven Scott An Algorithm for the Solution of a Class of Economic Models for the Design of Prestressed Concrete Bridges Colorado School of Mines 1989 Ph.D.
46 Kirk, John B. A Geometric Programming Based Algorithm for Preprocessing Nonlinear Signomial Optimization Programs Colorado School of Mines 1988 Ph.D.
An Algorithm for Solving a Class of Nonlinear, Unconstrained, Signomial Economic Models Using the Greening
47 Thome, James John, Sr. Colorado School of Mines 1988 Ph.D.
Technique
48 Plenert, Gerhard Johannes A Solution to Smith's General Case Bottleneck Problem Allowing for an Unlimited Number of Scheduled Jobs Colorado School of Mines 1987 Ph.D.
49 Chibani, Lahbib Integrated Optimal Design of Structures Subjected to Alternate Loads Using Geometric Programming University of Illinois-U-C 1987 Ph.D.
50 Muhammed, Abdelfatah A. Information Theory and Queueing Theory Via Generalized Geometric Programming North Carolina State 1987 Ph.D.
51 Mrema, Godwill E. E. C. Computation of Chemical and Phase Equilibria Universitetet I Trondheim 1987 Dr. Ing
52 Li, Xing-Si Entropy and Optimization University of Liverpool 1987 Ph.D.
53 Choi, Jae Chul Generalized Benders' Decomposition for Geometric Programming with Several Discrete Variables The University of Iowa 1987 Ph.D.
54 Arreola Contreras, Jose J. Semi-discrete Geometric Programming University of Pittsburgh 1986 Ph.D.
55 Tsai, Pingfang An Optimization Algorithm and Economic Analysis for a Constrained Machining Model West Virginia University 1986 Ph.D.
56 Rajopal, Jayant A Duality Theory for Geometric Programming Based on Generalized Linear Programming The University of Iowa 1985 Ph.D.
20. THESIS AND DISSERTATIONS ON GEOMETRIC PROGRAMMING

57 Murphy, Robert Craig A Dual-Dual Algorithm for the Solution of a Class of Nonlinear Resource Allocation Models Colorado School of Mines 1985 Ph.D.
58 Burns, Scott Allen Structural Optimization Using Geometric Programming and the Integrated Formulation University of Illinois-U-C 1985 Ph.D.
59 Djanali, Supeno Geometric Programming and Decomposition Techniques in Optimal Control University of Wisconsin 1984 Ph.D.

60 Rajasekera, Jayantha R. Perturbation Techniques for the Solution of Posynomial, Quadratic and LP Approximation Programs North Carolina State University 1984 Ph.D.

Figure 20.1: Continued. Thesis and Dissertations on Geometric Programming. Continues.

123

Author Title School Year Type

61 Tang, Che-Chung Mathematical Models and Optimization Techniques for Analysis and Design of Wastewater Treatment Systems University of Illinois-U-C 1984 Ph.D.
62 Ravanbakht, Camela Geometric Programming: An Efficient Computer Algorithmfor Traffic Assignmet North Carolina State University 1984 Ph.D
63 Chou, Jaw Huoy Contributions to Nondifferential Mathematical Programming North Carolina State University 1984 Ph.D
64 Kyparsis, Jerzy Sensitivity and Stability for Nonlinear and Geometric Programming: Theory and Applications George Washington University 1983 D.SC.
65 Dziuban, Stephen T. Ellipsoid Algorthim Variants in Nonlinear Programming Rensselaer Poly. Tech. 1983 Ph.D
Self-Teaching Enrichment Modules on Modern Applications of Mathemetics for Talented Secondary Students
66 Morningstar, Kay Gladys University of Delaware 1983 Ph.D
and Evaluation…
A Study of the Relationships Among Computer Programming Ability, Computer Program Content, Computer
67 Duggar, Cynthia Miller Georgia State University 1983 Ph.D
Programming Style…..
68 Ohanomah, Matthew O. Computational Algorithms for Multicomponent Phase Equilibria and Distillation Univ. of British Columbia 1982 Ph.D
69 Marin Gracia, Angel Contribuciones A La Teoria Matematica De La Planifcacion Del Transporte Y Sus Applicaciones Univ. Politecnica De Madrid 1982
70 Mayer, Robert Hall, Jr Cost Estimates from Stochastic Geometric Programs University of Delaware 1982 Ph.D
71 Brar, Guri Singh Geometric Programming for Engineering Design Optimization The Ohio State University 1981 Ph.D
72 Memon, Altaf Ahmed Environmental Noise Management: An Optimization Approach University of Pittsburgh 1980 Ph.D
73 Silberman, Gabriel Maurcio The Design and Evaluation of an Active Memory Unit SUNY-Buffalo 1980 Ph.D
74 Junna, Mohan Reddy Irrigation System Improvement by Simulation and Optimization Colorado State University 1980 Ph.D
75 Rosenberg, Eric Globally Convergent Algorithms for Convex Programming with Applications to Geometric Programming Stanford University 1979 Ph.D
Optimization of the Second-order Logarmithmic Machining Economics Problem by Extended Geometric
76 Hough, Clarence Lee, Jr. Texas A&M University 1978 Ph.D
Programming
The Design of Solid Waste Systems: An Application of Geo. Prog. to Problems in Municipal Solid Waste
77 Yale, Wilson Winant Lehigh University 1978 Ph.D
Management
78 Sheldon, Dan Nonlinear Optimization by Geometric Programming Methods with Applications UCLA 1978 Ph.D
79 Wong, Shek-Nam Danny Nonlinear Models for Systems Planning and Control: Solution and Analyses Via Geometric Programming The Pennsylvania State Univ. 1978 Ph.D
80 Maurer, Ruth Allene Lamb A Geometric Programming Approach to the Preliminary Design of Wastewater Treatment Plants Colorado School of Mine 1978 Ph.D
81 Ramamurthy, Subramanian Structural Optimization Using Geometric Programming Cornell University 1977 Ph.D
82 Grange, Franklin Earnest, II The Solution and Analysis of the Geometric Mean Portfolio Problem with Methods of Geometric Programming Colorado School of Mine 1977 Ph.D
83 Kise, Peter Ethan Systems of Optimization via Geometric Programming University of Pennsylvania 1976 Ph.D
84 Wu, Chin Chang Design and Modeling of Solar Sea Power Plants by Geometric Programming Carnegie-Mellon University 1976 Ph.D
Semi-infinite Programming Equivalents and Solution Techniques for Optimal Experimenjtal Design and GP
85 Gribik, Paul Robert Carnegie-Mellon University 1976 Ph.D
Problems..
86 McMasters, John Hamilton The Optimization of Low-speed Flying Devices by Geometric Programming Purdue University 1975 Ph.D
87 Hall, Michael Anthony A Geometric Programming Approach to HighwayNetwork Equilibria Northwestern University 1974 Ph.D
88 Emery, Sidney Williams, Jr. Transoceanic Dry-Bulk Systems Optimization by Geometric Programming Stanford University 1974 Ph.D
89 Plecnik, Joseph Matthew Optimization of Engineering Systems using Extended Geometric Programming The Ohio State University 1974 Ph.D
90 Zorachi, Michael John A Primal Method for Geometric Programming Rensselear Polytechnic Institute 1974 Ph.D
Figure 20.1: Continued. Thesis and Dissertations on Geometric Programming. Continues.
 124

Author Title School Year Type

91 Salinas-Pacheco, Juan Jose Minimum Cost Design of Concrete Beams Using Geometric Programming University of Calgary 1974 Ph.D
92 Smeers, Yves Marie Geometric Programming with Applications to Management Science Carnegie-Mellon University 1973 Ph.D
Deterministic and Stochastic G P Models for Optimal Engineering Design in Electrical Power Generation &
93 Wiebking, Rolf D. Rensselaer Poly. Institute 1973 Ph.D
Simulation
94 Ben-Tal, Aharon Contributions to Geometric Programming and Generalized Convexity Northwestern University 1973 Ph.D
95 Vandrey, Joan Betty W. Reversed Generalized Geometric Programming Northwestern University 1973 Ph.D
A Geometric Programming Approach to the Problem of Finding Efficient Algorithms for Constrained Location
96 Planchart, Alejo Northwestern University 1973 Ph.D
Prob..
97 McCarl, Bruce Alan A Computational Study of Polynomial Geometric Programming The Pennsylvania State Univ. 1973 Ph.D
N=Dimensional Transportation Problems: Algorithm for Linear Problems and Application of GP for non-linear
98 Mekaru, Mark Masanobu Arizona State University 1973 Ph.D
Probs.
99 Jefferson, Thomas Richard Geometric Programming with an Application to Transportation Planning Northwestern University 1972 Ph.D
100 Beck, Paul Alan A Modified Convex Simplex Algorithm for Geometric Programming with Subsidiary Problems Rensselaer Poly. Institute 1972 Ph.D
101 Dinkel, John Joseph A Duality Theory for Dynamic Programming Problems via Geometric Programming Northwestern University 1971 Ph.D
102 Unklesbay, Kenneth Boyd The Extension of Geometric Programming to Optimzl Mechanical Design Univ. of Missouri-Columbia 1971 Ph.D
103 Oleson, Gary K. Computational Aspects of Geometric Programming Iowa State University 1971 Ph.D
The Optimal Design of Water Quality Management Systems: An Application of Multistage Geometric
104 McNamara, John Richard Rensselaer Poly. Institute 1971 Ph.D
Programming
105 Staats, Glenn Edwin Computational Aspects of Geometric Programming with Degrees of Difficulty University of Texas-Austin 1970 Ph.D
106 Teske, Clarence Eugene Optimal Structural Design Using Geometric Programming Texas Tech University 1970 Ph.D
107 Kochenberger, Gary Austin Geometric Programming: Extensions to Deal with Degrees of Difficulty and Loose Constraints Univ. of Colorado-Bolder 1969 D.B.A.
108 Pascual, Luis De La Cruz Constrained Maximization of Posynomials and Vector-Valued Criteria in Goemetric Programming Northwestern University 1969 Ph.D
109 Ecker, Joseph George Geometric Programming: Duality in Quadratic Programming and LP Approximation The University of Michigan 1968 Ph.D
110 Passy, Ury Generalization of Geometric Programming: Partial Control of Linear Inventory Systems Stanford University 1966 Ph.D

111 Avriel, Mordecai Topics in Optimization: Block Search; Applied and Stochastic Geometric Programming Stanford University 1966 Ph.D
20. THESIS AND DISSERTATIONS ON GEOMETRIC PROGRAMMING

Figure 20.1: Continued. Thesis and Dissertations on Geometric Programming.

 125

Author’s Biography

DR. ROBERT C. CREESE

Dr. Robert C. Creese, Certified Cost Engineer (CCE), is Professor of Industrial and Manage-
ment Systems Engineering at West Virginia University, USA and recently has taught courses on
Engineering Economy, Advanced Engineering Economics, Cost and Estimating for Manufactur-
ing, Manufacturing Processes and Advanced Manufacturing Processes. He has previously taught at
The Pennsylvania State University (9 years), Grove City College (4 years), Aalborg University in
Denmark (3 sabbaticals) and at West Virginia University for over 31 years.
He is a Fellow of the Association for the Advancement of Cost Engineering, International
(AACEI), received the Charles V. Keane Service Award and Brian D. Dunfield Educational Service
Award presented by AACE, and has been treasurer of the Northern West Virginia Section of AACE
for more than 20 years. He is a Life Member of AACE International, ASEE (American Society for
Engineering Education) and ASM (American Society for Materials). He also is a member of ISPA,
SCEA, AIST, AWS, and AFS.
He obtained his B.S. Degree in Industrial Engineering from the Pennsylvania State University,
his M.S. Degree in Industrial Engineering from the University of California at Berkeley, and his
Ph.D. Degree in Metallurgy from the Pennsylvania State University.
He has authored the book Introduction to Manufacturing Processes and Materials (Marcel
Dekker-1999) and co-authored two books Estimating and Costing for the Metal Manufacturing
Industries(Marcel Dekker-1992) with Dr. M. Adithan of VIT University Vellore, India and Dr. B.
S. Pabla of the Technical Teachers’ Training Institute, Chandigarh, India and Strategic Cost Analysis
for Project Managers and Engineers (New Age International Publishers-2010) with Dr. M. Adithan,
VIT University, Vellore, India. He has authored and co-authored more than 100 technical papers.
 127

Index

Arithmetic Mean , 4 Dual Variables , 8, 9, 14, 18, 20, 22, 23, 29, 34,
36, 38–40, 44, 46, 48, 49, 60–62, 64,
Box design problem , 13 65, 69–71, 80, 81, 88–91, 95, 100,
105, 107, 108, 110–112
Cargo Shipping Box , 21, 95
Furnace Design , 37
Chvorinov’s Rule, 27, 67
Classical Problem, 95 Gas Transmission (Pipeline) design , 43
Cobb-Douglas Production Function, 47, 49 Geometric Mean , 4
Compressor Pressure Ratio Factor, 43 Geometric Programming, 3–7, 10, 13, 16, 17,
Condensation of Terms Approach, 95 21, 25, 27, 30, 31, 33, 36, 37, 47, 50,
Constrained Derivative Approach, 64, 99, 102, 55, 65, 75, 77, 87, 92, 95, 112, 115,
109, 112, 115 116, 119, 121–124
Cost Functions , 36, 112 Advantages, 115
Cylindrical Riser Design , 29, 30 Applications , 17, 115, 116
Engineering Design , 5, 6, 10
Derivative Approach, 64, 95–97, 99, 102, 109, History , 4, 5
112, 115 Pioneers , 5
Design of LPG (Propane gas) Cylinder Design
Inequality Constraints, 8, 9
(Propane Gas) Cylinder(LPG)
design , 77 Journal Bearing Design , 59
Design of LPG (Propane gas) Cylinder Design
(Propane Gas) Cylinder(LPG) Linear Programming , 4, 9
design Cylinder, 77 LPG (Propane gas) Cylinder Design , 77
Design of LPG (Propane Gas) Cylinder(LPG) Machining Economics , 55, 92
design , 77 Material Removal Economics , 51
Design of LPG (Propane Gas) Cylinder(LPG) Metal Casting, 5, 27, 31
design Cylinder, 77 Metal cutting (material removal) Economics ,
Design Relationships, 3, 24, 31, 75, 115, 116 51, 54, 55, 85, 86, 91, 92, 102
Dimensional Analysis Technique , 83 Metal cutting (Metal cutting (material
Dual Objective Function , 8, 47, 48, 64, 72, 81, removal) Economics ) Economics ,
82, 88, 90, 101, 105, 108, 110, 111 51, 55, 85, 86, 91, 92, 102
128 INDEX
Multiple Solutions, 85 Profit Maximization , 47, 103, 115
Propane Gas Cylinder(LPG) design , 77
Non-negativity Conditions , 8

Open Cargo Shipping Box , 21, 95 Riser Design, 5, 6, 27, 29–31, 67, 68, 75
Optimal Box Design Problem , 13 Cylindrical Riser , 29–31, 75
Optimization, 3–6, 10, 41, 55, 92
Scale of Production, 47
design optimization , 3
techniques of optimization (methods) , 3 Search Techniques , 108
Orthogonal Conditions , 8 Summary, 115

Primal and Dual Formulation, 54 Techniques of Optimization Methods , 3

Primal Objective Function , 8, 13, 17, 22, 47, Transformed Dual Approach, 109, 111, 112
96, 102, 105, 108, 109 Trash Can , 17, 20