Citation
On the automated optimal design of constrained structures

Material Information

Title:
On the automated optimal design of constrained structures
Creator:
Hornbuckle, Jerry Clyde, 1942-
Publication Date:
Language:
English
Physical Description:
ix, 254 leaves. : illus. ; 28 cm.

Subjects

Subjects / Keywords:
Approximation ( jstor )
Boundary conditions ( jstor )
Buckling ( jstor )
Cost functions ( jstor )
Eigenvalues ( jstor )
Initial guess ( jstor )
Mathematical variables ( jstor )
Mathematics ( jstor )
Maximum principle ( jstor )
Structural deflection ( jstor )
Buckling (Mechanics) ( lcsh )
Dissertations, Academic -- Engineering Sciences -- UF
Engineering Sciences thesis Ph. D
Girders ( lcsh )
Structural design ( lcsh )
Structural dynamics ( lcsh )
Structural engineering ( lcsh )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis -- University of Florida.
Bibliography:
Bibliography: leaves 244-253.
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
Jerry C. Hornbuckle.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
022919596 ( ALEPH )
14184224 ( OCLC )
ADB3751 ( NOTIS )

Downloads

This item has the following downloads:


Full Text













ON THE AUTOMATED OPTIMAL DESIGN


OF

CONSTRAINED STRUCTURES











BY

JERRY C. HORNBUCKLE


A Mi'.S;FTATION PRESENTED TO THE GR\DU.LITF CLOU'NCE OF
THE UNIVERSITY OF FLOr.IDA IN PARTIAL
FULFILLMENT OF THE REQLIRC1;S -r THE D'LE OF
DOCTOR OF PHILOSOPHY



[lilVEPSITY OF FLORIOA
1974
































Copyright by

Jerry C. Hornbuckle

1974




























DEDICATION



To my grandmother, Mrs. Florence Hornbuckle,

and my wife, Carolyn. Without the love, confidence, and

patient understanding of Granny Hornbuckle and Carolyn

my graduate studies would never have been attempted.

















ACKNOWLEDGMENTS


To Dr. Robert L. Sierakowski and Dr. William H. Boykin, Jr.,

for guiding my research and for being more than just advisors.

To Dr. Gene W. Hemp, Dr. Ibrahim K. Ebcioglu, and Dr. John

M. Vance, for their assistance, support, and for serving on my advisory

committee.

To Dr. Lawrence E. Malvern and Dr. Martin A. Eisenberg, for

always finding the time to offer advice and explanations on questions

related to solid mechanics and academics.

To the departmental office staff for their kind assistance

with administrative problems and clerical support.

To Randell A. Crowe, Charles D. Myers, and J. Eric Schonblom

for attentive discussions of many little problems and for assistance

in preparing for the qualifying examination.














TABLE OF CONTENTS

Page

ACKNOWLEDGMENTS . iv

ABSTRACT . viii

CHAPTER
I INTRODUCTION ,. .. 1

1.0 Survey Papers .. .. 1
1.1 Historical Development: Optimal Columns 6
1.2 Historical Development: Optimal Static Beams 9
1.3 Historical Development: Optimal Dynamical Beams 12
1.4 Scope of the Dissertation .. 14

II GENERAL PROBLEMS AND METHODS IN STRUCTURAL OPTIMIZATION 16

2.0 Introduction . .. 16
2.1 Problem Classification Criteria 16

2.1.0 Problem Classification Guidelines .. 18
2.1.1 Governing Equations of the System .. 18
2.1.2 Constraints ... 19
2.1.3 Cost Functionals ... 21
2.2 Methods: Continuous Systems ... 26
2.2.0 Special Variational Methods .. 27
2.2.1 Energy Methods .. 27
2.2.2 Pontryagin's Maximum Principle .. 29
2.2.3 Method of Steepest Ascent/Descent .. 31
2.2.4 Transition Matrix: Aeroelasticity Problems 31
2.2.5 Other Miscellaneous Methods .. 32

2.3 Methods: Discrete Systems ... 33

2.3.0 Mathematical Programming .. 34
2.3.1 Discrete Solution Approximations .. 35
2.3.2 Segmentwise-Constant Approximations .. 37
2.3.3 Complex Structures with Frequency
Constraints 38
2.3.4 Finite Element Approximations .. 40
2.4 Closure . .. 41







TABLE OF CONTENTS (Continued)


CHAPTER Page

III THEORETICAL DEVELOPMENT .. 43

3.0 Introduction . .. 43
3.1 Problem Statement and Necessary Conditions ... 44
3.2 Mathematical Programming: Gradient
Projection Method ... 48
3.3 Gradient Projection Methods Applied to
the Maximum Principle ... 53
3.4 Maximum Principle Algorithm .. 60
3.5 Solution Methods .. 63

IV CONSTRAINED DESIGN OF A CANTILEVER BEAM
BELiDIING DUE TO ITS OWN WEIGHT .. 66

4.0 Introduction . 66
4.1 Problem Statement .. 66
4.2 Structural System. .... 67
4.3 Unmodified Application of the Maximum Principle 70
4.4 Results: Geometric Control Constraints .. 79
4.5 Inequality Stress Constraints .. 89
4.6 Results: Stress Constraints Included. .. 93

V CONSTLr.I[:i) DESIGN FOR AN OPTIMAL EIGENVALUE P.ODLLF1I 101

5.0 Introduction . .. .. 101
5.1 Problem Statement .. 101
5.2 Structural System .. 102
5.3 Analysis of the Problem ... 109
5.4 Application of the Maximum Principle .. 121
5.5 Results: Geometric Control Constraints 132
5.6 Inequality Stress Constraints ... 148

VI FINITE ELEMENT METHODS IN ST U'ii i.i. OPf it...'T L.__:
AN EXAMPLE . 155

6.0 Introduction . .. 155
6.1 Finite Element Problem Statement. ... 155
6.2 Mathematical Programming: Gradient
Projection Method .. 157
6.3 Results ... 162

VII COMMENTS ON NUMERICAL INSTABILITY IN THE
(l'ir'SILINEARIZATION ALGORITHM. ... .. 173

7.0 Introduction .. 173
7.1 Computer Program Convergence Features .. 1.73
7.2 :!,.:rical Instatilibies for Cantilever
Beam Example . 175
7.3 Numerical Instabilities for Column
Buckling Example .... 183









TABLE OF CONTENTS (Continued)


CHAPTER Page
VIII CONCLUSIONS AND RECOMMENDATIONS ... 186

8.0 Summary and Conclusions ... 186
8.1 Recommendations ... 186

APPENDIXES
A HISTORICAL DEVELOPMENTS . 191

B A SIMPLE PROOF OF THE KUHN-TUCKER THEOREM .. 206

C COMPUTER SUBROUTINE LISTINGS ... 214


BIBLIOGRAPHY ..... .... 244

BIOGRAPHICAL SKETCH . ... 254








Abstract of Dissertation Presented to the Graduate Council
of the University of Florida in Partial Fulfillment of the Requirements
for the Degree of Doctor of Philosophy


ON THE AUTOMATED OPTIMAL DESIGN
OF CONSTRAINED STRUCTURES

By

Jerry C. Hornbuckle

August, 1974


Chair-man: Dr. William H. Boykin, Jr.
Co-Chairman: Dr. Robert L. Sierakowski
Major Department: Engineering Sciences


Pontryagin's Maximum Principle is applied to the optimal design

of elastic structures, subject to both hard inequality constraints and

subsidiary conditions. By analyzing the maximum principle as a non-

linear programming problem, an explicit formulation is derived for the

Lagrangian multiplier functions that adjoin the constraints to the cost

functional. With this result the usual necessary conditions for opti-

mality can themselves be used directly in an algorithm for obtaining

a solution.

A survey of general methods and problems in the optimal design

of elastic structures shows that there are two general problem types

depending upon whether or not the cost functional is an eigenvalue.

An example problem of each type is included with the solutions obtained

by the method of quasilinearization. In the first example, a minimum

deflection beam problem, classical Maximum Principle techniques are

used. The eigenvalue problem is exemplified by the maximization of the

buckling load of a column and uses the explicit multiplier function


viii








formulation mentioned above. Since the problem considered is conserva-

tive, it is therefore described mathematically by a self-adjoint system;

under this condition it is shown that the minimum weight problem is

identical to the maximum buckling load problem.

In order to demonstrate the theory for the programming techniques

used, the beam problem is also solved by using a finite element repre-

sentation of the structure. From a comparison to the maximum principle

solution it is shown that the form of the optimal solution obtained is

dependent upon the magnitude of the tolerance used with the numerical

solution scheme. Furthermore, it is shown that convergence by the

quasilinearization algorithm is related to the respective curvatures of

the initial guess and the solution.

Recommendations for additional investigations pertinent to this

study are also included.













CHAPTER I


INTRODUCTION



1.0 Survey Papers

It is exceedingly difficult to write a general introduction to

the field of structural optimization for two basic reasons: (i) there

is no conventionally accepted nomenclature, and (ii) there is also no

conventionally accepted classification of problem types or character-

istics. In marked contrast, when one considers the calculus of varia-

tions, "cost functional, system equations, kinematic and natural bound-

ary conditions, adjoint variables, Hamiltonian, etc.," all have well-

defined, universally accepted meanings. Additionally, there is no con-

fusion when speaking of the problem types of Mayer, Lagrange, and Bolza.

This common language and categorization of problems does not exist in

structural optimization.

Instead, the field tends to branch and fragment into very special-

ized sub-disciplines that are oriented towards applications. ':hile these

branches are related to the general field, the techniques and methods of

one branch can seldom be applied to another. Moreover, as a result of

the tendency to an applications orientation, solutions are generally ad

hoc and not useful for other problems even within the same branch. The

lack of any definitive unification of the subject cannot be blamed either

on being recently developed or in receiving too little attention.










This is readily seen by considering the survey papers described in the

following paragraphs.

The earliest comprehensive survey paper is Wasiutynski and

Brandt (1963). Although their excellent historical development is

dominated by Russian and Eastern European references, the authors do

include a higher percentage of papers by Western authors than is

encountered in the typical paper from Eastern Europe. A more funda-

mental criticism is that too little is said regarding problem types or

solution methods.

Chronologically, the next survey paper is Gerard (1966).

The theme of this review is aerospace applications, with a particular

orientation to the design-management, decision-making process. Most of

the papers cited treat specialized aerospace structures and applica-

tions; however, the author does try to generalize by introducing a

design index D, a material efficiency parameter M, and a structural

efficiency parameter S. After defining the expressions for M and S

corresponding to several structural elements, design charts are pre-

sented which show regions of possible application for various materials.

Unfortunately the design charts do not satisfy expectations aroused by

the introduction of the three general parameters.

Rozvany (1966) presents a similar paper pertaining to structures

in civil engineering. This paper is less comprehensive and more ori-

ented towards specific structural applications than either of the pre-

ceding survey papers. The author postulates several "interrelated

qu3nti.ies (or parameters)" which could perhaps be used to generalize










structural optimization into a more rational methodology. However,

these quantities--loading (L), material (M), geometry (G), initial

behavior (IB), and design behavior (SB)--are only applied to an abstract

discussion of concepts.

Barnett (1966) has a readable short survey of the field that

dwells more upon theoretical aspects. He postulates a general problem

in which the cost is minimized subject to a system in equilibrium with

its loads, while "behavior constraints on strength, stiffness, and

stability" are satisfied. Uniform strength design introduces the dis-

cussion of optimal trusses; virtual work theorems that were derived

originally for trusses are then applied to simultaneous plastic collapse

problems. Following a brief discussion of the plastic collapse load

bounding theorems, there is a short treatment of elastic stability

problems and material merit indices. Barnett's stiffness design example

exhibits several important features worth noting. Specifically, the

example is to minimize the weight of a beam subject to some given load

where the deflection at a certain point is specified. Virtual work is

used to handle the subsidiary deflection condition. Necessary condi-

tions are obtained from the calculus of variations, but more signif-

icantly, the Schwarz inequality provides a sufficient condition for

global optimality. Barnett concludes his survey with a section stating

that multiload designs satisfying "all three behavior criteria" are more

easily solved in "design space" by mathematical programming techniques

described by Schmidt (1966).

As a sequel to the comprehensive survey (1638 to 1962) by

Wasiutynski and Brandt, Sheu and Prager (1968b) present a complete review










of developments from 1962 to 1968. This paper contains three major

sections: general background, methodology, and specific problems.

In the first section they state that the "well-posed problem of optimal

structural design" requires specification of the

(1) purpose of the structure (load and environment),

(2) geometric design constraints (limits to design
parameters),

(3) behavioral design constraints (limits to the "state" of
the structure),

(4) design objective (cost functional).

The methodology section is not noteworthy, but the third section lists

what is in their opinion the specific problem types: static compliance,

dynamic compliance, buckling load, plastic collapse load, multipurpose/

multiconstraint structures, optimal layout (e.g., trusses), reinforced/

prestressed structures, and from the background section, probabilistic

problems. Their concluding remarks succinctly summarize the paramount

difficulty of the subject: realistic problems are too complicated for

precise analytical treatment. While progress is being made in analyt-

ical treatment of simple structures, the authors opine that realistic

problems require mathematical programming techniques. However, they do

Feel that analytical treatment is desirable to provide "a deeper insight

into the analytical nature of optimality."

A related survey by Wang (1968) on distributed parameter systems,

". .. whose dynamical behaviors are describable by partial differential

equations, integral eqlu-jtiLns, or functional differential equations,"

consists entirely of a bibliography. While not pertinent to the disser-

tation, it is mentioned here for completeness.










Prager (1970) provides another survey which is not comprehensive,

nor does he present any new results as claimed. For example, Prager

and Taylor (1968) used the principle of minimum potential energy

and the assumption that stiffness is proportional to specific mass to

prove global optimality. It could hardly be called a new development

in 1970. At the same time, the author does present an excellent example

of a multipurpose optimal design problem. Prager also treats "segment-

wise constant" approximations and the optimal layout of trusses.

The final survey paper, Troitskii (1971), is an unusual review

of methods in the calculus of variations. Whereas some of the earlier

surveys present lengthy lists of references but contain little method-

ology or theory, this survey is just the opposite. Part of what makes

it unique is that the author believed only eight articles merited cita-

tion--all of them by Troitskii. This shortcoming is more than overcome

by a thorough classification of optimal control problems in the calculus

of variations. Troitskii bases the classification on "certain character-

istics of control problems": types of constraints, properties of the

governing dynamical equations of the system, type of cost functional, and

possible state discontinuities. From these four criteria he postulates

five principal classes of problems; however, it is the criteria that

are important and not the specific problem type.

By comparing what the authors of the aforementioned surveys

believe to be the important types of problems, it is readily apparent

that there is little agreement on which characteristics of structural

optimization problems are significant.










1.1 Historical Development:
Optimal Columns

The beginning of structural optimization is generally attributed

to Galileo's studies in 1638 of the bending strength of beams. Accord-

ing to Barnett (1968), Galileo considered a constant-width cantilever

beam under a tip load as part of a study of "solids of equal resistance."

In requiring the maximum stress in each cross section to be constant

throughout the beam, the height must be a parabolic function of position

along the beam. While this appears to be the origin of the field, a prob-

lem that received more attention is the buckling of a column.

Using the newly developed calculus of variations, in 1773 Lagrange

attempted to apply variational techniques to the problem of finding that

distribution of a homogeneous material along the length of a column which

maximizes the buckling load. Truesdell (1968) relates that through an

insufficient mathematical formulation Lagrange showed the optimal form

to be a circular cylinder. Clausen (1851) provides the earliest known

solution to this problem for the simply supported case. As described in

Todhunter and Pearson (1893, pp. 325-329), Clausen minimized the volume

of the column with the differential equation for buckled deflection

treated as a subsidiary condition. Assuming all cross sections to be

similar, after several variable transformations and complicated manip-

ulation, he obtained an implicit, analytical solution.

The next development was Greenhill (1881), according to Keller

and Niordson (1966). Greenhill determined the height of a uniform

prismatic column, beyond which the column buckled due solely to its own

weight. Timoshenko and Gere (1961) reproduce the solution in which the










deflection is expressed as the integral of a Bessel function of the

first kind (of the negative one-third order).

Blasius (1913) introduces his paper with a uniform strength and

a minimum deflection beam problem. For a given load and amount of mate-

rial, the cross-sectional area distribution is determined which maxi-

mizes the buckling load of a circular column. The solution is identical

to that obtained by Clausen. In addition, Blasius also obtained solu-

tions for columns having rectangular cross sections and discussed the

effect of different boundary conditions on the results.

For the next few decades, structural optimization appears to

have been directed towards applications in the aircraft industry, where

aircraft structural problems and results are presented in the format of

a design handbook. Feigen (1952) is a good example of this, consider-

i:; the buckling of a thin-wall column. Given a constant load and wall

thickness, he required the variable inside diameter to be chosen such

that the buckling load is maximized. Wall thickness is selected to make

local buckling and Euler buckling occur at the same load. Solid tapered

columns having blunt ends are also treated for assumed stiffness dis-

tributions.

Renewed interest was aroused by Keller (1960), who examined the

problem from the point of view of the theory of elasticity, and in

choosing the cross-sectional shape to give the maximum stiffness.

Neglecting the r--.bht of the column, he obtained via the former that

twisting the column does not affect the buckling. Of all convex cross

sections, the equilateral triangle is shown to have the largest second










moment of area relative to a centroidal axis. Hence, from the defin-

ition of buckling load, the "best" cross-sectional shape is the equi-

lateral triangle. Keller also obtained Clausen's implicit, analytical

solution. Subsequently, Tadjbakhshand Keller (1962) generalized the

problem to a general eigenvalue problem and boundary conditions subject

to a subsidiary equality constraint. The latter corresponds to spec-

ifying the volume (or weight) of material to be distributed in an opti-

mal manner. Using the H6lder inequality they demonstrate global opti-

mality of the eigenvalue for the hinged-hinged column.

Keller and Niordson (1966) examine the case of a vertical column

fixed at the base, subject to a vertical load at the tip and the column's

own weight. It is also assumed that all cross sections are similar.

Their approach is to state the problem as a simultaneous, dual optimiza-

tion of the Rayleigh quotient. The eigenvalue is minimized with respect

to the eigenfunction and maximized with respect to cross-sectional area

distribution. Specifying the volume of material available is treated

as a subsidiary equality constraint. From the maximum lowest eigenvalue

the maximum height at buckling is determined. Solutions are obtained by

an iterative technique employed with integral equations.

In a brief note, Taylor (1967), suggests that ener, methods may

link optimum column problems to the traditional eigenvalue problems of

mechanics. Prager and Taylor (1968) classify problems in optimal struc-

tural design and demonstrate global optimality using energy principles.

Unfortunately their assumption of thin-wall construction limits the

results to structures where the stiffness is proportional to the specific










mass density. The consequence of this assumption is that in the energy

formulation the resulting control law and governing equations are

decoupled, and hence easily solved. Huang and Sheu (1968) apply this

same thin-wall assumption to the problem treated by Keller and Niordson.

However, the former seek the maximum end load instead of the maximum

height. The authors also attempt to limit the maximum allowable stress

and obtain solutions by a finite-difference method. Further discussion

of sandwich (thin-wall construction) columns is given by Taylor and Liu

(1968). Basically, this paper is an elaboration of the techniques

described by Prager and Taylor when applied to columns. Extensive

results are provided for various cases.

Post-buckling behavior for columns subject to conservative

loads is considered by Gajewski and Zyczkowski (1970). A nonconcerva-

tive problem is treated by Plaut (1971b). The first of these two papers

is lengthy but is much too narrow in scope to be particularly useful.

In the second paper, the Ritz method is applied to an energy functional,

obtaining the "best" form of the assumed approximation to the optimal

solution.


1.2 Historical Development:
ODtimal Static Beams

That beam problems played a role in the early developments of

structural optimization has already been indicated in the preceding sec-

tion. No attempt is made in what follows to present a complete history,

but merely to outline the type of problems that have been considered.










Opatowski (1944) has an outstanding paper that deals with

cantilever beams of uniform strength. Besides providing numerous refer-

ences to earlier studies, the author treats the problem with impres-

sive mathematical rigor. The beam is considered to deflect under its

own weight and a transverse tip load; bending deflection is described

by a Volterra integral equation which is solved exactly for various

assumed classes of variable cross-sectional geometry. This paper is

representative of earlier papers in that it contains extensive analysis

and analytical results, but little numerical data. Barnett's work (1961)

and its sequal (1963a) apply the calculus of variations to more real-

istic I-beams. One problem considered is maximizing the weight sub-

ject to general, unspecified loads for a specified deflection at a given

point. The Schwarz inequality is used to derive a sufficient condition

for global optimality. Also included is a comparison of uniform strength

beams to the minimum deflection beam for several different cases of

applied load and geometry. The paper is concluded with various minimum

weight design examples in which both bending and shear stiffness are con-

sidered.

Haug and Kirmser's (1967) work is one of the most comprehensive

studies of minimumweightbeam problem published. Wlnilc it may succeed in

handling any conceivable problem and in employing the most realistic

stress constraints, this very generality requires so many variables and

conditions that the mathematics is complicated almost beyond reason.

Another study of minimum weight beams (Huang and Tang, 1969) is

important for several reasons. By dividing the beam into many segments










having constant properties, and determining the necessary conditions

that must be satisfied by every segment, it appears that the authors

are using the same methods that were used to solve the Brachistochrone

problem in the seventeenth century. In their treatment, multiple

loads must produce a specified deflection at a given point; these

"segmentwise constant" approximations and multiple load problems have

recently received more attention. Of further interest in this paper is

the derivation of the multiple load optimality condition using

Pontryagin's Maximum Principle.

While no new results or techniques are contained in Citron's

(1969, pp. 154-166), the author gives a very readable minimum weight

beam example. The problem is simple, described in detail completely,

and provides an excellent example of how control theory is applied to

an optimal problem in structural design. An analysis of intermediate

beams which may form plastic hinges is provided by Gjelsvik (1971).

It shows that if hinges are placed at all points of the beam where the

bending moment is zero, this makes the plastic or elastic minimum

weight beam statically determinate. Both the elastic and plastic beam

designs are shown to be fully stressed, i.e., are uniform strength beams.

Application of generalized vector space techniques is character-

istic of more recent papers. Bhargava and Duffin's (1973) is such a study.

It treats the maximum strength of a cantilever beam on an elastic



Since these early papers are not readily available, this statement
is based upon descriptions of them given by historical references
cited in Appendix A.









foundation subject to an upper bound on weight. Although involving

more advanced mathematics than normally required by variational tech-

niques, vector space methods may also provide more powerful analyt-

ical tools.


1.3 Historical Development:
Optimal Dynamical Beams

The title of this section is a misnomer. On the basis of the

related literature, a more realistic terminology would be "Optimal

Quasistatic Beams." Papers dealing with the optimization of dynamical

beams ultimately contain some assumption or given condition that effec-

tively transforms the problem to an equivalent static case. Simple

harmonic motion is frequently assumed to remove time dependence from the

governing equation. For example, Barnett (1963b)minimizes the tip of

deflection of a cantilever beam accelerating uniformly upwards, with the

total '.Aicht of the beam specified. The optimal solution for cross-

sectional area distribution is specified by a nonlinear integral equa-

tion solved by successive approximations. Dynamics enters the problem

only as a time invariant inertia load which converts the problem to a

static beam subject to a body force.

Niordson (1964) finds the tapering of a simply supported beam

of given volume and length which maximizes the fundamental frequency of

free vibration. Assuming that all cross sections are similar, Niordson

expresses the desired frequency as the Rayleigh quotient. This is

obtained by the equation for spatial dependence associated with the

usual separation of variables technique. In this case it is assumed









that deflection, shear, and rotational inertia effects are small.

Solution of the conditions for optimality were obtained by successive

approximations. This approach results in a problem identical in form

to the eigenvalue problem associated with maximizing the Euler buckling

load of a column.

A very specialized problem is treated in Brach (1968). Cross-

sectional area and stiffness are proportional, and have both upper and

lower bounds; material properties and length of the beam are specified

constants. The object is to make the fundamental frequency of free

vibration stationary, without any constraints on the weight. Instead

of using the Rayleigh quotient, Brach uses the total potential energy.

His solution method is ad hoc and not generally useful. A more useful

approach is described by Icerman (1969) for structures subject to per-

iodic loads. Necessary and sufficient conditions for minimum weight

are obtained from the principle of minimum potential energy. Amplitude

and frequency of the applied load is specified as well as "dynamic

response," defined as the potential energy associated with the load

amplitude displaced a distance equal to the displacement amplitude at

the point of application. It is also required that the load's frequency

be less than the fundamental natural frequency. Subject to these con-

straints, the structure's weight is to be minimized. Trusses and

segmentwise constant approximations are also treated. Once again a

dynamical problem is effectively transformed into a quasistatic problem.

Realistic treatment of the dynamic response of beams subject to

time dependent loads is given in Plaut (1970). An upper bound on the









response of the beam is obtained from the "largest possible displace-

ment of the beam under a static concentrated unit load." This inequal-

ity is determined from the time derivative of the total energy of the

beam and the Schwarz inequality. Plaut minimizes the upper bound on

the response for specified total weight and relationship between specific

mass and stiffness. Despite the consideration of a truly dynamical

problem, this approach has two weaknesses indicated by the author.

First, the upper bound need not be close to the exact answer; secondly,

there is no demonstration that minimizing the upper bound also minimizes

the actual response of the beam subject to dynamic loading.

Another paper worthy of mention is Brach and Walters (1970).

They maximize the fundamental natural frequency which is expressed as

a Rayleigh quotient that includes the effect of shear. Standard varia-

tional methods are employed to derive necessary conditions, but no

solutions nor examples are given. The authors do, however, suicest

using the method of quasilinearization. This paper is another example

of a quasistatic, dynamical beam problem.


1.4 Scope of the Dissertation

This dissertation is primarily concerned with the application

of Pontryagip,'s Maximum Principle to problems in structural optimization.

Only elastic materials are considered; however, various types of con-

straints are treated.

The theoretical development of the dissertation pertains to

problems described by an ordinary differential equation but is based

upon a numerical technique normally used with systems described by










a finite number of discrete quantities. For this reason, examples are

included for both types of systems.

No new solution techniques have been developed for the nonlinear

two point boundary value problems which characteristically arise in

optimization problems for continuous systems. Solutions are obtained

by a standard quasilinearization method. However, a modified "feasible

direction" numerical algorithm for use with discrete systems is

described and an example included to demonstrate its operation. This

serves to illustrate the application of the theory on which the algo-

rithm is based to the theoretical development associated with contin-

uous systems. Furthermore, it provides a comparison between the

solution to a problem described as a continuous system, and alternately

by a discrete approximation.

Additionally, it is the intent of this dissertation to replace

some of the confusion in classification of problem types contained in

the various survey papers with an organization based upon mathematical

attributes. The result is a logical approach to the formulation of

optimization problems for elastic structures.













CHAPTER II

GENERAL PROBLEMS AND METHODS
IN STRUCTURAL OPTIMIZATION



2.0 Introduction

The first chapter presents a broad view of structural optimiza-

tion and the historical development of two general types of problems

that are used as examples in later chapters. Before the mathematical

theory is developed in the next chapters, general problems and methods

in structural optimization itself are briefly outlined in this chapter.

The classification of problem types vis-a-vis mathematical attributes

is discussed first. This is followed by short descriptions of the

major methods of structural optimization for both continuous and dis-

crete systems. Appropriate references are cited for each section.


2.1 Problem Classification Criteria

Perhaps the major source of difficulty in classifying struc-

tural optimization problems lies in the translation from the physics

involved to a mathematical representation. A single physical concept

when transformed to mathematics may become more than one mathematical

attribute. For example, consider the class of conservative problems.

Since energy is conserved this immediately prohibits dissipative mate-

rials, time varient constraints, and nonholonomic constants. More

important, consider the following statements from Lanczos (1962,

p. 226):










all the equations of mathematical physics which
do not involve any energy losses are deducible from
a "principle of least action," that is the principle
of making a certain scalar quantity a minimum or a
maximum .all the differential equations which are
self-adjoint, are deducible from a minimum-maximum
principle and vice versa.

However, it is shown in Chapter V that to be self-adjoint systems places

requirements on both the differential equation and the boundary condi-

tion. Thus, the single physics attribute of being a conservative sys-

tem is described mathematically by expressions involving the cost func-

tional, the governing system of differential equations, boundary condi-

tions, and constraints.

As a result of this lack of similarity in descriptions, a choice

must be made as to which realm will be used for classification of prob-

lems. Since all problems are ultimately transformed to mathematics,

mathematical characteristics are selected as the criteria. On the basis

of survey paper contents and the many related papers, it is felt that

the proper (not necessarily the best, nor all inclusive) characteristics

for classification of problem types are:

(i) cost functional

(ii) system equation and boundary conditions

(iii) control constraints

(iv) behavioral constraints (state and/or control constraints)

Subsequent discussion is in terms of these four characteristics.

Exceptions to these descriptors are readily acknowledged, e.g., whether

the problem is deterministic or probabilistic. An example of the latter

may be seen in Moses and Kinser (1967). These exceptions do not serve










as negating counterexamples but instead indicate the requirement for

additional descriptors and verify the difficulty of the task, suggest-

ing the need for further comprehensive study.


2.1.0 Problem Classification Guidelines

In the following sections each criterion is briefly discussed.

Some of the various characteristics of each are mentioned, and where

appropriate references exist the citation is given. It must again be

emphasized that the following is not all inclusive; it is an attempt to

categorize the types of problems existing in the literature according

to the four mathematical descriptors postulated. Moreover, the descrip-

tors are not discussed in the order given but instead are treated in

the order normally encountered during a problem solution.


2.1.1 Governing Equations of the System

The immediate question to be answered is whether the structural

system is described by a set of continuous functions or a set of dis-

crete constants. Bending deflection of simple structural elements is

an example of the former; design of trusses is a good example of the

latter. In general, variational techniques are employed with contin-

uous systems while mathematical progr-irming techniques are most fre-

quently applied to the discrete systems. However, variational methods

can be applied to the approximation of continuous systems by discrete

elements. This usually takes the form of either a finite element or

"segmentwise constant" approximation of the continuous structure.

References treating the different systems described above are included

in the section on methods.










2.1.2 Constraints

Two of the descriptors are postulated to be control constraints

and behavioral constraints. A further consideration is whether the

constraint is defined by an equality relationship or by an inequality

expression. Equality constraints are handled by a long-known technique

entitled "isoperimetric constraints." Valentine's (1937) work is known

to contain the initial development of a technique for converting inequal-

ity constraints to equality constraints. The introduction of slack

variables increases the number of variables in the problem to be solved

but at the same time permits all of the isoperimetric techniques to be

used. A detailed application of this approach is presented in Appendix B.

It should also be noted that isoperimetric constraints are sometimes

referred to as "accessory or subsidiary conditions."

Most real structural optimization problems possess an isoperi-

metric constraint as well as inequality constraints dependent upon the

control u and/or the state x of the structural system. Typically the

control constraints are the result of geometrical limitations or

restrictions to the types of available materials. Behavioral constraints

are related to the state of the system and may depend solely upon the

state (of deformation), or in the case of most stress constraints,

jointly upon the state and control variables. With this distinction the

constraints may be classified vis-a-vis the two criteria and optimal

control characteristics as follows:



By convention all vectors are column vectors unless indicated otherwise.








(i) unconstrained

(ii) 4(u) < 0 control constraint

(iii) O(x,u) < 0
Behavioral constraints
(iv) 0(x) 0)

All of this discussion pertains to both continuous and discrete systems.

No references for unconstrained optimization problems are given.

It may be that in some problems the unconstrained structural optimiza-

tion solutions have either infinite stiffness and finite weight, or

finite stiffness and zero weight. A discussion of this can be found in

Salinas (1968, pp. 23-26).

Investigation of control constraints led to the development of

the maximum principle. Although reserving discussion of the method for

a later section, the classical reference detailing the derivation of

the principle is given here for completeness. Rozonoer (1959) treats con-

trol constraints but only as related to the development of the maximum

principle.

Most of the literature concerns either bounded control problems

or a more general form of constraint which can be classified as a behav-

ioral constraint. The latter is a mixed constraint which depends upon

both the state and control variables. Breakwell (1959) is a lucid paper

dealing with this type of constraint. References which treat state

and/or mixed constraints are Bryson et al. (1963), and Speyer and

Bryson (1968). Constraints which depend upon only the state are not

treated in this dissertation; an example of such a constraint is to

determine the optimal solution for some problem subject to an upper

bound on deflection of the structure at any point.









2.1.3 Cost Functionals

There are two basic types of cost functionals that occur in the

field of structural optimization. They were first identified by Prager

(1969) but not for the proper reasons. Using Prager's notation, they are


J = Min f F(p) dV
V
f G(W) dV
JQ = Min
f H(J ) dV
V

where F, G, and H are scalar functionals of -. The latter functional

represents a Rayleigh quotient associated with an eigenvalue problem.

It can be reduced to the first type of functional shown above by choos-

ing a normalization of the eigenfunction such that the numerator

equals unity for all admissible variations (see section 5.3). This

normalization is thereafter treated as a subsidiary constraint.

What actually distinguishes the second functional from the

first is not that the functional is a quotient, but that the extrem-

ization of an eigenvalue requires a dual extremization (see section 5.3).

In terms of a state x and control u, the fundamental eigenvalue is

given by minimization of the Rayleigh quotient with respect to the

eigenfunction x, or

f G(x,u) dV
J = Min
x f H(x,u) dV
V

where u represents some specified design parameter. If the desired

result is to maximize the cost JQ with respect to all admissible u, it

is observed that a second extremization is required; for example, see










Keller and Niordson (1966). Thus, a more appropriate manner of classi-

fying cost functionals is on the basis of whether the problem statement

implies a single or a double extremization. Hence the two basic types

of cost functionals encountered in structural optimization are


J = Min f F(x,u) dV
u V

f G(x,u) dV
V
J = Max Minr
u x f H(x,u) dV
V

where x must satisfy an equilibrium condition of the state, and u is

subject to some admissibility requirements.

There is a special case related to these two in which the weight

is to be minimized for a specified eigenvalue. This problem is

treated in Icerman (1969) with a mathematical discussion of such a

variational problem presented in Irving and Mullineux (1959, p. 394).

In terms of the two cost functionals, the special case is


J = Min f {G(x,u) J H(x,u)} dV
uV V

where J is a specified constant. This approach is frequently employed

in eigenvalue problems to avoid the inherent difficulties associated

with the dual extremization problem.

There have also been many papers published that consider

"multi-purpose structures," e.g., Prager (1969), Martin (1970), and

Prager and Shield (1968). The cost function for such problems is

defined as









k
J = Z a.J.(x,u)
i=l


where a. are positive constants, serving as weighting parameters.

While perhaps demonstrating much potential, no significant results

obtained with this approach have so far been published. What problems

have been solved are too simple; indeed the authors indicate the need

for using a discrete approximation and mathematical programming tech-

niques in realistic applications.

A subject closely related to "multi-purpose structures" is

that of multiple constraints. It is mentioned here only because most

papers on the latter also include the former--see Martin (1971). The

idea of multiple constraints is not new; in both variational and math-

ematical programming fields there exist standard techniques for han-

dling multiple constraints.

A recent Russian paper (Salukvadze, 1971) suggests an alter-

nate to the "multi-purpose" cost functional. Instead of treating a

vector functional that requires the choosing of weighting coefficients,

it is suggested that the several functionals be combined into one.

Given a system and vector cost functionals

x = f(x,u,t)

J.[u] = J.(x,u,t) i = l,...,k


(i)
Let uOPT denote the optimal solution for which J. assumes the optimal
-OPT 1
value on the trajectory of the system. For each of the J. there is a
(i)
different u PT. These k values J. can be thought of as components of
-OPT 1








t
a vector r where

= (1)1 [ (kl )T
"I PTJ k"'"

For any arbitrary u the result

T
r[u] = {J [u] ... J [u]}

is just some vector functional.
*
Vector r represents a constant point in the space of

(J ,...'Jk). Since no choice of u can optimize all of the J. simul-

taneously, that is, to attain the point r in J.-space, the best alter-
*
native is to minimize the distance between r[u] and r That distance

is defined by the Euclidean norm. To avoid the question of incon-

sistent dimensions the functionals are reduced to dimensionless form.

Thus,
-_\

k k J[u] J.M
J[u] = i -- ___
(i)
i=1 J.
i PT

and PyT is that function u which minimizes the functional J[uj.

Mathematically speaking,

uOT = ARGMIN {J[u]}

This type of vector cost functional is much more appealing

than the type treated in the papers on multi-purpose structures.

It also suggests an entirely new field of study: the more realistic


t T
Superscript "T", e.g., u denotes the transpose of the vector u.









choice of cost functionals. The mathematics of a problem seldom

accommodates financial considerations. For example, the design which

requires the least material may reduce the cost of materials at an

overwhelming expense in manufacturing or fabrication. When aesthetic

appeal and environmental impact are included--as must be done in any

real, commercial application--the selection of an appropriate cost

functional is an almost insurmountable task. However, a simple exten-

sion of Salukvadze's composite cost functional may reduce the difficul-

ties to operations research considerations.

In problems where it is desired to optimize simultaneously

several different functionals, not all having the same dimensions, the

concept of a generalized inner product may prove useful. It is defined

in terms of a metric operator A; for some general vector z


z ||2= (z,z)A =TAz


and symbol means "is defined by." With reference to the vector

cost functional, A represents a set of scale factors which converts

all of the separate cost functionals to a common dimension. This is

where the operations research enters--relating material expense to

fabrication to sociological considerations and so forth--to determine

the metric A. For a vector c whose elements are functionals,

E = r[u] r

where r[u] and r are defined above, the composite cost functional is


J[u] = (E,Ac)









The weakness in this method is that an optimal solution must be

obtained for each individual cost functional prior to attempting a

solution to the composite problem. Additionally, some of the more

abstract cost objectives may be difficult to quantify in a meaningful

manner. Despite these shortcomings this approach does suggest inter-

esting applications.


2.2 Methods: Continuous Systems

The problems characterized by mathematical functions, in con-

trast to those represented by a set of discrete constants, are normally

treated by variational techniques. Many books on this subject have

been published; the better authors include Elsgolc (1961), Gelfand and

Fomin (1963), Dreyfus (1965), Hestenes (1966), Denn (1969), Luenberger

(1969), and Bryson and Ho (1969). An excellent summary paper is avail-

able in Berg (1962).

To see how these techniques are applied, three papers are

recommended. The first is Blasius (1913), which provides sufficient

detail and explanation to make it quite worthwhile. Although it does

include several examples that involve subsidiary conditions, no inequal-

ity constraints are treated. An example that includes inequality con-

straints to the control variable is contained in Brach (1968). A much

more general application of variational principles is presented in

Oden and Reddy (1974). In this paper a dual-complementary variational

principle is developed for a particular class of problems. It is shown

that the canonical equations obtained are the Euler-1.a-range equations

for a certain functional.









2.2.0 Special Variational Methods

Besides the ordinary variational method, more specialized

techniques have been developed to the point where they are recognized

as independent methods in their own right. In the following sections

these methods are identified and a number of representative references

given.


2.2.1 Energy Methods

The oldest of these methods is the energy method. It originated

with the principle of minimum potential energy, and was later extended

to include the representation of eigenvalues through the energy func-

tional. A good discussion of the former is available in Fung (1965) or

Przemieniecki (1968); the best general treatment of the latter is avail-

able in either Gould (1957, Chapter 4) or in Mikhlin and Smolitskiy

(1967, Chapter 3).

The principle of minimum potential energy is frequently used

with simple problems to prove that a necessary condition for optimality

is also sufficient. Prager and Taylor (1968) contains such a proof

for the global, maximum stiffness design of an elastic structure of

sandwich construction; two papers that also consider this problem are

Huang (1968) and Taylor (1969). Specific application of the energy

method to an eigenvalue problem is demonstrated in Taylor and Liu (1968).

A much more general discussion of the energy method is provided in

Salinas (1968). Further extensions of the method are presented in

Masur (1970), in which the principle of minimum complementary energy

is applied to problems of the optimum stiffness and strength of elastic










structures. In these problems a necessary condition for optimality

is that the strain energy density be constant throughout the structure.

This condition is also sufficient for optimality in certain classes of

structures that satisfy a specific relationship between the strain

energy density and design variables.

Many of the energy problems belong to the class of problems

having quadratic cost functionals. The significance of this character-

istic is that the Euler-Lagrange equations derived from such functionals

are linear.

A more recent energy method development is the concept of

"mutual potential energy." Mutual potential energy techniques resemble

those of the principle of minimum potential energy. In both methods

a cost functional is defined over the entire domain occupied by the

structure and optimized with respect to the control variable. If it

is desired that the optimal solution be required to have a specified

deflection at a certain point, this condition corresponds to a sub-

sidiary state constraint when using the principle of minimum potential

energy. The mutual potential energy method incorporates this type of

localized constraint into the cost functional which is defined over

the entire domain of the structure. By itself this alone is advanta-

geous; however, for certain types of problems the mutual potential

energy method also provides both a necessary and sufficient condition

for global optimality. In the way of a critical comment, either the

method has received too little attention, or else it does not effi-

ciently handle problems more difficult than the simple examples presented.









Four papers that are representative of the literature associated with

this method are Shield and Prager (1970), Chern (1971a, 1971b), and

Plaut (1971c).

Another recent development is that the class of problems for

which energy methods is applicable has been expanded to include cer-

tain types of nonconservative systems. Together with the mutual poten-

tial energy concepts, this suggests that perhaps the classical energy

method is a special case of a more generalized method. If a technique

can be developed which uses the adjoint variables to transform a gen-

eral nonconservative system into an equivalent self-adjoint form, the

method might be deduced. Some papers pertaining: to the subject are

Prasad and Herrmann (1969), Wu (1973), and Barston (1974).


2.2.2 Pontryagin's Maximum Principle

There are many textbooks which derive, explain, and give examples

for the maximum principle. The original (Pontryagin et al., 1962)

requires a knowledge of functional analysis. A condensed form of this

same material is available in Rozonoer (1959). Denn (1969) provides

another point of view in which the principle is derived from Green's

functions. In this manner, the sensitivity to variations is readily

observed. To understand Denn's treatment requires only a knowledge of

the solution of differential equations.

Shortly after Pontryagin's book was published, many papers

devoted to the theoretical aspects of the maximum principle were pub-

lished. Some of the more readable ones are Kopp (1962, 1963), Roxin

(1963), and Halkin (1963). Another early paper (Breakwell, 1959)









appears to be a completely independent derivation of the maximum prin-

ciple. Although quite general in the mathematical sense, the examples

presented are trajectory optimization problems and not a general type

of mathematical problem. This may be an explanation for what seems to

be a lack of recognition for a significant achievement.

The application of PMP to problems in structural optimization

is relatively recent. When the method is used, one of two difficulties

is often encountered. The first is errors in the formulation of the

optimal control problem; the second is that once a well-posed, non-

linear two-point boundary value problem (TPBVP) is obtained, it is dif-

ficult to solve. An example of the first is provided by Dixon (1968)--

the correction was given in Boykin and Sierakowski (1972). De Silva

(1972) provides a clear presentation on the application of PTIP to a

specific problem, but includes no data because a solution could not be

obtained. Despite the failure to determine the solution, this paper

is worthwhile for its lucid discussion of the PMP application. Another

paper that gives a good specific application of PMP is Maday (1973).

Although much analysis is presented very little is said regarding the

solution techniques.

All of the above references are applicable only to systems

that are described by ordinary differential equations, in contrast with

the calculus of variations which also handles problems described by

partial differential equations. Since many of the problems of mathe-

matical physics involve partial differential equations, an extension of

PMP to include this class of problems is the next logical development.










Some work has already been done, for example, Barnes (1971) and

Komkov (1972). A survey of these "distributed parameter systems"--

see Section 1.0--is presented in Wang (1968).


2.2.3 Method of Steepest Ascent/Descent

This method is frequently cited in the literature for trajec-

tory optimization, and occasionally in references related to optimal

structures. When the method of quasilinearization converged for the

dissertation example problems, there was no need to investigate other

methods such as the method of steepest ascent. Consequently, little is

said about it. According to the references, it is applied in a straight-

forward manner. Furthermore, the example problem solutions presented

seem to be real problems and not academically simple. The following

four papers treat the method in general with trajectory optimization

applications: Bryson and Denham (1962, 1964), Bryson et al. (1963), and

Hanson (1968). In Haug et al. (1969), the method of steepest ascent is

derived in detail, completely discussed, and compared to the maximum

principle. Several structural optimization problems are then solved

by the method of steepest ascent. Although no exciting results are

obtained, use of the method is clearly illustrated by the applications

to realistic structural problems.


2.2.4 Transition Matrix:
Aeroelasticity Problems

For the past few years a group at Stanford University has studied

the optimization of structures subject to dynamic or aerodynamic con-

straints. The general problem of their interest is that of minimizing










the weight of a given structure for specified eigenvalue, subject to

inequality constraints on control.

Three types of solution techniques are used once the necessary

conditions for minimum weight are determined. Exact solutions are

obtained for most of the problems because they are so simple that

analytical methods are applicable. More complicated problems are

solved by a "transition matrix" method described in Bryson and Ho (1969).

On the basis of convergence difficulties reported in the references,

this method should be used with caution. Results have been obtained

only for very simple problems. However, these results are corroborated

by data obtained from a discrete approximation method. Five papers

that are representative of this work are McIntosh and Eastep (1968),

Ashley and McIntosh (1968), McIntosh et al. (1969), Ashley et al. (1970),

and Weisshaar (1970).


2.2.5 Miscellaneous Methods

The preceding sections have briefly outlined the methods of

structural optimization most frequently encountered in the literature.

Appropriate, representative references have also been given. Not all

methods are listed; while some are omitted for not being generally use-

ful, others are omitted for not being generally used. Two examples of

the latter are the "modified quasilinearization" and "sequential gradient-

restoration" algorithms described in Miele et al. (1972) and in Hennig

and Miele (1972). At some later time these methods may be acknowledged

as major methods that are applicable to many different or important prob-

lems, but for now they are mentioned only in passing.









2.3 Methods: Discrete Systems

Discrete systems are described by a set of discrete constants

instead of the set of functions associated with continuous systems.

The classic example of a discrete system is a pin-connected truss, where

the state x. and control u. are the stress and cross-sectional area,
1 1
respectively, for each member i. A discrete system also arises in the

approximation of a continuous system.

Several references that present a good discussion of general

methods applied to discrete systems are available. Most of these exist

in the form of an edited collection of papers by various authors on the

topics of their acknowledged expertise. Four such publications are

Gellatly (1970), Gellatly and Berke (1971), Pope and Schmidt (1971),

and Gallagher and Zienkiewicz (1973). Another report, Melosh and Luik

(1967), provides a good exposition of the difficulties associated with

the analysis portion of least weight structural design. It also con-

tains a brief comparison of various mathematical programming methods.

McNeill (1971) is the last reference to be cited in the section

on general methods for the optimization of discrete systems. Minimum

weight design of general structures is treated in a mathematically

precise formulation. Legendre's necessary condition is combined with

the concepts of convex functions and sets to derive the necessary and

sufficient conditions for global optimality. Fully stressed designs

and constraints to eigenvalues are also discussed. In summary, this

paper provides a good example of the general mathematical problem that

must be solved in the optimization of discrete systems.









While certain variational methods may be applied to discrete

systems, the most frequently used technique is mathematical programming.

In the following sections, this method and other major methods are dis-

cussed and representative references cited.


2.3.0 Mathematical Programming

The general method of mathematical programming is discussed

in section 3.2 of the dissertation, and the solution of an example

problem using this method is detailed in Chapter VL. In the literature

related to this subject, a very readable textbook is available--Fox

(1971). This book complements the theory with numerous discussions

pertaining to numerical techniques and methods that can be employed to

overcome certain difficulties that may arise. Although it does contain

flowcharts of several algorithms, there are few specific examples given.

For a discussion of the general theory, two alternatives to this book

exist in the form of papers: Schmidt (1966, 1968). The first is

written in a conversational style, contains no mathematics, and is

intended to provide only a general description of the subject. The

latter paper is theoretical in content.

An excellent application to a realistic problem is to be found

in Stroud et al. (1971). Ths paper contains little discussion of the

method itself, but does demonstrate an application that allows a concep-

tual visualization of the solution. The approach is to assume the

solution to be a linear combination of specified functions, and to

choose the weighting coefficients to minimize the cost. Mathematical

programming is employed to determine the optimal set of coefficients.










This approach resembles Galerkin's method, and though not mathemat-

ically rigorous, it may provide a useful approximation to large,

unwieldy problems.


2.3.1 Discrete Solution Approximations

In the previous section a paper is cited that contains an

approximate solution obtained by Galerkin's method. The use of the

Galerkin or Rayleigh-Ritz approximate solution techniques is suffi-

ciently widespread to be considered a general method. For both methods,

the solution is assumed to be a linear combination of the solution to

the linear part of the governing equation and a set of prescribed func-

tions. This approximate solution does not satisfy the given equation

exactly but produces some residual function. A cost function that

depends upon the residual is then minimized with respect to the unknown

coefficients. The two weighted residual methods mentioned above have

different cost functions, but the methods are identical for linear

equations--see Cunningham (1958, p. 158).

The advantage to using these methods is that after assuming

the particular form of solution, the problem of solving for the weight-

ing coefficients may be much simpler than the original problem. In the

case of Stroud et al. (1971), the coefficients were obtained by mathe-

matical programming techniques. However, the weakness of the method

is the restricted function space of possible solutions. With the coef-

ficents obtained by these methods the resulting solution is the best

approximation that is possible from the set of solution functions










prescribed. There is no guarantee that the approximation even resembles

the true solution.

The flutter of a panel is solved using Galerkin's method in

Plaut (1971a). No general developments are presented and the assumed

solution functions are trivially simple. However, this paper does

provide an application of the method to obtain an approximate solution

to a very difficult optimization problem involving the stability of

a nonconservative system. A similar problem is treated in a more theo-

retical manner in Plaut (1971b) using a modified Rayleigh-Ritz method.

"Segmentwise-constant" control functions are assumed also; this partic-

ular approximation is discussed with more detail in the following sec-

tion. Additional nonconservative problems are treated in Leipholz

(1972), applying Galerkin's approximate solution to the energy method.

Several simple examples are included.

Nonconservative elastic stability problems of elastic continue

are treated in Prasad and Herrmann (1969) using adjoint systems. This

approach is more realistic than the segmentwise-constant control assump-

tion described in the following section. Solutions for the state and

adjoint system are assumed, such that approximation process resembles

the Rayleigh-Ritz method. However, only a single type of nonconserva-

tive system is considered. Extension to several other types of noncon-

servative elastic continue problems is given in Dubey (1970). Varia-

tional equations corresponding to both the Galerkin and Rayleigh-Ritz

methods are derived. Furthermore, the condition for equivalence of the

two methods is shown to be that the admissible velocity field must










satisfy a natural boundary condition over that portion of the body's

surface where tractions are prescribed.


2.3.2 Segmentwise-Constant Approximations

The definitive characteristic of this method is approximating

the structural system by a number of discrete segments, where within

each segment the control function has a constant value. In general,

the constant value of the control differs from segment to segment. For

the many papers on this method that have been published, the procedure

is the same. An optimality condition (necessary in all cases but also

sufficient in some) or cost functional is derived for the continuous

system. After defining the segmentwise-constant approximation, the

condition or functional is reformulated in terms of the discrete param-

eters. Most of the papers use so few elements that solving for the

discrete values of the control parameter poses no difficulties. Although

this method does simplify the mathematical problem to be solved, the

crudeness of the approximation is not appealing. Five papers which

treat a variety of problems using this approximation are cited below.

Minimum weight of sandwich structures subject to static loads

is discussed in Sheu and Prager (1968a). In Sheu (1968) the same type

of structure is considered. It differs from the first problem by

requiring point masses to be supported such that the total structure has

a prescribed fundamental frequency of free vibration. Icerman (1969)

treats the problem of elastic structures subject to a concentrated load

of harmonically varying amplitude. The minimum weight design is obtained










subject to a compliance constraint related to the applied load, and

which is effectively a boundary condition on displacement at the point

of application. A truss problem is also included.

The concept of a compliance constraint is pursued further in

Chern and Prager (1970). The minimum weight design for sandwich con-

struction beams under alternative loads is found, subject to this type

of constraint. The paper uses up to eight segments, thereby obtaining

a more realistic approximation to the continuous problem. Minimum weight

design of elastic structures subject to body forces and a prescribed

deflection is discussed in Chern (1971a). This investigation is no-

table in that it considers applied loads that are functions of the design

functions.


2.3.3 Complex Structures with
Frequency Constraints

On the basis of useful application, perhaps the most important

class of discrete structural optimization problems is the minimum weight

design of complex structures subject to natural frequency constraints.

Since most real structures are built with many structural elements of

various types, and are not realistically described by any single type,

this approach is more appropriate from the aspect of modeling the struc-

ture. Furthermore, many structures must be designed to avoid certain

natural frequencies because of resonance or self-induced oscillations;

this situation indicates that the natural frequency constraint is also

appropriate.

Many different solution schemes have been developed which are

usually based upon general mathematical programming techniques.









Typically, a design is iteratively altered to minimize the weight with

a subsequent increase in frequency until a constraint is violated.

At that point the design process uses an iteration which simultaneously

reduces both weight and frequency. These two processes are repeated

sequentially until no further weight reduction is possible.

Although circumstances may require the use of many elements,

the number of them may itself be a critical factor. Some of the schemes

require a matrix inversion as part of the eigenvalue problem solution

associated with the frequency constraint. If the number of elements

becomes too large, the size of the matrix to be inverted likewise be-

comes excessively large. When that occurs the matrix inversion can

require excessive amounts of computer time. Another possible difficulty

is that the inverse matrix itself is not sufficiently accurate, such

that the subsequent calculations are not acceptable. However, for

structures such as reinforced shells composed of different types of

structural elements, this method may be the most applicable.

Many papers have been published pertaining to this class of

structural optimization problem. Because the method is inherently

oriented towards applications, the references are cited in chronolog-

ical order without additional comments. Interested readers are referred

to: Turner (1967), Zarghamee (1968), Turner (1969), De Silva (1969),

Rubin (1970), Fox and Kapoor (1970), McCart et al. (1970), and Rudisill

and Bhatia (1971).









2.3.4 Finite Element Approximations

There is an unfortunate ambiguity to the label "finite elements"

that occurs because these words are used to describe two completely

different entities. In papers cited in the preceding section they are

used to indicate the discrete structural elements of finite dimensions

which comprise the complex structure. The analysis of such systems of

structural elements has been accomplished by ordinary matrix methods

during the last three decades. However, during the past decade another

method has been developed and named "the finite element method."

In this method a continuum is divided into small, finite ele-

ments over which a particular form of approximation of either the dis-

placement and/or force is assumed. A number of nodes common to one or

more element is prescribed; continuity is required to exist at these

nodes but not necessarily elsewhere. An equilibrium equation is derived

for each element, and then all of the individual equations are combined

into a single equilibrium equation for the entire system. lihc result-

ing equation is a linear algebraic equation whose unknowns are displace-

ments and/or forces at the nodes. Once the matrix equation is inverted,

the nodal displacements and/or forces are used with the assumed approx-

imation form to describe the state of the structure throughout each and

every element, and hence the system. Hereafter this method is referred

to as the "finite element method."

The most frequent application of the finite element method is

to problems having complicated loads, geometry, and response. Generally

speaking, the method is employed wherever the physical system is too










complex to be described adequately by a single differential equation

and boundary conditions. For a complete theoretical development of the

finite element method and numerous examples, see Zienkiewicz (1971).

With respect to structural optimization the method is employed

to simplify the problem to be solved. Very little has been published

on this subject, but the papers available cover a wide spectrum of tech-

niques. For example, Dupuis (1971) combines the finite element and

segmentwise-constant methods as applied to minimum weight beam design.

A similar application to column buckling is contained in Simitses et al.

(1973). Another paper, Wu (1973), is a study of two classical noncon-

servative stability problems. Although adapted to stability consider-

ations, this presentation is the best exposition available in the open

literature.

In Chapter VI a minimum deflection beam problem is solved with

the combined methods of finite elements and mathematical programming.


2.4 Closure

In the preceding sections of this chapter, general problem

types and methods are discussed. Only those methods that appear to

have attained some standard of acceptance are presented. It must be

acknowledged that other areas of important study exist but are perhaps

overlooked as not being pertinent to the general subject area of the

dissertation. As an example, Dorn et al. (1964) treats the optimal

layout of trusses--an important subject but not related to the general

problem to be considered in this dissertation. In addition only

elastic structures have been considered although there are numerous









publications on optimal design of inelastic structures. References

that are representative of this subject are: Drucker and Shield

(1957a, 1957b), Hu and Shield (1961), Shield (1963), Prager and Shield

(1967), and Mayeda and Prager (1967).

On considering the various references mentioned above it would

appear that there are two possible pitfalls in structural optimization

that should be avoided. The first is the confusing of method of opti-

mization with the solution techniques employed to obtain a solution to

the resulting TPBVP. In order to avoid possible errors the two should

be dealt with independently, unless it is clearly advantageous to

relate one to the other. Besides this it must be recognized that any

solution obtained is "optimal" only with respect to the given condi-

tions of the particular problem. Any change in the problem statement

invalidates the applicability of that solution. The change may lead to

a more desirable solution, but the original solution is no less valid.

Simitses (1973) is an example where this situation is not acknowledged.

In this paper the thickness of a thin reinforced circular plate of spec-

ified weight and diameter is determined such that the average deflec-

tion due to a uniform load is minimized. An earlier paper which did

not include stiffening is cited with the implication that the optimum

solution for the unstiffened plate is not correct. The point made above

is that both of these solutions are optimum under the respective condi-

tions of the two problems. Neither solution is more, or less, valid

than the other.













CHAPTER III


THEORETICAL DEVELOPMENT



3.0 Introduction

This chapter contains the development of two distinct methods

used in the theory of optimal processes, into a more general method.

The first section defines precisely the problem to be considered.

This includes the necessary conditions for an optimal solution given by

the calculus of variations. Several mathematical programming techniques

are described in the second section along with a numerical algorithm

called the gradient projection method. The application of this numer-

ical method to the solution of the necessary conditions from Pontryagin's

Maximum Principle (PMP) is detailed in Section 3.3. Results of this

approach are shown to be consistent with the necessary conditions, given

in Section 3.1; these results provide a clarifying insight to the math-

ematical processes entailed in the maximum principle, and an explicit

formulation for the Lagrangian multiplier functions. This explicit for-

mulation is used in the next section to show the necessary conditions

may then be regarded as an algorithm. The final section contains a brief

summary of solution methods.

The main theoretical development of the dissertation is contained

in the first three sections. It is well known that the problems encoun-

tered in the calculus of variations are equivalent to the optimization of










a functional (in the sense of mathematical programming problems) under

certain restrictions upon the variations. A good exposition of this is

available in Luenberger (1969). With this equivalence in mind, it is

noted that the PMP is itself worded as a constrained optimization prob-

lem. When treated with what is normally regarded as a numerical method,

the gradient projection method, an explicit formulation of the atten-

dant Lagrangian multipliers is obtained. This form satisfies all of the

calculus of variation necessary conditions and allows one to use them

in a most straightforward fashion. As a result, these necessary condi-

tions may be directly used in the form of an algorithm to obtain a solu-

tion. Furthermore, it is believed that treating the PMP as a mathemat-

ical programming problem in conjunction with the gradient projection

method helps to explain the effect of combined control-state constraints

upon the maximum principle.


3.1 Problem Statement and
Necessary Conditions

A general problem which represents a large class of structural

optimization problems is treated in the sequel. The functional

tF
J = / L0(x,u) dt (3.1.1)
0

is to be minimized with respect to the control u(t) where the state

x(t) must satisfy certain boundary conditions and a differential con-

straint; in addition, an inequality constraint involving both the state

and control must be satisfied. For










uT(t) = [u (t) u2(t) ... u (t)] (3.1.2)
-- 1 2 m


x (t) = [xl(t) x2(t) ... x (t)] (3.1.3)


the subsidiary conditions to minimizing the cost function J are:


x = f(x,u) (3.1.4)


Specified Boundary Conditions on x(t) (3.1.5)


(x,u) < 0 = l,...,q (3.1.6)


Terminal time tF is considered to be constant; allowing it to be

unspecified requires only a slight modification to the following

derivation.

This problem is a particular form of a very general one treated

by Hestenes (1966). His results are a set of necessary conditions which

must be satisfied by the optimal solution and include the maximum prin-

ciple. To obtain the necessary conditions, the inequality constraints

are converted to equality constraints in the manner of Valentine (1937).

These constraints and the differential constraints are then adjoined to

the cost function via Lagrangian multiplier functions ,j(t) and Pi(t)

respectively.

(x,u) + s2(t) = 0


where the slack variables s (t) are defined such that



The symbol denotes "is defined by."
The symbol "= denotes "is defined by."









A 1 ]
s(t) = [-(P(x,u)] 0 0

tF t
J = J Lo(x,.u)dt + p (t) [x-f(x,u)] dt +
0 0
tF
+ f P,(t) [PQ(x,u) + S2(t)] dt
0


Implied summation convention is used whenever a vector formulation leads

to possible ambiguities in later developments. Integrating the second

integral by parts gives a result that leads to the variational Hamiltonian.

t t
=T. TF dF
J = P- +/ [Lo-p- + -x 2] dt
0 0

Define:
A T
H(x,u,2) = L (x,u) p (t) f(x,u) (3.1.7)


the variational Hamiltonian, and H which will include terms arising

from the inequality constraint.


H =- H- H-

or

H = (t) f(x,u) Lo(x,u) Zi(t) pk(x,u)

Hence

J = T x F [H* + x pIs ] dt
0 0

With the e'.c:eption of the maximum principle, all of Hestenes' necessary

conditions are obtained from the requirement that the first variation

of the cost function vanish. In the following, "6x" designates





47



"the variation of x"; a subscript vector designates the partial deriva-

tive with respect to that vector, with the result itself a column vector.

Thus,


tF t
T-
6J = p 6x f
0 0


[x (H* ) +
x


T *
+ 6u H 2s6s] dt = 0
u k z k


To derive the PMP requires an extensive mathematical development and is

not included since it contributes nothing to the present discussion.

However, the necessary conditions are listed in order to be available

for later reference.


x = H = f(x,u)


P=-Hx


0 = Px t
1 1

0 = p.6x. t


Specified Boundary


= 0
- i

= t o
tF

Conditions on x(


0 =H
U


0 = uz(t) p((x,u) P(t) ( 0


H(x PT' ~PT' ) H(OpT' u, e)


The optimal solution must satisfy these six conditions together with the

inequality constraint (3.1.6).










The PMP states that along the optimal trajectory, each instant

of time t, state XOPT(t) and adjoint state p(t), treated as fixed, the

optimal control UPT(t) is that admissible control which minimizes the

variational Hamiltonian. In the present context, admissibility requires

that u(t) be piecewise continuous, the set of admissible controls being

denoted by Q. Hence the PMP indicates that


u (t) = ARGMIN [H(x O, u p)] (3.1.8)



Notice that the necessary conditions suggest nothing about how

a solution is obtained, but merely indicate certain functional relation-

ships that must be satisfied. However, equation (3.1.8) seems to inti-

mate that solution of the necessary condition of P!1P involves a mathe-

matical programming problem.


3.2 Mathematical Programming:
Gradient Projection Method

Having shown that the PMP from the calculus of variations

approach to an optimization problem may perhaps be related to a mathe-

matical programming problem, the latter will be discussed in general

terms. Consider a nonlinearly constrained optimization problem


-xo = ARGECMIN [F(x)]
xE~

subject to

gj(x) 0 j = l,...,m

where Q denotes the set of admissible state components x., i = l,...,n,

and to be admissible requires only the satisfaction of the m inequalities.









Necessary conditions which xOPT must satisfy are given in the Kuhn-Tucker

theorem:

(i) constraints are satisfied gj (xPT) < 0

(ii) multipliers exist such that X. 0

and for all j = l,...,m AXg (xPT) = 0
m
(iii) and VF(x T)+ E V g(x ) =0
Sj=l -


Observe that if I denotes the set of indices associated with active

constraints, the first two conditions may be written as


j e IA g.(x) = 0 and A. > 0


j IA gj.(x) < 0 and X. = 0
A J- J


Fox (1971, pp. 168-176) presents a very readable proof of this theorem;

a more mathematical proof using vector space concepts is available in

Luenberger (1969).

Many methods for obtaining a numerical solution to the nonlinear

programming problem described by the first two equations of this section

have been developed. The gradient projection method by Rosen (1960) is

used frequently in structural optimization. Basic to the method is the

orthogonal projection of the cost function gradient onto a subspace

defined by the normal vectors of the active constraints. An inherent

part of the algorithm is the concept of a "feasible," "usable" direction.

Any direction d is feasible if an increment x in that direction improves

the cost function, i.e., decreases F(x). Direction d is said to be

usable if it also satisfies the constraints. As long as a feasible,










usable direction exists, the cost function may be improved. A constrained

optimal solution xPT occurs at that point where no feasible direction

is also usable, i.e., any attempt to improve the cost violates a con-

straint. In Appendix B these concepts are used in a concise proof of the

Kuhn-Tucker conditions.

Fox (1971) derives the matrix P which projects the cost function

gradient into the subspace defined by vectors normal to the active con-

straints. This is equivalent to subtracting all components parallel to

vectors that are normal to surfaces of active constraints from the nega-

tive gradient of the cost function. Recalling the definition of set IA,

consider r constraints to be active such that


IA = {al ,a2... ,}


Define a vector whose elements are the corresponding nonzero

Lagrangian multipliers, and another vector whose elements are the active

constraint functions

A1 = [\ \ ... \ ]
1 C2 r
T
N [g g, ... g]
1 2 r

From the N vector, a matrix N is introduced, each column of xhiCh is the

gradient of an active constraint. Hence, N is an (nx r) matrix where

T
N N = [N..] i = 1,...,n

and (3.2.1)

N = = j = 1,. ,r
ij 1x. x.
1 1










With these definitions of A and N, the third Kuhn-Tucker condition can

be written as


V F(x) + N X = 0


At any feasible point x where g (x) : 0, the direction which best

improves the cost function is the negative gradient of the cost. If

those directions which lead to constraint violations are subtracted from

-V F(x), the projection matrix P is obtained. Directions causing a

constraint violation are specified by the gradients of active constraints,
T
i.e., the columns of N What is required of S, the projection of the
-x

gradient, is that


S = (-V F(x)) Nx A (3.2.2)
x x


where A are scalar coefficients to be determined such that S is ortho-

T
gonal to each column of N or
-x


(N)T S=0
(N )r S = 0
-x

T
When the matrix equivalent to N is used together with the S expres-
-x

sion (3.2.2), the result is


N (-V F(x) N X) = 0


such that the X which satisfies this orthogonality condition is:


A = (NTN)-1 NT(V F(x)) (3.2.3)


Unless the active boundary surface normals V xg(x) are linearly depen-

dent, the matrix (N N) is nonsingular. Conversely, if this matrix is










singular the active constraints are not linearly independent; however,

this is not a condition encountered in most real cases.

Substitution of the X expression into the S equation leads

directly to the projection matrix P:

S= -P V F(x)
X -

where

P = I N(NTN)-1 NT (3.2.4)

where I is the identity matrix. The direction S which best improves

the cost is given in terms of P, where P and N are given by (3.2.4)

and (3.2.1). If no constraints are active at a point x, then N is a

null matrix, P reduces to an identity matrix, and the direction of best

improvement is coincident with the direction of steepest descent.

In the algorithm associated with this method the starting point

must be a feasible point where g.(x) : 0 for all j = l,...,m. The

design then proceeds in the S direction until the solution is satisfied

to within a specified position tolerance E. Necessary conditions

generally programmed in a computer program are:

Is. i i = 1,...,n

A. > 0 j A


A. = 0 j I


It is readily seen that for S, P, and A defined as above, these are

completely equivalent to the Kuhn-Tucker conditions.









3.3 Gradient Projection Method Applied
to the Maximum Principle

Based upon the preceding discussion, the similarity between PMP

and the-mathematical programming problem can be discussed. The maximum

principle states that the optimal control uOPT minimizes the variational

Hamiltonian with respect to all admissible u. Or, at each time

0 t t tF, uPT minimizes H(x,u,p) with respect to u for given x and p

and where OPT is subject to constraints 4 (x,u) < 0, k = l,...,q.

Treating this as a mathematical programming problem, the following

correspondences are recognized

x U

F(x) % H(x,u,p)

gj(x) % (x,u)

j v<(t)

V F(x) 'b H
x- u

S H plus constraints
u

Continuing to identify corresponding quantities, at each time t, let

IA denote the set of active constraints, taken to be r in number.


IA = {al, 02' ...' r }


Then NT T = [ U ..1. ** ]


T T
A' = [) (t) PQ (t) ... (t)0
1 2 r



N T H j] = Jj = l,...,n
3 j = 1,... ,m






54



Furthermore, define (H ) to be the gradient of H with respect to u

where all components that cause a constraint violation have been removed.

Since projection matrix P removes cost function gradient components that

lead to constraint violations, consider its use in the maximum principle.

T
With the correspondent to N identified as then
--U'

P = I Tu (3.3.1)
u -- u
and

(Hu) = PHu (3.3.2)


From the Kuhn-Tucker conditions, this implies that along the

optimal trajectory (t,XOPT, u OPT)


= (Hu)p = 0 (3.3.3)


Similarly, at each time


g(t) (- T )- T H (3.3.4)
T-- T u
u u -


from which it follows


(. T )) + H T Hu = 0
u u -

T
u + Hu = 0


u
(H + _To)u = 0 (3.3.5)


( -H ) 0


Or, H = 0 (3.3.6)
u










Hence the control law from Hestenes' necessary conditions can be derived

from the PMP condition by treating it as a nonlinearly constrained math-

ematical programming problem. While using the gradient projection method

in the derivation, it is seen that equation (3.3.5) is equivalent to the

third Kuhn-Tucker condition. The second Kuhn-Tucker condition is iden-

tical to Hestenes' necessary condition on the Lagrangian multipliers used

to adjoin the inequality constraints to the cost function. Satisfaction

of the inequality is implied by requiring the first Kuhn-Tucker condition

to be fulfilled, where

X. > 0 i(t) > 0 (3.3.7)


Sj g(x) =0 -+ (t)>(x,u) = 0 (3.3.8)


g.(x) W 0 z 4 (x,u) : 0 (3.3.9)

Thus by treating the solution of the necessary conditions of the max-

imum principle as a programming problem with inequality constraints,

using the gradient projection matrix, and by requiring satisfaction of

the Kuhn-Tucker conditions, an explicit formula for Hestenes'

Lagrangian multiplier functions has been derived. It is further demon-

strated that with the u (t) so defined satisfaction of the extremum con-

trol law condition is implied. However, before this treatment can be

accepted as valid, it must also be shown that the system of canonical

differential equations is unchanged.

Consider the state system equations

x = H = f(x,u)
--









where

H = T (t)f(x,u) L (xu) (t) (x,u)

It is obvious that the explicit form of p(t) has absolutely no effect

upon the state system equation expressed in canonical form.

Demonstrating that the adjoint system equation is unchanged

requires the method described by Bryson et al. (1964). Consider the

general problem of Section 3.1 again, but with only the differential

constraints adjoined to the cost function, i.e.,
tF t
Min {J = px + f (H xT)dt} (3.3.10)
u 0 0

subject to: pk(x,u) < 0 9 = l,...,q (3.1.6)

where H(x,u,p) = L (x,u) T (t)f(x,u) (3.1.7)

and x = f(x,u) (3.1.4)

Again let IA denote the set of indices associated with r active

constraints at any time t


IA = {al a, ...,ar


4 (
k(x,u) = 0 E IA A-> v(t) > 0

and p is defined as before

T
-1 a 2 r

The problem can then be thought of as minimizing (3.3.10) subject to

p(x,u) = 0









While on the constraint surfaces defined by this equation the variations

in control 6u(t) and state 6x(t) are not independent but instead are

related through the subsidiary requirement that

6O(x,u) = 0

or

Sx 4 (x,u) + 6uT T(x,u) = 0 (3.3.11)
--- -- __U -

This imposes a restriction to the admissible variations. For cost func-

tion (3.3.10) to be a minimum, it is necessary that its first variation

vanish, i.e.,

tF tI
6J = 6x + f 6xT H + 6u H 6xT] dt = 0 (3.3.12)
-0 0 u

It has already been shown that

T
H = (H + ) = 0
u u

which will be used to advantage shortly, after having added and sub-
T T
tracted the term 6u (1u 4T) from the integrand of (3.3.12).







+ 6UT RT1 6uT(u
t t
T F ^F
0=E6x + f 6xT(H uTH +


+ uT ~( )u ~T (T4) )udt

Rearranging terms gives
tF tF
T F F
0 =px +f pxT(H i) +
0 0

T T T T).dt
+ 5u (H + (j 4i)u) cu (ii) dt





58



But,
T T *
(H + (iT) = (H + I )u =(-H) =0
0
and
T T
(Ti )u =

Hence,


0 = T tF + F 6x T(H ) u dt
0 0x -


It is here that the restrictions imposed by the active constraints are

applied; from (3.3.11)

T T T T
Su t = 6x x

such that


T~6x + F _xT(Hx- ) + x _A dt
0 0

t t
0 = x + f/ 6x (H p + ) dt
+f x x-
0 0 -

Applying Euler's lemma, for arbitrary variations in the state which

satisfies the constraints,

(H- + P) =


which by the following manipulations is shown to be the adjoint system

equation of Hestenes.

i = + (T)x


x-
(-H- _Tx

p=-H
x










Thus, the explicit formulation for pL(t) obtained by applying the gradient

projection method to the PMP satisfies all the necessary conditions of

Hestenes.

It may happen that in some cases the constraint upon control

does not depend upon the state. It can be shown that the y(t) explicit

formulation is equally valid in this instance. By examination of

equations (3.3.1) through (3.3.9) it can be verified that all the neces-

sary conditions except the adjoint system equation are satisfied. To

demonstrate the latter, recall that when on a constraint boundary the

first variation of both the cost functional and the constraint function

must vanish. That is, for

K(u) = 0

both

J = 0

and

65 = uT 6 T = 0 (3.3.13)
-u

To derive the desired equivalence, the same term must be added and

subtracted from the integrand of 6J as before, again arriving at the

result
t t
T I / T T T
0 = p 6x, + f Fx (H P) 6Tu I dt
0 0 -0

When the constraint variation (3.3.13) is introduced into this last

equation, then by Euler's lemma


(Hx p) = 0
A










Since it was stipulated that (u) is not a function of x, the equation

may be written

T
(H +1 4)_ = 0

S-H )-


x


Thus, the expression for I(t) is valid when the constraint inequality

depends only upon the control u(t).


3.4 Maximum Principle Algorithm

In the introduction to this chapter it was stated that the

Lagrange type problem from the calculus of variations is equivalent to

an ordinary mathematical programming problem based on the Kuhn-Tucker

conditions. Furthermore, when inequality constraints are present the

necessary conditions are equivalent to the Kuhn-Tucker conditions.

It was demonstrated in the preceding section that if the PMP is itself

treated as a mathematical programming problem, application of the

gradient projection method provides an explicit solution for the

Lagrangian multipliers associated with active constraints. This explicit

solution for P(t) also satisfies all of the other necessary conditions

for an optimal solution. The ability to determine i (t) explicitly

in terms of parameters and functions that describe the problem suggests

the possibility of converting the necessary conditions of an optimal

solution into an algorithm for obtaining it.








Ensuing discussion of the algorithmic form of the necessary

conditions contains the implicit assumption that all equations are valid

along the optimal trajectory. It is further assumed that the problem

under consideration is that one described in equations (3.1.1 3.1.6).

The algorithm requires that x(t) and p(t) be known at each time

0 ( t < tF for which the solution procedure is as follows.

(i) Use PMP on the variational Hamiltonian to determine an optimal

control u (t) independent of constraints.

u (t) = ARGMIN [H(x,u,p)]

Evaluating the inequality constraints with u = u reveals which of

the k = l,...,q constraints are active. Let r denote the number

of active constraints and IA the set of indices associated with

them.

IA = {a a2, ., a }

9 (x,u) = 0 C IA

P (x,u) < 0 I1 IA

From this the vectors whose elements are the nonzero Lagrangian

multipliers and corresponding constraint functions are defined,

respectively, at the instant of time t.

T
PT(t) = [u J ... ]
a 1 42 r

T (x'u) = [ a1 a2 a' a I


(ii) Having identified which of the q constraints are active, r

components of OPT are specified by _(x,u) = 0. They may be

solved by using the implicit function theorem, which requires









T to be of rank r. This in turn requires the r constraints
-Iu

which are active at point x(t) to be linearly independent.

To determine the remaining (m- r) components of uOPT requires

that be known at time t, but


(t) =- ( T )-1 T Hu
T -- T u
u U -


This value of is used to determine the "constrained" Hamil-

tonian,

T
H = (H + P )
*
(iii) With the nonzero Lagrangian multipliers V known and H conse-

quently defined, the remaining (m-r) unknown components of

OPT are determined from the control law for the constrained

system, i.e,, from


H =0
u

Once uOPT is completely known, the adjoint system equations

are determined by


x

The process outlined above then allows uOPT to be written as


u = ARGMIN [H(x Tu,)
-OPT OPT


since the u obtained in this fashion satisfies (x,u) < 0 which is the

only requirement for being admissible. However it must be recalled

that these equations are valid along the optimal trajectory; it remains

to be shown that this algorithm may be employed in some manner to obtain

that optimal trajectory and to demonstrate their satisfaction along it.










3.5 Solution Methods

Necessary conditions from the calculus of variations provides

a Two Point Boundary Value Problem (TPBVP) to be solved, which is in

general, nonlinear. For all but the most simple problems no analytical

solution is possible and if any solution is to be obtained a computer

must be used with some numerical method. A discussion of the available

methods and their relative advantages/disadvantages is not included here

due to the availability of such discussions in the literature, e.g.,

Bullock (1966). All of the methods involve some iterative scheme, and

for optimal control can be separated into two general categories.

(i) Indirect methods. Schemes which require an initial guess of

the state's solution: In these methods the starting point is

an initial guess of the time history of the solution. The con-

trol associated with the solution is a subsequent calculation.

Iteration continues until the state satisfies some criterion

connoting convergence; the final control history at conver-

gence is the optimal control.

(ii) Direct methods. Schemes which require an initial guess of the

control function: The starting point for these methods is an

initial guess of the control time history. For this class of

methods the state associated with the control is a subsequent

calculation. Iteration continues until the control satisfies

some convergence criterion.

The method of quasilinearization was selected, based upon the

success of Boykin and Sierakowski (1972) in applying it to a constrained









structural optimization problem. Excellent convergence for their

problem, the capability to handle nonlinear systems, and the avail-

ability as an IBM SHARE program, ABS QUAS1, dictated its selection.

In the application to the examples in Chapters IV and V the program

required no modification. As a result, a detailed discussion of the

method of quasilinearization is not included.

The problem discussed in preceding sections of this chapter

falls into the general class of problems that QUAS1 handles, that is,

Y = (Y,t)

with the boundary condition of the form

B Y(O) + B Y(t ) + C = 0
k- r F -Q

where tF, square matrices B and Br, and vector C are specified,

constant quantities. The specific form of B B and C depend upon the
2 r o-Q
given boundary conditions. As described in algorithm form


x = f(x-, oPT(xP)) = Gl(x,E)

P = -H (x P (x,)) = G (x,p)
x -OPT -2

In terms of the general QUASI nomenclature,


= g(Y, t) = _
1G2 ('-)


Boundary conditions are determined by those specified for the original

system and by the necessary conditions outlined in the first section

of this chapter.

In Chapter VI the problem treated by Boykin and Sierakowski (1972)

is solved by the gradient projection method applied to a finite element





65



formulation for the description of the structural system. This is

a method of the second kind mentioned above. Results of the two methods

are compared.













CHAPTER IV


CONSTRAINED DESIGN OF A CANTILEVER BEAM
BENDING DUE TO ITS OWN WEIGHT



4.0 Introduction

A structural optimization problem has been selected for its

simplicity and stated as an optimal control problem. The maximum prin-

ciple is applied, giving a nonlinear TPBVP of the Mayer type. Among the

earliest expository papers on the maximum principle, Rozonodr (1959) gives

an excellent treatment to a similar type of problem; his technique is

used to obtain both the variational Hamiltonian and adjoint variable

boundary conditions. It is shown that no finite solution exists for the

situation of unconstrained control. Numerical solutions for constrained

control are obtained by the method of quasilinearization. Constraints

include both geometric limitations to control as well as maximum stress

limits that become mixed constraints depending upon both state and con-

trol variables.


4.1 Problem Statement

A cantilever beam of variable rectangular cross section is to be

designed for minimum tip deflection due solely to its own weight. The

material is specified to the extent that the modulus E and density p are

constants. Length L is specified but the design variables, height h(x)

and width w(x), may be chosen independently of each other, subject to










hard constraints upon the allowable dimensions. That is


a < w(x) < c
(4.1.1)
b < h(x) d d

If Y(x) denotes the deflection of the centerline, the problem is:

given E, p, L, and the constraints, find h(x) and w(x) to minimize

Y(L). The particular form of differential constraints to be satisfied

will be derived in the next section.


4.2 Structural System

Small deflections are assumed in order to use linear Bernoulli-

Euler bending theory. Basic conventions assumed for this example are

depicted in Figure 4.1; with these conventions the governing equation

is derived using standard strength of materials considerations. The

result is


EIB(x)Y"(x) = M(x) (4.2.1)

where
L
MB(x) = yf (T-x)w(T) h(T)d
x

1
IB(x) = 1 w(x) h3(x)


and y = pg. Kinematic boundary conditions to be satisfied by the solu-

tion of (4.2.1) are:

Y(O) = 0

Y'(0) = 0















/ x h (x)


T



/ w(x)
dW = yw(x)h(x)dx











V (x)




MB (x)





Note: Y(x) is centerline deflection, positive downward.


Structural Conventions


Figure 4.1









Design variables and the related constraints (4.1.1) are put into

dimensionless form such that


h(x)
v= = h + b/d < v < 1
1 d 1

w(x)
v = W a/c s v2 1
2 c 2


IB(x) = 1 cd3 v 2
B 12 1 2


Replacing the independent variable with a dimensionless equivalent, and

using the control components allows the governing equation to be put

into a dimensionless form. For

x
t = -
L


1 E d2
(2)2 u = f (T-t) U1(T)u2(T)dT
yL2 t
where


(4.2.2)


u.(t) = vi(x(t))

':-ien constant CB is defined, the usual kinematical relationships for

a beam may be written in a simple dimensionless form; that is, let

1 E d 2 -1
C (d) (Units = Length )
B 12 yL2


x2 = CB
31 B 12









d
x3 = CBUU2Y





"l2 = CB 2t (u1u2Y)
u1d2 B (u1u2Y)
dt2


Deflection

Slope

Moment


Shear


Load









These state component definitions are used with the natural boundary

conditions to obtain

x3(1) = 0

x4(l) = 0

From (4.2.2), the state component definitions, and the above boundary

conditions it follows that


Xl = x2

2 = x3/uuu2

x3 = 4

x4 = UlU2


x1(0) = 0

x2(0) = 0

x3(1) = 0

x4(l) = 0


These equations and boundary conditions are used in the following section

to precisely state the problem. The solution and results are given later.


4.3 Unmodified Application of
the Maximum Principle

In terms of the state variables defined in the preceding section,

the problem can be stated with more mathematical precision. Find


Up (t) = ARGMIN [x (1)]
subject to:

subject to:


(i) differential constraints

(ii) kinematic boundary conditions

natural boundary conditions


(iii) hard geometric constraints


x = f(x,u)

xl(0) = x2(0) = 0

x3(1) = x4(l) = 0


b/d : ul(t) 5 1

a/c < u2(t) < 1









According to terminology in the calculus of variations this is a Mayer

type problem. Among the early papers concerning the PMP, Rozonoer (1959)

applies the PMP to a similar problem giving a geometric interpretation

to the function of the adjoint variables.

In Rozonoir's problem the cost is a generalization of the

ordinary Mayer problem, in the sense that the cost function is a linear

combination of the terminal state components. It can be shown via the

calculus of variations that to minimize


where c is
-r


J = c x(tF)
-r-- F

a vector of prescribed constants, the necessary conditions are

A T
H = T(t)f(x,u,t)

x= H = f

T
p = -H =

0 =H


o = p (0) x(0)


0 = (tF) [r + P(tF)



H(OPT', uOPT' p,t) H(xOPT' u, t)

That is,

Min[J] Max [H]
u u









If we consider the cantilever beam problem, the forms in the

necessary conditions are

c = [1 0 0 ]T
--r

H = pl2 + P2X3/U 2 + 3X4 + p UU2

t =1


rx2


x3/uu2
f(x,u)
x



from which
from which


l = 0


P2 = -P1


= -P2/ulU2


P4 = -P3


pl(l) = -1


P2(1) = 0

P3(0) = 0

4(0) = 0


Adjoint variables Pl(t) and p2(t) can be integrated by inspection


Pl(t) = -1


p3(t) = (l-t)


0 < t 1


such that


H = x (l-t)x3/uu2 + P3X4 + PUU2


(4.3.1)









With this result, the necessary condition for control to minimize tip

deflection is

Hu = 3(l-t)x3/uu2 + 2 = 0


H = (l-t)x3/u3u2 + p41 = 0
u2 3 12


At first this appears to be a contradiction since the two equations can

be satisfied only by the trivial solution because they have equivalent

forms,

p4u = -3(1-t)x3
P4 1 2 3
(4.3.2)

p u = (1-t)x
P4UU2 3

Further examination, however, leads to the conclusion that when the

control is completely unconstrained there is no horizontal tangent

plane to the surface H = H(ul,u2).

When the geometric constraints to the control are included,

a constrained minimum may exist. If such is the case, the maximum

value of H occurs on the boundary of admissible control space. To that

end, PMP is employed along the control space boundary to determine

uPT (t) at each time t. Before detailing this procedure, it is neces-

sary to first consider some structural aspects of the problem.

By definition the control components are positive, which in

turn implies

Load: 4 > 0 k i4 = u l2

Shear: x4 < 0 k x4 >0 x (1) = 0

Moment: x3 0 x < 0 x3(1) = 0








Furthermore, since p2(t) 5 0 from (4.3.1)


P3 > 0 3 > 0 P3(0) = 0

p4 < 0 4 P4 < 0 P4(0) = 0

This exercise makes it possible to use the information of the sense for

x3 and p4 to simplify the search for OPT on the control boundary. By

arranging the Hamiltonian in the following fashion


H = -2 + 34 + P4[u1U2 (l-t)x3/p4ul2]


it is observed that both terms in the bracketed expression are positive.

This and the p4 outside the leading bracket allows the following

equivalence:

Max [H] I Min [ul2 (1-t)x3/P4UU2]
uE 9U u eaU
or,

u =PT ARGMIN [$(u)]

where

[(u) = [u2 + F2(t)/uu2 ] (4.3.3)

and

F2(t) = (l-t)x3/P4 > 0


At each position t along the beam, the state and adjoint variables must

satisfy the appropriate differential equations, and u is specified by

the preceding three equations.

Control space boundary rU is illustrated in Figure 4.2, where

the ul axis is treated as the ordinate since ul(t) and u2(t) correspond

















b/d < u 1 1


a/c < u < 1




/
1










ul
U1










b/d -


0
0 a/c 1
U2
"i2


Figure 4.2 Admissible Control Space









to the height and width, respectively, of the cross section of the

beam at position t.

Along the constant u edges of 3U, let uI = u where u has the
1 1 c c
value of either b/d or unity. If $ (Uc,u2) has a minimum point


dl d2$
--= 0 and -- > 0
du2 du2
where
where


The value of


D(u ,u ) = u u + F2(t)/u3u

2 c 22 2
du u F (t)/u3u2
d c c 2



du2 c2
2


u2 which satisfies the first


u = F(t)/u2
2 c


Furthermore, it is observed that only one extremum of Q(u) exists along

ul = Uc and that it is a minimum. Hence, either 4(uc,U ) has a minimum

on the constant u1 edge or is monotonically decreasing/increasing. If

either


u 2 a/c


1 i u2


then along the constant ul edge, H has its maximum value at a corner of

the rectangular 3U. On the other hand, if

a/c < u2 < 1

then H has its maximum value on the line u = u interior to the
1 c
endpoints.


condition is









Similarly, along the u2 edges of U, denote u2 = u where u

has the value of either a/c or unity. If $(ul,uc) has a minimum point


de d2,
S0 and 7- > 0
du1 du

where

$(Ulu ) = 1u + F2(t)/u3u


d- u 3F2(t) /uu
du1 c 1c
d21
d2 12F2(t)/ulu
du? c
1

It is observed that $(u1,uc) has only one extremum along u2 = uc,

it is a minimum, and occurs at the point ul = ul where


u = + {3F2(t)/u2}
1 c


Thus, by the same argument posed in the preceding paragraph, if either

*
U1 < b/d or 1 ul


then along the constant u2 edge, H has its maximum value at a corner of

the rectangular DU. Wherever


b/d < u1 < 1


H has its maximum value on the line u2 = u interior to the endpoints.
2 c









On the basis of these arguments, the following system was

solved by the method of quasilinearization:

X1 = x2 x1(0) = 0

2 = x3/uu2 x2(0) = 0


3 = x4 x3(1) = 0

4 = ul2 x4(1) = 0

3 = -P2/3U2 ''P3(0)= 0

4 = -P3 P4(0) = 0


u = ARGMIN [UU2 (l-t)x3/p4ulu2]
U 1U


The beam is represented by 100 intervals composing the range 0 < t 5 1,

which is separated by 101 "mesh points." An initial guess of the

solution x(t) and p(t) is chosen; it is selected to satisfy the bound-

ary conditions. This guess is not a solution and does not satisfy the

differential equations. The and j equations are linearized about the

initial guess, then the resulting linear TPBVP is solved to obtain new

x(t) and p(t) functions which more closely satisfy the differential

equations. At each time t corresponding to a mesh point, H is numer-

ically evaluated along each of the four straight line segments com-

posing 3U to determine 4OPT. The point (ul,u2) on 3U which gives

H(u;x,p,t) its maximum value is u T. This process is repeated until

the x(t) and p(t) iterate satisfies the differential equations to

within a specified tolerance. The equations necessary to use the IBM

program available are given in Appendix C, in the form of a subroutine

listing.





79



4.4 Results: Geometric Control Constraints

For the most part, no major difficulties were encountered in

using quasilinearization to obtain a solution to the sixth order sys-

tem derived in the previous section. Certain parameter values did

engender numerical instability. These cases, the source of the diffi-

culty, and its circumvention are discussed in Chapter VII. Moreover,

all calculations were done in double precision as necessitated by matrix

inversion accuracy requirements.

Parameter values selected to illustrate the solution method are:

u : b/d = 0.25
(4.4.1)
u2 : a/c = 0.20


The measure of error of satisfaction of the differential equations in

the TPBVP is in terms of the general system

dY.
= f.(Y,x) Y Y.(x), i = ,...,n
dx 1 -1

E-1. i = Max IdY. f.(Y,x)dx| (4.4.2)
i

Deflection of a uniform beam due to its own weight was used to infer

an initial guess which satisfies all boundary conditions:

xl(t) = t4

x2(t) = t3

x3(t) = 1 t2
0 < t < 1 (4.4.3)
x4(t) = -1 + t

P3(t) = t2

P4(t)= -t3









With these specified parameter values and initial guess of the solu-

tion, the program converged to a solution in five iterations. From

this run a tolerance was selected for all subsequent cases; the follow-

ing tabulation provides the data used in its selection:

Iteration ERROR Tip Deflection (Cost)

1 .2028 .7387749327 x 10-

2 .1704 .3192152426 x 10-

3 .6533 x 10-1 .2847993812 x 10-

4 .3031 x 10-5 .2853731846 x 10-

5 .1129 x 10-10 .2853719983 x 10-1


It is seen from these tabular data that there is little improvement in

cost (tip deflection) as a result of the fifth iteration. For this
-o
reason a value for the tolerance was selected as 0.5 x 10 which corre-

sponds to about six significant digits in the cost functional.

Recall from the previous section that no unconstrained minimum

exists. With the control bounds included, the intuitive solution is

one in which the cross-sectional area is maximum near the root, and

reduces to a minimum at the tip. Recalling that for the optimal control,

u =NMax [H] Min [i]



A sequence of illustrations in Figure 4.3 demonstrates the location of

uPT on 5U for several stations along the beam. Constant contours of

D(u) are plotted on the admissible control space at five distinct posi-

tions. If an extremal point exists interior to DU some lines of constant

0(u) contours must be closed curves in u-space. This is impossible for

this example.










x/L .1







t---I


0 1 1


1



U1






0


x/L = .7


O Minimum $(u) u-pT




O Maximum O(u)


-->- Direction of Increasing 0(u)


01


Figure 4.3 Contour Plots of 0(u) at Various Stations
Along the Beam


x/L = .5


01


u2



x/L = 1.


x/L = .3









The first illustration is for the t = 0.1 cross section, near

the root of the beam. Since uPT occurs at the point of minimum 0(u),

the optimum value for both ul and u2 is unity, the maximum allowable

dimensions for both height and width. Constant contour lines indicate

that ((u) is mathematically decreasing in either direction of u-space.

Lines of constant 0(u) are also plotted for the cross section of

t = 0.3. The optimum control has the maximum admissible value for

height ul but u2 has a value somewhat less than unity. However, there

are still no contour lines which are closed curves.

At the midpoint cross section the minimum 0(u) point occurs

at ul = 1 and u2 = a/c. Although the surface Q(u) forms a scoop-like

shape, there are still no closed curve contours, and hence no extremal

interior to admissible u-space. The next cross section at which 0(u)

is displayed occurs at t = 0.7. On this section, u2 is still at its

lower bound but ul is no longer at the maximum allowable value of unity

as shown in Figure 4.3. In the last of the sequence, d(u) contours for

the cross section at the tip of the beam are displayed. The point of

minimum q(u) occurs where both components of control have their minimum

allowable values. Again, no contour lines of constant 0(u) form a

closed curve indicating the existence of an interior extremal point.

This sequence of illustrations indicates two things. First,

the lack of closed curve contour lines of <(u) verifies that uPT exists

on U. With further study it may be possible to obtain some condition

on H = 0 which implies the equations corresponding to (4.3.2) can

never yield a finite, unconstrained optimum. Such a condition would

define the class of structures whose unconstrained solution is the










"zero volume solution" frequently described in the literature on struc-

tural optimization. Secondly, at t = 0 the point u PT occurs at
T
u = [1,1] the point of maximum cross-sectional area; as t increases

from zero to one, the point uPT moves along the ul = 1 boundary of aU

to the u2 lower bound, and then down the u2 = a/c boundary of DU to ul

lower bound. By the time t = 1 the optimal cross-sectional area is the

minimum allowable area. As a result of the prescribed form of 3U,

if u (t) follows this particular path as t increases from zero to one,
-OPT
each component of u (t) has its own distinct region of transition.
-OPT
That is, at any value of t, if b/d < ul < 1, then u2 must be on either

its upper or lower bound. Conversely, if u2 is in transition where

a/c < u2 < 1, then ul must be on one of its bounds.

This effect is seen most clearly in Figure 4.4, where uOPT is

displayed for the example case parameter values specified by (4.4.1).

The profiles are displayed on a two-view drawing as a plan-form of the

beam might appear. State components corresponding to this beam are

shown in Figure 4.5, representing dimensionless deflection, slope,

moment, and shear, respectively. As observed in Figure 4.4, there are

five distinct regions of the beam:


(i) 0 < t .25

uI =1
controls onupper bounds
u2 = 1

(ii) .25 < t < .52

u = 1
u2 transition
a/c





















x/L


a/c = .20


TOP VIEW

SIDE VIEW


Figure 4.4 Plan-form Views of Optimal Solution for
b/d = .20 and a/c = .25


"u2


2 u




















Shear


0





-.2
x4



-.4





.10



.05
x3

0



.05


x1



0



.02








0


0 .2 .4 .6 .8 1.0


Figure 4.5


State Components of Optimal Solution for
b/d = .20 and a/c = .25


-


Moment


Slope







-Deflection





Deflection





86



(iii) .52 < t < .59

uI = 1 on upper bound

u2 = a/c on lower bound


(iv) .59 < t < .90

b/d 1 u transition
u = a/c


(v) .90 < t < 1.00

ul = b/d
controls on lower bounds
u = a/c


The curves that show the intercept locations as a function of

parameter values b/d have been called "correlation curves" in earlier

studies. When the width is allowed to vary also, the second parameter

a/c is introduced. For the sake of comparison to previous studies, the

intercept/correlation curves are plotted as dependent upon b/d and

parametric in a/c. However, it would be just as correct to do the

opposite.

Intercept location curves described above are shown in

Figure 4.6. The heavy black curve is the case where a/c = 1, a beam

of constant uniform width--the case cited from earlier literature.

Another special case is represented by dashed lines, correspondin;- to

a/c = 0 which is the case corresponding to a minimum allowable thick-

ness equal to zero. Dashed lines are used because these data are an

extrapolation: convergence problems encountered for parameter values

less than 0.1 prevented obtaining numerical results.

































O
0

rJ


o c

r1-


a 0


co
(d 4


Lr)





0









A discussion on the convergence difficulties experienced by

the quasilinearization algorithm for parameter values approaching zero

is presented in the chapter on numerical instabilities. In that dis-

cussion, isolation of the source of difficulty is reported; it is pos-

sible that this difficulty may be a general result applicable to all

problems to be solved by the method of quasilinearization. A solu-

tion for this case is later obtained by finite element techniques.

Note that since 0 < a/c < 1 these two cases represent limits to the

solutions of the problem. In addition, if the four intercept locations

are plotted versus a/c and parametric in b/d, curves r and r appear
a c

as "horizontal vees" with rb and rd lines that are nearly parallel.

It is interesting to note from the figure depicting the solu-

tion of this case as a plan-form, that in the central region of the

beam, the height is greater than the width. This result can be antic-

ipated since such a configuration gives a greater bending resistance

per unit weight.

With further reference to Figure 4.4, the transition of ui(t)

is seen to be almost a linear taper, whereas the u2(t) transition

exhibits a much more pronounced curvature. To generalize from this

specific case of given values of b/d and a/c to arbitrary values

requires the introduction of four quantities characterizing the solu-

tion. These quantities are the values of t at the points where the

transitions intercept the bounds on ul and u2; since t represents a

normalized position x/L, these quantities can be thought of as an

intercept location expressed as percent of the beam's length. They










are defined with reference to the five distinct regions of the beam

previously given, where


rb design

rd design

ra design

r design


such that the five

(i) 0 < t

(ii) rc < t

(iii) r < t
a -

(iv) rd < t

(v) rb t
b -


tes u (t)

tes ul(t)

tes u2(t)

tes u2(t)


regions

< r
- c

< r
a
< rd
-d

< rb

< 1.0


intercept with lower bound

intercept with upper bound

intercept with lower bound

intercept with upper bound


are:

control on upper bounds

u2 transition

control on upper/lower bounds

u1 transition

control on lower bounds


4.5 Inequality Stress Constraints

This section treats an inequality limit to allowable normal

and shear stresses associated with bending. Using the ordinary

strength of materials formulations it can be shown that for the rec-

tangular cross section these constraints take the form


1 MB(x)
Sh(x) MAX
2 IB(X) MAX


1 VB(x)
8 IB(X) h2(x) MAX
B


When dimensionless quantities are introduced

6 Lx3 u u2 < MAX

3 < 2
2 YL x4/ulu2 MAX









the inequalities may be written in the required form for mixed con-

straints, i.e., as a function of both control and state components:


i1(xu) = x3/u u2 0 (4.5.1)


h2(xu) = x4/uu2 To 0 (4.5.2)

where
1 MAX d
((--
0 6 yL

2 'MAX
T0 3 yL


The two stress constraints place restrictions upon the minimum

cross-sectional dimensions to keep the normal and shear stresses less

than prescribed values. Specifically, from the constraints (4.5.1)

and (4.5.2), two control inequalities must be satisfied at each station

t, and these inequalities depend upon the state of the structural sys-

tem. The inequalities are:

ulu2> x3/o0


uu2 > x4/T x4(t) < 0


from which can be derived boundary arcs in u-space:


ul(u2) = (x3/O0u2) 0
(4.5.3)

uT(U2) = 4/TO > 0

Both of these boundary arcs are hyperbolas restricted to the first

quadrant of u-space. Depending upon the location of the arcs, vis-a-vis

the rectangular 3U, inclusion of stress constraints has one of three










effects in determining what u is admissible. At any station t in some

structural state x(t),

(i) if 0 ,T0 is too small the stress boundary arc lies entirely

above rectangular 3U; all geometrically admissible u violate

the stress constraints.

(ii) if 0,T0 is too large the stress boundary arc lies entirely

below rectangular DU; all geometrically admissible u satisfy

the stress constraints.

(iii) for some range of 0,T 0 the stress boundary arc divides the

rectangular DU into two regions: the upper region consists of

geometrically admissible u that satisfy the stress constraint,

u in the lower region are geometrically admissible but violate

the stress constraint.


Inclusion of stress constraints alters the admissible control

space from the rectangular shape previously considered to a shape that

may contain a stress boundary arc as part of its boundary. Consider the

normal stress boundary arc specified by (4.5.3) to be a part of 4U.

Then to find uPT in the manner outlined in Section 4.3, (u) must be

evaluated along ul = u l(u2). If a minimum exists along the orthogonal

projection of u l(u2) on the O(u) surface, then

^a$(u)
S=0

Min (u) U -
ul ul- a20(u)
> 0
lu2
1




Full Text

ON THE AUTOMATED OPTIMAL DESIGN
OF
CONSTRAINED STRUCTURES
BY
JERRY C. HORNBUCKLE
A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF
THE UNIVERSITY OF FLORIDA IN PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1974

Copyright by
Jerry C. Hornbuckle
1974

DEDICATION
To my grandmother, Mrs. Florence Hornbuckle,
and my wife, Carolyn. Without the love, confidence, and
patient understanding of Granny Hornbuckle and Carolyn
my graduate studies would never have been attempted.

ACKNOWLEDGMENTS
To Dr. Robert L. Sierakowski and Dr. William H. Boykin, Jr.,
for guiding my research and for being more than just advisors.
To Dr. Gene W. Hemp, Dr. Ibrahim K. Ebcioglu, and Dr. John
M. Vance, for their assistance, support, and for serving on my advisory
committee.
To Dr. Lawrence E. Malvern and Dr. Martin A. Eisenberg, for
always finding the time to offer advice and explanations on questions
related to solid mechanics and academics.
To the departmental office staff for their kind assistance
with administrative problems and clerical support.
To Randell A. Crowe, Charles D. Myers, and J. Eric Schonblom
for attentive discussions of many little problems and for assistance
in preparing for the qualifying examination.
iv

TABLE OF CONTENTS
Page
ACKNOWLEDGMENTS iv
ABSTRACT viii
CHAPTER
I INTRODUCTION , 1
1.0 Survey Papers 1
1.1 Historical Development: Optimal Columns 6
1.2 Historical Development: Optimal Static Beams ... 9
1.3 Historical Development: Optimal Dynamical Beams . . 12
1.4 Scope of the Dissertation 14
II GENERAL PROBLEMS AND METHODS IN STRUCTURAL OPTIMIZATION . 16
2.0 Introduction 16
2.1 Problem Classification Criteria 16
2.1.0 Problem Classification Guidelines 18
2.1.1 Governing Equations of the System 18
2.1.2 Constraints 19
2.1.3 Cost Functionals 21
2.2 Methods: Continuous Systems 26
2.2.0 Special Variational Methods 27
2.2.1 Energy Methods 27
2.2.2 Pontryagin's Maximum Principle 29
2.2.3 Method of Steepest Ascent/Descent 31
2.2.4 Transition Matrix: Aeroelasticit.y Problems . 31
2.2.5 Other Miscellaneous Methods 32
2.3 Methods: Discrete Systems 33
2.3.0 Mathematical Programming 34
2.3.1 Discrete Solution Approximations 35
2.3.2 Segmentwise-Constant Approximations 37
2.3.3 Complex Structures with Frequency
Constraints 38
2.3.4 Finite Element Approximations 40
2.4 Closure 41
v

TABLE OF CONTENTS (Continued)
CHAPTER Page
IIITHEORETICAL DEVELOPMENT 43
3.0 Introduction 43
3.1 Problem Statement and Necessary Conditions 44
3.2 Mathematical Programming: Gradient
Projection Method . . . . 48
3.3 Gradient Projection Methods Applied to
the Maximum Principle 53
3.4 Maximum Principle Algorithm 60
3.5 Solution Methods 63
IVCONSTRAINED DESIGN OF A CANTILEVER BEAM
BENDING DUE TO ITS OWN WEIGHT 66
4.0 Introduction . 66
4.1 Problem Statement 66
4.2 Structural System 67
4.3 Unmodified Application of the Maximum Principle . . 70
4.4 Results: Geometric Control Constraints 79
4.5 Inequality Stress Constraints 89
4.6 Results: Stress Constraints Included 93
VCONSTRAINED DESIGN FOR AN OPTIMAL EIGENVALUE PROBLEM . . 101
5.0 Introduction 101
5.1 Problem Statement 101
5.2 Structural System 102
5.3 Analysis of the Problem 109
5.4 Application of the Maximum Principle 121
5.5 Results: Geometric Control Constraints 132
5.6 Inequality Stress Constraints 148
VIFINITE ELEMENT METHODS IN STRUCTURAL OPTIMIZATION:
AN EXAMPLE 155
6.0 Introduction 155
6.1 Finite Element Problem Statement . 155
6.2 Mathematical Programming: Gradient
Projection Method 157
6.3 Results 162
VIICOMMENTS ON NUMERICAL INSTABILITY IN THE
QUASILINEARIZATION ALGORITHM 173
7.0 Introduction 173
7.1 Computer Program Convergence Features 173
7.2 Numerical Instatilibies for Cantilever
Beam Example 175
7.3 Numerical Instabilities for Column
Buckling Example 183
vi

TABLE OF CONTENTS (Continued)
CHAPTER Page
VIII CONCLUSIONS AND RECOMMENDATIONS 186
8.0 Summary and Conclusions 186
8.1 Recommendations 186
APPENDIXES
A HISTORICAL DEVELOPMENTS 191
B A SIMPLE PROOF OF THE KUHN-TUCKER THEOREM 206
C COMPUTER SUBROUTINE LISTINGS 214
BIBLIOGRAPHY 244
BIOGRAPHICAL SKETCH 254
vii

Abstract of Dissertation Presented to the Graduate Council
of the University of Florida in Partial Fulfillment of the Requirements
for the Degree of Doctor of Philosophy
ON THE AUTOMATED OPTIMAL DESIGN
OF CONSTRAINED STRUCTURES
By
Jerry C. Hornbuckle
August, 1974
Chairman: Dr. William H. Boykin, Jr.
Co-Chairman: Dr. Robert L. Sierakowski
Major Department: Engineering Sciences
Pontryagin's Maximum Principle is applied to the optimal design
of elastic structures, subject to both hard inequality constraints and
subsidiary conditions, By analyzing the maximum principle as a non¬
linear programming problem, an explicit formulation is derived for the
Lagrangian multiplier functions that adjoin the constraints to the cost
functional. With this result the usual necessary conditions for opti¬
mality can themselves be used directly in an algorithm for obtaining
a solution.
A survey of general methods and problems in the optimal design
of elastic structures shows that there are two general problem types
depending upon whether or not the cost functional is an eigenvalue.
An example problem of each type is included with the solutions obtained
by the method of quasilinearization. In the first example, a minimum
deflection beam problem, classical Maximum Principle techniques are
used. The eigenvalue problem is exemplified by the maximization of the
buckling load of a column and uses the explicit multiplier function
viii

formulation mentioned above. Since the problem considered is conserva¬
tive, it is therefore described mathematically by a self-adjoint system;
under this condition it is shown that the minimum weight problem is
identical to the maximum buckling load problem.
In order to demonstrate the theory for the programming techniques
used, the beam problem is also solved by using a finite element repre¬
sentation of the structure. From a comparison to the maximum principle
solution it is shown that the form of the optimal solution obtained is
dependent upon the magnitude of the tolerance used with the numerical
solution scheme. Furthermore, it is shown that convergence by the
quasilinearization algorithm is related to the respective curvatures of
the initial guess and the solution.
Recommendations for additional investigations pertinent to this
study are also included.
ix

CHAPTER I
INTRODUCTION
1.0 Survey Papers
It is exceedingly difficult to write a general introduction to
the field of structural optimization for two basic reasons: (i) there
is no conventionally accepted nomenclature, and (ii) there is also no
conventionally accepted classification of problem types or character¬
istics. In marked contrast, when one considers the calculus of varia¬
tions, "cost functional, system equations, kinematic and natural bound¬
ary conditions, adjoint variables, Hamiltonian, etc.," all have well-
defined, universally accepted meanings. Additionally, there is no con¬
fusion when speaking of the problem types of Mayer, Lagrange, and Bolza.
This common language and categorization of problems does not exist in
structural optimization.
Instead, the field tends to branch and fragment into very special¬
ized sub-disciplines that are oriented towards applications. While these
branches are related to the general field, the techniques and methods of
one branch can seldom be applied to another. Moreover, as a result of
the tendency to an applications orientation, solutions are generally ad
hoc and not useful for other problems even within the same branch. The
lack of any definitive unification of the subject cannot be blamed either
on being recently developed or in receiving too little attention.
1

2
This is readily seen by considering the survey papers described in the
following paragraphs.
The earliest comprehensive survey paper is Wasiutynski and
Brandt (1963). Although their excellent historical development is
dominated by Russian and Eastern European references, the authors do
include a higher percentage of papers by Western authors than is
encountered in the typical paper from Eastern Europe. A more funda¬
mental criticism is that too little is said regarding problem types or
solution methods.
Chronologically, the next survey paper is Gerard (1966) .
The theme of this review is aerospace applications, with a particular
orientation to the design-management, decision-making process. Most of
the papers cited treat specialized aerospace structures and applica¬
tions; however, the author does try to generalize by introducing a
design index D, a material efficiency parameter M, and a structural
efficiency parameter S. After defining the expressions for M and S
corresponding to several structural elements, design charts are pre¬
sented which show regions of possible application for various materials.
Unfortunately the design charts do not satisfy expectations aroused by
the introduction of the three general parameters.
Rozvany (1966) presents a similar paper pertaining to structures
in civil engineering. This paper is less comprehensive and more ori¬
ented towards specific structural applications than either of the pre¬
ceding survey papers. The author postulates several "interrelated
quantities (or parameters)" which could perhaps be used to generalize

3
structural optimization into a more rational methodology. However,
these quantities—loading (L), material (M), geometry (G), initial
behavior (IB), and design behavior (SB)—are only applied to an abstract
discussion of concepts.
Barnett (1966) has a readable short survey of the field that
dwells more upon theoretical aspects. He postulates a general problem
in which the cost is minimized subject to a system in equilibrium with
its loads, while "behavior constraints on strength, stiffness, and
stability" are satisfied. Uniform strength design introduces the dis¬
cussion of optimal trusses; virtual work theorems that were derived
originally for trusses are then applied to simultaneous plastic collapse
problems. Following a brief discussion of the plastic collapse load
bounding theorems, there is a short treatment of elastic stability
problems and material merit indices. Barnett's stiffness design example
exhibits several important features worth noting. Specifically, the
example is to minimize the weight of a beam subject to some given load
where the deflection at a certain point is specified. Virtual work is
used to handle the subsidiary deflection condition. Necessary condi¬
tions are obtained from the calculus of variations, but more signif¬
icantly, the Schwarz inequality provides a sufficient condition for
global optimality. Barnett concludes his survey with a section stating
that multiload designs satisfying "all three behavior criteria" are more
easily solved in "design space" by mathematical programming techniques
described by Schmidt (1966).
As a sequel to the comprehensive survey (1638 to 1962) by
Uasiutynski and Brandt, Sheu and Prager (1968b) present a complete review

4
of developments from 1962 to 1968. This paper contains three major
sections: general background, methodology, and specific problems.
In the first section they state that the "well-posed problem of optimal
structural design" requires specification of the
(1) purpose of the structure (load and environment),
(2) geometric design constraints (limits to design
parameters),
(3) behavioral design constraints (limits to the "state" of
the structure),
(4) design objective (cost functional).
The methodology section is not noteworthy, but the third section lists
what is in their opinion the specific problem types: static compliance,
dynamic compliance, buckling load, plastic collapse load, multipurpose/
multiconstraint structures, optimal layout (e.g., trusses), reinforced/
prestressed structures, and from the background section, probabilistic
problems. Their concluding remarks succinctly summarize the paramount
difficulty of the subject: realistic problems are too complicated for
precise analytical treatment. While progress is being made in analyt¬
ical treatment of simple structures, the authors opine that realistic
problems require mathematical programming techniques. However, they do
feel that analytical treatment is desirable to provide "a deeper insight
into the analytical nature of optimality."
A related survey by Wang (1968) on distributed parameter systems,
". . . whose dynamical behaviors are describable by partial differential
equations, integral equations, or functional differential equations,"
consists entirely of a bibliography. While not pertinent to the disser¬
tation, it is mentioned here for completeness.

5
Prager (1970) provides another survey which is not comprehensive,
nor does he present any new results as claimed. For example, Prager
and Taylor (1968) used the principle of minimum potential energy
and the assumption that stiffness is proportional to specific mass to
prove global optimality. It could hardly be called a new development
in 1970. At the same time, the author does present an excellent example
of a multipurpose optimal design problem. Prager also treats "segment-
wise constant" approximations and the optimal layout of trusses.
The final survey paper, Troitskii (1971), is an unusual review
of methods in the calculus of variations. Whereas some of the earlier
surveys present lengthy lists of references but contain little method¬
ology or theory, this survey is just the opposite. Part of what makes
it unique is that the author believed only eight articles merited cita¬
tion—all of them by Troitskii. This shortcoming is more than overcome
by a thorough classification of optimal control problems in the calculus
of variations. Troitskii bases the classification on "certain character¬
istics of control problems": types of constraints, properties of the
governing dynamical equations of the system, type of cost functional, and
possible state discontinuities. From these four criteria he postulates
five principal classes of problems; however, it is the criteria that
are important and not the specific problem type.
By comparing what the authors of the aforementioned surveys
believe to be the important types of problems, it is readily apparent
that there is little agreement on which characteristics of structural
optimization problems are significant.

6
1.1 Historical Development:
Optimal Columns
The beginning of structural optimization is generally attributed
to Galileo's studies in 1638 of the bending strength of beams. Accord¬
ing to Barnett (1968), Galileo considered a constant-width cantilever
beam under a tip load as part of a study of "solids of equal resistance."
In requiring the maximum stress in each cross section to be constant
throughout the beam, the height must be a parabolic function of position
along the beam. While this appears to be the origin of the field, a prob¬
lem that received more attention is the buckling of a column.
Using the newly developed calculus of variations, in 1773 Lagrange
attempted to apply variational techniques to the problem of finding that
distribution of a homogeneous material along the length of a column which
maximizes the buckling load. Truesdell (1968) relates that through an
insufficient mathematical formulation Lagrange showed the optimal form
to be a circular cylinder. Clausen (1851) provides the earliest known
solution to this problem for the simply supported case. As described in
Todhunter and Pearson (1893, pp. 325-329), Clausen minimized the volume
of the column with the differential equation for buckled deflection
treated as a subsidiary condition. Assuming all cross sections to be
similar, after several variable transformations and complicated manip¬
ulation, he obtained an implicit, analytical solution.
The next development was Greenhill (1881), according to Keller
and Niordson (1966). Greenhill determined the height of a uniform
prismatic column, beyond which the column buckled due solely to its own
weight. Timoshenko and Gere (1961) reproduce the solution in which the

7
deflection is expressed as the integral of a Bessel function of the
first kind (of the negative one-third order).
Blasius (1913) introduces his paper with a uniform strength and
a minimum deflection beam problem. For a given load and amount of mate¬
rial, the cross-sectional area distribution is determined which maxi¬
mizes the buckling load of a circular column. The solution is identical
to that obtained by Clausen. In addition, Blasius also obtained solu¬
tions for columns having rectangular cross sections and discussed the
effect of different boundary conditions on the results.
For the next few decades, structural optimization appears to
have been directed towards applications in the aircraft industry, where
aircraft structural problems and results are presented in the format of
a design handbook. Feigen (1952) is a good example of this, consider¬
ing the buckling of a thin-wall column. Given a constant load and wall
thickness, he required the variable inside diameter to be chosen such
that the buckling load is maximized. Wall thickness is selected to make
local buckling and Euler buckling occur at the same load. Solid tapered
columns having blunt ends are also treated for assumed stiffness dis¬
tributions .
Renewed interest was aroused by Keller (1960), who examined the
problem from the point of view of the theory of elasticity, and in
choosing the cross-sectional shape to give the maximum stiffness.
Neglecting the weight of the column, he obtained via the former that
twisting the column does not affect the buckling. Of all convex cross
sections, the equilateral triangle is shown to have the largest second

8
moment of area relative to a centroidal axis. Hence, from the defin¬
ition of buckling load, the "best" cross-sectional shape is the equi¬
lateral triangle. Keller also obtained Clausen's implicit, analytical
solution. Subsequently, Tadjbakhsh and Keller (1962) generalized the
problem to a general eigenvalue problem and boundary conditions subject
to a subsidiary equality constraint. The latter corresponds to spec¬
ifying the volume (or weight) of material to be distributed in an opti¬
mal manner. Using the Holder inequality they demonstrate global opti¬
mality of the eigenvalue for the hinged-hinged column.
Keller and Niordson (1966) examine the case of a vertical column
fixed at the base, subject to a vertical load at the tip and the column's
own weight. It is also assumed that all cross sections are similar.
Their approach is to state the problem as a simultaneous, dual optimiza¬
tion of the Rayleigh quotient. The eigenvalue is minimized with respect
to the eigenfunction and maximized with respect to cross-sectional area
distribution. Specifying the volume of material available is treated
as a subsidiary equality constraint. From the maximum lowest eigenvalue
the maximum height ac buckling is determined. Solutions are obtained by
an iterative technique employed with integral equations.
In a brief note, Taylor (1967), suggests that energy methods may
link optimum column problems to the traditional eigenvalue problems of
mechanics. Prager and Taylor (1968) classify problems in optimal struc¬
tural design and demonstrate global optimality using energy principles.
Unfortunately their assumption of thin-wall construction limits the
results to structures where the stiffness is proportional to the specific

9
mass density. The consequence of this assumption is that in the energy
formulation the resulting control law and governing equations are
decoupled, and hence easily solved. Huang and Sheu (1958) apply this
same thin-wall assumption to the problem treated by Keller and Niordson.
However, the former seek the maximum end load instead of the maximum
height. The authors also attempt to limit the maximum allowable stress
and obtain solutions by a finite-difference method. Further discussion
of sandwich (thin-wall construction) columns is given by Taylor and Liu
(1968). Basically, this paper is an elaboration of the techniques
described by Prager and Taylor when applied to columns. Extensive
results are provided for various cases.
Post-buckling behavior for columns subject to conservative
loads is considered by Gajewski and Zyczkowski (1970). A nonconcerva-
tive problem is treated by Plaut (1971b). The first of these two papers
is lengthy but is much too narrow in scope to be particularly useful.
In the second paper, the Ritz method is applied to an energy functional,
obtaining the "best" form of the assumed approximation to the optimal
solution.
1.2 Historical Development:
Optimal Static Beams
That beam problems played a role in the early developments of
structural optimization has already been indicated in the preceding sec¬
tion. No attempt is made in what follows to present a complete history,
but merely to outline the type of problems that have been considered.

10
Opatowski (1944) has an outstanding paper that deals with
cantilever beams of uniform strength. Besides providing numerous refer¬
ences to earlier studies, the author treats the problem with impres¬
sive mathematical rigor. The beam is considered to deflect under its
own weight and a transverse tip load; bending deflection is described
by a Volterra integral equation which is solved exactly for various
assumed classes of variable cross-sectional geometry. This paper is
representative of earlier papers in that it contains extensive analysis
and analytical results, but little numerical data. Barnett's work (1961)
and its sequal (1963a) apply the calculus of variations to more real¬
istic I-beams. One problem considered is maximizing the weight sub¬
ject to general, unspecified loads for a specified deflection at a given
point. The Schwarz inequality is used to derive a sufficient condition
for global optimality. Also included is a comparison of uniform strength
beams to the minimum deflection beam for several different cases of
applied load and geometry. The paper is concluded with various minimum
weight design examples in which both bending and shear stiffness are con¬
sidered .
Haug and Kirmser's (1967) work is one of the most comprehensive
studies of minimum weight beam problem published. While it may succeed in
handling any conceivable problem and in employing the most realistic
stress constraints, this very generality' requires so many variables and
conditions that the mathematics is complicated almost beyond reason.
Another study of minimum weight beams (Huang and Tang, 1969) is
important for several reasons. By dividing the beam into many segments

11
having constant properties, and determining the necessary conditions
that must be satisfied by every segment, it appears that the authors
are using the same methods that were used to solve the Brachistochrone
JL
A
problem in the seventeenth century. In their treatment, multiple
loads must produce a specified deflection at a given point; these
"segmentwise constant" approximations and multiple load problems have
recently received more attention. Of further interest in this paper is
the derivation of the multiple load optimality condition using
Pontryagin’s Maximum Principle.
While no new results or techniques are contained in Citron's
(1969, pp. 154-166), the author gives a very readable minimum weight
beam example. The problem is simple, described in detail completely,
and provides an excellent example of how control theory is applied to
an optimal problem in structural design. An analysis of intermediate
beams which may form plastic hinges is provided by Gjelsvik (1971).
It shows that if hinges are placed at all points of the beam where the
bending moment is zero, this makes the plastic or elastic minimum
weight beam statically determinate. Both the elastic and plastic beam
designs are shown to be fully stressed, i.e., are uniform strength beams
Application of generalized vector space techniques is character¬
istic of more recent papers. Bhargava and Duffin's (1973) is such a study
It treats the maximum strength of a cantilever beam on an elastic
Since these early papers are not readily available, this statement
is based upon descriptions of them given by historical references
cited in Appendix A.

12
foundation subject to an upper bound on weight. Although involving
more advanced mathematics than normally required by variational tech¬
niques, vector space methods may also provide more powerful analyt¬
ical tools.
1.3 Historical Development:
Optimal Dynamical Beams
The title of this section is a misnomer. On the basis of the
related literature, a more realistic terminology would be "Optimal
Ouasistatic Beams." Papers dealing with the optimization of dynamical
beams ultimately contain some assumption or given condition that effec¬
tively transforms the problem to an equivalent static case. Simple
harmonic motion is frequently assumed to remove time dependence from the
governing equation. For example, Barnett (1963b) minimizes the tip of
deflection of a cantilever beam accelerating uniformly upwards, with the
total weight of the beam specified. The optimal solution for cross-
sectional area distribution is specified by a nonlinear integral equa¬
tion solved by successive approximations. Dynamics enters the problem
only as a time invariant inertia load which converts the problem to a
static beam subject to a body force.
Niordson (1964) finds the tapering of a simply supported beam
of given volume and length which maximizes the fundamental frequency of
free vibration. Assuming that all cross sections are similar, Niordson
expresses the desired frequency as the Rayleigh quotient. This is
obtained by the equation for spatial dependence associated with the
usual separation of variables technique. In this case it is assumed

13
that deflection, shear, and rotational inertia effects are small.
Solution of the conditions for optimality were obtained by successive
approximations. This approach results in a problem identical in form
to the eigenvalue problem associated with maximizing the Euler buckling
load of a column.
A very specialized problem is treated in Brach (1968). Cross-
sectional area and stiffness are proportional, and have both upper and
lowed bounds; material properties and length of the beam are specified
constants. The object is to make the fundamental frequency of free
vibration stationary, without any constraints on the weight. Instead
of using the Rayleigh quotient, Brach uses the total potential energy.
His solution method is ad hoc and not generally useful. A more useful
approach is described by Icerman (1969) for structures subject to per¬
iodic loads. Necessary and sufficient conditions for minimum weight
are obtained from the principle of minimum potential energy. Amplitude
and frequency of the applied load is specified as well as "dynamic
response," defined as the potential energy associated with the load
amplitude displaced a distancesqual to the displacement amplitude at
the point of application. It is also required that the load's frequency
be less than the fundamental natural frequency. Subject to these con¬
straints, the structure's weight is to be minimized. Trusses and
segmentwise constant approximations are also treated. Once again a
dynamical problem is effectively transformed into a quasistatic problem.
Realistic treatment of the dynamic response of beams subject to
time dependent loads is given in Plaut (1970). An upper bound on the

14
response of the beam is obtained from the "largest possible displace¬
ment of the beam under a static concentrated unit load." This inequal¬
ity is determined from the time derivative of the total energy of the
beam and the Schwarz inequality. Plaut minimizes the upper bound on
the response for specified total weight and relationship between specific
mass and stiffness. Despite the consideration of a truly dynamical
problem, this approach has two weaknesses indicated by the author.
First, the upper bound need not be close to the exact answer; secondly,
there is no demonstration that minimizing the upper bound also minimizes
the actual response of the beam subject to dynamic loading.
Another paper worthy of mention is Brach and Walters (1970) .
They maximize the fundamental natural frequency which is expressed as
a Rayleigh quotient that includes the effect of shear. Standard varia¬
tional methods are employed to derive necessary conditions, but no
solutions nor examples are given. The authors do, however, suggest
using the method of quasilinearization. This paper is another example
of a quasistatic, dynamical beam problem.
1.4 Scope of the Dissertation
This dissertation is primarily concerned with the application
of Pontryagin,'s Maximum Principle to problems in structural optimization.
Only elastic materials are considered; however, various types of con¬
straints are treated.
The theoretical development of the dissertation pertains to
problems described by an ordinary differential equation but is based
upon a numerical technique normally used with systems described by

15
a finite number of discrete quantities. For this reason, examples are
included for both types of systems.
No new solution techniques have been developed for the nonlinear
two point boundary value problems which characteristically arise in
optimization problems for continuous systems. Solutions are obtained
by a standard quasilinearization method. However, a modified "feasible
direction" numerical algorithm for use with discrete systems is
described and an example included to demonstrate its operation. This
serves to illustrate the application of the theory on which the algo¬
rithm is based to the theoretical development associated with contin¬
uous systems. Furthermore, it provides a comparison between the
solution to a problem described as a continuous system, and alternately
by a discrete approximation.
Additionally, it is the intent of this dissertation to replace
some of the confusion in classification of problem types contained in
the various survey papers with an organization based upon mathematical
attributes. The result is a logical approach to the formulation of
optimization problems for elastic structures.

CHAPTER II
GENERAL PROBLEMS AND METHODS
IN STRUCTURAL OPTIMIZATION
2.0 Introduction
The first chapter presents a broad view of structural optimiza¬
tion and the historical development of two general types of problems
that are used as examples in later chapters. Before the mathematical
theory is developed in the next chapters, general problems and methods
in structural optimization itself are briefly outlined in this chapter.
The classification of problem types vis-h-vis mathematical attributes
is discussed first. This is followed by short descriptions of the
major methods of structural optimization for both continuous and dis¬
crete systems. Appropriate references are cited for each section.
2.1 Problem Classification Criteria
Perhaps the major source of difficulty in classifying struc¬
tural optimization problems lies in the translation from the physics
involved to a mathematical representation. A single physical concept
when transformed to mathematics may become more than one mathematical
attribute. For example, consider the class of conservative problems.
Since energy is conserved this immediately prohibits dissipative mate¬
rials, time varient constraints, and nonholonomic constants. More
important, consider the following statements from Lanczos (1962,
P. 226):
16

17
. . . all the equations of mathematical physics which
do not involve any energy losses are deducible from
a "principle of least action," that is the principle
of making a certain scalar quantity a minimum or a
maximum . . . all the differential equations which are
self-adjoint, are deducible from a minimum-maximum
principle . . . and vice versa.
However, it is shown in Chapter V that to be self-adjoint systems places
requirements on both the differential equation and the boundary condi¬
tion. Thus, the single physics attribute of being a conservative sys¬
tem is described mathematically by expressions involving the cost func¬
tional, the governing system of differential equations, boundary condi¬
tions, and constraints.
As a result of this lack of similarity in descriptions, a choice
must be made as to which realm will be used for classification of prob¬
lems. Since all problems are ultimately transformed to mathematics,
mathematical characteristics are selected as the criteria. On the basis
of survey paper contents and the many related papers, it is felt that
the proper (not necessarily the best, nor all inclusive) characteristics
for classification of problem types are:
(i)cost functional
(ii)system equation and boundary conditions
(iii)control constraints
(iv)behavioral constraints (state and/or control constraints)
Subsequent discussion is in terms of these four characteristics.
Exceptions to these descriptors are readily acknowledged, e.g., whether
the problem is deterministic or probabilistic. An example of the latter
may be seen in Moses and Kinser (1967). These exceptions do not serve

18
as negating counterexamples but instead indicate the requirement for
additional descriptors and verify the difficulty of the task, suggest¬
ing the need for further comprehensive study.
2.1.0 Problem Classification Guidelines
In the following sections each criterion is briefly discussed.
Some of the various characteristics of each are mentioned, and where
appropriate references exist the citation is given. It must again be
emphasized that the following is not all inclusive; it is an attempt to
categorize the types of problems existing in the literature according
to the four mathematical descriptors postulated. Moreover, the descrip¬
tors are not discussed in the order given but instead are treated in
the order normally encountered during a problem solution.
2.1.1 Governing Equations of the System
The immediate question to be answered is whether the structural
system is described by a set of continuous functions or a set of dis¬
crete constants. Bending deflection of simple structural elements is
an example of the former; design of trusses is a good example of the
latter. In general, variational techniques are employed with contin¬
uous systems while mathematical programming techniques are most fre¬
quently applied to the discrete systems. However, variational methods
can be applied to the approximation of continuous systems by discrete
elements. This usually takes the form of either a finite element or
"segmentwise constant" approximation of the continuous structure.
References treating the different systems described above are included
in the section on methods.

19
2.1.2 Constraints
Two of the descriptors are postulated to be control constraints
and behavioral constraints. A further consideration is whether the
constraint is defined by an equality relationship or by an inequality
expression. Equality constraints are handled by a long-known technique
entitled "isoperimetric constraints." Valentine's (1937) work is known
to contain the initial development of a technique for converting inequal¬
ity constraints to equality constraints. The introduction of slack
variables increases the number of variables in the problem to be solved
but at the same time permits all of the isoperimetric techniques to be
used. A detailed application of this approach is presented in Appendix B.
It should also be noted that isoperimetric constraints are sometimes
referred to as "accessory or subsidiary conditions."
Most real structural optimization problems possess an isoperi¬
metric constraint as well as inequality constraints dependent upon the
control _u and/or the state x of the structural system. Typically the
control constraints are the result of geometrical limitations or
restrictions to the types of available materials. Behavioral constraints
are related to the state of the system and may depend solely upon the
state (of deformation), or in the case of most stress constraints,
jointly upon the state and control variables. With this distinction the
constraints may be classified vis-á-vis the two criteria and optimal
control characteristics as follows:
By convention all vectors are column vectors unless indicated otherwise.

20
(i)
uncons trained
(ii)
o
VI
/—s
-0-
(iii)
$ (x, u) < 0
(iv)
(x) < 0
control constraint
behavioral constraints
All of this discussion pertains to both continuous and discrete systems.
No references for unconstrained optimization problems are given.
It may be that in some problems the unconstrained structural optimiza¬
tion solutions have either infinite stiffness and finite weight, or
finite stiffness and zero weight. A discussion of this can be found in
Salinas (1968, pp. 23-26).
Investigation of control constraints led to the development of
the maximum principle. Although reserving discussion of the method for
a later section, the classical reference detailing the derivation of
the principle is given here for completeness. Rozonoér (1959) treats con¬
trol constraints but only as related to the development of the maximum
principle.
Most of the literature concerns either bounded control problems
or a more general form of constraint which can be classified as a behav¬
ioral constraint. The latter is a mixed constraint which depends upon
both the state and control variables. Breakwell (1959) is a lucid paper
dealing with this type of constraint. References which treat state
and/or mixed constraints are Bryson et al. (1963), and Speyer and
Bryson (1968). Constraints which depend upon only the state are not
treated in this dissertation; an example of such a constraint is to
determine the optimal solution for some problem subject to an upper
bound on deflection of the structure at any point.

21
2.1.3 Cost Functionals
There are two basic types of cost functionals that occur in the
field of structural optimization. They were first identified by Prager
(1969) but not for the proper reasons. Using Prager's notation, they are
J = Min / F(ip) dV
V
/ G0I0 dV
V
JQ = Min
/ H(i|>) dV
V
where F, G, and H are scalar functionals of ifi. The latter functional
represents a Rayleigh quotient associated with an eigenvalue problem.
It can be reduced to the first type of functional shown above by choos¬
ing a normalization of the eigenfunction such that the numerator
equals unity for all admissible variations (see section 5.3). This
normalization is thereafter treated as a subsidiary’ constraint.
What actually distinguishes the second functional from the
first is not that the functional is a quotient, but that the extrem-
ization of an eigenvalue requires a dual extremization (see section 5.3).
In terms of a state x and control u, the fundamental eigenvalue is
given by minimization of the Rayleigh quotient with respect to the
eigenfunction x, or
= Min
/ G(x,_u) dV
V
^ ü / H(x,u) dV
V
where ii represents some specified design parameter. If the desired
result is to maximize the cost with respect to all admissible _u, it
is observed that a second extremization is required; for example, see

22
Keller and Niordson (1966). Thus, a more appropriate manner of classi¬
fying cost functionals is on the basis of whether the problem statement
implies a single or a double extremization. Hence the two basic types
of cost functionals encountered in structural optimization are
J = Min ( F(x>Jj) dV
u V
/ G(x,_u) dV
V
J = Max Min——■ - ■■ ■ ■———
• u x / H(x,u) .dv
V
where _x must satisfy an equilibrium condition of the state, and u is
subject to some admissibility requirements.
There is a special case related to these two in which the weight
is to be minimized for a specified eigenvalue. This problem is
treated in Icerman (1969) with a mathematical discussion of such a
variational problem presented in Irving and Mullineux (1959, p. 394).
In terms of the two cost functionals, the special case is
J = Min J (G(x,u) - d^lKxju)} dV
u V ^
where is a specified constant. This approach is frequently employed
in eigenvalue problems to avoid the inherent difficulties associated
with the dual extremization problem.
There have also been many papers published that consider
"multi-purpose structures," e.g., Prager (1969), Martin (1970), and
Prager and Shield (1968). The cost function for such problems is
defined as

23
k
J = Z a.J. (_x>iO
i=l 1 X
where a_^ are positive constants, serving as weighting parameters.
While perhaps demonstrating much potential, no significant results
obtained with this approach have so far been published. What problems
have been solved are too simple; indeed the authors indicate the need
for using a discrete approximation and mathematical programming tech¬
niques in realistic applications.
A subject closely related to "multi-purpose structures" is
that of multiple constraints. It is mentioned here only because most
papers on the latter also include the former—see Martin (1971). The
idea of multiple constraints is not new; in both variational and math¬
ematical programming fields there exist standard techniques for han¬
dling multiple constraints.
A recent Russian paper (Salukvadze, 1971) suggests an alter¬
nate to the "multi-purpose" cost functional. Instead of treating a
vector functional that requires the choosing of weighting coefficients,
it is suggested that the several functionals be combined into one.
Given a system and vector cost functionals
x = f_(x,u, t)
J [u] = J^0x,u,t) i = l,...,k
Let uiP denote the optimal solution for which J. assumes the optimal
—OPT ^ i F
value on the trajectory of the system. For each of the J_^ there is a
different These k values J. can be thought of as components of
—OPT i

24
* t
a vector r where
For any arbitrary ui the result
rjuj = { J [u] ... Jk[u]}T
is just some vector functional.
k
Vector _r represents a constant point in the space of
(J , ...,J. ). Since no choice of u can optimize all of the J. simul-
1 k — i
*
taneously, that is, to attain the point _r in J^-space, the best alter-
k
native is to minimize the distance between _r[u] and r_ . That distance
is defined by the Euclidean norm. To avoid the question of incon¬
sistent dimensions the functionals are reduced to dimensionless form.
Thus,
J[u] =
k
Z
i=l
i[Hl -
J.
1
' (i)
-HIPT
3
and u.__ is that function u which minimizes the functional J[uj.
—OPT — —
Mathematically speaking,
-opt = ARGMIN
This type of vector cost functional is much more appealing
than the type treated in the papers on multi-purpose structures.
It also suggests an entirely new field of study: the more realistic
f T
Superscript T , e.g., ju , denotes the transpose of the vector tu

25
choice of cost functionals. The mathematics of a problem seldom
accommodates financial considerations. For example, the design which
requires the least material may reduce the cost of materials at an
overwhelming expense in manufacturing or fabrication. When aesthetic
appeal and environmental impact are included—as must be done in any
real, commercial application—the selection of an appropriate cost
functional is an almost insurmountable task. However, a simple exten¬
sion of Salukvadze's composite cost functional may reduce the difficul¬
ties to operations research considerations.
In problems where it is desired to optimize simultaneously
several different functionals, not all having the same dimensions, the
concept of a generalized inner product may prove useful. It is defined
in terms of a metric operator A; for some general vector _z
||_Z || 2 = (_Z, z)A = z^kz
and symbol "=" means "is defined by." With reference to the vector
cost functional, A represents a set of scale factors which converts
all of the separate cost functionals to a common dimension. This is
where the operations research enters—relating material expense to
fabrication to sociological considerations and so forth—to determine
the metric A. For a vector e_ whose elements are functionals,
e_ = - jr
where r[u] and r are defined above, the composite cost functional is
J t u] = (c,Ac)

26
The weakness in this method is that an optimal solution must be
obtained for each individual cost functional prior to attempting a
solution to the composite problem. Additionally, some of the more
abstract cost objectives may be difficult to quantify in a meaningful
manner. Despite these shortcomings this approach does suggest inter¬
esting applications.
2.2 Methods: Continuous Systems
The problems characterized by mathematical functions, in con¬
trast to those represented by a set of discrete constants, are normally
treated by variational techniques. Many books on this subject have
been published; the better authors include Elsgolc (1961), Gelfand and
Fomin (1963), Dreyfus (1965), Hestenes (1966), Denn (1969), Luenberger
(1969), and Bryson and Ho (1969). An excellent summary paper is avail¬
able in Berg (1962).
To see how these techniques are applied, three papers are
recommended. The first is Blasius (1913), which provides sufficient
detail and explanation to make it quite worthwhile. Although it does
include several examples that involve subsidiary conditions, no inequal¬
ity constraints are treated. An example that includes inequality con¬
straints to the control variable is coontained in Brach (1968). A much
more general application of variational principles is presented in
Oden and Reddy (1974). In this paper a dual-complementary variational
principle is developed for a particular class of problems. It is shown
that the canonical equations obtained are the Euler-Lagrange equations
for a certain functional.

27
2.2.0 Special Variational Methods
Besides the ordinary variational method, more specialized
techniques have been developed to the point where they are recognized
as independent methods in their own right. In the following sections
these methods are identified and a number of representative references
given.
2.2.1 Energy Methods
The oldest of these methods is the energy method. It originated
with the principle of minimum potential energy, and was later extended
to include the representation of eigenvalues through the energy func¬
tional. A good discussion of the former is available in Fung (1965) or
Przemieniecki (1968); the best general treatment of the 3_atter is avail¬
able in either Gould (1957, Chapter 4) or in Mikhlin and Smolitskiy
(1967, Chapter 3).
The principle of minimum potential energy is frequently used
with simple problems to prove that a necessary condition for optimality
is also sufficient. Prager and Taylor (1968) contains such a proof
for the global, maximum stiffness design of an elastic structure of
sandwich construction; two papers that also consider this problem are
Huang (1958) and Taylor (1969). Specific application of the energy
method to an eigenvalue problem is demonstrated in Taylor and Liu (1968).
A much more general discussion of the energy method is provided in
Salinas (1968). Further extensions of the method are presented in
Masur (1970), in which the principle of minimum complementary energy
is applied to problems of the optimum stiffness and strength of elastic

28
structures. In these problems a necessary condition for optimality
is that the strain energy density be constant throughout the structure.
This condition is also sufficient for optimality in certain classes of
structures that satisfy a specific relationship between the strain
energy density and design variables.
Many of the energy problems belong to the class of problems
having quadratic cost functionals. The significance of this character¬
istic is that the Euler-Lagrange equations derived from such functionals
are linear.
A more recent energy method development is the concept of
"mutual potential energy." Mutual potential energy techniques resemble
those of the principle of minimum potential energy. In both methods
a cost functional is defined over the entire domain occupied by the
structure and optimized with respect to the control variable. If it
is desired that the optimal solution be required to have a specified
deflection at a certain point, this condition corresponds to a sub¬
sidiary state constraint when using the principle of minimum potential
energy. The mutual potential energy method incorporates this type of
localized constraint into the cost functional which is defined over
the entire domain of the structure. By itself this alone is advanta¬
geous; however, for certain types of problems the mutual potential
energy method also provides both a necessary and sufficient condition
for global optimality. In the way of a critical comment, either the
method has received too little attention, or else it does not effi¬
ciently handle problems more difficult than the simple examples presented.

29
Four papers that are representative of the literature associated with
this method are Shield and Prager (1970), Chern (1971a, 1971b), and
Plaut (1971c).
Another recent development is that the class of problems for
which energy methods is applicable has been expanded to include cer¬
tain types of nonconservative systems. Together with the mutual poten¬
tial energy concepts, this suggests that perhaps the classical energy
method is a special case of a more generalized method. If a technique
can be developed which uses the adjoint variables to transform a gen¬
eral nonconservative system into an equivalent self-adjoint form, the
method might be deduced. Some papers pertaining: to the subject are
Prasad and Herrmann (1969), Wu (1973), and Barston (1974).
2.2.2 Pontryagin's Maximum Principle
There are many textbooks which derive, explain, and give examples
for the maximum principle. The original (Pontryagin et al., 1962)
requires a knowledge of functional analysis. A condensed form of this
same material is available in Rozonoér (1959). Denn (1969) provides
another point of view in which the principle is derived from Green's
functions. In this manner, the sensitivity to variations is readily
observed. To understand Denn's treatment requires only a knowledge of
the solution of differential equations.
Shortly after Pontryagin's book was published, many papers
devoted to the theoretical aspects of the maximum principle were pub¬
lished. Some of the more readable ones are Kopp (1962, 1963), Roxin
(1963), and Halkin (1963). Another early paper (Breakwell, 1959)

30
appears to be a completely independent derivation of the maximum prin¬
ciple. Although quite general in the mathematical sense, the examples
presented are trajectory optimization problems and not a general type
of mathematical problem. This may be an explanation for what seems to
be a lack of recognition for a significant achievement.
The application of PMP to problems in structural optimization
is relatively recent. When the method is used, one of two difficulties
is often encountered. The first is errors in the formulation of the
optimal control problem; the second is that once a well-posed, non¬
linear two-point boundary value problem (TPBVP) is obtained, it is dif¬
ficult to solve. An example of the first is provided by Dixon (1968)—
the correction was given in Boykin and Sierakowski (1972). De Silva
(1972) provides a clear presentation on the application of PMP to a
specific problem, but includes no data because a solution could not be
obtained. Despite the failure to determine the solution, this paper
is worthwhile for its lucid discussion of the PMP application. Another
paper that gives a good specific application of PMP is Maday (1973).
Although much analysis is presented very little is said regarding the
solution techniques.
All of the above references are applicable only to systems
that are described by ordinary differential equations, in contrast with
the calculus of variations which also handles problems described by
partial differential equations. Since many of the problems of mathe¬
matical physics involve partial differential equations, an extension of
PMP to include this class of problems is the next logical development.

31
Some work has already been done, for example, Barnes (1971) and
Komkov (1972). A survey of these "distributed parameter systems"—
see Section 1.0—is presented in Wang (1968).
2.2.3 Method of Steepest Ascent/Descent
This method is frequently cited in the literature for trajec¬
tory optimization, and occasionally in references related to optimal
structures. When the method of quasilinearization converged for the
dissertation example problems, there was no need to investigate other
methods such as the method of steepest ascent. Consequently, little is
said about it. According to the references, it is applied in a straight¬
forward manner. Furthermore, the example problem solutions presented
seem to be real problems and not academically simple. The following
four papers treat the method in general with trajectory optimization
applications: Bryson and Denham (1962, 1964), Bryson et al. (1963), and
Hanson (1968). In Haug et al. (1969), the method of steepest ascent is
derived in detail, completely discussed, and compared to the maximum
principle. Several structural optimization problems are then solved
by the method of steepest ascent. Although no exciting results are
obtained, use of the method is clearly illustrated by the applications
to realistic structural problems.
2.2.4 Transition Matrix:
Aeroelasticity Problems
For the past few years a group at Stanford University has studied
the optimization of structures subject to dynamic or aerodynamic con¬
straints. The general problem of their interest is that of minimizing

32
the weight of a given structure for specified eigenvalue, subject to
inequality constraints on control.
Three types of solution techniques are used once the necessary
conditions for minimum weight are determined. Exact solutions are
obtained for most of the problems because they are so simple that
analytical methods are applicable. More complicated problems are
solved by a "transition matrix" method described in Bryson and Ho (1969) .
On the basis of convergence difficulties reported in the references,
this method should be used with caution. Results have been obtained
only for very simple problems. However, these results are corroborated
by data obtained from a discrete approximation method. Five papers
that are representative of this work are McIntosh and Eastep (1968),
Ashley and McIntosh (1968), McIntosh et al. (1969), Ashley et al. (1970),
and Weisshaar (1970).
2.2.5 Miscellaneous Methods
The preceding sections have briefly outlined the methods of
structural optimization most frequently encountered in the literature.
Appropriate, representative references have also been given. Not all
methods are listed; while some are omitted for not being generally use¬
ful, others are omitted for not being generally used. Two examples of
the latter are the "modified quasilinearization" and "sequential gradient-
restoration" algorithms described in Miele et al. (1972) and in Hennig
and Miele (1972). At some later time these methods may be acknowledged
as major methods that are applicable to many different or important prob¬
lems, but for now they are mentioned only in passing.

33
2.3 Methods: Discrete Systems
Discrete systems are described by a set of discrete constants
instead of the set of functions associated with continuous systems.
The classic example of a discrete system is a pin-connected truss, where
the state x. and control u. are the stress and cross-sectional area,
i i
respectively, for each member i. A discrete system also arises in the
approximation of a continuous system.
Several references that present a good discussion of general
methods applied to discrete systems are available. Most of these exist
in the form of an edited collection of papers by various authors on the
topics of their acknowledged expertise. Four such publications are
Gellatly (1970), Gellatly and Berke (1971), Pope and Schmidt (1971),
and Gallagher and Zienkiewicz (1973). Another report, Melosh and Luik
(1967), provides a good exposition of the difficulties associated with
the analysis portion of least weight structural design. It also con¬
tains a brief comparison of various mathematical programming methods.
McNeill (1971) is the last reference to be cited in the section
on general methods for the optimization of discrete systems. Minimum
weight design of general structures is treated in a mathematically
precise formulation. Legendre's necessary condition is combined with
the concepts of convex functions and sets to derive the necessary and
sufficient conditions for global optimality. Fully stressed designs
and constraints to eigenvalues are also discussed. In summary, this
paper orovides a good example of the general mathematical problem that
must be solved in the optimization of discrete systems.

34
While certain variational methods may be applied to discrete
systems, the most frequently used technique is mathematical programming
In the following sections, this method and other major methods are dis¬
cussed and representative references cited.
2.3.0 Mathematical Programming
The general method of mathematical programming is discussed
in section 3.2 of the dissertation, and the solution of an example
problem using this method is detailed in Chapter VL. In the literature
related to this subject, a very readable textbook is available—Fox
(1971). This book complements the theory with numerous discussions
pertaining to numerical techniques and methods that can be employed to
overcome certain difficulties that may arise. Although it does contain
flowcharts of several algorithms, there are few specific examples given
For a discussion of the general theory, two alternatives to this book
exist in the form of papers: Schmidt (1966, ]968). The first is
written in a conversational style, contains no mathematics, and is
intended to provide only a general description of the subject. The
latter paper is theoretical in content.
An excellent application to a realistic problem is to be found
in Stroud et al. (1971). Ths paper contains little discussion of the
method itself, but does demonstrate an application that allows a concep
tual visualization of the solution. The approach is to assume the
solution to be a linear combination of specified functions, and to
choose the weighting coefficients to minimize the cost. Mathematical
programming is employed to determine the optimal set of coefficients.

35
This approach resembles Galerkin's method, and though not mathemat¬
ically rigorous, it may provide a useful approximation to large,
unwieldy problems.
2.3.1 Discrete Solution Approximations
In the previous section a paper is cited that contains an
approximate solution obtained by Galerkin's method. The use of the
Galerkin or Rayleigh-Ritz approximate solution techniques is suffi¬
ciently widespread to be considered a general method. For both methods,
the solution is assumed to be a linear combination of the solution to
the linear part of the governing equation and a set of prescribed func¬
tions. This approximate solution does not satisfy the given equation
exactly but produces some residual function. A cost function that
depends upon the residual is then minimized with respect to the unknown
coefficients. The two weighted residual methods mentioned above have
different cost functions, but the methods are identical for linear
equations—see Cunningham (1958, p. 158).
The advantage to using these methods is that after assuming
the particular form of solution, the problem of solving for the weight¬
ing coefficients may be much simpler than the original problem. In the
case of Stroud et al. (1971), the coefficients were obtained by mathe¬
matical programming techniques. However, the weakness of the method
is the restricted function space of possible solutions. With the coef-
ficents obtained by these methods the resulting solution is the best
approximation that is possible from the set of solution functions

36
prescribed. There is no guarantee that the approximation even resembles
the true solution.
The flutter of a panel is solved using Galerkin's method in
Plaut (1971a). No general developments are presented and the assumed
solution functions are trivially simple. However, this paper does
provide an application of the method to obtain an approximate solution
to a very difficult optimization problem involving the stability of
a nonconservative system. A similar problem is treated in a more theo¬
retical manner in Plaut (1971b) using a modified Rayleigh-Ritz method.
"Segmentwise-constant" control functions are assumed also; this partic¬
ular approximation is discussed with more detail in the following sec¬
tion. Additional nonconservative problems are treated in Leipholz
(1972), applying Galerkin's approximate solution to the energy method.
Several simple examples are included.
Nonconservative elastic stability problems of elastic continua
are treated in Prasad and Herrmann (1969) using adjoint systems. This
approach is more realistic than the segmentwise-constant control assump¬
tion described in the following section. Solutions for the state and
adjoint system are assumed, such that approximation process resembles
the Rayleigh-Ritz method. However, only a single type of nonconserva¬
tive system is considered. Extension to several other types of noncon¬
servative elastic continua problems is given in Dubey (1970). Varia¬
tional equations corresponding to both the Galerkin and Rayleigh-Ritz
methods are derived. Furthermore, the condition for equivalence of the
two methods is shown to be that the admissible velocity field must

37
satisfy a natural boundary condition over that portion of the body’s
surface where tractions are prescribed.
2.3.2 Segmentwise-Constant Approximations
The definitive characteristic of this method is approximating
the structural system by a number of discrete segments, where within
each segment the control function has a constant value. In general,
the constant value of the control differs from segment to segment. For
the many papers on this method that have been published, the procedure
is the same. An optimality condition (necessary in all cases but also
sufficient in some) or cost functional is derived for the continuous
system. After defining the segmentwise-constant approximation, the
condition or functional is reformulated in terms of the discrete param¬
eters. Most of the papers use so few elements that solving for the
discrete values of the control parameter poses no difficulties. Although
this method does simplify the mathematical problem to be solved, the
crudeness of the approximation is not appealing. Five papers which
treat a variety of problems using this approximation are cited below.
Minimum weight of sandwich structures subject to static loads
is discussed in Sheu and Prager (1968a). In Sheu (1968) the same type
of structure is considered. It differs from the first problem by
requiring point masses to be supported such that the total structure has
a prescribed fundamental frequency of free vibration. Icerman (1969)
treats the problem of elastic structures subject to a concentrated load
of harmonically varying amplitude. The minimum weight design is obtained

38
subject to a compliance constraint related to the applied load, and
which is effectively a boundary condition on displacement at the point
of application. A truss problem is also included.
The concept of a compliance constraint is pursued further in
Chem and Prager (1970). The minimum weight design for sandwich con¬
struction beams under alternative loads is found, subject to this type
of constraint. The paper uses up to eight segments, thereby obtaining
a more realistic approximation to the continuous problem. Minimum weight
design of elastic structures subject to body forces and a prescribed
deflection is discussed in Chern (1971a). This investigation is no¬
table. in that it considers applied loads that are functions of the design
functions.
2.3.3 Complex Structures with
Frequency Constraints
On the basis of useful application, perhaps the most important
class of discrete structural optimization problems is the minimum weight
design of complex structures subject to natural frequency constraints.
Since most real structures are built with many structural elements of
various types, and are not realistically described by any single type,
this approach is more appropriate from the aspect of modeling the struc¬
ture. Furthermore, many structures must be designed to avoid certain
natural frequencies because of resonance or self-induced oscillations;
this situation indicates that the natural frequency constraint is also
appropriate.
Many different solution schemes have been developed which are
usually based upon general mathematical programming techniques.

39
Typically, a design is iteratively altered to minimize the weight with
a subsequent increase in frequency until a constraint is violated.
At that point the design process uses an iteration which simultaneously
reduces both weight and frequency. These two processes are repeated
sequentially until no further weight reduction is possible.
Although circumstances may require the use of many elements,
the number of them may itself be a critical factor. Some of the schemes
require a matrix inversion as part of the eigenvalue problem solution
associated with the frequency constraint. If the number of elements
becomes too large, the size of the matrix to be inverted likewise be¬
comes excessively large. When that occurs the matrix inversion can
require excessive amounts of computer time. Another possible difficulty
is that the inverse matrix itself is not sufficiently accurate, such
that the subsequent calculations are not acceptable. However, for
structures such as reinforced shells composed of different types of
structural elements, this method may be the most applicable.
Many papers have been published pertaining to this class of
structural optimization problem. Because the method is inherently
oriented towards applications, the references are cited in chronolog¬
ical order without additional comments. Interested readers are referred
to: Turner (1967), Zarghamee (1968), Turner (1969), De Silva (1969),
Rubin (1970), Fox and Kapoor (1970), McCart et al. (1970), and Rudisill
and Bhatia (1971).

40
2.3.4 Finite Element Approximations
There is an unfortunate ambiguity to the label "finite elements"
that occurs because these words are used to describe two completely
different entities. In papers cited in the preceding section they are
used to indicate the discrete structural elements of finite dimensions
which comprise the complex structure. The analysis of such systems of
structural elements has been accomplished by ordinary matrix methods
during the last three decades. However, during the past decade another
method has been developed and named "the finite element method."
In this method a continuum is divided into small, finite ele¬
ments over which a particular form of approximation of either the dis¬
placement and/or force is assumed. A number of nodes common to one or
more element is prescribed; continuity is required to exist at these
nodes but not necessarily elsewhere. An equilibrium equation is derived
for each element, and then all of the individual equations are combined
into a single equilibrium equation for the entire system. The result¬
ing equation is a linear algebraic equation whose unknowns are displace¬
ments and/or forces at the nodes. Once the matrix equation is inverted,
the nodal displacements and/or forces are used with the assumed approx¬
imation form to describe the state of the structure throughout each and
every element, and hence the system. Hereafter this method is referred
to as the "finite element method."
The most frequent application of the finite element method is
to problems having complicated loads, geometry, and response. Generally
speaking, the method is employed wherever the physical system is too

41
complex to be described adequately by a single differential equation
and boundary conditions. For a complete theoretical development of the
finite element method and numerous examples, see Zienkiewicz (1971).
With respect to structural optimization the method is employed
to simplify the problem to be solved. Very little has been published
on this subject, but the papers available cover a wide spectrum of tech
ñiques. For example, Dupuis (1971) combines the finite element and
segmentwise-constant methods as applied to minimum weight beam design.
A similar application to column buckling is contained in Simitses et al
(1973). Another paper, Wu (1973), is a study of two classical noncon¬
servative stability problems. Although adapted to stability consider¬
ations, this presentation is the best exposition available in the open
literature.
In Chapter VI a minimum deflection beam problem is solved with
the combined methods of finite elements and mathematical programming.
2.4 Closure
In the preceding sections of this chapter, general problem
types and methods are discussed. Only those methods that appear to
have attained some standard of acceptance are presented. It must be
acknowledged that other areas of important study exist but are perhaps
overlooked as not being pertinent to the general subject area of the
dissertation. As an example, Dorn et al. (1964) treats the optimal
layout of trusses—an important subject but not related to the general
problem to be considered in this dissertation. In addition only
elastic structures have been considered although there are numerous

42
publications on optimal design of inelastic structures. References
that are representative of this subject are: Drucker and Shield
(1957a, 1957b), Hu and Shield (1961), Shield (1963), Prager and Shield
(1967), and Mayeda and Prager (1967).
On considering the various references mentioned above it would
appear that there are two possible pitfalls in structural optimization
that should be avoided. The first is the confusing of method of opti¬
mization with the solution techniques employed to obtain a solution to
the resulting TPBVP. In order to avoid possible errors the two should
be dealt with independently, unless it is clearly advantageous to
relate one to the other. Besides this it must be recognized that any
solution obtained is "optimal" only with respect to the given condi¬
tions of the particular problem. Any change in the problem statement
invalidates the applicability of that solution. The change may lead to
a more desirable solution, but the original solution is no less valid.
Simitses (1973) is an example where this situation is not acknowledged.
In this paper the thickness of a thin reinforced circular plate of spec¬
ified weight and diameter is determined such that the average deflec¬
tion due to a uniform load is minimized. An earlier paper which did
not include stiffening is cited with the implication that the optimum
solution for the unstiffened plate is not correct. The point made above
is that both of these solutions are optimum under the respective condi¬
tions of the two problems. Neither solution is more, or less, valid
than the other.

CHAPTER III
THEORETICAL DEVELOPMENT
3.0 Introduction
This chapter contains the development of two distinct methods
used in the theory of optimal processes, into a more general method.
The first section defines precisely the problem to be considered.
This includes the necessary conditions for an optimal solution given by
the calculus of variations. Several mathematical programming techniques
are described in the second section along with a numerical algorithm
called the gradient projection method. The application of this numer¬
ical method to the solution of the necessary conditions from Pontryagin's
Maximum Principle (PMP) is detailed in Section 3.3. Results of this
approach are shown to be consistent with the necessary conditions, given
in Section 3.1; these results provide a clarifying insight to the math¬
ematical processes entailed in the maximum principle, and an explicit
formulation for the Lagrangian multiplier functions. This explicit for¬
mulation is used in the next section to show the necessary conditions
may then be regarded as an algorithm. The final section contains a brief
summary of solution methods.
The main theoretical development of the dissertation is contained
in the first three sections. It is well known that the problems encoun¬
tered in the calculus of variations are equivalent to the optimization of
43

a functional (in the sense of mathematical programming problems) under
certain restrictions upon the variations. A good exposition of this is
available in Luenberger (1969). With this equivalence in mind, it is
noted that the PMP is itself worded as a constrained optimization prob¬
lem. When treated with what is normally regarded as a numerical method,
the gradient projection method, an explicit formulation of the atten¬
dant Lagrangian multipliers is obtained. This form satisfies all of the
calculus of variation necessary conditions and allows one to use them
in a most straightforward fashion. As a result, these necessary condi¬
tions may be directly used in the form of an algorithm to obtain a solu¬
tion. Furthermore, it is believed that treating the PMP as a mathemat¬
ical programming problem in conjunction with the gradient projection
method helps to explain the effect of combined control-state constraints
upon the maximum principle.
3.1 Problem Statement and
Necessary Conditions
A general problem which represents a large class of structural
optimization problems is treated in the sequel. The functional
tF
J = / L (x,u) dt (3.1.1)
0
is to be minimized with respect to the control u.(t) where the state
x(t) must satisfy certain boundary conditions and a differential con¬
straint; in addition, an inequality constraint involving both the state
and control must be satisfied. For

45
uT(t) = [u (t) u„(t) ... u (t)] (3.1.2)
— l ¿ m
xT(t) = [x^Ct) x2(t) ... *n(t)] (3.1.3)
the subsidiary conditions to minimizing the cost function J are:
x = f_(x,_u) (3.1.4)
Specified Boundary Conditions on x(tz) (3.1.5)
(J>£ (x,u) <0 l = 1,. ..,q (3.1.6)
Terminal time t is considered to be constant; allowing it to be
r
unspecified requires only a slight modification to the following
derivation.
This problem is a particular form of a very general one treated
by Hestenes (1966). His results are a set of necessary conditions which
must be satisfied by the optimal solution and include the maximum prin¬
ciple. To obtain the necessary conditions, the inequality constraints
are converted to equality constraints in the manner of Valentine (1937).
These constraints and the differential constraints are then adjoined to
the cost function via Lagrangian multiplier functions p (t) and p.(t)
A/ 1
respectively.
<^£(x,u) + s^(t) = 0
J.
where the slack variables are defined such that
The symbol "="
denotes "is defined by."

46
s£(t) = [-£(x,u) ] 2 > 0
tp tp t
J = / Ln(x,u)dt + / p (t) [x-f(x,u)]
0 U 0
tF
+ / P£(t) [uA(x,u) + s~(t)] dt
dt +
Implied summation convention is used whenever a vector formulation leads
to possible ambiguities in later developments. Integrating the second
integral by parts gives a result that leads to the variational Hamiltonian.
J
0 0
[Iv
T
El +
dt
Define:
H(x,u,£) =
Lq(x>H.)
- P (t) f(x,u)
(3.1.7)
the variational Hamiltonian, and H which will include terms arising
from the inequality constraint.
H = - H -
or
H = £T(t) _f(x_,u) - Lq (x,u) - y£(t) Hence
J =
T
£. Ü
J F EH* + xTp
0
dt
With the exception of the maximum principle, all of Hestenes’ necessary
conditions are obtained from the requirement that the first variation
of the cost function vanish. In the following, "Sjx" designates

47
"the variation of _x"5 a subscript vector designates the partial deriva¬
tive with respect to that vector, with the result itself a column vector.
Thus,
ÓJ = J3 6_x
tp tp T *
- / [6x (H - ¿) +
0 0 -
T *
+ óu H
— u
yA2S£6S£] dt = °
To derive the PMP requires an extensive mathematical development and is
not included since it contributes nothing to the present discussion.
However, the necessary conditions are listed in order to be available
for later reference.
x = H = f (x, u)
- _p_
*
0 = p ,<5x. , t = 0 ^
1 1 i
0=p.6x. , t = r l
i F f
Specified Boundary Conditions on x_^(t) ,
0 = H"
jj
0 = u^Ct) > P£(t) > 0
d OPT ’ -^OPT ’ - 11 ^OPT’ — ’ ^
The optimal solution must satisfy these six conditions together with the
inequality constraint (3.1.6).

48
The PMP states that along the optimal trajectory, each instant
of time t, state x^ (t) and adjoint state p_(t), treated as fixed, the
optimal control u^p^Ct) is that admissible control which minimizes the
variational Hamiltonian. In the present context, admissibility requires
that _u(t) be piecewise continuous, the set of admissible controls being
denoted by Q. Hence the PMP indicates that
UQPpCt) = ARGMIN [HCxQpp, u , p)] (3.1.8)
uefi
Notice that the necessary conditions suggest nothing about how
a solution is obtained, but merely indicate certain functional relation¬
ships that must be satisfied. However, equation (3.1.8) seems to inti¬
mate that solution of the necessary condition of PMP involves a mathe¬
matical programming problem.
3.2 Mathematical Programming:
Gradient Projection Method
Having shown that the PMP from the calculus of variations
approach to an optimization problem may perhaps be related to a mathe¬
matical programming problem, the latter will be discussed in general
terms. Consider a nonlinearly constrained optimization problem
-^OPT
= ARGMIN [F(x)]
x e Í2
subject to
gj (x) < 0 j = 1 m
where Í2 denotes the set of admissible state components x^, i = l,...,n,
and to be admissible requires only the satisfaction of the m inequalities.

49
Necessary conditions which x^p^ must satisfy are given in the Kuhn-Tucker
theorem :
(i)
(Ü)
(iii)
constraints are satisfied
multipliers exist such that
and for all j = 1,...,m
m
and
V(W
E
j=l
?j ^OPT^
A.
J
A. V
3
j sj (^opt
8 j ^OPT
< 0
> 0
= 0
= 0
Observe that if I. denotes the set of indices associated with active
A
constraints, the first two conditions may be written as
3 £ IA gj (x) = 0
and
A .
3
j i I, g. (x) < 0
A J
and
A .
3
Fox (1971, pp. 168-176) presents a very readable proof of this theorem;
a more mathematical proof using vector space concepts is available in
Luenberger (1969).
Many methods for obtaining a numerical solution to the nonlinear
programming problem described by the first two equations of this section
have been developed. The gradient projection method by Rosen (1960) is
used frequently in structural optimization. Basic to the method is the
orthogonal projection of the cost function gradient onto a subspace
defined by the normal vectors of the active constraints. An inherent
part of the algorithm is the concept of a "feasible," "usable" direction.
Any direction d is feasible if an increment x in that direction improves
the cost function, i.e., decreases F(x). Direction d is said to be
usable if it also satisfies the constraints. As long as a feasible,

50
usable direction exists, the cost function may be improved. A constrained
optimal solution x^pp occurs at that point where no feasible direction
is also usable, i.e., any attempt to improve the cost violates a con¬
straint. In Appendix B these concepts are used in a concise proof of the
Kuhn-Tucker conditions.
Fox (1971) derives the matrix P which projects the cost function
gradient into the subspace defined by vectors normal to the active con¬
straints. This is equivalent to subtracting all components parallel to
vectors that are normal to surfaces of active constraints from the nega¬
tive gradient of the cost function. Recalling the definition of set I ,
consider r constraints to be active such that
1^ — , • - • ,n^_}
Define a vector whose elements are the corresponding nonzero
Lagrangian multipliers, and another vector whose elements are the active
constraint functions
A = [A A ... A
— a. a„ a
12 r
n " tg g ... ga ]
12 r
From the N vector, a matrix N is introduced, each column of whi£h is the
gradient of an active constraint. Hence, N is an (n x r) matrix where
T
N = N
—x
and
3N,
N. . =
ij 9x.
1 =
[N. .]
ij
5gai
9x. J
i
i = 1, . . . , n
j = 1,•••,r
(3.2.1)

51
With these definitions of A_ and N, the third Kuhn-Tucker condition can
be written as
V F(x) + N A_ = 0
At any feasible point x where g(x) < 0, the direction which best
improves the cost function is the negative gradient of the cost. If
those directions which lead to constraint violations are subtracted from
-V^F(x), the projection matrix P is obtained. Directions causing a
constraint violation are specified by the gradients of active constraints,
T -
i.e., the columns of N^. What is required of S, the projection of the
gradient, is that
S = (-V F(x)) - NT A (3.2.2)
_x — —x —
where A_ are scalar coefficients to be determined such that S is ortho-
T
gonal to each column of N^, or
T T *
(N ) S = 0
—x
T .
When the matrix equivalent to is used together with the S expres¬
sion (3.2.2), the result is
NT(-vxF(x) - N X) = 0
such that the which satisfies this orthogonality condition is:
A = - (NTN) 1 N1(V F(x))
— x —
(3.2.3)
Unless
dent,
the active boundary surface normals (x)
T
the matrix (N N) is nonsingular. Conversely,
are linearly depen-
if this matrix is

52
singular the active constraints are not linearly independent; however,
this is not a condition encountered in most real cases.
Substitution of the _X expression into the S equation leads
directly to the projection matrix P:
S = -P V F(x)
x —
where
P = I - N(NTN)-1 NT (3.2.4)
where I is the identity matrix. The direction S which best improves
the cost is given in terms of P, where P and N are given by (3.2.4)
and (3.2.1). If no constraints are active at a point then N is a
null matrix, P reduces to an identity matrix, and the direction of best
improvement is coincident with the direction of steepest descent.
In the algorithm associated with this method the starting point
must be a feasible point where g^. (jc) < 0 for all j = l,...,m. The
design then proceeds in the S direction until the solution is satisfied
to within a specified position tolerance e. Necessary conditions
generally programmed in a computer program are:
isp
* E,
i
1, . . . , n
A .
J
> o ,
j
£
XA
A.
3
= o ,
j
i
*A
It is readily seen that for S, P, and _A defined as above, these are
completely equivalent to the Kuhn-Tucker conditions.

53
3.3 Gradient Pro jection Method Applied
to the Maximum Principle
Based upon the preceding discussion, the similarity between PMP
and the'mathematical programming problem can be discussed. The maximum
principle states that the optimal control minimizes the variational
Hamiltonian with respect to all admissible u_. Or, at each time
0 < t < tp, .Hqpp minimizes H(jc,ii,p) with respect to u for given _x and j)
and where _ti is subject to constraints p < 0, £ = l,...,q.
Treating this as a mathematical programming problem, the following
correspondences are recognized
y: 'v- _u
F(x) ^ H(x,u,jp)
gj (x) ^ £ (x,u)
A. ^ p (t)
J 1
V F (x) H
— U
S ^ H plus constraints
_u
Continuing to identify corresponding quantities, at each time t, let
I. denote the set of active constraints, taken to be r in number.
A
1^ • • • j
Then
N1 'v éT = [4 ó ... 4i ]
a, a„ a
12 r
at - PT = [p (t) P (t) ... P (t)]
— a, ct„ a
12 r
N ^
^ aj
3x.
L J -
j = 1,. . ., n
j l,...,m

54
Furthermore, define (H )p to be the gradient of H with respect to u
where all components that cause a constraint violation have been removed
Since projection matrix P removes cost function gradient components that
lead to constraint violations, consider its use in the maximum principle
T
With the correspondent to N identified as

and
P - 1 - T^-1 4. T
— u — u
(3.3.1)
(H ) = P H (3.3.2)
u P _u
From the Kuhn-Tucker conditions, this implies that along the
optimal trajectory (t.^pp, jj0PT)
§ = - (Hu)p = 0 (3.3.3)
Similarly, at each time
1*00 - (A T JÍíV1 1 T Hu (3.3.4)
u u —
from which it follows
(i T & M + 1 T Hu = 0
u — u —
â– *u * + Hu - 0
(H + /<{>) = 0
u
Or,
( - H )
0
h'‘
u
0
(3.3.5)
(3.3.6)

55
Hence the control law from Hestenes' necessary conditions can be derived
from the PMP condition by treating it as a nonlinearly constrained math¬
ematical programming problem. While using the gradient projection method
in the derivation, it is seen that equation (3.3.5) is equivalent to the
third Kuhn-Tucker condition. The second Kuhn-Tucker condition is iden¬
tical to Hestenes’ necessary condition on the Lagrangian multipliers used
to adjoin the inequality constraints to the cost function. Satisfaction
of the inequality is implied by requiring the first Kuhn-Tucker condition
to be fulfilled, where
>*
(—1.
IV
o
> o
(3.3.7)
Ajg.Cx) = o
py (t)£(x,u)
= 0
(3.3.8)
g.(x) < 0
< o
(3.3.9)
Thus by treating the solution of the necessary conditions of the max¬
imum principle as a programming problem with inequality constraints,
using the gradient projection matrix, and by requiring satisfaction of
the Kuhn-Tucker conditions, an explicit formula for Hestenes'
Lagrangian multiplier functions has been derived. It is further demon¬
strated that with the p£(t) so defined satisfaction of the extremum con¬
trol law condition is implied. However, before this treatment can be
accepted as valid, it must also be shown that the system of canonical
differential equations is unchanged.
Consider the state system equations
*k
x = H = f(x,u)
-

56
where
«U rp rj-1
H = p_ (t)_f (x,u) - Lq(x,u) - jj_ (t)¿(x,u)
It is obvious that the explicit form of y(t) has absolutely no effect
upon the state system equation expressed in canonical form.
Demonstrating that the adjoint system equation is unchanged
requires the method described by Bryson et al. (1964). Consider the
general problem of Section 3.1 again, but with only the differential
constraints adjoined to the cost function, i.e.,
Min {J =
u
(H - xTp.)dt}
(3.3.10)
subject to: (3.1.6)
where
T
H(x,u,¿) = LQ(x,ji) - 2. (t)_f(x,u)
(3.1.7)
and
X = f (x, u)
(3.1.4)
Again let I. denote the set of indices associated with r active
A
constraints at any time t
ta ~ {ai3
2’
,a }
r
< 0 > & ^ TA = °
4>^= 0 > £ E IA -> y£(t) > 0
and jj>_ is defined as before
• <í>0 ]
r
“l “2
The problem can then be thought of as minimizing (3.3.10) subject to
(j>(x,u) = 0

57
While on the constraint surfaces defined by this equation the variations
in control 6_u(t) and state <5x(t) are not independent but instead are
related through the subsidiary requirement that
6^(2£,_u) = 0
or
<5x%T (x,u) + óuT<}>T(x,u) = 0
X — —U — —
(3.3.11)
This imposes a restriction to the admissible variations. For cost func¬
tion (3.3.10) to be a minimum, it is necessary that its first variation
vanish, i.e.,
F
F
, T
, T T
„ T-
+
óx H
+ 5u H -
- 6x p
0
0
— X
— u
-
<5J = p 6 x.
It has already been shown that
dt = 0
(3.3.12)
H* = - (H + yT) = 0
u u
which will be used to advantage shortly, after having added and sub-
T T
tracted the term 6u (_U 40 from the integrand of (3.3.12).
0 = p 6 x
ÃœF tF
+ Í
0 0
óx^(H — p) + Su H +
— x — u
+ SuT(pT¿) - 6uT(pTD
— c-u I-U
dt
Rearranging terms gives
0 = p o >c
fcF fcF
0 0
+ / 6xT(H - p) +
+ 6uT(H + (yTj.) ) - 5u1(pT£)
— u ^ u — L u
dt

58
But,
and
Hence,
(H + (yT<{>) = (H + yT!) = (-h“) = 0
u ... — u u
(lLTi)u = ^ £
0 = R <5x|
i F F
T. „ T T
_^o
+
o
6x (H — j>) — 6u y
dt
It is here that the restrictions imposed by the active constraints are
applied; from (3.3.11)
T T T T
- ou — -Hi — —x
such that
0 = jp. 6x
0 = ja á^x
+ / ÓXT(H - V) + 5xT y
0 0 L - - -
dt
tF üf f
+ /
6xT(H - p + 4>T y)
— x x —
dt
0 0 L
Applying Euler's lemma, for arbitrary variations in the state which
satisfies the constraints.
(Hx “ P + £) =
0
which by the following manipulations is shown to be the adjoint system
equation of Hestenes.
¿ = Hx + (y ±)x
- (-H - V ^
R = - H
X

59
Thus, the explicit formulation for y_(t) obtained by applying the gradient
projection method to the PMP satisfies all the necessary conditions of
Hestenes.
It may happen that in some cases the constraint upon control
does not depend upon the state. It can be shown that the _p(t) explicit
formulation is equally valid in this instance. By examination of
equations (3.3.1) through (3.3.9) it can be verified that all the neces¬
sary conditions except the adjoint system equation are satisfied. To
demonstrate the latter, recall that when on a constraint boundary the
first variation of both the cost functional and the constraint function
must vanish. That is, for
$(u) = 0
both
5 J = 0
and
6~ = 0 (3.3.13)
— — —u
To derive the desired equivalence, the same term must be added and
subtracted from the integrand of 6J as before, again arriving at the
result
Wien the constraint variation (3.3.13) is introduced into this last
equation, then by Euler's lemma
0 = p_ <5x
6xT(H
- p) -
* T T
6u * p
(H - p) = 0
x

60
Since it was stipulated that j£(u) is not a function of x, the equation
may be written
(H + yV>x - £ = 0
( - H* )x - ¿ = 0
Thus, the expression for _p(t) is valid when the constraint inequality
depends only upon the control ji(t) .
3.4 Maximum Principle Algorithm
In the introduction to this chapter it was stated that the
Lagrange type problem from the calculus of variations is equivalent to
an ordinary mathematical programming problem based on the Kuhn-Tucker
conditions. Furthermore, itfhen inequality constraints are present the
necessary conditions are equivalent to the Kuhn-Tucker conditions.
It was demonstrated in the preceding section that if the PMP is itself
treated as a mathematical programming problem, application of the
gradient projection method provides an explicit solution for the
Lagrangian multipliers associated with active constraints. This explicit
solution for 1^(0 also satisfies all of the other necessary conditions
for an optimal solution. The ability to determine p^(t) explicitly
in terms of parameters and functions that describe the problem suggests
the possibility of converting the necessary condtions of an optimal
solution into an algorithm for obtaining it.

61
Ensuing discussion of the algorithmic form of the necessary
conditions contains the implicit assumption that all equations are valid
along the optimal trajectory. It is further assumed that the problem
under consideration is that one described in equations (3.1.1 - 3.1.6).
The algorithm requires that jx(t) and ]3(t) be known at each time
0 < t < t„ for which the solution procedure is as follows.
(i) Use PMP on the variational Hamiltonian to determine an optimal
JL
control _u (t) independent of constraints.
u"(t) = ARGMIN [H(x,u,p)]
Evaluating the inequality constraints with jj = jj reveals which of
the £ = 1,...,q constraints are active. Let r denote the number
of active constraints and I. the set of indices associated with
^ = 0 1 e IA
? (x,u) <0 a i lA
From this the vectors whose elements are the nonzero Lagrangian
multipliers and corresponding constraint functions are defined,
respectively, at the instant of time t.
(ii)
P (t) = [u p ... p ]
a a a
12 r
(x,u) = [
al a2
Having identified which of the q constraints are
components of u^ are specified by jj>_(x,u) = 0.
active, r
They may be
solved by using the implicit function theorem, which requires

62
T
—u
which are active at point _x(t) to be linearly independent.
To determine the remaining (m-r) components of u^pT requires
that p be known at time t, but
y (t) = - (i T 1 — — I u — i u
u — u —
This value of _p is used to determine the "constrained" Hamil¬
tonian,
H* = - (H + uT£)
k
(iii) With the nonzero Lagrangian multipliers jj known and H conse¬
quently defined, the remaining (m-r) unknown components of
are determined from the control law for the constrained
system, i.e,, from
*
H = 0
ij
Once ÃœQp.p is completely known, the adjoint system equations
are determined by
JL
• o
£. = ~ H
x
The process outlined above then allows UqPP to be written as
uQpT = ARGMIN [H(xopT,u,£)]
u eSJ
since the _u obtained in this fashion satisfies <|>£ (x,_u) ¿ 0 which is the
only requirement for being admissible. However it must be recalled
that these equations are valid along the optimal trajectory; it remains
to be shown that this algorithm may be employed in some manner to obtain
that optimal trajectory and to demonstrate their satisfaction along it.

63
3.5 Solution Methods
Necessary conditions from the calculus of variations provides
a Two Point Boundary Value Problem (TPBVP) to be solved, which is in
general, nonlinear. For all but the most simple problems no analytical
solution is possible and if any solution is to be obtained a computer
must be used with some numerical method. A discussion of the available
methods and their relative advantages/disadvantages is not included here
due to the availability of such discussions in the literature, e.g.,
Bullock (1966). All of the methods involve some iterative scheme, and
for optimal control can be separated into two general categories.
(i) Indirect methods. Schemes which require an intial guess of
the state's solution: In these methods the starting point is
an initial guess of the time history of the solution. The con¬
trol associated with the solution is a subsequent calculation.
Iteration continues until the state satisfies some criterion
connoting convergence; the final control history at conver¬
gence is the optimal control.
(ii) Direct methods. Schemes which require an initial guess of the
control function: The starting point for these methods is an
initial guess of the control time history. For this class of
methods the state associated with the control is a subsequent
calculation. Iteration continues until the control satisfies
some convergence criterion.
The method of quasilinearization was selected, based upon the
success of Boykin and Sierakowski (1972) in applying it to a constrained

64
structural optimization problem. Excellent convergence for their
problem, the capability to handle nonlinear systems, and the avail¬
ability as an IBM SHARE program, ABS QUASI, dictated its selection.
In the application to the examples in Chapters IV and V the program
required no modification. As a result, a detailed discussion of the
method of quasilinearization is not included.
The problem discussed in preceding sections of this chapter
falls into the general class of problems that QUASI handles, that is,
1 = £(X>t)
with the boundary condition of the form
B£Y(0) + Br Y(tF) + = 0
where t_, square matrices B and B , and vector C , are specified,
r 3c r Q
constant quantities. The specific form of B^, B^ and depend upon the
given boundary conditions. As described in algorithm form
• J-
p = -h^(2£>R0PT(x>R)) = G2(x,p)
In terms of the general QUASI nomenclature,
G1 (*>£.)
G2
Y =
£(Y,t) =
Boundary conditions are determined by those specified for the original
system and by the necessary conditions outlined in the first section
of this chapter.
In Chapter VI the problem treated by Boykin and Sierakowski (1972)
is solved by the gradient projection method applied to a finite element

65
formulation for the description of the structural system. This is
a method of the second kind mentioned above. Results of the two methods
are compared.

CHAPTER IV
CONSTRAINED DESIGN OF A CANTILEVER BEAM
BENDING DUE TO ITS OWN WEIGHT
4.0 Introduction
A structural optimization problem has been selected for its
simplicity and stated as an optimal control problem. The maximum prin¬
ciple is applied, giving a nonlinear TPBVP of the Mayer type. Among the
earliest expository papers on the maximum principle, Rozonoér (1959) gives
an excellent treatment to a similar type of problem; his technique is
used to obtain both the variational Hamiltonian and adjoint variable
boundary conditions. It is shown that no finite solution exists for the
situation of unconstrained control. Numerical solutions for constrained
control are obtained by the method of quasilinearization. Constraints
include both geometric limitations to control as well as maximum stress
limits that become mixed constraints depending upon both state and con¬
trol variables.
4.1 Problem Statement
A cantilever beam of variable rectangular cross section is to be
designed for minimum tip deflection due solely to its own weight. The
material is specified to the extent that the modulus E and density p are
constants. Length L is specified but the design variables, height h(x)
and width w(x), may be chosen independently of each other, subject to
66

67
hard constraints upon the allowable dimensions. That is
a < w(x) < c
(4.1.1)
b < h(x) < d
If Y(x) denotes the deflection of the centerline, the problem is:
given E, p, L, and the constraints, find h(x) and w(x) to minimize
Y(L). The particular form of differential constraints to be satisfied
will be derived in the next section.
4.2 Structural System
Small deflections are assumed in order to use linear Bernoulli-
Euler bending theory. Basic conventions assumed for this example are
depicted in Figure 4.1; with these conventions the governing equation
is derived using standard strength of materials considerations. The
result is
where
L
Mg(x) = y/ (t-x)w(t) h(t)d r
x
and y = pg. Kinematic boundary conditions to be satisfied by the solu¬
tion of (4.2.1) are:
Y(0) = 0
Y' (0) = 0

68
Nota: Y(x) is centerline deflection, positive downward.
Figure 4.1 Structural Conventions

69
Design variables and the related constraints (4.1.1) are put into
dimensionless form such that
v = b/d < v < i
id 1
and
w(x) .
= —— -> a/c < v„ < 1
2 c 2
VX) ' 12 Cd3 V1V2
Replacing the independent variable with a dimensionless equivalent, and
using the control components allows the governing equation to be put
into a dimensionless form. For
x
- L
~(-^)2 u|u?Y = / (x-t) u1(T)u2(x)dT
(4.2.2)
where
U.(t) = v.(x(t))
When constant C is defined, the usual kinematical relationships for
B
a beam may be written in a simple dimensionless form; that is, let
c = JL _JL, (-á)2
b 12 yl2 V
(Units = Length B
xi = cby
X2 = CBY
Deflection
Slope
X3 = C8U1U2Y
x4 â–  CB at (uiu2Y)
Moment
Shear
U1U2 = CB 77 dtz
Load

70
These state component definitions are used with the natural boundary
conditions to obtain
x^Cl) = 0
V1} = 0
From (4.2.2), the state component definitions, and the above boundary
conditions it follows that
x, = x„
x„ =
*3/u2U2
= X,
x, =
ulu2
x1(0) = 0
x2(0) = 0
x3(l) = 0
x^(l) = 0
These equations and boundary conditions are used in the following section
to precisely state the problem. The solution and results are given later.
4.3 Unmodified Application of
the Maximum Principle
In terms of the state variables defined in the preceding section,
the problem can be stated with more mathematical precision. Find
UQpjCt) = ARGMIN [x1(l)]
U£Í1
subject to:
(i)differential constraints
(ii)kinematic boundary conditions
natural boundary conditions
X = _f (x,_u)
x^CO) = x2(0) = 0
x3(l) = x^(l) = 0
(iii)hard geometric constraints
b/d < u^(t) < 1
a/c < u2(t) < 1

71
According to terminology in the calculus of variations this is a Mayer
type problem. Among the early papers concerning the PMP, Rozonoér (1959)
applies the PMP to a similar problem giving a geometric interpretation
to the function of the adjoint variables.
In Rozonoér's problem the cost is a generalization of the
ordinary Mayer problem, in the sense that the cost function is a linear
combination of the terminal state components. It can be shown via the
calculus of variations that to minimize
J
where c is a vector of prescribed constants, the necessary conditions are
—r
H = £T(t)f_(x,u, t)
x = H
f
£
x
0 = H
u
0 = /(0) x(0)
0 = £ (tpMllj. + £^tp) ]
T
n^OPT’ -^OPT
p,t) > H(xnpT, u, £,t)
That is
Min[J] Max [H]
u
u

72
If we consider the cantilever beam problem, the forms in the
necessary conditions are
c = [1 0 0 0]T
—r
H = P;Lx2 + P2x3/u|u2 + P3x4 + P^U1U9
from which
"N
f(x,u) /
x3/ulu2
VU1U2 J
pi = °
Pl(l) = -1
P2 = "Pi
P2(l) = 0
p3 = -P2/u3u2
P3(0) = 0
P4 = -p3
p4(0) = 0
Adjoint variables p^(t) and p2(t) can be integrated by inspection
pL(t) = “I
P3(t) = - (1-t)
0 < t ¿ 1
(4.3.1)
such that
H = - x2 - (l-t)x3/uju2 + p3x4 + P4U]Lu2

73
With this result, the necessary condition for control to minimize tip
deflection is
Hux = 3(l-t)x3/u^u2 + p4u2 = 0
Hu2 = (l-t)x3/u|U2 + p4ux = 0
At first this appears to be a contradiction since the two equations can
be satisfied only by the trivial solution because they have equivalent
forms,
P4U1U2 = -3(1-t)x-3
4 2
P4U1U2 =
(l-t)x.
(4.3.2)
Further examination, however, leads to the conclusion that when the
control is completely unconstrained there is no horizontal tangent
plane to the surface H = H(u3,u?).
When the geometric constraints to the control are included,
a constrained minimum may exist. If such is the case, the maximum
value of H occurs on the boundary of admissible control space. To that
end, PMP is employed along the control space boundary to determine
ucipT(t) at each time t. Before detailing this procedure, it is neces¬
sary to first consider some structural aspects of the problem.
By definition the control components are positive, which in
turn implies
Load:
—4
>
0
-f-
U1U2
Shear:
X4
<
0
<-
h *
0 ,
x4(l) = 0
Moment:
X3
>
0
X3
0 ,
x3(l) = 0

74
Furthermore, since p2(t) - 0 from (4.3.1)
P3 > 0 -e p3 > 0 , P3(0) = 0
p4 < 0 P4 < 0 > P4(°) = 0
This exercise makes it possible to use the information of the sense for
x3 and p^ to simplify the search for u^^ on the control boundary. By
arranging the Hamiltonian in the following fashion
H = -x2 + p^ + p4[u3u2 - (l-t)x3/p4u3u2]
it is observed that both terms in the bracketed expression are positive.
This and the p^ outside the leading bracket allows the following
equivalence:
Max [H]
u e 9U
or,
-^OPT
where
*(u)
Hir, [UjU2 - (l-t)x3/p4u|u2]
US 3U
ARGMIN [$(u)]
u e 3U
[U;Lu2 + F2(t)/u3u? ]
and
F2(t) = - (l-t)x3/p4 > 0
(4.3.3)
At each position t along the beam, the state and adjoint variables must
satisfy the appropriate differential equations, and u. is specified by
the preceding three equations.
Control space boundary SU is illustrated in Figure 4.2, where
the Up axis is treated as the ordinate since Up(t) and u.-(t) correspond

75
b/d < < 1
a/c < u2 < 1
1 "
b/d--
a/c
Figure 4.2 Admissible Control Space

76
to the height and width, respectively, of the cross section of the
beam at position t.
Along the constant u edges of 3U, let u, = u where u has the
1 1 c c
value of either b/d or unity. If $(uc,u2) has a minimum point
d$
du„
= 0
and
d2$
du2
> 0
where
$(u ,u ) = u u + F2(t)/u3u
c 2 c 2 c 2
= u - F (t)/u3u2
du0 c c 2
d~ du2
2F2(t)/u3u3
c z
The value of which satisfies the first condition is
u = F(t)/u2
2 c
Furthermore, it is observed that only one extremum of $(u) exists along
U-, = u and that it is a minimum. Hence, either $(u ,u„) has a minimum
I c c ¿
on the constant u^ edge or is monotonically decreasing/increasing. If
either
* *
u^ < a/c or 1 < u^
then along the constant u edge, H has its maximum value at a corner of
the rectangular 3U. On the other hand, if
a/c < u2 < 1
then H has its maximum value on the line u,= u interior to the
1 c
endpoints.

77
Similarly, along the edges of 3U, denote
has the value of either a/c or unity. If $(u.,u ) has
1 c
= u where u
c c
a minimum point
where
d2$
and —2" > 0
du
1
u,u + F2(t)/u2u
1 c 1 c
u - 3F2(t)/uju
c 1 c
12F2(t)/u^u
1 c
It is observed that has only one extremum along u2 = uc5
it is a minimum, and occurs at the point where
U;L = + { 3F2 (t) /u2 }
Thus, by the same argument posed in the preceding paragraph, if either
k k
u^ < b/d or 1 < u^
then along the constant u? edge, H has its maximum value at a corner of
the rectangular 3U. Wherever
b/d < u^ < 1
H has its maximum value on the line u„ = u interior to the endpoints.
2 c

78
On the basis of these arguments, the following system was
solved by the method of quasilinearization:
■ ¿2
*3
*4
X2
x1(0) = 0
= x3/u|u2
x2(0) = 0
il
X
x3(l) = 0
= U1U2
x4(l) = 0
= -P2/u];u2
P3(0) = 0
= "P3
P4(0) = 0
--OPT = ARGM1N ^uiu2 ~ (l_t)x3/P^u|u2^
u e 9U
The beam is represented by 100 intervals composing the range 0 < t < 1,
which is separated by 101 "mesh points." An initial guess of the
solution x(t) and p(t) is chosen; it is selected to satisfy the bound¬
ary conditions. This guess is not a solution and does not satisfy the
differential equations. The x and ¿ equations are linearized about the
initial guess, then the resulting linear TPBVP is solved to obtain new
x(t) and p(t) functions which more closely satisfy the differential
equations. At each time t corresponding to a mesh point, II is numer¬
ically evaluated along each of the four straight line segments com¬
posing 3U to determine u^^. The point (u^,u^) on 9U which gives
H(u.;x,p,t) its maximum value is • This process is repeated until
the Ct) and _p(t) iterate satisfies the differential equations to
within a specified tolerance. The equations necessary to use the IBM
program available are given in Appendix C, in the form of a subroutine
listing.

79
4.4 Results: Geometric Control Constraints
For the most part, no major difficulties were encountered in
using quasilinearization to obtain a solution to the sixth order sys¬
tem derived in the previous section. Certain parameter values did
engender numerical instability. These cases, the source of the diffi¬
culty, and its circumvention are discussed in Chapter VII. Moreover,
all calculations were done in double
precision as necessitated by matrix
inversion accuracy requirements.
Parameter values selected to
illustrate the solution method are:
u : b/d = 0.25
(4.4.1)
u2 : a/c = 0.20
The measure of error of satisfaction of the differential equations in
the TPBVP is in terms of the general system
dY .
—^ = f. (Y,x) Y = Y. (x),
dx i — —i
i = 1,.. . ,n
ERROR = Max 1dY. - f.(Y,x)dx|
1 l l —
X
(4.4.2)
Deflection of a uniform beam due to its own weight was used to infer
an initial guess xihich satisfies all boundary conditions:
xx(t) = t4
x2(t) = t3
x3(t) = 1 - t2
0 < t < .1 (4.4.3)
x^(t) = -1 + t
P3(t) = t2
P4(t) = ~t3

80
With these specified parameter values and initial guess of the solu¬
tion, the program converged to a solution in five iterations. From
this run a tolerance was selected for all subsequent cases; the follow¬
ing tabulation provides the data used in its selection:
Iteration ERROR Tip Deflection (Cost)
1
.2028
.7387749327
X
10“1
2
.1704
.3192152426
X
10“!
3
.6533 x
10“1
.2847993812
X
10“1
4
.3031 x
10“ 5
.2853731846
X
10_1
5
.1129 x
10“10
.2853719983
X
10-1
It is seen from these tabular data that there is little improvement in
cost (tip deflection) as a result of the fifth iteration. For this
reason a value for the tolerance was selected as 0.5 x 10 ° which corre¬
sponds to about six significant digits in the cost functional.
R.ecall from the previous section that no unconstrained minimum
exists. With the control bounds included, the intuitive solution is
one in which the cross-sectional area is maximum near the root, and
reduces to a minimum at the tip. Recalling that for the optimal control,
IV -Max [H] 5 Min [ ]
—OPT n
_u e 0 ti e Q
A sequence of illustrations in Figure 4.3 demonstrates the location of
Uqpt on for several stations along the beam. Constant contours of
<5(u) are plotted on the admissible control space at five distinct posi¬
tions. If an extremal point exists interior to 3U some lines of constant
<&(u_) contours must be closed curves in jj-space. This is impossible for
this example.

81
O Minimum $(u.) uopT
□ Maximum á>(u)
Direction of Increasing $(u)
Figure 4.3 Contour Plots of $(u) at Various Stations
Along the Beam

82
The first illustration is for the t = 0.1 cross section, near
the root of the beam. Since _u^p occurs at the point of minimum $(vi),
the optimum value for both u^ and u^ is unity, the maximum allowable
dimensions for both height and width. Constant contour lines indicate
that $(u) is mathematically decreasing in either direction of _u-space.
Lines of constant $ (u3 are also plotted for the cross section of
t = 0.3. The optimum control has the maximum admissible value for
height but has a value somewhat less than unity. However, there
are still no contour lines which are closed curves.
At the midpoint cross section the minimum $(u) point occurs
at Up = 1 and u^ = a/c. Although the surface $(_u) forms a scoop-like
shape, there are still no closed curve contours, and hence no extremal
interior to admissible _u-space. The next cross section at which ?(_u)
is displayed occurs at t = 0.7. On this section, is still at its
lower bound but u^ is no longer at the maximum allowable value of unity
as shown in Figure 4.3. In the last of the sequence, $(u) contours for
the cross section at the tip of the beam are displayed. The point of
minimum i>(u) occurs where both components of control have their minimum
allowable values. Again, no contour lines of constant $ (_u) form a
closed curve indicating the existence of an interior extremal point.
This sequence of illustrations indicates two things. First,
the lack of closed curve contour lines of $(u) verifies that exists
on 3U. With further study it may be possible to obtain some condition
on = 0 which implies the equations corresponding to (4.3.2) can
never yield a finite, unconstrained optimum. Such a condition would
define the class of structures whose unconstrained solution is the

83
"zero volume solution" frequently described in the literature on struc¬
tural optimization. Secondly, at t = 0 the point u^p occurs at
T
_u = [1,1] , the point of maximum cross-sectional area; as t increases
from zero to one, the point u^p^ moves along the u^ = 1 boundary of 9U
to the Up lower bound, and then down the Up = a/c boundary of 3U to u^
lower bound. By the time t = 1 the optimal cross-sectional area is the
minimum allowable area. As a result of the prescribed form of 8U,
if u^p,j,(t) follows this particular path as t increases from zero to one,
each component of u^p^(t) has its own distinct region of transition.
That is, at any value of t, if b/d < u^ < 1, then Up must be on either
its upper or lower bound. Conversely, if Up is in transition where
a/c < Up < 1, then u^ must be on one of its bounds.
This effect is seen most clearly in Figure 4.4, where u^p is
displayed for the example case parameter values specified by (4.4.1).
The profiles are displayed on a two-view drawing as a plan-form of the
beam might appear. State components corresponding to this beam are
shown in Figure 4.5, representing dimensionless deflection, slope,
moment, and shear, respectively. As observed in Figure 4.4, there are
five distinct regions of the beam:
(i) 0 < t < .25
(ii) .25 < t < .52
controls onupper bounds
a/c < Up < 1
Up transition

84
TOP VI Eli
SIDE VIEW
Figure 4.4 Plan-form Views of Optimal Solution for
b/d = .20 and a/c = .25

85
Figure 4.5 State Components of Optimal Solution for
b/d = .20 and a/c = .25

86
(iii) .52 < t < .59
U1 = 1
u2 = a/c
(iv) .59 < t < .90
b/d < < 1
u2 = a/c
(v) .90 < t < 1.00
u^ = b/d
u2 = a/c
on upper bound
on lower bound
u^ transition
controls on lower bounds
The curves that show the intercept locations as a function of
parameter values b/d have been called "correlation curves" in earlier
studies, liben the width is allowed to vary also, the second parameter
a/c is introduced. For the sake of comparison to previous studies, the
intercept/correlation curves are plotted as dependent upon b/d and
parametric in a/c. However, it would be just as correct to do the
opposite.
Intercept location curves described above are shown in
Figure 4.6. The heavy black curve is the case where a/c = 1, a beam
of constant uniform width—the case cited from earlier literature.
Another special case is represented by dashed lines, corresponding to
a/c = 0 which is the case corresponding to a minimum allowable thick¬
ness equal to zero. Dashed lines are used because these data are an
extrapolation: convergence problems encountered for parameter values
less than 0.1 prevented obtaining numerical results.


88
A discussion on the convergence difficulties experienced by
the quasilinearization algorithm for parameter values approaching zero
is presented in the chapter on numerical instabilities. In that dis¬
cussion, isolation of the source of difficulty is reported; it is pos¬
sible that this difficulty may be a general result applicable to all
problems to be solved by the method of quasilinearization. A solu¬
tion for this case is later obtained by finite element techniques.
Note that since 0 < a/c < 1 these two cases represent limits to the
solutions of the problem. In addition, if the four intercept locations
are plotted versus a/c and parametric in b/d, curves r^ and r^ appear
as "horizontal vees" with r. and r, lines that are nearly parallel.
b a
It is interesting to note from the figure depicting the solu¬
tion of this case as a plan-form, that in the central region of the
beam, the height is greater than the width. This result can be antic¬
ipated since such a configuration gives a greater bending resistance
per unit weight.
With further reference to Figure 4.4, the transition of u* (t)
is seen to be almost a linear taper, whereas the u^(t) transition
exhibits a much more pronounced curvature. To generalize from this
specific case of given values of b/d and a/c to arbitrary values
requires the introduction of four quantities characterizing the solu¬
tion. These quantities are the values of t at the points where the
transitions intercept the bounds on u^ and u^; since t represents a
normalized position x/L, these quantities can be thought of as an
intercept location expressed as percent of the beam's length. They

89
are defined with reference to the five distinct regions of the beam
previously given, where
r^ designates u-^(t) intercept with lower bound
designates u^(t) intercept with upper bound
r^ designates u^Ct) intercept with lower bound
r^ designates u^Ct) intercept with upper bound
such that the five regions are:
(i)
0
<
t
<
r
c
control on upper bounds
(ii)
r
<
t
<
r
u„ transition
c
a
2
(iii)
r
a
<
t
<
rd
control on upper/lower bounds
(iv)
rd
<
t
<
rb
u^ transition
(v)
rb
<
t
<
1.0
control on lower bounds
4.5 Inequality Stress Constraints
This section treats an inequality limit to allowable normal
and shear stresses associated with bending. Using the ordinary
strength of materials formulations it can be shown that for the rec¬
tangular cross section these constraints take the form
1 Vx)
2 I (x) h(x) - °MAX
£>
and
i Vx) . 2..
SOrf ” (x) - TMAX
Jd
When dimensionless quantities are introduced
6 kr X3',“iU2 - “max
2 YL x4/ulu2 “MAX

90
the inequalities may be written in the required form for mixed con¬
straints, i.e., as a function of both control and state components:
^(Xju) = x3/u^u2 - < 0 (4.5.1)
4>2 ^ = “ x4y/uiu2 " tq - 0 (4.5.2)
where
1 gMAX ,d
6 yL
2 TMAX
3 yL
The two stress constraints place restrictions upon the minimum
cross-sectional dimensions to keep the normal and shear stresses less
than prescribed values. Specifically, from the constraints (4.5.1)
and (4.5.2), two control inequalities must be satisfied at each station
t, and these inequalities depend upon the state of the structural sys¬
tem. The inequalities are:
ulu2 : X3/o0
uiu2;
- VT0
x4(t) < 0
from which can be derived boundary arcs in u-space:
%
ula(u2} = (x3/O0U2)2 - 0
U1t(u2) = " X4/T0U2 - °
(A.5.3)
Both of these boundary arcs are hyperbolas restricted to the first
quadrant of u-space. Depending upon the location of the arcs, vis-a-vis
the rectangular 3U, inclusion of stress constraints has one of three

91
effects in determining what _u is admissible. At any station t in some
structural state x(t),
(i)if °q5tq is too small the stress boundary arc lies entirely
above rectangular 3U; all geometrically admissible u violate
the stress constraints.
(ii)if Gq,Tq is too large the stress boundary arc lies entirely
below rectangular 3U; all geometrically admissible ju satisfy
the stress constraints.
(iii)for some range of Oq,Tq the stress boundary arc divides the
rectangular 3U into two regions: the upper region consists of
geometrically admissible ju that satisfy the stress constraint,
u in the lower region are geometrically admissible but violate
the stress constraint.
Inclusion of stress constraints alters the admissible control
space from the rectangular shape previously considered to a shape that
may contain a stress boundary arc as part of its boundary. Consider the
normal stress boundary arc specified by (4.5.3) to be a part of OU.
Then to find u^p in the manner outlined in Section 4.3, 0(ii) must be
evaluated along u^ = u^C^). If a minimum exists along the orthogonal
projection of u-^Cup) on t^ae ®(n) surface, then
Min
U1
(30(u)
320(_u)
^ 3u2
0
0

92
Along the normal stress boundary arc U2a^u2^’
U1U2 = X3/O0 " U2 = X3/0OUl
Substituting for U2 in'$(_u) it follows from (4.3.3) that
$(u)
(x3/oQ + F2(t)a())/u;L
Since the expression in parentheses is positive, it is readily observed
that the value u^ = + “ minimizes $(u) along u^^iu^). From (4.5.3), the
value of U2 at this point in ja-space is zero.
The same result exists for (_u) evaluated along u^_; on the
shear stress boundary arc
ulu2 “ - VT0 * u2 " - VVl
Substituting for ^ in (11) it follows from (4.3.3)
$(u)
Ul = (_X4/t0) + (_x4/t0)_1 p2(t)/ui
lx
that
Since both the expression in parentheses and Fz(t) are positive, it can
be argued as above that the point of minimum ®(_u) along occurs
at the point (u, = + 00, u„ = 0) .
At any cross section there exists a single minimum of 0(u) along
either stress constraint boundary arc. Since these minima occur at the
point u: (+°°,0), as one proceeds along either boundary in the direction
of increasing u^, reason, along the stress constraint boundary arc, it is necessary to
evaluate $(u) at only one point. That point is the intersection of the
stress boundary constraint arc with the rectangular boundary having the

93
larger value of u^. It is possible that the two stress boundaries
intersect, with the coordinates of that point given by
Even though this adds a complication to the boundary of what is admis¬
sible _u, the monotonically decreasing property of both stress boundary
arcs still results in the necessity of evaluating $(u) at only a single
point along the stress boundary arcs. To illustrate this, Figure 4.7
depicts an admissible control region determined by a combination of
geometric constraints and a composite stress constraint boundary,
together with the point at which $(u_) is a minimum along the composite
stress boundary. With continued reference to the figure, u^ is that
point along line ABC which minimizes §(u_).
4.6 Results: Stress Constraints Included
For the purpose of comparison, the parameter values used in the
case with only geometric constraints were also used in the cases where
stress inequality constraints are present. The results are quite simply
stated: neither stress constraint is ever active. In general it was
found that the critical section occurs at the root. At that section
the greatest portion of geometrically admissible control space is pro¬
hibited by the stress constraints. The further along the beam towards
the tip, the less the prohibited geometrically admissible control space.

a/c 1
U2
Figure 4.7 Behavior of $(jj) Along Composite Stress Boundary
VO

95
Both and -x^ decrease monotonically to zero at t = 0, and is the
reason for this phenomenon.
Each of the parameters and thus has three important
critical values. For the first pair of values the maximum allowable
stresses are sufficiently large that all geometrically admissible
control contained by the rectangular 3U satisfy the stress constraints.
In this case
oQ = 9.128
tq = 8.658
and the stress constraints intersect the lower left-hand corner of 9U
at t = 0 and lie below 9U for all t > 0. For the range of values
.5705 < aQ < 9.128
2.1645 < t0 < 8.658
the stress boundary arcs intercept the u^ = .2 portion of rectangular
9U. If they are exactly equal to the lower values, respectively, both
arcs pass through the point (1.,.2) at t = 0, and lie below that point
for all t > 0. However, these values are not as important as either
the first pair or the next. For the range of values
.1141 < oQ < .5705
.4329 < tq < 2.1645
both arcs intercept the u^ = 1 portion of rectangular 9U. When equal to
the lower values both arcs pass through the upper right-hand corner of
3U at t = 0. If the values of

96
oQ < .1141
tq < .4329
then at t = 0 there is no control which satisfies both geometric
constraints and stress constraints. The problem has inconsistent
constraints such that no solution is possible.
A sequence of plots showing the admissible control space is
shown in Figure 4.8 for six separate stations along the beam. The
value of Oq is .1141 for this case such that in section t = 0 the set
of admissible controls is the single point jj: (1, 1). That portion of
the geometrically admissible control space disallowed by the stress
constraint is indicated by dashed lines. Additionally, the point cor¬
responding to is included for each section. By observing the
sequence of plots as t increases, both the stress boundary and ate
seen to move closer to the origin of _u-space. However, never over¬
takes the stress boundary, thus the normal stress boundary never
becomes active. Figure 4.9 depicts the same case for the shearing
stress constraint.
Two additional figures are included with the intermediate
critical values of and t . The normal stress constraint case
Oq = 0.5705 is shown in Figure 4.10; shear stress constraint case
Tq = 2.1645, in Figure 4.11. No illustrations are provided for the
cases associated with the largest critical values since for t > 0 all
geometrically admissible controls satisfy the stress constraints.

t
.3
t = .5
Figure 4.8 Admissible Control: o =
o
•^OPT
J
.1141

t = .1
I
L_
t = .7
t = .3
= .4329

Figure 4.10 Admissible Control: a =
o
VO
VO
.5705

Figure 4.11 Admissible Control: tq
U2
= 2.1645
100

CHAPTER V
CONSTRAINED DESIGN FOR AN OPTIMAL
EIGENVALUE PROBLEM
5.0 Introduction
A column is to be designed for maximum buckling load in order
to demonstrate (i) an application of the theory developed in Chapter III,
and (ii) an optimal eigenvalue problem. Difficulties pertinent to this
class of problems are discussed as well as a method to obtain a solu¬
tion by using a theorem concerning self-adjoint problems. Use of this
theorem with the maximum principle overcomes the difficulty associated
with the simultaneous dual optimization required for problems where the
cost functional is an eigenvalue expressed as a Rayleigh quotient.
Numerical solutions to the combined set of state and adjoint variables
are obtained by the method of quasilinearization. Characteristics of
the solutions for a variety of control bounds are presented, along with
a discussion of attempts to also include a mixed, inequality constraint
related to stress.
5.1 Problem Statement
A vertical column, fixed at the base and free at the tip, is to
be designed such that the buckling load is maximized. Length L is
a specified constant as is the total weight W. The dependence of cross-
sectional area A(x), material modulus E(x), and density p(x), upon
101

102
position x along the column is to be such that the vertical end load P
at buckling is maximized. Weight is not to be neglected and all cross-
sectional shapes are similar.
In more specific terms, a given weight of material is to be
arranged with variable properties and geometry such that the result¬
ing mass distribution M(x) and stiffness distribution S(x) maximizes
the vertical tip load the column can support when the effect of weight
is included. Upper and lower bounds to all design variables
< A(x) < Ay
El 1 E(*) < Eu (5.1.1)
\ - P(x) < RU
are also specified. Governing equations for the structural system are
derived in the next section.
5.2 Structural System
In the subsequent analysis, small deflection bending theory is
assumed: the result is a linear differential equation with variable
coefficients. Without this assumption, the governing equation is that
of the nonlinear elástica problem. Additionally, axial compression
effects are neglected as second order. Let Y(x) denote the deflection
of the centerline at position x along the beam as depicted in Figure 5.1.
Since the system is conservative, the governing equations can be
derived quite simply by energy techniques. In terms of stiffness S(x)
and mass per unit length, the equation and boundary conditions are:

103
Vx)
Figure 5.1 Structural Conventions

104
(S(x)Y")" + PY" + g(Y' / M(Od£)' = 0 (5.2.1)
C=x
Y(x) = Y'(x) = 0 x = 0
(S(x)Y") = (S(x)Y") + PY' = 0 x = L
Stiffness and mass distributions for the column are
S(x) = E(x)I(x)
M(x) = p(x)A(x)
where I(x) is the second moment of area about the neutral axis.
It can be shown that assuming similar cross sections provides
a relationship between I(x) and A(x). If f(x) represents the depen¬
dence of one dimension of the cross section with the position along the
beam, then for some reference area A^,
A(x) = Aq f2(x) (5.2.2)
Choosing a reference area moment Iq related to A^ through a radius of
gyration for the cross-sectional shape such that
ro - koAo
it can easily be shown that
ko
I(x) = — A2(x)
A dimensionless form of the differential equation is obtained
by introducing the following dimensionless quantities

105
t =
x/L
n =
Y/L
a(x) =
A(x)/Aq
e(x) =
E(x)/Eq
P(x) =
R(x)/Rq
m(x) =
M(x)/MQ
s(x) =
S(x)/SQ
where Ag, E^, designate some reference value and
Mo =
poAo
II
O
CO
Vo
and
m(x) =
a(x)r(x)
s(x) =
a2(x)e(x)
Design variables and the given geometric constraints (5.1.1) are also
converted to dimensionless form:
V 1
a(x) < ay
eL -
e(x) < e
rL; p where the transformed bounds have been divided by the appropriate refer¬
ence value. Replacing variables in (5.2.1) by their dimensionless
equivalent gives

106
dt"
(s(t)n) + An + k ^f| / m(£)d£;j= 0
n = ñ = 0
(s(t)n) = — (s(t)n) + xfi = 0
t = 0
t = 1
where
(5.2.3)
X =
PLZ
*0*0
k =
E0I0
Integrating by parts and using the given boundary conditions,
equivalent formulations to (5.2.3) are obtained which are used to
convert the governing equation to a state component representation.
Equivalent formulations are:
1
(s(t)h) = X[n(l) - n(t)] + k / m(0[n(0 - n(t)]d£
S=t
, 1
— (s(t)h) = - Xf) - kfi / m(C)d^ (5.2.4)
£=t
”7 (ñ 1-^ (s(t)h)j = km(t)
Let
x = n Deflection
y*2 = Ó Slope
= s(t)q Moment
= ñ ^ (s(t)h)
Shear * Slope

107
then, from this choice of state components and the last of the
equations (5.2.4) it follows that
¿2 = x3/s(t)
X3 = X2X4
x, = km(t)
4
To complete the transformation of the structural problem to an optimal
control problem, the design parameters are treated as the components
of control u(t) where
u1(t) = a(x(t))
u2(t) = e(x(t))
u3(t) = r(x(t))
Hence with the definition of control ja(t) and the representation of the
system’s state by a four-dimensional vector, the governing equations are
X1 X2
x0 = xj
u?u.
3 1 2
x„ = x„x.
3 2 4
x. = ku.,u„
4 13
x^CO) = 0
x2(0) = 0
x3(l) = 0
x^(l) = -A
(5.2.5)
To complete the conversion of the structural problem to an
optimal control problem, the specified weight condition must be con¬
sidered. In terms of the specific mass distribution M(x), that condi¬
tion is

108
or
L
W = g / M(x)dx
0
1
w = P A gL / m(t)dt
This introduces yet another dimensionless parameter to the system in
that
1
1 = y / u (t)u (t)dt
0
where
y
poAo
w
gL
With this last development, the buckling problem can be stated
with more mathematical precision. Find
uQpT = ARGMAX [-x4(l)] , -x4(l) = A
U E Í!
subject to:
(i) differential constraints
(ii) kinematic boundary conditions
natural boundary conditions
(iii) hard geometric constraint
(iv) hard material constraints
(v) subsidiary constraint
x = f(x,u)
x1(0)
= 0
x2(0)
= 0
x3(l)
= 0
x4d)
= -A
aL -
UpCt) <
a
U
V I
CD
u2(t) <
e
U
r, <
u0(t) <
r
L -
3
U
1
n /
0
Ulu3dt .
= 1

109
A cursory examination of this problem statement reveals the inherent
difficulty of optimal eigenvalue problems. If the eigenvalue X is the
cost functional, the cost functional appears in the governing equation
and/or the boundary conditions. For the state components selected,
the eigenvalue X appears only as a boundary condition. However, as
stated in (5.2.3), the eigenvalue appears in both the equation and
boundary conditions.
5.3 Analysis of the Problem
In the preceding section the governing equation for column
buckling is expressed as either a second, third, or fourth order equa¬
tion with appropriate boundary conditions. State components are then
defined from structural quantities, one of which is the shear divided by
the slope. Since the slope vanishes at t = 0, it must be shown that
this state variable, x^(t), is not indeterminate at the point in ques¬
tion. Moreover, in order to use the theoretical techniques developed
in preceding chapters the first eigenvalue (cost) must be expressed as
some functional of the state and design parameters.
According to Bolotin (1963, p. 22), conservative systems are
described by self-adjoint boundary value problems. Classical elastic
stability theory is conventionally restricted to conservative problems
in which the buckling load is the fundamental eigenvalue. It is shown
in what follows that the column buckling problem is self-adjoint and
that the eigenvalue may be obtained via the total potential energy of
the system from the Rayleigh quotient. Sufficiency for x^(t) to be deter¬
minate at t = 0 is that the problem be a self-adjoint eigenvalue problem.

110
Consider a dependent variable in the third-order representa¬
tion of the governing equation (5.2.4) and appropriate boundary con¬
dition. If
z = n
another valid representation of the system is
_d_
dt
(sz) + (A+k f m(5)d£)z = 0
0
z = 0 ,
s(t)z = 0 ,
t = 0
t = 1
Using the notation of Lovitt (1924), who treats the general problem
(pu') + (q 4- Ar)u = 0
(puu')
0
= 0
the corresponding quantities are
u n» z
x ^ t
p(x) n, s ( t)
1
q(x) ^ k / m(5)d?
t
r(x) ^ 1
When written as a linear differential operator L( ) and its adjoint
4. 4.
A A
L ( ), the operator L( ) is said to be self-adjoint if L ( ) H L( ).
If the TPBVP is to be self-adjoint for any admissible function
cp and ip, then by Green's Formula

Ill
b j, b
/ {ipL(»i/j)
a a
0
where B(4>,^) is the bilinear concomitant of L( ). In terms of a general
second order equation
L() = Pj^" + P2 l W = (Pj'i')" - Cp2^)' + (q + Ar)^
The necessary and sufficient conditions that L( ) be self-adjoint is
that p2 = p|. If L( ) is a self-adjoint operator and the bilinear
concomitant vanishes, the problem is said to be self-adjoint. For the
general operator,
b
B( = p1((J)> - a
or in terms of Lovitt's equation,
operator,
which obviously has a self-adjoint
= p(<}>'il; - (pip')
1
0
it is seen that the problem is self-adjoint if the boundary conditions
of the adjoint and given systems are identical.
There is a well-developed theory associated with self-adjoint
eigenvalue problems, where the eigenvalue problem itself is the result
of minimizing an energy functional. Along this line, Lovitt shows that
under the requirements that p(x) > 0 and is piecewise smooth in (0,1),
and that q(x) is a well-behaved function which is finite everywhere on
(0,1), there are an infinite number of real positive eigenvalues which
may be arranged as

112
0 < A1 < A2 <
< A <
n
If p(x) and q(x) are specified functions which satisfy the above con¬
ditions, the fundamental eigenvalue is given by
A = Min [D(u) ]
u(x)
1
D(u) = / {p(x)(u')2 - q(x)u2} dx
0
where
u e C2
puu' = 0
I 0
and
1
/ r(x)u2dx = 1
0
2
Lovitt shows for u e G and p(x), q(x), r(x) bounded, that A^ is finite.
Higher ordered eigenvalues are obtained by using the ortho¬
gonality property of eigenfunctions as additional subsidiary condi-
For example,
A
n
Min
„n(x)
[D(u )1
n
1
D(iO = / (p(x)(u^)2 - q (x) u2 } dx
tions.

113
for
u e C2
n
.1
pu u' =0
n nlo
1
/ r(x)u2 dx = 1
0 n
and
1
I r(x)u u.dx = 0 i = 1,...,n-l
0 n 1
The immediate use of the self-adjoint property of the column
buckling problem is that the fundamental eigenvalue is finite. This
is used to prove that x^(0) is determinate. Recall the definition
x
4
.-1
n
From the third-order formulation of the governing equation (5.2.4)
-Id 1
H “jj: (sri) = - A - k / m(£)d£
t
such that
1
x, (0) = - A - k / m(£;)d£; = - 3
0
where g is a positive constant to be determined. By substituting the
specified weight subsidiary condition for the integral
x^(0) = - (A + k/p) = - 6
Both k and p are specified, finite parameters of the system. If A is
finite then x^(0) is also. In fact, the buckling load \ is the

114
fundamental eigenvalue A^ which is shown by Lovitt to be finite for
the conditions prescribed. Hence, x^(0) is determinate.
In the literature of classical elastic stability problems,
the eigenvalue to be optimized is expressed as a Rayleigh quotient.
This quotient can be found from the total potential energy. For the
general equation of Lovitt, the quotient is
1
/ (p(u')2 - qu2} dx
/ ru2dx
0
or in terms of the column buckling parameters and components (5.2.5)
1 1
/ (x2/s(t) - k / m(£)d£ x2} dt
, .. 0 t
A____ -
/ x2 dt
0
On reversing the order of integraion for the double integral contained
in the numerator, and introducing design parameters, the Rayleigh
quotient becomes
1 t
/ (x2/u2u - ku u / x2(£)d£} dt
0 1J0
A = 1
/ x2 dt
0
The normalization condition of Lovitt requires that the
quotient's denominator equals unity for all admissible solutions, or
1
/ x2(t)dt = 1
0

115
This particular normalization is selected in order to obtain the
simplest expression for the functional A. Also, recall that
x^(0) = - (A + k/p)
and
x^(l) = -A
It is seen from these boundary conditions and the definition
x^ = ku^u^ that the subsidiary weight condition can be written as a
mixed boundary condition on x^(t):
x^(l) - x^(0) = k/p
This, together with converting the normalization condition to another
state component with boundary conditions specified at both ends,
transforms the problem to:
\vIax
Max {J[u]}
u e Q
1
Jtu] = / {x^/u^u2 - ku^u^x^} dt
subject to
¿1
= X2
x1(0)
= 0
¿2
= x3/u2u2
x2(0)
= 0
*3
X2X4
x3(1)
= 0
= kulu3
x4(i)
- x4(0)
X5
= *2
x5(0)
= 0
x5(l)
= 1
k/p

116
where _u e means that _u £ C2 and satisfies the geometric control
inequalities. For reasons as yet undetermined, the quasilineariza¬
tion algorithm did not converge when the latter two state boundary
conditions were used, even though they belong to the class of problems
to which the algorithm is theoretically applicable. This might be due
solely to the mixed boundary condition being dependent on the value of
the cost functional being minimized.
On failing to obtain a solution by minimization of the Rayleigh
quotient, the next attempt was suggested by the state boundary condi¬
tions. From the condition
x4(l) = - A
and definition of A, whereby A > 0, it follows that
Min [x4(l)] = Max [A]
u u
Thus, the problem can be cast as a Mayer type of optimal control problem
Vx = Mi" [x4(1)1
u e Q
X1 = X2
X2 = X3/U1U2
X3 = X2X4
X4 = kulU3
x1(0) = 0
x2(0) = 0
x3(l) = 0
X4 (0) = -£
where B is a positive constant to be determined. The quasilineariza¬
tion algorithm did not converge for this formulation of the problem
also, again for undetermined reasons.

117
Consider from the choice of state variables that
x. = km(t) and x,(1) = - A
4 4
In an earlier derivation, g was defined such that
1
x,(0) = - A - k | m(t) dt = - g < 0
0
In another form,
1
- B + k j m(t) dt = - A
0
(Constant) + (Weight) = - (Load)
or in terms of the weight
A = B - kW
Therefore, if the weight W is minimized subject to constraints, this
is the same as maximizing A, the load, subject to the same constraints.
Notice that constant 3 is an undetermined boundary condition, i.e.,
a free condition, so that the corresponding adjoint variable's boundary
condition is known.
At this point the results of a doctoral dissertation—Salinas
(1968)—were used in another formulation of the same problem. Salinas
considered the application of the energy method to self-adjoint prob¬
lems. He showed that the following problems are identical:
(i) Maximum load, specified weight
u0PT = ARGMAX {Jp[u]>
U £ ft
Jp = the buckling load cost functional
J = a constant, the weight
w

118
(ii) Minimum weight, specified load
UqPT = ARGMIN {Jw [u]}
li e Í2
Jw = the weight cost functional
Jp = a constant, the load
Both problems are subject to the same differential constraints,
boundary conditions, and control admissibility constraints. A concise
way of stating the equivalence is
u = ARGMAX {J [u] : J specified}
—OPT _ r — W
u e Í2
= ARGMIN (J [u] : Jp specified}
_u e
The equivalence of these two problems is used to state the
column buckling problem as
u^ = ARGMIN [Jw]
UEÍ1
JW = U1U3 dt
0
X1 " x2
Xl(0) = 0
x2 x3' ul"*2
,/u?u,
x2(0) = 0
X3 X2X4
.3(1) - 0
x, = ku,u.
4 13
x.(1) = - A
4 s
where the subscript s on the eigenvalue indicates a specified load.
With reference to the statement of the problem as a Mayer type, notice
that

119
JW = k - x4^0^ Min tx^(l)]
u £ ft
with x^(0) = - 3, a free boundary condition. Furthermore, since
x^(l) = - X and X =
observe that
- J(r = 3 - JD
y W P
Admissibility of control (u e ft) requires that u.(t) is piecewise
continuous through the second derivative, and that the control inequal¬
ity constraints are not violated. For every value Xg assigned a priori
to the eigenvalue X there is a solution (in , ) that minimizes
OPT ^OPT
the weight with respect to u e ft and satisfies the eigenvalue problem.
However, an arbitrary Xg need not satisfy the specified weight condi¬
tion and in general does not. To fulfill the given total weight require¬
ment the value of X^ was adjusted such that
1
Jw 1% 3 = n/ uiu3dt = 1
OPT 0
When the cost is not only a minimum but is also equal to unity, all
conditions of the buckling problem as stated in section 5.2 are satis¬
fied. By Salinas' theorem the Xg causing this satisfaction is the
maximum lowest eigenvalue, i.e., the maximum buckling load. This effect
is illustrated in Figure 5.2 and is mathematically stated by
Max
{JP[-yy ]
OPT
V^W ] = 1}
OPT
= ARGMIN ÍJtI[u] : X = X }
W — s
U£il
where

Figure 5.2 Relationship Between Optimal Eigenvalue
and Minimum Weight Cost Functional
120

121
5.4 Application of the Maximum Principle
In the preceding analysis difficulties associated with the
optimal eigenvalue problem are discussed. Circumvention of such dif¬
ficulties is described for those problems that are self-adjoint.
It is shown that the column buckling belongs to that class of prob¬
lems. However, before the theoretical developments presented in the
third chapter may be applied, the particular forms of parameters and
functions must be identified. Accordingly, for the optimization of
the column buckling load
x ^ (x, x„ x„ x.)
— 1 2 j 4
T
J ^ JW = / V*>u)dt
L0(x,u) = puiu3
%(0) = 0
x„ (0)
Boundary
Conditions
s

122
(au-u1)(u1-aL)
£(2£>u) % \ ~ ^eU~u2^ul_eL-)
(rU~U3) (VrL)
where the maximum value of the As is that value such that
1
JW = p / u u dt = 1
0
(5.4.1)
Based upon these particular forms for the general parameters,
the variational Hamiltonian is
H = - íp-^^2 + P2X3^UÍU2 + P3X2X4 + P4kulu3^ + pulu3
k
Adjoint variables p/t) are defined from the Hamiltonian H which con-
tains the constraint functions. Since the constraint functions ^ are
independent of x for the case of only geometric constraints
JL.
p = - h'’ = H
x x
From this result and from the state boundary conditions it follows that
P1 = 0 P-j^l) = o
P2 = ~pix2 " P3X4 p2(1) = 0
P3 - -p2/x|u2 P3(0) = 0
P^ = -P3X2 P4(°) = 0
Variable p^(t) may be integrated by inspection:
p^(t) = 0 for all t > 0
which immediately reduces the number of adjoint variables to be deter¬
mined and simplifies the Hamiltonian as well. Thus,

123
H = - {p3x2x4 + p2x3/u“Ul + (kp^y^u }
where
p2 = - p3x4 P9(l) = 0
p3 = - p2/u2u2 P3(0) = 0
P4 = - P3x2 P4(°) = 0
Optimal control as determined by the PMP stated in (3.1.8) is
UqPT = ARGMIN [H(x0pT,u,p)] (5.4.2)
U £ Q
where the admissibility restriction in Pontryagin's derivation
requires only that _u(t) be piecewise continuous. Applying Salinas'
theorem further restricts the control to _u e C2. Considering for the
moment that no further constraints on the control exist, the uncon¬
strained UQPp is that choice of (up(t), u7(t), u3(t)) which minimizes
the variational Hamiltonian H. Recognizing that the physical quantities
represented by u have meaning only if they are positive, assume this to
be the case. With non-negative controls, the state equations and
boundary conditions jointly imply that
x.(t) > 0 i = 1,2,3
x4(t) < 0
Furthermore, by means of (5.4.2) and the expression for H
(P2X3) > 0 -> u2 = 0
(p2x3) <0 -* u2 = <*>

124
and
(kp^-y) >0 -* u^ = 00
(kp^-y) < 0 -> = 0
Control up remains to be determined.
Assuming that u2>u^ are known quantities, then u
independently to minimize the variational Hamiltonian.
Hu - - {-2p2x3/u^u2 + (kp^-y)u^} = 0
H = - {6p x-Zu'iu } > 0
upup 23 12
The unconstrained optimal control variable up is defined
first condition; the second condition indicates that p2(
tive for H(2copT,uopT,£) to be a minimum. So,
ui ‘
2p2x3
,1/3
»p4-v)u2u3_
and
P9(t) < 0
which leads to the following argument on the signs of pp
p2
>
0 -e
P2(t)
< 0
and
P2(l)
= 0
P3(t)
>
0 «-
P2
= "P3X4
and
x4(t)
< 0
h
<
0
p4
= 'P3X2
and
x2(t)
> 0
p4(t)
<
0
K
< 0
and
P4(°)
= 0
p must be chosen
In that context
to satisfy the
t) must be nega-
(t):

125
It is noted that with these signs if (kp^-y) < 0 the optimal uncon¬
strained control is
r
ui
â– ^OPT ^
V.0 y
k
where the existence of u^ is discussed in the next section. What is
important is that the solution requires a material of infinite strength
and zero density. Some structural analysts commonly refer to this
hypothetical material as Bolognium.
In the course of an actual design process, control bounds corre¬
sponding to the constraint functions are specified. On the basis of
the preceding analysis with given bounds on controls, optimal u^(t) and
u^(t) are bang-bang controls defined by
u2(t) = %(eJ+eu) + %(eL-eu) SGN (p2x3) (5.4.3)
u3(t) = %(rI+ru) - %(rL~ru> SGN (kp^-y) (5.4.4)
where for an arbitrary argument z of function SGN ( )
r+l
SGN (z) = < 0
z > 0
z = 0
-1
z < 0
k
Depending upon the magnitude of u vis-á-vis a and a , there are three
.Li U
possibilities for u at each point (t,jx(t) ,£(t)) :

126
^OPT
■ “2
^OPT "\U2 /
- “2
That class of problems having no mixed constraints requires computing
the Lagrangian multiplier functions jj(t) of (3.3.4) only as a check.
Each individual multiplier function must be non-negative for all
0 < t < 1. A positive value in a mathematical programming application
indicates that the usable, feasible direction is attempting to leave
an active constraint surface. With the similarity between this and
PMP demonstrated in Chapter III, the existence of a negative multiplier
function indicates a similar inconsistency in active constraints and
the constrained direction of steepest descent of H with respect to _u.
In the usual application of PMP to a problem, the constraints
are adjoined to the cost function and another functional defined:
H

127
Each constraint 4>£(x>_y) which is active such that
f (x,u) = 0
is solved for a component of control. The resulting non-zero multi¬
plier function p? (t) must be determined from the necessary condition
*
H =0
_u
This p^(t) is used in turn to determine the effect upon the system
from the solution's lying on a mixed constraint surface which by
definition depends upon the state. This effect is manifested by the
JL
A
adjoint differential equation ¿ which can be written as
¿ = - + ^0
If constraints ^ depend only upon the control, the second term is
T
unnecessary because ^ = 0.
Such is the case of the column buckling problem possessing only
control bounds. For this problem, whenever the constraints are active
the multiplier functions determined by the above procedure are
P1(t)
-2P2x3/u|u2 + (kp4-y)u,
2„1 - CaL + O
P2(t)
-2P2X3/U!U2
2u2 - P3(t)
(kp^-u)u1
2u - (r +r )
3 u L
(5.4.5)
(5.4.6)
(5.4.7)

128
These same parameters will now be determined from the explicit formula¬
tion of jj(t) derived in Chapter III.
Recall that where I, denotes the set of active constraints which
A
are r in number:
IA = (a^, o^,. . . .a^}
= { ...

a, a„ a
12 r
4= [v
3$,
8u.
i
i = 1,...,m
j = 1,-••,r
and that in terms of and H, jj(t) is defined as
-1
y(t) i i Hu
\ U — J u —
(3.3.4)
For the column buckling problem subject to constraints (5.4.1),
order to calculate _p the following quantities must be defined:
r c ~\ r ~\
*1,1
0
0
*l,u= <
0
> *2,u <
'**2,2
>
(
*3,u" <
0 Í
0
J
0
^ J
fu-
ui
/'-2p2K3/u3„2
+
(kp/-y)u3*N
H =
u 1
hu2
>- - <
'2P2X3/U1U2
+
0
^HU3y
^ o
+
(kp4-y)u1^
in

129
with
, = 2u -
(a +aT)
1,1 1
U L
= 2u„ -
W
2,2 2
„ = 2u„ -
(rU+rL)
3,3 3
Consider first the case where u^ is not on a control boundary.
Only the latter two constraints are active since ^ and are bang-
bang controls, thus = 0 and both ^ anc^ Í3 are determined by (3.3.4)
From the preceding paragraph's definitions
r = 2 -> IA = = {2,3}
(i To
U —
0 *2,2 °
0 0
2,2
3,3.
(1 xO"1
u —
-2
h,2
jj =
^3
0 (f>
2,2
0 0
0
0 4>
-2
3,3
0
4>
3,3.
r -\
H
U1
H
u2
H
Vu3y
\\ h*
2
H / ^u3 3,3

130
Hence,
_ ~P2X3/U1U2
"2 " 2ul ' (eu+eL)
(kp^-y)u1
2u3 ” (rU+rL)
which are identical to the expressions (5.4.6) and (5.4.7) obtained by
the usual method.
Whenever the constraint is active, u^ is prescribed and y^ is
nonzero. If the derivation in the preceding paragraph is expanded to
include this additional constraint, the resulting expression for y^ is
identical to (5.4.5).
The result of this analysis is a nonlinear TPBVP whose solution
is the desired optimal solution:
¿1=
X2
x]_(0)
= 0
X2 =
x3/ul“2
x2(0)
= 0
X3 =
X2X4
x3d)
= 0
*•
ii
kulu3
x4(d
= -A
s
p2 =
-P3x4
p2(1)
= 0
p3 =
"P2/UÍU2
p3(o)
= 0
II
• a
-P3X2
p4(°)
= 0
-^OPT
= APvGMIN
ue!!
H = - (p^x^ + P2x3/|fuiu2 +
A
Max
^JP^OPT-* : JW^OPT^
1}

131
where ui e requires that ui satisfy constraints (5.4.1) and Lagrangian
multiplier functions given by (5.4.5) through (5.4.7) must be non¬
negative. A solution to this problem is obtained by the method of
quasilinearization; a subroutine listing of the equations required by
the IBM program is given in Appendix C.
One additional preliminary item must be attended to before pre¬
senting results. In order to provide solutions subject to comparison
standards, an eigenfunction normalization must be prescribed. Most
Rayleigh quotient problems in the open literature normalize the eigen¬
function such that the denominator of the quotient is always equal to
unity. For the column buckling problem this is equivalent to
1
/ x2(t) dt = 1 (5.4.8)
0
The unmodified results of solution via quasilinearization do not
satisfy this condition; instead,
1
/ x2(t) dt = A2
0
where A is some constant. A transformation which satisfies normaliza¬
tion (5.4.8) such that the differential equations, boundary conditions,
and optimal control are unaffected by the transformation, is given by
-* x^/A
x2 -> x2/A p2 -* Ap2
x3 * VA p3 + Ap3
x4 x4 p4 P4

132
With a specified normalization of the eigenfunction, the state solu¬
tion Xqp^, may be compared to any other similar solution for which the
normalization is known. All subsequent results are normalized as above
to satisfy (5.4.8).
5.5 Results: Geometric Control Constraints
Three dimensionless parameters of the system are derived in
section 5.3. The first is A, the eigenvalue to be maximized. The
second is k, which is a relative measure of the weight of a uniform
column to its stiffness—how strongly inclusion of the weight affects
the solution. The third parameter, p, is a measure of the weight of
a uniform column described by the reference conditions relative to the
weight of the column having distributed geometry and properties.
Values of k and y describe a particular problem statement and must be
specified a priori.
Since the present discussion is directed towards technique, as
opposed to results for particular geometries, loading conditions, and
materials, a single value of both y and k is used for all cases. Most
of the literature compares the optimal eigenvalue to that obtained for
a uniform cylinder of identical weight. This corresponds to a value
of unity for y, selected so that the results may be compared. Param¬
eter k involves both material and geometric properties. For convenience
the material selected is structural steel and the Euler buckling slim¬
ness ratio (L/k^) value of 60 is assumed. With these assumptions the
values used throughout are
y = 1
k = .012

133
At this point a comment upon the effect that the magnitude of
k has upon the nature of Ug (t) is appropriate. In the last section
it was shown that u^ and u^ are bang-bang controls dependent upon the
sign of switching functions (?2X3^ an<^ (kp^-y) , respectively. Intui¬
tively one expects the strongest, least dense material to be optimal,
corresponding to
(p2x3) < 0 - u2
e
U
(kp^-y) <0 -»■ u^ = rL
Also shown in the last section is that for Hamiltonian H to be minimized
JL.
A
by an unconstrained value of u^ = u^ at some time, it is necessary that
P2(t) < 0. Note that if this does not occur somewhere in 0 < t < 1
the control is uniform throughout, and the system described is a uniform
column; unless = tt2/4 the solution of the state is the trivial solu¬
tion.^ A direct consequence of p2 being non-positive is that p^(t)
is also. Hence any positive value of k is sufficient to make (kp^-y) <0,
so that the minimum density is required.
This insignificant requirement that a nontrivial, optimal solu¬
tion to the buckling problem exist provides an unusual result. When
the variable signs required for a nontrivial solution are used in the
PMP expressions (5.4.3) and (5.4.4), the intuitive solution is shown
mathematically to be the optimal solution. It is for this reason that
all subsequent cases were run with control bound values
^Coding errors produced precisely this case, and indeed,to
within the numerical limits of the computer the algorithm converged
to the trivial solution.

134
eu = 1-°
e = 0.8
JLj
ru = 1'1
r = 1.0
J_i
Unity values were selected to simplify results; nonunity control bound
values were included to reveal unforeseen switching in and u^. None
were encountered. Effectively, the problem as stated has a single
control component u^(t) = a(x/L).
To illustrate the operation of the quasilinearization algorithm
detailed results are presented for one case. Data for this case are:
a = 0.9
â– Li
(5.5.1)
Just as in the cantilever beam example, the initial
from the uniform solution and must satisfy only the
Initial guesses are in terms of
C(t) = cos (%7Tt)
0 < t < 1
S(t) = sin (%rrt)
guess is
boundary
taken
conditions
such that
x1(t)
x2(t)
x3(t)
x4(t)
x5(t)
x6(t)
x7(t)
= ~ (l-C(t))
7T
= 2S(t)
= C(t)
= - X - (l-t)k/p
S
■ - í c«
= - s(t)
TT

135
Given this initial guess, the value of A^ which satisfies the
weight condition is A._ =2.75. This is demonstrated by the follow-
Max
ing values of A^ and which represent specific points on a curve
like that of Figure 5.2:
A JT7
s w
2.74108 0.998S19
2.74396 0.999210
2.75000 1.000000
For A>r = A such
Max s
that J__ = 1
W
the convergence
is measured
in terms of
ERROR as defined in Section 4.4
, and shown in the following tabulation:
Iteration
ERROR
Component
(Position)
Cos
Jw
•f-
JP
0
.696*10"2
*2(o)
1.00000
2.75
1
.151*10~2
x2(.630)
1.00017
2.75
2
.128*10"4
x^(.555)
1.00003
2.75
3
.250*10“12
x2(.550)
1.00000
2.75
Deflection of the
centerline is
shown in Figure
5.3 which
contains
three curves: the initial guess, a plot of subsequent iterates (coin¬
cident for the scale used), and the normalized solution. The optimal
control is displayed in Figure 5.4. Control functions for iterations
subsequent to the initial guess plot coincident to a single curve.
All non-zero components of the state and adjoint variables are shown
in Figure 5.5. Shear (represented by the product x^^) a-*-so included
to give a complete set of structural variables.

13b
tte*
ates
Solu‘
ci°e.
Guess
* 0.9
5-3
T5/Í>ic
al
tlec
tl°a
o5
Ceí1
tet

Figure 5.4 Typical Optimal Control Function
137

138
Figure 5.5 Typical State and Adjoint Variable Soltuions

139
Notice that for the case where density and modulus are equal
to unity, the condition on specified weight simplifies to
1
JTT = / u, (t)dt = 1 (5.5.2)
w 1
where implicitly y = 1. As the value of increases, it reaches an
upper limit of unity: values of a^ larger than unity violate the
weight condition (5.5.2). Thus, for no upper bound on u^ there are two
limiting cases associated with lower bound values,
a = 0 and a = 1.0
L J-i
Values of a not contained in this range automatically violate some
J-j
given condition. Each value of 0 < a < 1.0 has a corresponding optimal
eigenvalue A^ arid design profile u., (t) a, a(x/L) . Several of these
1
profiles are shown in Figure 5.6. In all but the uniform column limit¬
ing case for a =1.0, the optimal profile has a transition region
* *
ere = up- The point where decreasing u^(t) intersects the lower
wh
bound a occurs at a time designated t . Optimal control is on the
Li i-J
lower bound for t > t .
Li
A comparative measure for the effectiveness of the optimiza¬
tion is the ratio of A., to the eigenvalue for a uniform column of
Max
equal weight. This latter eigenvalue is the classical Euler buckling
load denoted A-,, and corresponds to the case a = 1.0. Thus,
h Li
> Vl ii
E" Vo ' 4
The dependence of the effectiveness ratio A^^/A^, and the lower bound
intercept tT upon aT is shown in Figure 5.7. As is typical of simple
Li Li

Figure 5.6 Optimal Area Distributions Parametric in Lower Bound:
No Upper Bound
140

141
\,a /A*
Max n
CINT
Figure 5.7 Effectiveness Ratio and Lower Bound Intercept:
No Upper Bound

142
problems of this type, the most improvement occurs for the smallest
excursions from the uniform case (.6 < a < 1.0). To allow a to be
"" Li ~ L
reduced from .6 to 0 provides much less improvement in the cost.
Observe that the maximum value for a(x) occurs at t = 0 for
a^ = 0, and is 1.333. Thus for a^ > 1.333 no optimal profile in
Figure 5.6 is constrained by the upper bound on u^(t). For a^ = 1.333,
at one point t = 0,
a(0) = a^
and
a(t) < au t > 0
Therefore, instead of being unbounded in (5.5.1), it is only necessary
that a^ > 1.333. An a^ this large is essentially unbounded for the
given weight constraint.
Results similar to those above have been obtained for similar
problems by other methods. The next examples are for a case not yet
reported in the open literature. These cases could not be obtained
without specifying both upper and lower control bounds. The first
example has a lower bound specified with a range of various upper
bounds, decreasing to the limiting case of a uniform solution. The
first example of this case is
1.0 < afJ < 1.333
Since the nature of the solution methods differ only in details, no
solution methods are presented. Optimal profiles are presented for
various values of a^ given a^ = 0. In Figure 5.8 the limiting cases

Figure 5.8 Optimal Area Distributions Parametric in Upper Bound:
Lower Bound a =0.0
iu
143

144
are shown with three intermediate cases as well. One case is displayed
as a dashed line, since it is obtained as an approximation. As a^-*1.0
the numerical procedure becomes unstable for reasons discussed in
Chapter VII; the case for a^ = 1.05 must approach the uniform cylinder
associated with a^ = 1.0. Because the transition region where
a(x/L) < a^ is approaching a vertical straight line in the limit as
a^ ->â–  1.0, the approximation is taken to be:
a(x/L) = ay 0 < x/L < ty
a(x/L)
a-V j
t^ < X/L < l
That value of t^ which satisfies the specified weight condition under
the assumed form of the approximation is given by
t
U
1
No further use is made of this approximation; its sole reason for
existence is to provide an additional curve in Figure 5.8 to more effec¬
tively demonstrate the transition from the case with no upper bound to
the uniform beam case.
The characteristic form of solutions is indicated by a plot of
tjj versus a^ shown in Figure 5.9 together with the effectiveness ratio
A., /A^. Just as observed in the case possessing only a lower bound,
Max E
when only an upper bound is present the first excursions from a uni¬
form column (1.0 < a^ < 1.2) provides the greatest cost improvement.
In the preceding case, although only t^ is presented, the
implication that tT does not exist is incorrect. By observation of
Li

145
AMax/AE'
Figure 5.9 Effectiveness Ratio and Upper Bound Intercept:
Lower Bound a-. =0.0

146
Figure 5.8 and the definition of t it is obvious that for all cases
l_i
t = 1.0. This occurs as a consequence of selecting a = 0.0; in order
Li Lj
to demonstrate a more general behavior of the solution process, a non¬
zero lower bound is selected in order to provide a non-constant t^.
The third configuration of the column buckling problem to be
considered is for the control bounds
aL = 0.3
1.0 < ajj < 1.327
where the upper limit to a^ is such that for a^ > 1.327 no upper bounds
are ever active as u^ is essentially unbounded from above. Optimal
profiles for the limiting and three intermediate cases are shown in
Figure 5.10. The dashed curve is an approximation composed of three
straight line segments (numerical instability was again encountered as
the uniform column limiting case was approached). Mathematically, the
approximation is
a(t)
U
< Í cu
(t tu)
' a« ■ (vl>
t <
t
"L
such that the case which satisfies the specified weight condition
requires the control bound intercepts of the approximation to satisfy
'U
+ fcL =
1 -
aTT -
a
a
L
L
2

Figure 5.10 Optimal Area Distributions Parametric in Upper Bound:
Lower Bound a =0.3
Li
147

148
From an extrapolation of intercept values associated with higher values
of a in Figure 5.10, t is found to be about 0.985. The corresponding
U i-i
value of t is thus 0.882. Just as in the preceding case it must be
emphasized that this approximation is included only to more clearly
illustrate the transition of a(x/L) from one limiting case to the other
as a^ decreases from 1.327 to 1.0.
Effectiveness of design optimization is revealed in the plot
of A,, /A„ versus a_T (see Figure 5.11). Characteristic solution forms
Max E U
are also depicted there in the plot of t and t^ dependence upon a^.
It is to be noted that tT deviates little from the value of unity but
very definitely has a functional dependence upon a^.
5.6 Inequality Stress Constraints
Having obtained solutions for various control constraints, the
next problem configuration attempted involves a mixed constraint.
Initially it was decided to restrict the maximum allowable normal
stress due to bending:
K(X)I < aMax 0 < x < L
Recognizing that for any cross section the maximum value of Og(x) occurs
at the "outermost fibres," the distance d(x) of these fibres from the
bending axis must be determined. On the basis of assuming similar cross
sections, the function f(x) appearing in the A(x) expression (5.2.1)
also determines d(x), i.e.,
d (x) = dQf(x) = dQaLi(x)

149
Figure 5.11 Effectiveness Ratio and Control Bound Intercepts:
Lower Bound a^ = 0.3

150
where d^ is some reference value. In terms of Bernoulli-Euler bending
equation, the inequality constraint on bending normal stress is
laB(x)
(x)d(x)
V”
<
- o Max
(EIY")d(x)
Vx)
<
- a Max
E(x) d(x)Y"|
<
- a Max
By introducing dimensionless variables and squaring to remove the
absolute value signs,
e2 (t)a(t)ri
<
a
2
Max
which is manipulated into the desired mixed constraint form when written
in terms of state and control variables:
0
where
(5.6.1)
L o
Max
Eodo
Besides satisfying the geometric constraints, the u^ control
component must also satisfy a state dependent constraint
ul(t) > (x3/o0)2/3 > 0
Results for this case of mixed constraints are not presented. To be
a valid demonstration, the optimal control ought to be constrained by
the mixed constraint (5.6.1) on at least part of the interval 0 < t < 1.

151
In the present example, for small values no solution exists that can
fulfill both the stress and geometric constraints. Large values of
result in all geometrically admissible control simultaneously satisfy¬
ing the mixed constraint. Intermediate values produce solution iter¬
ates in the quasilinearization process that were constrained by (5.6.1)
over some interval. However, the final iterate u^^Ct) is not con¬
strained by the mixed constraint at any point 0 < t < 1. For this rea¬
son no mixed constraint examples are presented. Theoretical consider¬
ations of other possible stress constraints are given in the following
paragraphs.
A more realistic stress constraint is to limit the stress
component o(x) normal to a cross section, composed of a term Og(x) due
to bending and a term Op(x) attributed to the end load P. With refer¬
ence to the definition of 0 shown in Figure 5.1, a(x) and its constitu¬
ents are:
o(x) = oB(x) + cr (x)
Og (x)
PL (x)d(x)
"Vr-
ap(x)
P cos 6(x)
A(x)
Differential calculus and trigonometry provide two identities that are
used to evaluate cos (0) as a function of x in terms of the slope of
the centerline. They are
dY
= tan 0
dx
tan20 + 1 = sec20

152
From this,
cos 0 = < 1 +
(gj
2
Thus, the normal stress inequality becomes
M (x)d(x) r
a(x) = - + -f- h +
IB(x) A(x) ^ \dx /
<
Max
By introducing dimensionless state and control variables, after
much manipulation the following mixed constraint inequality is obtained
, k,3 , a
= - (up + -
0
"<(u?> + x3/a0
Three parameters exist for this derivation,
L a.
an =
Max
0 Eodo
the Euler buckling slimness ratio
— a, a geometrical slimness ratio
JL
A simpler constraint function can be obtained by taking Op(x) as merely
°p(x) "iw
for which
(x, u)
UJ + x3/o0

153
This function can be analytically solved for whenever
%
= 0* What must be solved is a cubic equation in u^.
(up3 - A (up - B = 0
* â–  t ft' ft
(5.6.2)
B = x3/a0
A reduced cubic equation with real coefficients
y3 + py + q = 0
has a discriminant
A = - 108R, R = (¿ p)3 + (~ q)2
that determines the nature of the roots to the equation. For,
A < 0 -> one real, two complex roots
A = 0 -*â–  two real, equal roots
A > 0 -> three distinct real roots
Consider a typical problem where the cross section is circular, and
having typical parameter values:
k.
1
2
o„ =
.06
it follows from

154
and
A = - 4p3 - 27q3
/11.664 o ) x 10
X3 x10 5/7
Thus for typical parameter values, there is a single real u^ value
whenever
x3 > X3 x 10“5/7 = 4 x 10~5
or
x3 > (X3 x10~5/7)^ * 6.2 x l(f3
In other words, near the tip where both t and x^ approach zero, three
real distinct roots to (5.6.2) exist. Unless this possibility is
excluded by a sufficiently large minimum bound to u^, the question of
which root is appropriate must be answered. Having demonstrated how
complicated the analysis becomes for this particular stress constraint,
no further considerations are presented. The developments are included
to show that even simple stress constraints can be too complicated for
ordinary methods of analysis.

CHAPTER VI
FINITE ELEMENT METHODS IN STRUCTURAL
OPTIMIZATION: AN EXAMPLE
6.0 Introduction
In order to provide a contrasting comparison to the foregoing
PMP examples the cantilever beam problem is solved by using finite
elements in conjunction with mathematical programming techniques. It
is assumed the beam has a constant specified width in order to better
exemplify the concepts contained in this chapter. Numerical solutions
obtained by a feasible directions algorithm are compared to the PMP
results. This example serves to illustrate the principles used in the
theoretical derivations.
6.1 Finite Element Problem Statement
The same cantilever beam problem solved in Chapter IV is treated
here with different techniques. In what follows the structural system
is modeled by and solved with finite elements. A direct consequence
of this mathematical representation is that the optimization problem is
transferred from a vector space of continuous functions to a space of
finite dimension. Since no contributions to finite element theory are
made, the method itself is not discussed. The reader is referred to
Zienkiewicz (1971) for details of the method. Bernoulli-Euler Bending
Theory is applied to a fixed length cantilever beam of constant width
155

156
and material properties, which is symmetric about its central axis.
The problem is to find the variable height h(x) that gives the minimum
tip deflection due to its own height, subject to hard constraints
of maximum and minimum allowable heights.
A higher ordered element is used to obtain better accuracy;
both the displacement and its derivative at the nodes are prescribed
to be the independent variables. For this simple problem, n nodes divide
the beam into (n-1) elements of equal length.
Hence, the displacement and its derivative are the deflection
and rotation of the beam's centerline at the locations of the nodes.
It is also assumed that each element is linearly tapered, whereby the
Bernoulli-Euler bending equation for element "i" takes the form
EI (x) = M(x)
dx2
I(x) = J2 wh3(x)
Position along the beam x, height h(x), and displacement of the beam
y(x), and slope 9(x) of the centerline, are approximated by vectors
composed of discrete elements:
T
x = [x x . .. x ]
— 1 z n
h(x) = [h(x ) h(x ) ... h(x Í]1
1 Z n
z(x) = [yix^ y(x2> ... y(xn)]T
0(x) = [0(x ) 6(x ) ... e(x )]T
1 z n
Whenever a height distribution is specified, a finite element solution
for the beam's slope and deflection can be obtained.

157
6.2 Mathematical Programming:
Gradient Projection Method
With the structural analysis given in the form of a finite
element solution, the mathematical statement of the optimization problem
is
Given: (i) material properties p,E
(ii) dimensions L, w
(iii) number of nodes n
Find: Min [y(x )]
h(x) n
subject to b < h(x^) < d i = l,...,n
This may be stated as a nonlinear mathematical programming problem,
whose optimality conditions are prescribed by the Kuhn-Tucker Theorem
given in section 3.2. For this particular beam problem corresponding
quantities of the theorem are
x ^ h_
F(x) ^ v(x)
x=L
rh(x ) - d i = 1,...,n j = 1
g (x) ^ <
J v_b - h(>:^) i = l,...,n j = i+n
The numerical algorithm used to obtain the optimal solution
satisfying the Kuhn-Tucker conditions is a modified gradient projec¬
tion method described by Kowalik (1970). Briefly speaking, it uses
projections of the cost function gradient into a subspace satisfying
currently active constraints. Starting from a feasible design point
h^, a sequence of feasible designs is generated such that

158
where
■V-! ’-2’ • • * • • • ’^OPTIMAL
h t = h + a S
-q+1 ~q q
S = - PVF
a >0
q
T -1 T
P = I - N(N N) N
A derivation of projection matrix P is found in Kowalik (1970).
Columns of N are the gradient vectors of the active constraints; thus,
N is an (n * r) matrix where r is the number of active constraints and
0 reduces to the identity matrix I. With P so defined, those components
of -VF that lead to a constraint violation are subtracted from the
direction of steepest descent. In this manner S is the direction which
best improves the cost function without leaving the region of feasible
designs where g^ (x) < 0. From the form of g^(x) in this particular
problem, the typical element N of N is defined as:
1K.
N., =
ik 9x
i = 1,...,n k=l,...,r
where a, are those indices .1 < j < 2n associated with the r active
k - J -
constraints for which g^. (x) = 0. Hence,
N., = 0
xk
i ^ a
r i
i = a.
N
ik
l-l i = a.
and
and
a, < n
k -
a, > n
k

159
There is no difficulty in calculating the gradients required
by N; since the g^(x) are specified functions the gradients can be
determined explicitly. Derivatives required to evaluate VF are not
so easily obtained. To circumvent the lack of an explicit formulation,
the derivatives are obtained as follows. The i*"^ component of VF is
found numerically by
3F
9x.
i
F(x + Ax) - F(x)
Ax
where the only nonzero component of Ax is
Ax.
i
JL_
10
x.
i
To avoid instability as the solution is approached, the increment is
modified to
10
kx.
x
where k is the maximum percent change of all components of from the
previous iteration. No numerical instabilities were encountered with
this scheme; however, if too small a value of numerical tolerance is
chosen, the convergence criteria are never satisfied and the algorithm
begins to "hunt" in the neighborhood of the optimal solution. A sim¬
plified flowchart is shown in Figure 6.1.
To furnish a comparison standard, the Bernoulli-Euler equation
for a beam with a linearly tapered center section was integrated anal¬
ytically. This revealed some interesting qualitative results which are
discussed below, with quantitative results given in the following sec¬
tion. First, the expression for the second derivative of y(x) is

160
Figure 6.1 Simplified Flowchart of Algorithm

161
only piecewise smooth due to I(x). To integrate the equations for the
linearly tapered beam required dividing the range of x into three
reqions. In the region 0 < x < the height h(x) = d and y(x) is
a quartic polynomial in x. Variable height complicates the equation in
the region x^ < x < x^; however, the centerline deflection is still
a function of a single variable, u(x), defined as
u(x) = 1 - ( l ~ L
xi
X 1
where
A = 2 — tan a, a constant
d
tan a =
1 /LV1 (1 " 2 Id
X2 X1
such that
u(x2) = ¿
One of the complicated terms of the expression for y(x) in this region
of the beam is
12 In (u (x) ) - u (x) ]
which is undefined at x = when b/d = 0. Within the third region
x^ < x < L, the height h(x) = b and the centerline deflection is a
quartic polynomial in (x - x^) •
It is shown in Hornbuckle et al. (1974) that in all three
regions the deflection could be written in nondimensionalized form

162
iM=fct
, b.
(x’d}
y
o
(6.2.1)
This result supports Dixon's statement that d/L is not a parameter of
the dimensionless system. These equations indicate that dimensionless
results indeed depend only upon b/d. However, for a particular beam of
given dimensions, the actual tip deflection does increase as the square
of the slimness ratio L/d.
6.3 Results
Before any attempts were made to obtain results, certain basic
questions had to be answered. First, the integration inherent to the
finite element solution as described in Zienkiewicz (1971) is done
numerically, thereby requiring determination of the proper step size.
It was found that with the range of parameters used in this problem,
ten integration intervals per element gave the desired accuracy without
excessive computation. Secondly, there was a question of how many nodes
are required to give a valid representation of the beam. Because the
system is so simple, as few as two finite elements produce good struc¬
tural results vis-a-vis tip deflection. However, unless eight or six¬
teen elements were used, the approximation of h(x) by h(x) was not
acceptable. The results for eight or sixteen elements were almost iden¬
tical. (For the latter two cases the execution times on an IBM-370/165
were 15 seconds and 110 seconds, respectively.)

163
Basic operation of the algorithm and convergence to the optimal
solution is illustrated in Figure 6.2 for a typical value of b/d. The
profile is displayed at each stage in the design iteration process,
including the final design. A "stage" in the design process occurs
whenever another constraint is encountered, requiring the recomputation
of VF, P, and S. To simplify the figure only nine nodes were used; the
optimal profile for the beam having linear taper is shown for comparison
The listing below indicates how constraints are encountered as the numer
ical process approaches the optimal design. Tip deflections for each
design iterate are included. Thus, for the design process depicted by
Figure 6.2,
No. of Active
Cons traints
0
1
2
3
4
5
Comparison
Tip Deflection
.108879
.027556
.020673
.017103
.016646
.015351
.015332
Initial Design
Final Design
Linear Taper Design
In the earlier studies, the problem was found to have a single
parameter b/d; however, two different beam profiles h(x) were obtained
for this problem by various authors. It was found with the finite
element method that for numerical precision corresponding to slide
rule accuracy the optimal profile is an almost linearly tapered beam.
Greater numerical accuracy provides a distinct curvature in the tapered
portion of the beam. This is shown in Figure 6.3 by the two curves
generated with the finite element method. The profile depicted by the

Intermediate Feasible Design
Constrained Optimal Design
Figure 6.2 Design Iteration to Optimal Profile
164

b/d = .25
L/d = 20
h(x/L)
d
Figure 6.3 Optimal Profile: Comparison of Methods and Numerical Tolerance
165

166
dots is taken from the work by Boykin and Sierakowski (1972), and is
presented for corroboration. For this latter curve the absolute magni¬
tude of the error in satisfying the differential equations is less
, __—15
than 10
Since tip deflection is the cost function, the tip deflection
from the finite element algorithm was compared to that from the linearly
tapered beam (Figure 6.4). Differences were less than .5% except near
b/d = 0. This is attributed to the natural logarithm term in the linear
taper solution discussed previously. No computer results were obtained
for the linear taper solution with b/d almost zero because the natural
logarithm term became excessively large. This may explain why all
earlier studies using this example problem contain no results for
b/d < 0.1. However, the finite element algorithm converged to a solu¬
tion for b/d = 0 with no difficulty.
Further analysis centered on a statement in which Bellamy
and West (1969), claim: ". . .as b/d increases, the midsection of
the beam profile reduces in length and increases in slope." Having
found quite close agreement with the linear tapered tip deflection,
we expected at least qualitative agreement for the angle of taper a.
Angle of taper for the linearly tapered beam and the finite element
results are shown in Figure 6.5. Notice that both curves approach zero
in the limit as b/d approaches one. Of greater importance is the fact
that the angle of slope for both methods is reasonably close only for
b/d < 0.5. Figure 6.6 shows that for b/d > 0.5, if the beam is divided
into sixteen finite elements, the tapered section contains only three

Y
TIP
(feet)
Figure 6.4 Tip Deflection Versus b/d
167

Figure 6.5 Angle of Taper Versus b/d
168

h(x/L)
d
Figure 6.6 Optimal Beam Profiles via Finite Elements
169

170
nodes because it spans such a short section of beam. This is believed
to be insufficient for accurate representation. The same shortcoming
is also apparent in Figure 6.7, which shows the optimal profile dimen¬
sions reported in the various studies. For b/d > 0.5, the value of r.
d
obtained by the finite element method diverges from the other solution
method results.
To verify independence of the solution to L/d, a series of runs
were made with fixed b/d and the slimness ratio L/d ranging from five
to eighty. Resulting profiles at first suggested that L/d is a param¬
eter, but closer analysis revealed there was an oversight in the use of
numerical tolerances. Initially a single tolerance was applied to both
geometrical constraints and the Kuhn-Tucker test. The former is inde¬
pendent of relative beam shape, but the latter is not; this is implied
by the form of the nondimensionalization coefficient given by (6.2.1).
Basically, the projection matrix P deletes those components
from -VF which are normal to an active geometrical constraint boundary.
Since all numerical schemes contain inherent accuracy limits, the Kuhn-
Tucker test is considered to be satisfied when at those points not on
a boundary
3y(xn)
3x. < CG
i
Yet as the slimness ratio L/d increases, for any point on the beam the
magnitude of the component of the gradient also increases. By requir¬
ing the Kuhn-Tucker test to satisfy the same tolerance for all values
of L/d, as L/d increases the solution is effectively forced to become
more accurate. In fact, more accuracy than is possible from the

171
Digital (Dixon)
•— Analog - Hybrid
Quasilinearization
Figure 6.7 Dependence of Intercept Location upon b/d

172
algorithm was required and the system began hunting around the solution,
failing to converge. By numerical experimentation and curve fitting
it was found—Hornbuckle et al. (1974)—that uniform accuracy could be
obtained for this particular problem when the tolerance used in the
Kuhn-Tucker test is
N0
£g = £o + ko(L " V
^U5
d
Nq = 2.77825
kQ = 1.184045 x 10-6
Subscript zero indicates values associated with reference length corre¬
sponding to L^/d; the value of five was chosen because it represents an
approximate lower bound for the validity of Bernoulli-Euler bending
theory. In uniform beams whenever L/d < 5, the effect of shear is no
longer negligible. As an example of why care should be exercised in
selecting the range of parameters to generate data, Dixon (1967, 1968)
obtained results for 0.2 < b/d < 4; having assumed the effect of shear
negligible, he considered only the range of L/d where it is significant.

CHAPTER VII
COMMENTS ON NUMERICAL INSTABILITY IN
THE OUASILINEARIZATION ALGORITHM
7.0 Introduction
Quasilinearization is an indirect type of numerical scheme in
which a solution is obtained for the differential equations that must
be satisfied by the optimal solution. Since a standard IBM SHARE pro¬
gram is used without modification, the method is not discussed. Inter¬
ested readers are referred to any standard textbook on numerical tech¬
niques, or to Ghosh (1973).
This type of algorithm requires an initial guess of the state
and then iterates from that guess to the solution. Control functions
associated with the final iterate are the optimal control functions
actually sought. If the process converges at all, the convergence is
rapid and typically requires fewer than ten iterates. However, the
paramount difficulty of using this method is to select an initial guess
which is "good" enough to yield convergence. This chapter contains some
numerical data related to specific cases of the example problems where
convergence difficulties are encountered.
7.1 Computer Program Convergence Features
As explained in section 4.4, convergence is measured by whether
or not the iterate satisfies the differential equation. In terms of
173

174
the quantity ERROR defined by (4.4.2), in subroutine QUASI the test for
numerical instability involves three program parameters:
E1MAX the value of ERROR for the initial guess
E2MAX the value of ERROR for some subsequent iteration
BLO a specified instability parameter
Whenever the ratio of E2MAX to E1MAX exceeds BLO the program terminates
on an error exit. Typical values of BLO are 10 to 20. The primary mode
of convergence is for E2MAX, which decreases with each iteration under
normal circumstances, to become smaller than a specified value (assumed
0
to be 0.5 x 10 ). Other special modes of convergence contained in
QUASI were never used by any of the cases run for either example problem.
The purpose of this test is twofold. In some instances the
initial guess is not "good" enough and the first few iterations possess
E.2MAX values greater than E1MAX, but is sufficiently good for the algo¬
rithm to ultimately converge. The other purpose is to prevent excessive
computation for those cases where the initial guess is so bad that the
numerical process is divergent.
Another related convergence test, is required by the subroutine
RECIP which performs required matrix inversion. This test involves
a program parameter SUM which is defined in terms of a matrix to be
inverted, and the inverse C 1 obtained numerically by RECIP. If C .
and C denote appropriate elements respectively, then
SUM = E T. ICT,1 C, . I
. ... ' xk kj 1
k i/j

175
If the matrix has been inverted exactly the value of SUM is zero.
Whenever SUM is less than program parameter OFFDG the inversion is con¬
sidered to be sufficiently accurate. The value selected for all cases
-15
for both example problems is 5 x 10 . Whenever the inverted matrix
fails to satisfy the test on SUM, the subprogram iterates upon the
inverse to improve it. If the improved inverse satisfies the accuracy
requirement, the result is returned to QUASI. After five unsuccessful
improvement attempts, results of the fifth attempt are returned to QUASI
A singular matrix results in immediate termination of all calculations
via an error exit.
During the operation of the quasilinearization algorithm two
different matrices must be inverted. One of these, D., exists for every
—i
element i and each separate _D^ must be inverted. From all of the Ih
another matrix is defined which also requires inversion. The specific
form of these matrices is not important; however, their definitions may
be found in Ghosh (1973, pp. 52-58). It is noteworthy that in the cases
run for the example problems no difficulties were experienced in the
inversion of any I) matrix. All matrix inversion problems were asso¬
ciated with the J3 matrix.
7.2 Numerical Instabilities for
Cantilever Beam Example
An initial guess for the beam problem was obtained from the
solution for a uniform beam, that is, for parameter values a/c = b/d = 1.
Cases were then run for a range on both parameters from 0.1 to 1. As
would be expected, no convergence difficulties were encountered with

176
combinations of parameter values corresponding to a nearly uniform
beam. In fact, with the initial guess based upon the uniform beam,
only three cases failed to converge. For this initial guess given by
(4.4.3) the following three cases exhibited numerical instability.
a/c = 0.1 b/d = 0.3
= 0.2
= 0.1
(7.2.1)
All other combinations of parameter values resulted in convergence by
the quasilinearization algorithm.
Since all of those cases which did converge showed very little
change in structural state components x^.(t), i = 1,...,4, the uniform
beam initial guess was replaced by a guess based upon the case where
a/c = 0.1 and b/d = 0.4. It was suspected that the adjoint variable
guesses were not sufficiently "good." Hence, the equations (4.4.3)
were replaced by:
Xi(t) = .025 * t2
x2(t) = .040 * t
x (t) = .100 * (1-t)
x,(t) = -.400 * (1-t)2
P3(t) = 5.00 * t3
P4(t) = -1.25 * t3
0 < t < 1
(7.2.2)
With this initial guess only one case corresponding to the three sets
of parameter values (7.2.1) failed due to numerical instability; the
first two cases of these cases converged.

177
Having succeeded in forcing two additional cases to converge
by improving the initial guess, this same approach was tried with the
third case as well. The initial guess was based upon the converged
solution for the case with a/c = 0.1 and b/d = 0.2. Equations (7.2.2)
were replaced by
x (t) = .020 * t2
x2(t) = .040 * t
x3(t) = .080 * (1-t)3
x.(t) = -.350 * (1-t)3
4
P3(t) = 13.0 * t4
p.(t) = -2.00 * t4
4
0 < t < 1
(7.2.3)
With this initial guess the case for a/c = .1 and b/d = .1 converged
also. This last case gave a complete data set for parameter values
.1 < a/c < 1
.1 < b/d < 1
in increments of .1. No cases for parameter values equal to zero were
run since the parameters appear in the denominator of various expressions.
As mentioned earlier, indirect methods such as quasilineariza¬
tion generally converge rapidly if the initial guess is sufficiently
good. To try to determine why the three cases failed to converge, all
of the cases were closely examined. Two characteristics were observed:
(i) There is little change in the maximum value of each separate
state component, regardless of the parameter values. Curva¬
ture of x3(t) and x^(t) did change for lower parameter values;

178
this is evident in different expressions used for the
initial guess of these two variables. However, the var¬
iable that actually caused the instability termination
was p^(t) and not a state component.
(ii) The maximum value of both adjoint variables increases with
decreasing parameter values, with an associated increase
in curvature.
With p^(t) isolated as the source of convergence difficulties for cases
with small parameter values, p^(t) was plotted for several combinations
of small parameter values.
The increased curvature mentioned above is shown in Figure 7.1.
All cases that failed to converge exhibit a "numerical instability
termination" for which p^(t) becomes divergent. Maximum error in satis¬
fying the p^ equation occurs in the region .755 < t < .965 for these
divergent runs. From the figure it can be seen that this is the region
of greatest change in the curvature of the solution p^(t) - This sug¬
gests that the linearization gradients in the quasilinearization algo¬
rithm may not be valid. If that is the case, then the linear expansion
about the known iterate cannot be expected to result in an improvement.
A modification developed by Ghosh (1973) to inhibit instabilities asso¬
ciated with too-large improvements has no effect upon these numerically
divergent cases. Removal of this possible source of numerical instabil¬
ity further implies the difficulty is related to the linearity assumptions.
An explanation for the increasing slope is available from the
differential equation for the adjoint variables, where

179
Figure 7.1 Solutions for Adjoint Variable p^(t)
for Certain Cases of Interest

180
P3 = (l-t)/u13u2
P3(0) = 0
(7.2.4)
P4(0) = 0
Since both and u^ are equal to their minimum allowable values near
t = 1, p>3 increases with decreasing parameter values in this region.
However, the term (1-t) forces j>3 to zero at t = 1. For unknown rea¬
sons, the large curvature associated with this effect seemed to pose
no difficulties for the numerical algorithm. As seen in equations
(7.2.4), if P3(t) can be determined, then p^(t) is readily obtained.
Curves for p4(t), corresponding to cases presented in Figure 7.1, are
given in Figure 7.2. All six cases exhibit positive curvature and
possess none of the inflection points seen in their counterparts of
Figure 7.1. The increase in amplitude of these too variables for
decreasing values of b/d is illustrated in Figure 7.3. In this figure
it is seen that with respect to increasing amplitude as b/d -»- 0, P3(t)
is the critical variable.
The major disadvantage of the indirect solution method is that
little is known about the connection between convergence and require¬
ments upon the initial guess. For this specific example, only one of
the seven variables causes convergence difficulties, and the result¬
ing numerical instability occurs in a region of large, rapidly changing
curvature. No convergence problems are encountered when the curvature
of the initial guess more closely resembles the final solution. All of
this suggests that the question of whether or not an initial guess
allows the quasilinearization algorithm to converge, may depend upon the
respective curvatures of the initial guess and the solution.

181
Figure 7.2 Solutions for Adjoint Variable p^(t)
for Certain Cases of Interest

P3(t)
-10 p^(t)
•2 .4 .6 .8 1.0
b/d
Figure 7.3 Terminal Point Amplitude of p^(t) and p^(t):
Dependence upon Parameter b/d Value
182

183
7.3 Numerical Instabilities for
Column Buckling Example
Another type of convergence failure was experienced with the
column buckling problem. In these cases which did not converge, each
of the variables was divergent, increasing with each iteration begin¬
ning with the initial guess. All of these numerically unstable cases
were for parameter values which resulted in a solution that approached
the uniform column limiting solution. Attempts to force convergence
were unsuccessful.
The first attempt was to improve the initial guess used as a
starting point by the algorithm. Although this technique succeeded
with the beam problem, it failed with the column problem. The next
attempt to obtain convergence was to employ the instability inhibiting
modification developed by Ghosh. The only result was to prolong the
numerical divergence. At this point it was suspected that perhaps the
source of difficulty was related to normalization of the eigenfunction.
However, when each iterate was normalized as indicated in section 5.4,
the sole result was that the state variables maintained reasonable
amplitudes while the amplitudes of the adjoint variables increased more
than without normalization. Doubling the number of intervals in the
finite difference approximation (from 100 to 200) was also ineffective.
A possible source of the numerical difficulty was the matrix
inversion. For those cases failing to converge each of the JD matrices
was inverted accurately but the inversion of C was questionable, as
indicated by the value of SUM. Since each iterate depends directly

184
upon the inverse of JC, a degradation in would cause a correspond¬
ing degradation in the "solution" at each iteration.
To illustrate what happens in these cases consider a specific
example: a = 0, a =1.1. A plot of values of SUM versus t , the
ij U u
intercept of the upper bound, is shown in Figure 7.4. These data points
represent the C_ ^ accuracy at different iterations for various initial
guesses—acceptable accuracy is indicated by the straight line. Basi¬
cally, if an initial guess yielded a t of approximately one-half, each
iteration gives an increased value of t^ with a correspondingly less
accurate C) As t^ approaches one, the solution approaches the limit¬
ing uniform column.
The result of this is that may become ill-conditioned if the
parameters are chosen such that the solution is similar to a uniform
column. There are no explicit data or derivations to indicate this is
the situation, only the implicit suggestion of the data in Figure 7.4.
However, as stated in the preceding section, there is little known
about why certain initial guesses fail to converge in the indirect
solution algorithms.

SUM
1 x 10
-12
5
2
1 x 10
-13
5
2
1
a = 0.0
!j
©
©
o
au = 1,1
©
o
©
©
©
/
/ / /
/
/ /
/
/
/ / / /
/
0
\
©
Acceptable Matric
Inversion
0
.4
.6
!
.8
1.
t
U
Figure 7.4
Degradation of C ^ Accuracy as Iterate Approaches a Solution
Similar to the Uniform Column Limiting Solution
185

CHAPTER VIII
CONCLUSIONS AND RECOMMENDATIONS
8.0 Summary and Conclusions
This dissertation treats the optimal design of elastic structures
subject to both hard inequality constraints and subsidiary conditions.
It is shown that structural problems can be classified into two general
types depending upon whether or not the cost functional is an eigenvalue.
For the conservative systems considered, which are described by self-
adjoint differential equations, it is shown that the minimum weight
problem is identical to the maximum buckling load problem.
Pontryagin's Maximum Principle is analyzed as a nonlinear
programming problem in Chapter III. Specifically, the theory of the
gradient projection method is applied to the maximum principle, with
the control subject to inequality constraints. From this an explicit
formulation is obtained for the Lagrangian multipliers that adjoin the
constraints to the variational Hamiltonian.
Additional characteristics of the maximum principle result from
the gradient projection method analogy. In the latter, the projected
gradient is that specific direction which minimizes the cost without
violating the constraints in effect at any point in question. As
applied to the maximum principle, the projected gradient is
* â– p
H = - (H + _¿)ü which in u-space denotes the direction of constrained
186

187
maximum descent whenever jj is given by equation (3.3.4). While this
attribute is not pursued further in the dissertation, it may prove use¬
ful in subsequent applications of the maximum principle. To further
illustrate the nonlinear programming principles used in the theoret¬
ical development, the example of Chapter IV is approximated by finite
elements, and solved by a gradient projection algorithm. Operation of
the algorithm is illustrated and the results compared to those obtained
with the maximum principle. When a sufficient number of finite ele¬
ments are used to adequately describe the structure, the results of the
two methods are identical.
A problem concerning the minimum deflection of a beam is opti¬
mized by using classical maximum principle techniques. It is shown
that no finite solution is obtained for unconstrained control; this also
holds for the column buckling problem treated in Chapter V. Further¬
more, it is shown that for the conservative system corresponding to the
buckling problem, the minimum weight and maximum buckling problems are
identical. Another result of using the maximum principle is that for
the problem described, in order to have a nontrivial solution the modu¬
lus and density must be the maximum and minimum allowable constant
values, respectively. This is a mathematical verification of an intui¬
tive result.
Numerical instabilities experienced by the quasilinearization
algorithm are detailed in Chapter VII. While no concrete results are
obtained, from the data presented it appears that the question of con¬
vergence is related to the respective curvatures of the initial guess
and the solution.

188
8.1 Recommendations
Further work suggested in the course of dissertation research
falls into three categories: structural considerations, mathematical
theory, and numerical solution techniques. Under the first category
the most obvious extension is to not invoke the linear bending assump¬
tion. Since the quasilinearization algorithm handles nonlinear TPBVP,
theoretically there is no need for the assumption. In contrast to the
first of equations (5.2.4) the complete, nonlinear bending problem is
described by
(s(t)h) = M(t)[l + Ú2]3/2 , 0 < t < 1
1
M(t) = A[n(l) - u(t)] + k / m(£)[n(£)-u(t)]d£
t
n (0) = ñ(0) =0
Another interesting possibility is to include the effect of shear.
These structural aspects pertain to specific problems. A more
general result, and hence more interesting, is related to expanding the
class of problems that can be treated by energy techniques. It is sug¬
gested that the mutual potential energy method and the application of
adjoint systems to nonconservative problems be investigated to see if
a "general energy method" exists. Such a development would expand the
class of problems now considered to be amenable to the energy method
and its well-developed theory.
In terms of mathematical theory, the double optimization
problem associated with extremization of eigenvalues for self-adjoint
systems is an inviting prospect. No significant contributions have been

189
published in recent years. Any new discovery of a general nature would
be a significant development. Another useful area of mathematical
research would be the further study of the nonlinear programming aspect
of the maximum principle. Beyond this, it might also prove fruitful to
combine the latter with a comparison to dynamic programming.
Finally, the discussion of numerical instabilities of Chapter VII
indicates the possibility that the respective curvatures of the initial
guess and the solution determine convergence. Any development that
could indicate convergence, or the lack thereof, for an initial guess
required by an indirect solution technique, would be a significant
development. One possibility is a study of the sensitivity of the
solution increment function, je(t), to error sources such as initial
guess, step size, etc., for a case with a known solution. Little is
known about this subject.

APPENDIXES

APPENDIX A
HISTORICAL DEVELOPMENTS
The dissertation subject matter originated from two papers:
a short survey of structural optimization problems by Prager and Taylor
(1968) and a technical note by Boykin and Sierakowski (1972) .
A survey of the literature began with the former and revealed that what
were presumed to be recent developments were in fact two hundred years
old. Besides engendering a real sense of humility, my discovery aroused
a fear that some fundamental part of my work might be a duplication of
a much earlier study. I also found the historical development to be
quite intriguing; for that reason this appendix has been included to
give the reader a brief history of the separate development of mechanics
and the calculus of variations. Neither is intended to be comprehensive
however, the reader may enjoy being able to associate some recognizable
equation or method with a specific person and period of time.
The appendix consists of three sections, where the first is an
anecdotal discussion of the major early contributors to the calculus
of variations and a table listing major developments in chronological
order. This is follox^ed by a similar account for mechanics, also having
a table of chronologically ordered developments. Concluding the appen¬
dix is a short biographical table that lists the life span and national¬
ity of most of the people mentioned.
191

192
Newton posed a problem in Book II of the Principia
which requires the techniques of the calculus of variations for solution.
His goal was to find the solid of revolution which has the minimum
resistance in axial flow. No apparent significance was attached to the
problem of either Newton or his contemporaries.
The calculus of variations actually begins with two Swiss
mathematicians, James Bernoulli and his younger brother John. Until
1690 the latter was a student of James, but soon became a rival.
In June, 1696, John Bernoulli posed the brachistochrone problem in
Acta Eruditorum:
A New Problem, to the Solution of Which
Mathematicians Are Invited
Given two points A and B in a vertical plane, to find for
the movable (particle) M, the path AMB descending along which
by its own gravity, and beginning to be urged from the point A,
it may in the shortest time reach the other point B.
According to various sources, solutions were offered by Newton,
Leibnetz, de l'Hopital, and the Bernoullis. At Leibnetz' request
the solution was withheld to encourage others to consider the problem.
In January, 1697, the problem was reannounced. In the following May
issue of Acta Eruditorum, the solutions of the Bernoullis and the
Marquis de l'Hopital were published. John's is alleged to be the most
readable but since it is solved as an analog to an optics problem, the
method is restricted to a small class of problems. The solution of
James is quite geometrical, treating the curve as a polygon with sides
of infinitesimal length. It was also assumed that whatever optimal

193
property the entire curve possessed, was also possessed by each part.
The solution of de l'Hopital was given, but without proof.
At the conclusion of his paper, James posed two more difficult
problems:
First: find the curve of quickest descent from a fixed point A
to a given vertical line. This version of the brachistochrone
problem involves conditions for minimizing a definite integral
where variations at the limits are permitted. Such conditions
lead to natural boundary conditions, or transversality condi¬
tions, and were not obtained until some years later by Lagrange.
Second: among all curves with given length and base, find the
the curve such that a second curve, whose coordinates are
some function of the first, contains a maximum or minimum area.
This subsidiary condition is called an isoperimetrical constraint.
Besides inviting all mathematicians to attempt solutions, James specif¬
ically offered his brother John a prize of fifty ducats for the correct
solution—according to Bliss (1925, p. 12). John claimed the award but
James refused, showing that the assumption of a uniform optimality prop¬
erty for each infinitesimal element was not valid. James published the
correct solutions in the May, 1701, issue of Acta Eruditorum; to include
the effect of the constraint, he used three adjacent polygon elements of
infinitesimal length instead of two that sufficed for the brachistochrone,
and no fewer than four solution techniques:
(1) Equilibrium of forces on an element
(2) Equilibrium of moments on an element
(3) Principle of virtual work
(4) Principle of minimum potential energy
Whether or not the prize existed in immaterial; it is true the problem
embroiled the brothers in a bitter quarrel, and together with the brach¬
istochrone problem marks the beginning of the calculus of variations.

194
All of this work by the Bernoullis was based upon complicated
geometrical results. A student of theirs, Leonhard Euler, expanded and
generalized their initial developments. His first major contribution
to the calculus of variations was the differential equation bearing his
name. Euler suggested that for continuous functions a definite integral
vanishes identically if the integrand vanishes. When the integrand is
the product of some function M(x,x,t) and an arbitrary variation Sx(t)
the function must vanish if the variation is to be arbitrary. Euler's
equation is M(x,x,t) = 0; up to the first part of this century this
vanishing integrand was referred to as "Euler's Fundamental Lemma"—
"Euler's equation" is the more recent appellation. He also was the first
to indicate the two general classes of optimal solutions obtained from
his technique: absolute and relative. After reading a memoir by Lagrange
which introduced the symbol "6," Euler adopted its use for his own
studies and named it in an article entitled "Elementia Calculi Varia-
tionum." Other investigations related to this include combining the cal¬
culus techniques of Newton and Leibnetz via differential elements, and
what is apparently the earliest study of integrability. Euler's list of
886 books and articles also inclucbs the introduction of "£" to denote
summation, and an intuitive description of continuous functions—all of
which are necessary elements of the calculus of variations.
As mentioned in the previous paragraph, Lagrange introduced the
concept and symbol "6"; using it enabled him to derive Euler's equation

195
without resorting to the infinitely-sided polygon and the accompanying
complicated geometry. It also facilitated the evaluation of problems
having definite integrals with unspecified conditions at the limits—
these were called "the definite equations," and later "terms at the
limits," but are currently referred to as "natural boundary conditions."
This represented a generalization on the type of problem originally
posed by James Bernoulli. Lagrange introduced yet another generaliza¬
tion by using "indeterminate multipliers" to enforce implicit constraints,
e.g., differential equations. The final major contribution by Lagrange
to be cited here was the extension to problems having double integrals
with fixed conditions at the limits.
Following a study of the major works of Newton, Laplace, and
Lagrange, William Rowan Hamilton applied the analytical method of
Lagrange to his own study of optics. In the "Theory of Systems of Rays"
(1828) and its supplements, Hamilton deduced all optical properties
from a single "Characteristic Function" using the "Law of Varying Action,"
a generalization of the "Law of Least Action" due to P. L. M. de Maupertis.
When applied to dynamical systems in "On General Methods of Dynamics"
(1834) this method is recognized as Hamilton's Principle. The canonical
equations of motion derived from this method are called Hamilton's equa¬
tions, and were used by Jacobi to later derive the Hamilton-Jacobi equa¬
tion.
Euler derived the first necessary condition for an optimal solu¬
tion in the calculus of variations: in the manner developed by Lagrange,
the first variation must vanish. But this may be satisfied by either

196
a minimum, maximum, or an "inflection point." Legendre made the first
attempt to distinguish what the case may be by investigating the second
variation in 1786. Using a transformation of variables, he arrived at
what was believed to be a sufficient condition (for either a maximum or
a minimum) for fifty years. However, in 1837 Jacobi discovered that
Legendre's condition failed in some cases. The "conjugate points" asso¬
ciated with these cases were found from derivatives of the solution to
Euler's equation with respect to the constants of integration. As a
result, Legendre's condition was determined to be only necessary, along
with the newer Jacobi's condition, and not sufficient as originally
thought.
Derivation of a true sufficient condition was done by Weierstrass
in 1879. It occurred as a result of his re-examination of earlier works,
formulating the problems very precisely. Both his condition and his
"corner condition" for a discontinuous solution are stated in terms of
the "Weierstrass function."
Further elaborations, involving geodesics and the existence of
integrals, are attributed to Kneser and Hilbert about 1900. Twenty years
later Bliss applied the "adjoint system" to the solution of a variational
problem. The "maximum principle" was suggested in 1937 by Valentine but
is generally attributed to Pontryagin in 1956.
All of these basic developments are presented in Table A.l,
together with less pertinent details. According to Bliss (1925, p. 181)
the classical memoirs of the Bernoullis, Euler, Lagrange, Legendre, and
Jacobi are contained in Ostwald's "Klassiker der exacten Wissenschaften,"

197
TABLE A.1
MAJOR DEVELOPMENTS
IN THE CALCULUS OF VARIATIONS
1686 Newton
first to pose a problem of a type requir¬
ing the calculus of variations; not
treated as significant at the time.
1696 John Bernoulli
posed the brachistochrone problem. Solu¬
tions were obtained by James Bernoulli,
Leibnetz, and Newton, as well as by John
Bernoulli and de l'Hopital. Accompany¬
ing his solution were two additional
problems proposed by James Bernoulli,
one involving natural boundary conditions,
another with an isoperimetrical constraint.
1701 James Bernoulli
solved the isoperimetrical constraint
problem posed in his solution to the
brachistochrone.
1722 Euler
extended the work of the Bernoulli brothers
to several classes of problems with var¬
ious constraints.
1740 Euler
introduced the technique that setting
coefficient of 6Y in integrand to zero
satisfies necessary condition. The
result is called Euler's Equation.
1744 Euler
divided optimization problems into two
general classes: absolute and relative.
This is the first systematic exposition;
all preceding work represented ad hoc
solutions to specific problems.
1766 Euler
published one tract giving the name
"calculus of variations," and another
containing the first study of conditions
of integrability.
* Lagrange
introduces the symbol "6" to distinguish
between variation and derivative.
* Lagrange
deduced Euler's Equation without using adja¬
cent elements of an infinitely-sided polygon
No specific dates are available; two memoirs on the general subject from
which these might originate were published in 1762 and 1770.

198
TABLE A.l (Continued)
* Lagrange
determined the method to treat unspec¬
ified end conditions; the results are
now called natural boundary conditions.
* Lagrange
... introduced enforcement of implicit con¬
straints by the use of indeterminate
multipliers.
* Lagrange
... extended the calculus of variations to
include double integrals with fixed
boundary conditions.
1786 Legendre
... investigated the second variation of
the cost function to find a criterion
with which maxima and minima could be
distinguished. First results were not
really a sufficient condition, because of
non-unique solutions demonstrated by
Lagrange
1810 Brunaci
... extended Legendre’s work to double
integrals, retaining the same general
flaw.
1834 Hamilton
... developed the canonical equations and
Hamilton's Principle based upon a general¬
ization of the Law of Least Action.
1837 Jacobi
... studied the transformation of Legendre,
discovering how to determine when the
method failed. The resulting conjugate
points are determined from derivatives
of solutions to Euler's equation with
respect to the constants of integration.
1842 Sarrus
... won the French Academy of Science prize
for mathematics by obtaining the natural
boundary conditions for a general multiple
integral having variable limits. His
technique used a new substitution sign to
designate a particular value for general
variable.
1844 Cauchy
... simplified the work of Sarrus to the form
used currently.
No specific dates are available; two memoirs on the general subject
from which these might originate were published in 1762 and 1770.

199
TABLE A.l (Concluded)
1852 Brioschi
1858 Clebsch
1871 Todhunter
1877 Erdmann
1879 Weierstrass
1919 Bliss
1937 Valentine
was the first to apply determinants to
the investigation of second order terms.
generalized Jacobi's Theorem to several
dependent variables, with or without
connecting equations, and for multiple
integrals.
suggested that variations might be of
restricted sign, allowing the possibility
of discontinuous solutions.
derived the conditions that must be
satisfied by a discontinuous solution,
called the Weierstrass-Erdmann conditions
helped to make the theory of the calculus
of variations more precise, in general.
Specifically, he found a new necessary
condition in terms of his function
E(X,Y,P,Y') and very clearly determined
those conditions which are sufficient.
was perhaps the first to use the adjoint
system of variational equations.
derived the earliest equivalent to a
strong minimum principle, using the
Weierstrass condition and slack variables
1956 Pontryagin
together with colleagues, derived the
Pontryagin maximum principle.

200
Nos. 46 and 47. Historical developments are treated by Carll (1881),
Todhunter and Pearson (1893), Bolza (1904), and Bliss (1925, 1946).
Biographical data are found in Truesdell (1968), Bergamini (1963),
Ireland (1962), and Asimov (1960).
Analytical mechanics became a viable method with the introduc¬
tion of calculus by Leibnetz and Newton. For example, Galileo Galilei
observed in 1585 that the free fall of various objects of different size
and weight is given by the formula Y = 16t2. Newton is reported to have
differentiated twice to discover that all objects fall with an acceler¬
ation of thirty-two feet per second squared. Yet there were differences
between the calculus of Newton and Leibnetz; the former designated a
derivative by a dot above the variable, and the latter used the form
"dY/dt" to indicate the derivative of "Y" with respect to "t." Newton
also deduced that maxima/minima occur at points where the rate of change
is zero, whereas Leibnetz insisted such points occurred where the tangent
to the curve has zero slope.
To show that the work of these two was equivalent was one of
the many accomplishments of Euler. In using this discovery he
demonstrated that much of what was then considered to be pure physics
was transformed quite simply into problems of mathematics. Such work led
him to sufficiently important developments that specific equations bear
his name in the separate fields of the calculus of variations, differ¬
ential equations, solid mechanics, and rigid body dynamics. According to
Truesdell (1968), Euler is the major source of rational analysis. Cer¬
tainly he is one of the most prolific, having published 886 books and
articles, and fathering thirteen children.

201
A more detailed list of the basic developments of mechanics
is given in Table A.2; a few biographical data are presented in
Table A.3.
TABLE A.2
MAJOR DEVELOPMENTS IN ANALYTICAL MECHANICS
1694
James
Bernoulli
1704
James
Bernoulli
James Bernoulli
1704 James Bernoulli
Leibnetz
1727 Euler
published the very first paper on the
mathematical theory of elasticity.
concluded a thirteen year study which
included an investigation of the
catenary (results were published in 1744).
Problem was also solved by John Bernoulli
using a theorem concerning the center of
gravity of an arc.
declared elasticity is generally non¬
linear, with linear elasticity a special
case. He also derived the differential
equation for the elástica.
first characterized a material with stress
as a function of strain instead of the
then accepted load versus displacement
for a particular specimen. He avoided
the E of linear elasticity, suggesting
E^ by implication.
performed the first analysis of a loaded
beam, assuming that some fibres were in
tension. That the tension varied lin¬
early gave the result that the moment
was proportional to the second area mo¬
ment of the cross section.
derived the Bernoulli-Euler equation for
bending of beams while a student of John
Bernoulli. The work was not published
until 1862.
•k
No specific dates are available from ordinary sources; items are
located with respect to related entries of the table.

202
TABLE A.2 (Continued)
1729 Musschenbroek
... performed the first comprehensive exper¬
iments on the strength of materials. He
observed that beams and columns bend
before breaking and that breaking strength
varies inversely with the square of the
length and also depends upon depth and
breadth.
1736 Euler
... introduced the study of mechanics, specif¬
ically the concepts point mass, acceler¬
ation, and vectors (which he called
"geometrical quantity").
1739 Daniel Bernoulli
... developed the principle of minimum poten¬
tial energy, although he is best known
for his work in fluid mechanics.
1740 _ Euler
... learned how to solve analytically, linear
ordinary differential equations with con¬
stant coefficients. Prior to that time
only series solutions were available.
* Euler
... after making some calculations on an
actual beam deflecting under its load,
discovered that most real applications
result in small strain. All previous
work had presumed that deformations were
finite.
1743 Euler
... determined that the equation of the
elastic curve could be obtained from the
minimum potential energy.
1743 Euler
... derived the column buckling formula; the
results were compared to Musschenbroek's
experiments in 1765.
1743 John Bernoulli
... obtained the first differential equation
for the motion of a system by studying
the dynamics of a weighted cord using
lumped masses.
No specific dates are available from ordinary sources; items are
located with respect to related entries of the table.

203
TABLE A.2 (Continued)
1743 D'Alembert
... introduced the first partial differential
equations. He also was the first to
separate constraint forces from external
forces, but never stated "D'Alembert's
Principle," which was given later by Euler
and Lagrange.
1744 Euler
published the earliest example of what
came to be called the "Newtonian Method."
Also defined were linear momentum and
kinetic energy derived by integration.
1747 Euler
... was the first to derive "Newton's Equa¬
tions" for a discrete system. By apply¬
ing them to each part of the system, all
the required equations were obtained.
The example used was the three-body
problem of celestial mechanics.
1750 Euler
... extended the preceding study to "mechan¬
ical systems of all kinds," i.e., to con¬
tinuous systems. Examples used included
rigid body motion and transverse vibra¬
tions of a rod.
1754 Riccati
... suggested that the work done to deform
an elastic body was recoverable in part.
1760 Euler
... developed an entire theory of rigid body
motion. He defined center of mass and
distinguished it from center of gravity.
Additionally, he obtained the inertia
tensor and described its elements with
respect to the mass.
* Lagrange
determined the infinite sequence of theo¬
retical buckling loads.
1773 Lagrange
... attempted to find the column shape which
would have the largest buckling load for
given height and volume. His analysis was
incorrect; this results, a circular
cylinder.
No specific dates are available from ordinary sources; items are
located with respect to related entries of the table.

204
TABLE A.2 (Concluded)
1778
Euler
• • •
determined the height at which a uniformly
heavy column buckles.
1788
Lagrange
—
published one of the most important works
in mechanics: "Méchanique Analitique."
In this book he systematized mechanics,
using the calculus of variations to
derive general equations which apply to all
problems of mechanics. Generally speak¬
ing, he obviated the use of complicated
geometry, replacing it with pure algebra.
More specifically, ha introduced the
misnomer "D'Alembert's Principle," gen¬
eralized coordinates, and Lagrange's
equations.
1877 Gauss ... proved the fundamental theorem of algebra
in his doctoral dissertation, requiring
the introduction of complex numbers. He
later recognized their vector character
and applied them to the solution of
complicated mechanics problems.

205
TABLE A.3
MAJOR CONTRIBUTORS IN THE DEVELOPMENT OF
THE CALCULUS OF VARIATIONS AND ANALYTICAL MECHANICS
NEWTON, Sir Isaac
(1642-1727)
English mathematician
LEIBNETZ, Baron von
Gottfried Wilhelm
(1646-1716)
German mathematician
BERNOULLI, John (Jean,
Johan)
(1667-1748)
Swiss mathematician
BERNOULLI, James (Jacques,
Jakob)
(1654-1705)
Swiss mathematician
MUSSCHENBROEK, Pieter van
(1692-1761)
Dutch physicist and
mathematician
BERNOULLI, Daniel
(1700-1782)
Swiss mathematician
EULER, Leonhard
(1707-1783)
Swiss mathematician
ALEMBERT, Jean Le Rond D'
(1717-1783)
French astronomer and
mathematician
LAGRANGE, Comte Joseph
Louis
(1736-1813)
French mathematician
LEGENDRE, Adrien Marie
(1752-1833)
French mathematician
GAUSS, Karl Friedrich
(1777-1855)
German astronomer and
mathematician
HAMILTON, Sir William
Rowan
(1805-1865)
Irish mathematician
JACOBI, Karl Gustav Jacob
(1804-1851)
German mathematician
ERIOSCHI, Francesco
(1824-1897)
Italian mathematician
CLEBSCH, Alfred Rudolf
Friedrich
(1833-1872)
German mathematician
TODHUNTER, Isaac
(1820-1884)
English mathematician .
historian
WEIERSTRASS, Karl Theodor
(1815-1897)
German mathematician
BLISS, Gilbert Ames
(1876-1951)
American mathematician
PONTRYAGIN, Lev S.
(1908- )
Russian mathematician

APPENDIX B
A SIMPLE PROOF OF THE KUHN-TUCKER THEOREM
Consider the minimization of a general unspecified functional
with respect to an n-dimensional vector x, subject to m < n inequal¬
ity constraints,
functional:
Min [F(x)]
x
inequality constraints: g^(x) < 0
j 1, .. . , m
where
x = [x, x„ ... x ]
— 12 n
Inequality constraints are converted to equality constraints by intro¬
ducing "slack" variables s^ 00 defined such that
gj(x) + s2(x)= 0 j = l,...,m
(B.l)
or
^2
sj (x) = [_8j Cx) ] 2 - 0
(B .2)
Since for all j the left-hand side of equation (B.l) sums to zero,
for arbitrary values of A.
J
m
£ A.[g.(x) + s?(x)] = 0
3=1 2 2 2
(B.3)
it follows that:
m
$(x,£,A_) = F00 + £ A. [g . (x) + s2. (x) ] = F 00
3=1 2 2 2 "
206

207
Necessary conditions for the adjoined cost function $ to be
a minimum are derived from the requirement that the first variation
vanish.
s$
. T W , . T 9$ , ,,T 9$
6x — + ós — + SX —
— 9x — 9s — 9A
0
If all the variations are to be independent, then
V F(x) + A. V g.(x) =0 j summed
x - J x J ~
2A. s.(x) = 0 j = 1,.. . ,m
J J ~
gj(x) + Sj(x) = 0 j = l,...,m
where repeated indices in a product imply summation. The independence
of j3 and _A can be argued heuristically. Slack variables are an arti¬
ficial contrivance used to force a desired condition without altering
the original problem. Any change in s^. (x) which satisfies the given
constraint must be accompanied by a change in (x) such that, the net
effect on $ is zero; thus $Cx,_s,_A) is stationary with respect to j3.
Furthermore, since the sum in (B.3) is equal to zero for any arbitrary
values of A^, any arbitrary variation of $ with respect to A^ must also
be zero.
Given the definition of s.(x) in (B.2), the three necessary
J “
conditions may be expressed as
V [F(x) + A.g.(x)] = 0
x — j J ~
AjgjM = 0
Sj(x) < 0
(B. 4)
(B .5)

208
The first condition will be used with the concept of a feasible, usable
direction to derive the final condition remaining before the Kuhn-Tucker
condition may be stated. Any direction in x-space is feasible if an
increment Ax in that direction reduces the cost function F(_x) . Direc¬
tion d is said to also be usable if an increment Ax: in that direction
causes no constraint violation.
Consider first the concept of a feasible direction; for any
direction d the directional derivative is defined to be
dF
f- = d • V F(x)
ds x —
where
lid || = 1
In addition to this restriction on the norm of direction á, let the
increment dx be prescribed by
dx = (ds)d (B.6)
and the vector scalar product be written as the equivalent inner product
such that
d • V F E dT V F H (V F, d)
X X X
In terms of an inner product notation
dF a
dl â– 
For an increment in direction d of length ds, the change in the cost
dF = (V F, d) ds
x
function is

209
By the linearity property of the inner product
dF = - (-V F, d) ds
Hence the cost F(x) is decreased by an increment in any direction d
which satisfies
(-V F, d) > 0 (B.7)
Conversely, a direction d is feasible if (-V^F, d) > 0, i.e., the angle
between d and the direction of steepest descent is acute.
For any direction to be usable, an increment Ax in that direc¬
tion must cause no constraint violation. On the constraint boundaries
g. (x) = 0 for j £ I as defined in Chapter III. With any direction again
1 A
denoted by d, the change in g^. (x) with respect to an increment dx is
defined through the directional derivative:
where
such that
(Vv g, (x) 3)
Ü J
I!
= i
d) ds
Since ds denotes a vector norm by way of the definition of djc given in
equation (B.6), it is a positive quantity. Hence the sense of change
dg. is prescribed by the inner product (V g., d) for which there are
J 2£ J
three possibilities.

210
(i) (V g., d) <0 -* dg. <0
2£ J J
It was assumed that active constraints are being discussed, for
which g_. (x) = 0. The constraints require all g.. (>c) to be non¬
positive. An increment dx in the direction d used in the inner
product satisfies the constraint with the inequality sign.
Because the direction does not lead to a constraint violation
it is by definition usable. Actually, for such a direction d,
even though point x lies on the boundary g^(x) = 0, the point
(x + dx) does not.
(ii) (V g , d) = 0 -* dg = 0
J J
Under the assumption that g^. (x) is active where g^ (x) = 0, when
the increment die in direction d causes no change in g^ , d is
tangent to the constraint surface. With x + dx used in the
inner product, the point (x + dx) may still lie on the constraint
boundary. With such an increment the constraint may perhaps not
be violated, and if not, the direction is therefore usable.
Note that for linear constraints, the word "may" is deleted from
the previous sentence. It is included because nonlinear con¬
straints require an argument calling for infinitesimal incre¬
ments; a finite increment leads to a point that is not on the
constraint boundary. A finite increment from a point on a non¬
linear constraint boundary results in that constraint either
becoming inactive on being violated, depending upon whether the
constraint surface is concave or convex, respectively.

211
(iii) (V g., d) > O -> dg. > 0
A J J
That the constraint (x) is active requires C2S) = 0.
If an increment d_x in the direction d used in the inner
product results in a positive dg^, the constraint has been
violated. Therefore, d is not a usable direction.
The result of these possibilities is that for any point x on
a constraint boundary, for a direction d to be usable it must satisfy
(V g., d) < 0 j e I (B.8)
A J A
If the set of all directions d which are feasible contains no usable
directions at the point x in question, then x = x^pp and x is the
constrained minimum point.
Conditions for a usable, feasible direction given by (B.8) and
(B.7) are used with the gradient condition (B.4) to prove that the
Lagrangian multipliers of (B.5) must be non-negative at the point where
x is a constrained optimum. Condition (B.4) states that at the con¬
strained minimum point, the negative gradient of the cost function lies
in a subspace defined by the gradients of the active constraints, i.e.,
- yw I b b b
In the following, implied summation is replaced by the symbol E, and
the dependence of F and g. upon is implied. Resolving -V F into com-
J A
ponents along the V g. directions via the inner product gives
X .1

212
Thus
m (-VXF, Vxg )
-V F = £ = =-J- V g.
- 3 = 1 ||V 8j || * J
A. =
J
<-y. y.j)
II Vj II
3 e 1,
This condition relating the Lagrangian multipliers to an inner
product is used with the conditions necessary for the existence of a
constrained minimum to show that A. > 0. At a constrained optimal point,
for every feasible direction d, the cost decreases and
(-V F, d) > 0 d is feasible
x
the constraints are violated (otherwise not optimal),
(V g,, d> > 0 <- d is not usable
2£ 3
Assume that some j, A^ is negative and a feasible but not usable direc¬
tion á exists. If d is taken to be coincident with -V F, this direc-
x
tion is feasible and usable, contrary to the original assumption.
That A^ > 0 is proved by contradiction. Mathematically stated,
Given:
(-V F, d) > 0
for all d
(vxgj, d) > 0
for all j e I
Assume: A . < 0
3
for some j e I
A
(i) A. < 0 (-V F, V g.) < 0
3 x x J

213
(ii)
Consider d = -V F
X
then
(_VxF, d) = (-VxF, -VxF) = ||-VxF ||2
>0 -> d is feasible
(iii)
But
(V g., d) = (V g., -V F)
X J X J X
<0 -> d is usable
by virtue of (i). This says the direction of steepest descent
is a feasible direction which is also usable, contrary to the
given condition.
(iv)
Therefore,
X. > 0, j e T Q.E.D.
J A
This completes the proof of the Kuhn-Tucker conditions for the
existence of constrained minimum of F(x) subject to g. (x) < 0
j = 1» •
..,m. They are
(i)
g (x ) < 0
SjV±0PT' “
(ii)
X. > 0
J
(iii)
Ajgj(2k)PT) = °
in
+ 3, xj vxVW â– 0
- J=1 -

APPENDIX C
COMPUTER SUBROUTINE LISTINGS
In order to use the IBM SHARE quasilinearization program, three
user-supplied subroutines must be combined with three SHARE subroutines
The latter three are written in terms of a general problem described in
section 3.5 and do not change from one problem to another. Their names
and functions are:
QUASI
- a quasilinearization algorithm
RECIP
- a matrix inversion subroutine
OUTPUT
- a subroutine that prints out different, specified
combinations of problem variables
In addition to
these IBM subroutines, the user must provide three sub¬
routines which
convert a specific problem into the general formulation
required by the above subroutines. Their names and functions are:
MAIN
- reads data, establishes initial guess, calls QUASI,
and terminates execution with additional output if
desired
DIFEQ
- is called by QUASI to evaluate the differential
equation and linearization gradients at every point
for each iteration
CORRC
- corrects specified boundary conditions after each
iteration
These six subroutines are combined into a single computer program for
solving the given TPBVP.
214

215
Listings of nine subroutines are contained in this appendix.
The first six represent a program for solving the minimum deflection
beam problem subject to stress constraints. The order of presenta¬
tion is:
MAIN
DIFEQ
CORRC
QUASI
RECIP
OUTPUT
Following these are the three use-supplied subroutines required to
solve the maximum buckling load problem subject to only geometric and
material bounds. They are presented in the following order:
MAIN
DIFEQ
COPTIC

IOC
IMPLICIT REAL*8(A-H,0-Z )
CGMMCN/QUCOMM/X.ERRORsOFFOGjHjElMAX.tZMAXrBLQ
CGNMCN/DATA/8D,AC,SIG,TAU
Cl MENS ION GI6),M6),T(1C1I*P[1).AI6,6],B(6,6),BLI6,6) »
1ER(6,6),C(6,6),CG{6,6),R(101J6),Y{101,6),G16,6,1CC)
CIMENSION PH 118 )
I FLAG
= 0
BC
= .25
AC
= .20
S IG
= 8.75
TAU
= 2.31
N
= 6
M
= 100
NC
= 6
M
= 100
NP
= 101
NPAR
= 1
MT
= 15
IPRM
= 1
ERRCR
= 0.5C-6
CFFCG
= 5.0C-15
ELC
= 10.
CO 100 1=1,6
CC 100 J-1,6
e l < i,
J) = 0.0
BR t 1 »
J ) =0.0
CONTINUE
ELI 1,
1 ) = 1.0
GL12,
2) = 1.0
EL 1 5,
5) = 1.0
DU 6 ,
6) = 1.0
ERl 3,
3) = 1.0
e R 1 A ,
A) = 1.0
CO 200 I=a,NP
216

Til) = ÍI-1.J/M
Z = Til)
Y tI» 1 ) = .0 20 *(Z * *2 )
Y(1,2) = .O 4 O * Z
Y( I , 3) = .080*1 (l.-Z)**3)
Y( I ,4) =-.035*1 (l.-Z ) * * 3 i
Yí 1,5) = 13•O*(Z * *4 )
Y ( I j 6 ) = -2.00*(Z**4 )
200 CONTINUE
CALL CUASI ( N, M, NC, N I, N P ,.NP AR, N I T, IPRN T , G , W, T, P , A, B , B L, B R, C , DG » R,
1Y,D,+2000 )
IF(E2MAX.LE.ERROR) CO TO 9GC
800 CONTINUE
WRITE (6,810) E2MAX,ERROR
810 FOR Y AT ( * 1 • /*0•,10X,•DIC NOT CONVERGE, E2MAX =. • , DI 2.5 , ' . GT. ERROR
li* , C12.5,// )
IFLAC = -1
GO TC 920
900 WRITE (6,910) E2MAX,ERROR
910 FORMAT í •11/* O', 10X,5 CONVERGED, E2MAX = ’,012.5,*.LE. ERROR =’,D12
1.5,//)
920 CONTINUE
WRITE(6,701)
701 FORMAT ( •O', 10X, 9 ... X AND YJ(X) AT EXIT QUASI ...',//)
WRITE(6,7010 )
7010 FORM AT ( 9X1HX, 8X 2HY 1,8X 2HY 2,8X2HY3,8X2H Y4,8X2H Y5,8X2H Y6,8X2HY7 ,
18X2FY8,8X2HY9,7X3HY10//)
WRITE(6,702) (T( I 5, (Yt I»J),J=1,N)»I=1»NP)
70 2 FORMAT(5X,F7.4,6F10.4)
I F(IFLAG.LT.0) GO TO 30CG
W R I T E(6, 1000 )
1000 FORMAT ('1'/'0’, 2X , 5 CONTROL DEPENDENT PORTION OF HAMILTONICN’»//)
WRITE (6,1100)
1100 FORMAT(T7,’X•,T 1 5, *U1 *,T23, ’U2=.20',T37, 902 = .30’,T51,'U2=.40’ , T65,
217

1 »U2=. 50'»T79,»U2=.60', T9 3» »U2 = .70*,T107» 'U2=.80•,T121,»U2 = .90‘)
CC 1600 1=2,11
Z = ( 1-1. 1/10.
II = 10 *(I-1) + 1
F2X =-(l.-Z)«Y( I I 3 ) / Y { 11,6)
CG 1600 J = 1, 16
U1 = .25 + .05 *(J-l.)
CO 1AGO K = 1,8
U2 = .20 + .1*(K-1. )
1400 PHI1K) = U 1*U 2 + F2X/( (U1 * * 3)* U2)
WRITE (6,1500) Z,U1,(PHI(L),L=1,8)
1500 FORMAT l »0' ,2X,F6.3,3X, F6.3,8D14.5)
1600 CONTINUE
GO TC 3000
2000 CONTINUE
WRITE(6,2100)
2100 FORMAT('0»,1CX,'ERROR EXIT C STRESS CONSTRAINT EXCLUDES ALL GEOMET
1RICALLY ADMISSIBLE CONTROL»)
3000 CONTINUE
STOP
ENC

SUBROUTINE DIFEG (NfNCfNPAR,IM,ITER,G»W,P,DGt*)
IMPLICIT REALMS(A-H»Q-Z )
COMNON/QUCOMM/X,ERRORfOFFDGfHtE1MAX,E2MAX,8L0
COMMON/DATA/ 80, AC,.S IG, TAU
DIMENSION G(NC) ,W(NC ) ,P(NPAR ) â–ºDG(NC,NC)
DIMENSION S I GBl 10 ) » TAUB(10)
FUNCiUl»U2»F2X) = U1*U2 * F2X / { (Ul** 3)*U2 )
F2X
= -( l.-X )*W(3)/W(6)
100
FLL
= FUNCtBC,AC, F2X )
Ul
= BD
U2
= AC
PHI
= PLL
MPI
= 0
110
US2
= (F2X/(BD«*4).}**.5
IFtUS2.LT.AC.OR.US2.CT.1. } GO TO 120
PH I S= FUNCtBC,US2,F2X )
I F{PHIS.GT.PHI ) GO TO 12C
U2 = US 2
PHI = PHIS
MPI = 2
120 FLU = FUNCtBC,l.»F2X)
IFtPLU.GT .PHI ) GO TO 130
0 2 * 1.
PHI = PLU
MPI = 0
130 CONTINUE
US1 = (3.*F2X)**.25
IFtUSl.LT .BD.CR.US1.GT . 1. ) CO TO 1400
P HI S = FUNCÍUS1»1.»F2X)
I F(PHIS.GT.PHI) GO TO 1400
Ul = US1
U 2 =1.
PHI = PHIS
MPI = 1
219

no no oooooo non
1400
1401
140 2
1403
CONTINUE
N 8 =0
INB = 0
U1BNSL = l S I G * W ( 3 ) /A C 1 * * • 5
U1BSSL = -TAU*W(4)/AC
U10NSU = (SIG*W{3 ) ) **. 5
U1BSSU = -T AU *W(4 )
IFÍU1BNSU.LE.1. .ANC.U1BSSU.LE.1.) GO TO 1402
STRESS CONSTRAINTS ACMIT M0 GEOMETRICALLY ADMISSIBLE
NB = -1
WRITE(6,1401) X,U1BNSU.U1BSSU
FORMAT('010X, 'X =• , F1C . 4, 'U1BNSU ='» D 1 5.6 »'U1BSSU *
GO TO 500
CONTINUE
STRESS CONSTRAINTS ACMIT GEOMETRICALLY ADMISSIBLE
IF(UIBNSL.GE.BD.OR.UIBSSL.GE.BD) GO TO 1403
STRESS CONSTRAINTS ACMIT ALL GEOMETRICALLY ADMISSIBLE
NB =0
GO TO 140
CONTINUE
U28NSU = S I G*W(3 )
U2BSSU = -T AU*W(4)
IF1U2BSSU.GT .U2DNSU) GO TO 1404
IF(U2BNSUoLT.AC ) GO TO 1410
CONTROL
»D15.6)
CONTROLS
CONTROL
220

on oooo or> oooo
C LI UPPER BOUNDARY IS 0 U2LB .LE. U2 .LE. 1.0
C BOUNDARY IS DEFINED BY MAXIMUM NORMAL STRESS
U2LE = U2BNSU
INB = I
NB =2
GO TC 140
C
c
1404 CONTINUE
I F(U2BSSU.LT.AC ) GO TO 1410
U1 UPPER BOUNDARY IS 0 U2LB .LE. U2 .LE. 1.0
BOUNDARY IS DEFINED BY MAXIMUM SHEAR STRESS
U2LB = U2BSSU
INB = 2
NB = 2
GG TC 140
1410 CONTINUE
IFÍU1BSSL.GT oUlENSL ) GO TO 1415
U2 LOWER BOUNDARY IS C UlLB .LE. U1 .LE. 1.0
BOUNDARY IS DEFINED BY MAXIMUM NORMAL STRESS
L1LB = U1BNSL
INB = 1
NB = 1
GC TC 140
1415 CONTINUE
C
221

o o o
U2 LCWER BOUNDARY IS O UlLB .LE. U1 .LE. 1.0
BOUNDARY IS DEFINED DY MAXIMUM SHEAR STRESS
L1LB = U1BSSL
INB = 2
ND = 1
C
C
C
C
c
INB
= 1
... UKSIG, AC ) .GT.UK TAUt AC)
c
INB
= 2
... U1(SIG,AC).LT.UKTAU,AC)
c
NB
= -l
... NC ADMISSIBLE CONTROL
c
NB
= 0
... NO ACTIVE STRESS CONSTRAINT
c
NB
= 1
... UlLB .LE. Ui .LE. 1 AND
U2 .EQ. AC
c
NB
= 2
... U2LB .LE. U 2 .LE. 1 AND
U1 .EQ. 1.0
C
C
c
c
140 PUU = FUNCIl.tl.jF2X)
I F(PLU . GT . PH I ) GO TO 15C
U1 =1.
U2 =1.
PHI = PUU
MPI = O
150 CONTINUE
LS2 = (F 2 X)**.5
IFÍNE.EQ.2) CO TO 155
IF(US2.LT.AC.OR.US2.GT. 1. ) GO TO 16C
153 PHIS= FUNC( 1»,US2». F2X)
I F(PHIS.GT.Phl ) GO TO 160
U1 = 1.
L2 = US 2
222

CÜ TO 180
FHI = PHIS
NPI = 2
IFÍNB.EQ.2)
GC TC 16C
155 IFIU2LB.LT.US2.AND.US2.LT .1» ) GO TO 153
FU L B = FUNC (1.j U 2 L B»F2X)
IF(PULB.GT.PHI) GO TO 180
U1 =1.
U2 = U2LB
FHI - PULB
PPI = 3
GO TC 180
160 PUL = FUNCÍ1.,AC»F2X )
I F(FUL.GT.PHI ) GO TO 170
U1 = 1.
U2 = AC
FHI = PUL
KPI = 0
170 CONTINUE
LSI = (3.*F2X/(iC**2).)**.25
IF(NE.EQ.l) GO TO 175
IFtUSl.LT .BD.CR.US1.GT. 1. ) GO TO 180
173 PH IS= FUNC(US1,AC,F2X)
I F{PHIS.GT.PHI ) GO TO 180
LI = US1
L2 = AC
PHI = PHIS
FPI = 1
GO TC 180
175 IF(U1LB.LT.US1.AND.US1.LT.1 . ) GO TO 173
PUL 0= FUNCIU1LB,AC»F2X)
IF(PLLB.GT.Phi ) GO TO 180
LI = U1L B
L2 = AC
223

PHI = PULO
y pi -• 4
180 CONTINUE
200 G(1) = W(2)
G(2) = W(3)/Í(U1**3)*U2}
G ( 3 ) = W ( 4 )
G(4 ! = U 1*U 2
G(5 ) = ( l.-X )/( (U1**3)*U2)
G Í 6 )
=-W¡5)
CO 220 1=1,6
CO 220 J = 1,6
2 20
CG ( I
jJ ) = 0.0
LIU
= 0.0
CG 11
,2) = 1.0
CG C 3
,4 ) = 1.0
CG ( 6
,5) =-1.0
C
c
c
c
NPI
“0 « « o
U1
ANC
U2 ARE CONSTANTS
c
y p i
“ 1 9 0 9
U1
IS
NOT CONSTANT
c
ppi
~ 2 * • •
U 2
IS
NOT CONSTANT
c
y pi
= 3 tt 9 •
U 2
IS
ON A STRESS CONSTRAINT
BOUNDARY
c
ypi
= 4
U1
IS
ON A STRESS CONSTRAINT
BOUNDARY
c
c
c
NPIG
= MPI + 1
GO TC (230,240*250»260,2705» MPIG
230 CG{2 » 3) = l./t(U1**3)*U2)
GO TC 290
240 CONTINUE
LIU = l./( (Ul**4 )*U2)
CU1W3 = .25*U1/W(3)
224

250
2 60
265
270
CU1W6
C G í -4 y 3 ) =
CG(4,6 ) =
CG(5 j 3 ) =
CG(5 » 6 ) =
CG(2,3 ) =
C G ( 2 » 6 ) =
CO TC 290
CONTINUE
UIU
CU2W3
CU2W6
C G (4 » 3 ) =
C G ( 4 » 6 ) =
CG(5,3 ) =
CG(5,6) =
CG(2 » 3) =
CG(2,6 ) =
GO TC 290
CONTINUE
IF{INB.EC
C G ( 2 » 3 ) =;
CG(4 » 3 ) =
CG Í 5 » 3 ) =.
CC TC 290
CONTINUE
C G ( 2,3 J =
CG{2 t 4) =
CG(4 j 4) =
C G ( 5 j 4 ) =
GO TC 290
CONTINUE
IF ( INB.EC
CG { 2,3) =
.25*U1/W(6)
U2*DU1W3
U 2 * D L 1W 6
-3. »( ic-X )*UIU*DU1W3
OG Í5,3 )*CU1W6/DU1W3
-3.*W(3 )*UIU*DU1W3 + l./l(U1**3)*U2)
-3.«W(3 )*UIU*DU1W6
1v/{(Ul»*3)*(U2**2))
.5*U2/W(3 )
-♦5»U2/W(6 )
U1*CU2W3
U1#CU2W6
-(1- —X)*UIU»DU 2W3
DG(5 T 3)»CU2W6/DU2W3
-W(3)*UIU*DU2W3 + UIU*U2
“W(3 ) *UIU*DU2H6
2) GO TO 265
-S IG«G(2)/U2 + 1./U2
S IG
-S IG*G(5)/U2
1 ,/U2
TAU*C-( 2 ) /U2
-TAU
TAU*G(5)/U2
2) GO TO 275
-1.5*SIG*G(2)/(Ul*Ul*AC) + I./({Ul**3)*AC)
225

CG(4,3 > = O • 5*S IG/U1
CG(5» 3) = -1.5*SIG*GÍ51/(U1*U1*AC)
GC TG 290
275 CCMINUE
C G(2,3) = 1./{(U1*«3)*AC)
C G(2,4 ) = 3.*TAU*G(2 ) /í U1* AC )
CG{4 14) = -TAU
CC(5 »4) = 3 .*TAU*G(5)/ÍU1*AC)
290 CONTINUE
XCS = SIG * W(3)
XCT = -T AU*W{4 )
U1BI= XCS/XCT
U2B1= X CT» * 2/X C S
CC 295 IX = 1, 10
XU2 = 0.1*IX
SIGE(IX) = CSQRT(XC5/XU2 )
TAUBIIX) = XCT/XU2
295 CCMINUE
Z = X + .5 *H
WRITEÍ6»3C0 ) Z»U1,U2,F2X
30C FOR N AT( 2X, • X= ' , D13.6,6X , ' U 1=; ’ , D15.6,6X , ’ U2 = ' , D1 5.6,5 X, 'F2X= ’ t D15 .
16 )
VvRI TEÍ6» 310 ) U1BUU2BI, INB,NB,MPI
310 FORMAT(26X,D15.6,9X,D15.6„58X,3 I 3)
WRITEC6, 320) {S IGB( 1X5, IX = 1,10)
URITEÍ6,321) (TAUBÍ IX)» I X = 1»10)
320 FORMAT(2X,10C12.3 )
321 FORMAT(2X,10C12.3,// )
RETURN
500 CGNTINUE
RETURN 1
ENC
226

SUBROUTINE CCRRC (N,M,NC,NP,NPAR,I TER,T,P,8L,BR,Y)
IMPLICIT REAL»8(A-H,0-Z )
COI* NON/QUCOMM/X» ERROR »OFFDG, H, E1MAX,E2M AX, BLO
CONNGN/DATA/.BD, AC,.$IG,TAU
Cl MENS ION T(NP) ,P ÍNPAR),BL(NC,NC),BR(NC,NC) ,Y(NP,NC)
V(1, 1 ) = 0.0
Y(1 * 2 ) = 0.0
Y(101,3) = 0.0
Yl101,4) = 0.0
Y(1 , 5 ) = 0.0
Y(1,6) = 0.0
RETURN
ENC

1
SU'd RCUT I N E QUAS 1 (N , M »NC , NI»IMP ,.N PAR» NI T. I PRNT , G,W,T,P,A,Q,BL,BR,C,
1 C G » R » Y , D * * )
IMPLICIT REAl.«8(A-H,Q-Z ) 3
CCMMCN/QUCOMM/X,ERROR,OFFDG,H,EIMAX,E2MAX,BLO 4
COMMCN/DATA/BC,AC»SIG,TAU
DIMENSION G(NC) ,W(NC)»T(NP),P(NPAR),A(NC,NC),B(NC,NC), 6
IBL(NC»NC).BR(NCjNC),C(NC »NC ) ,DG{NC,NC),R(NP,NC) , 7
2Y(NPfNC)*D(NC»NC»NI ) 8
M1 = M + I 9
CALL OVER FL{JJJ ) 10
IT£R=0 II
1 IM = C 12
CALL CQRRC (N,M,NC,NP,NPAR,I TER,T,P,BL,BR,Y) 13
I F( I PRNT) 99,99,9B 14
98 CALL OUTPUT (N, M, NC ,NP, IMI PRNT, I TER,G, W , T,DG ,.Y ) 15
99 E2MAX = O.ODO 16
MPAS = 3 17
VvRITE (6,1200)
1200 FORMAT('1'/'0',10X,'CONTROL VECTOR COMPONENTS*,//)
DC 30 IM =1,M 18
IM1=IM +1 19
H=T( IM1)-T(IM) 20
X=T(IM)+0.5*F 21
CO 2 1=1,N 22
2 W(I ) =0 • 5 *(Y( IM, I )+Y( IM1, I ) ) 23
CALL CIFEC (N,NC,NPAR,IM,ITER,G,W,P,DG,+9997)
KKK =1 25
CALL OVERFKJJJ ) 26
IF!JJJ-2) 1000,1001,1001 27
1000 WRITE (6,110) KKK,.ITER,IM 28
110 FORMAT<17HOOVERFLOW OF TYPE 13, 12H ITERATION 14, 1CH INTERVAL 29
114///) 30
1001 DC 29 1=1,N 31
W(I)=Y(I Ml,I )-Y( IM,I) 32
ro
S3
co

G(I)=h*G(I) 33
G = DABS(W(I)-G{I)) 34
IF(E2MAX-G) 6,29,29 35
6 E2f'AX=G 36
12 I = IM 37
J2J = I 33
29 CONTINUE 39
IFUPRNT-1) 28,28,27 40
27 CALL CUT FUT (N,P,NC»NP»IM,IPRNT,I TER,G»W,T,DG,Y) 41
28 H=0• 5 »H 42
CC 5 1=1,N 43
DO 5 J = 1»N 44
5 A(I,J)=H*CG(I» J ) 45
CO AA 1=1,N 46
A( I , I ) = 1 .+M I, I ) 47
44 G(I )=K( I )-G( I ) 48
CALL RECIP (N,M,NC,IM,IPRNT,ITER,MPAS,W,A,B,C,DG) 49
K K K = 2 50
CALL CVERFL(JJJ ) 51
IFUJJ-2) 1002,1003,1003 52
1002 WRITE (6,110) Ki 1003 CO 30 1=1,N 54
R(IM,I)=0• 55
CO 8 J=1,N 56
R(IN,I) = R(IM, I ) + A( I,J ) *G(J ) 57
8 C(I,J,IM) = -2.*A(I,J) 58
30 C(I,I,IM) = 1.+ C(I,I,IM) 59
IF(IPRNT) 34,34,35 60
35 X = G.5*(T{ I 2 I) + T( 121 + 15 ) 61
WRITE (6,105) E2MAX,J2J,X 62
105 FOR R AT (16HOMAX * ABS„ ERR .= E15.6,6H FOR V 12,6H AT X=,F7.4) 63
WRITE (6,1201) Y(N P,1 )
1201 FORPATt'O',10X,'TIP DEFLECTION -*,020.12)
34 IF(ITER) 31,32,31 64
229

32 E1MAX = E2MAX 65
GC TC 37 66
31 IF ( E2MAX - BLC*E1MAX) 26,36,96 67
96 WRITE (6,106) ITER 68
106 F0RMAT125H0INSTABILITY IN ITERATION 13//) 69
WRITE(6,1311)
WRITE(6,1312) (T( I ) , l Y( I,J),J=1,N),I=1»NP)
1311 FORMAT (4X1HX,8X2HY1,8X2HY2,8X2HY3,8X2HY4,8X2HY5,8X2HY6,8X2HY7,
18X2HY8,8X2HY9,7X3HY10//)
1312
FORMAT (1X.F7.4
, 6F10.4 )
97
RETURN
70
36
IF ( E2MAX - ERROR) 57,97,37
71
37
IF ( ITER - NIT)
137,97,97
72
137
IF (E2MAX - .01
» E 1M A X ) 38,38,33
73
38
IF {E2MAX - PRERR) 33,32,97
74
33
ITER=ITER+1
75
FRERR = E2MAX
76
MPAS = 5
77
CO 111 1=1,N
78
G ( I ) =<0 •
79
CC 1C J = 1,N
80
10
A(I,J)=C.
81
111
A ( I , I ) = 1 .
82
CO 12 IM= 1» M
83
IF (MOD( IN,2 ) )
212,312,212
84
212
X = -1.
85
GO TC 152
86
312
X= 1.
87
152
CO 52 1=1,N
88
CO 52 J= 1, N
89
52
G ( I ) = G ( I ) -X * A ( I
,J)*R(IM,J)
90
CO 56 1=1,N
91
CO 56 J=1,N
92
8 ( I ,J)=0 .
93
230

CG 56 K=1j N 94
56 B(I,J)= B tI * J ) + A( I,K)*C(K,J, I M ) 95
CC 12 1=1, N 96
CC 12 J=UN 97
12 A{ I , J) = B( I,J ) 98
IF (F'OC( K,2 ) ¡4 12,512,412 99
412 X = -1. 100
GO TO 157 101
512 X = 1. 102
157 CC 57 1=1,N 103
Rtf'l , I ) = 0. 104
CO 57 J=1,N 105
R(M1,I)=R(M1»I)-BL( IJ J *G ( J ) 106
57 0(I»J)=X»A(I,J) 107
CO 67 1=1,N 108
CO 67 J=1,N 109
A(I,J)=0 • 110
CO 68 K=1,N 111
68 MI, J)=A( I, J )+BL( I,K)*B(K,J ) 112
67 A(I , J ) = A t I,J ) +BR( I,J ) 113
KKK=3 114
CALL CVERFL(JJJ) 115
IFUJJ-2) 1004, 1005,1005 116
1004 WRITE (6,110) KKK,ITER»M1 117
1005 CALL RECIP (N,M,NC,M1,IPRNT,ITER,MPAS»W,A,B,C,DG) 118
K K K = 4 119
CALL OVER FL(JJJ ) 120
IFUJJ-2) 1006, 1007,1007 121
1006 WRITE (6,110) KKK, ITER,K1 122
1007 CO 14 1 = 1, N 123
G( I ) =0. 124
CO 14 J = 1 ,N 125
14 G( I )=G( I )+A( I, J )*R( M1,.J ) 126
CO 15 1=1,N 127
231

Y(Ml, I) = Y(Ml, I) + Gil)
15 R(M1, I ) = G( I )
CO 40 IM=1,M
ID=M1-IM
ID1= ID + 1
CO 16 1=1,N
G(I)=0 .
CC 16 J = 1,N
16 G(I) = G(I) - ü( I,J, ID) *R( IC1,J )
CO 40 1 = 1,N
R(IC 11> = R(ID, I )+G( I )
40 Y( IC, I ) = Y(ID, I ) + R( IC, I )
KKK = 5
CALL CVERFL(JJJ )
IFtJJJ-2) 1008,1,1
1003 WRITE (6,110) KKK,ITER,M1
CO TC 1
9997 RETURN 1
ENC
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
232

SUBROUTINE RECIF ( N »M ,NC , IM , IPRNT, I TERt MPAS , W,.A , B,C ,DG ) 146
IMPLICIT REAL#8(A-H » O-Z) 147
COMNCN/QUCOMM/X» ERROR,OFFOG,H,E 1MAX,E2MAX ,810 148
COMMCN/DATA/BD,AC,SIG,TAU
DIMENSION W{NC),A(NC,NC),B(NC,NC),C(NC,NC),DG(NC,NC) 150
CG 1 1=1,N 151
CO 1 J=1,N 152
1 C(I ,J) = A( I,J ) 153
CC 7C 1=1,N 154
FACT = A{ I , I ) 155
A(I,I)=1. 156
IF(FACT) 72,34,72 157
34 WRITE (6,100) I, IM 158
100 FORMAT(22H0ZER0 ON DIAGONAL ROW 13, 7HMATRIX 13///) 159
STCP 160
72 DO 71 J=1»N 161
71 A(I,J)=A(I,JJ/FACT 162
CO 70 K=1,N 163
IF(K-I) 73,70,73 164
73 I F(A(K , I) ) 74,70,74 165
74 F ACT = A í K » I ) 166
A(K , I ) =0 . 167
CO 75 J=1,N 168
75 A{K,J)=A(K,J)-FACT*A( I,J ) 169
70 CONTINUE 170
N P A S = 1 171
14 SUM =0. 172
CO 2 1=1,N • 173
CO 2 J = 1,N 174
CG(I,J)=0.0 175
CO 2 K=1,N 176
2 CG(I , J) = CG(I,J)+A(I,K)*C(K,J5 177
M = N-1 178
CC 4 1=1,N1 179
233

11=1+1 180
CO 4 J = 11»N 181
4 SUM = SUM+ DABS(CG( I» J ) ) + DABS(DG(J,I>) 182
IF{ IPRNT-3) 31,32,32 183
32 WRITE (6,101) SUM 184
101 FORMAT(30H SUM OF ABS OF OFFDIAG TERMS E15.8) 185
31 IF (SUM - OF FOG ) 13,13,12 186
12 IF (NPAS -1) 9,5,15 187
15 PRERR = 2.*PRERR 183
IF (SUM - PRERR) 8,200,200 189
200 CO 201 1=1,N 190
CO 201 J= 1» N 191
201 A( I , J ) = B < I,J ) 192
IF ( IPRNT - 3 ) 13,203,203 193
203 WRITE (6,102) 194
102 FORM AT ( 30FI USED NEXT TO LAST RECIPROCAL.) 195
GO TO 13 196
8 IF (NPAS-MPAS ) 9,13,13 197
9 PRERR = SUM 198
NPAS = NPAS +1 199
CO 202 I = 1,N 200
CO 2C2 J = 1,N 201
202 B(I , J) = A( I, J ) 202
CO 7 1=1,N 203
7 CG C I , I) = CG(I,I)-1. 204
CO 1C 1=1,N 205
CG 11 J=1, N 206
W(J)=C. 207
CO 11 K=1,N 208
11 W(J)=W(J)+A(K,I )« DG(J »K ) 209
CO 10 J=1»N 210
10 A(J,I) = A(J, I )-W(J ) 211
GO TC 14 212
13 RETURN 213
234

SUBROUTINE OUTPUT ( N, M, NO , NP , I Ml, IPRNT, I TER, G , W , T , DG , Y) 215
IMPLICIT R£Al*8(A-H, O-Z ) 216
COMMCN/QUCOMM/X,ERROR,OFFDG,H,E1MAX,E2MAX.BLO 217
COMMCN7DATA/BC,AC,SIG,TAU
DIMENSION G(NC),W(NC)»T(NP)»DG(NC,NC)»Y(NP,NC) 219
IF (IM) 1,1,16 220
1 WRITE (6,100) ITER 221
WRITE (6,102) 222
Ml = N+l 223
CO 5 1=1,Ml 224
6 WRITE (6,104) T( I), (Y(I»J),J = 1»N) 225
IF (IPRNT-2) 21,11,11 226
11 WRITE (6,106) ITER 227
WRITE (6,108) 228
WRITE (6,110) 229
GO TO 21 230
16 WRITE (6,112) X,(C< J ) , J = 1 ,-N ) 231
WRITE (6,112) X, (W(J ) ,J = 1,N ) 232
IF (IPRNT-3) 21,21,18 233
13 CO 19 1=1,N 234
19 WRITE (6,114) (CG(I,J ) ,J = l,N) 235
21 RETURN 236
100 FORMAT (25H1S0LUT ION AFTER ITERATION 13//) 237
102 FORMAT (4X1HX,8X2HY1,8X2HY2,8X2HY3,8X2HY4,8X2HY5,8X2HY6,8X2HY7, 238
18X2HY8,8X2HY9»7X3HY10// ) 239
104 FORMAT ( 1XF7 .4 , 10F1C .4 ) 240
106 FGRMAT !19H1ERRORS IN ITERATE 14//) 241
108 FORMAT (4X1HX,6X4HH*G1,6X4HF*C2,6X4HH*G3,6X4HH*G4,6X4HH*G5, 242
16X4hF*G6,6X4HH*G7,6X4HH*G0,6X4HH*G9,5X5HH*G10) 243
110 FORMAT (4X1HX,7X3HDY1,7X3HDY2,7X3H0Y3,7X3HDY4,7X3HDY5,7X3HDY6, 244
17X3FCY7,7X3HCY8,7X3FCY9,6X4FDY10// / ) 245
112 FORMAT (1XF7.4,10F10.4) 246
114 FORMAT l1X10 F10.3 ) 247
END 248
235

IMPLICIT REAL*8(A-H,Q-Z ) I
REAL*8 MU tKM»ML »L AM 2
COMMCN/QUCOMM/X, ERROR, QF FOG, H»E IMAX , E2MAX ,610 3
COMMCN/DATA/ AU»AL,EU»EL,RU,RL»MU*KM,DU»SU»LAM 4
COM MCN/TCC/ Ul, US 1. ASF, XhT 5
DIMENSION G(7),W(7),T(2C1),P(10),A(7,7),B(7,7), 6
1 BL(7,7 ) ,BR(7,7 ) ,C(7,7 ) , DG (7,7 ) ,R(201,7) , 7
2 Y(201»7).D(7,7»20C) 8
CIMENSION DYS AV E(20 1,7 ) 9
C ... MUST SPECIFY ML = FCT(X) 10
ML = 0.0 11
Ul = 1.0 12
AU = 1.1
EU = 1.0 14
RU = 1.0 15
AL = 0.9 16
EL = 1.0 17
RL = 1.0 18
MU = 1.0
CU = 2.5 20
SU = 0.5 21
KM = .012 22
N =7 23
NC = 7 24
M = 200 25
M = 2C0 26
l\P =201 27
N PAR = 1 28
NIT = 5
IPRNT =1 30
ERROR = l.OD-8 31
CFECC = 5.0D-15 32
BLC = 10. 33
CG ICO 1=1,7 34
NJ
O'

CG ICO J = 1» 7 35
EL( ItJ) = 0.0 36
ORI I fJ) = 0.0 37
100 CONTINUE 38
E L ( 1 , 1 ) = 1.0 39
BL(2,2) = 1.0 40
8 R C 3 ♦ 3 ) = 1.0 41
ER(4,4) = 1.0 42
D R ( 5 » 5 ) = 1.0 43
EL(6,6) = 1.0 44
0L{7 j 7 ) = 1.0 45
150 CONTINUE 46
FI = 3.141586 49
FI2 = P I * 2 50
P02 = .5 » PI 51
LAN = 2.74950
CO 2C0 1=1,NP 53
T(I) = (I-l.J/M 54
Z = T(I ) 55
CT = DCCS(PG2#Z) 56
ST = DSIN(P02*Z ) 57
V( I , 1) = 1.0 - CT 58
Y(1,2) = 1.75#ST 59
Yl I,3J = 2.50CT 60
Y( I , 4 I = -LAN - ( 1. -Z )* K M/MU 61
Y(I,5) = 5C0»CT 62
Y(1 ,6) = .350*ST 63
Yd, 7 ) = ~.275#ST 64
200 CONTINUE 69
CALL CUASI (N,M,NCtNI,NP,.NPAR,NIT,IPRNT,G,W,T,P,A,B,BLrBR,C,DG,R, 79
1 Y , C , OYSAV E) 80
IF{E2MAX .LE. ERROR ) GO TO 90C 81
000 CONTINUE 82
WRITE (6,810) E2NAX,ERROR 83
237

810 FORMAT (» 1*/'O' , 1CX, ' CIC NOT CONVERGE, E2MAX = ' ,D12« 5 , '.GT. ERROR 84
1=4,C12.5) 85
CC TC 920 86
900 WRITE (6,910) E2MAX,ERROR 87
910 FORMAT (* 1•/*0•,10X, 'CONVERGED, E2MAX = • ,012.5, • .LE. ERROR =*,012 88
1.5) 89
920 CONTINUE 90
CIF = -( LAM + Y(l,A) }*MU/KM 91
XHT = l./M 47
YI2 = 0.0 48
CC 930 1=1,NP
CYT = Y( I,2)*Y( I,2)*XHT 65
IFCNP-I) 925,925,930
925 CYT = .5*DYT
930 Y I 2 = Y I 2 + CYT
ASF = DSGRT(Y I 2) 70
AIN = l./ASF 71
CC 935 J=1,N 72
CC 935 1=1,NP 73
GO TC (931,931,931,935,923,933,935), J 74
931 Y ( I , J ) = A IN *Y ( I,.J ) 75
GC TC 935 76
933 Y l I , J ) = ASF*Y(I,.J) 77
935 CONTINUE 78
WRITE (6,945) CIF 92
945 FORMAT ( *1*/* 0 * , 20X,* STATE VARIABLES ... COST FUNCTION =',D15.6) 93
WRITE (6,946) Y I 2 »• A S F 94
946 FORMAT!»0*,20X, 'NORM SQUARED ...*,D15.6,5X,'NORMALIZATION CONSTANT 95
1 ...',C15.6,//) 96
WRITE (6,947) 97
947 FORMAT (9X1HX,8X2hY1,8X2hY2,8X2HY3,8X2HY4,8X2HY5,8X2HY6,8X2HY7, 98
18X2FY8,8X2HY9,7X3HY1C//) 99
WRITE (6,950) ( T ( I ) , ( Y ( I ,. J ) , J = 1,7 ) , I = 1, N P ) 100
950 FORMAT(5 X,F7.4,7F10.4) 101
238

STGP
ENC

SUBROUTINE D I F EG (N»NC,NPAR,IM,I TER,G,W,P,DG) 104
IMPLICIT REAL*8(A~H, G-Z ) 105
REAL*8 MU,KM,ML,LAM 106
COM MCN/QUCOMM/X,ERROR,0FFDG,H,E1MAX,E2MAX,BLO 107
COM M CN/UAT A/ AU,AL,EU,EL»RU,RL,MU,KM,DU,SU,LAM 108
COMMCN/TCO/ Ul,ÜSlf ASF»XI-T 109
DIMENSION G(NC),W(NC),.P(NPAR),DG(NC»NC) 110
C ... MUST SPECIFY ML = FCT(X) 111
ML = 0.0 112
EXU = 1./3. 113
US2 = 1.0 114
US3 = 1.0 115
XN = 2 • *W ( 5 ) * W ( 3 ) 116
XD = (KM*W(7) - MU)*US2*US3 117
ARG = XN/XD 118
Z1 = DSIGNIUS2,ARG) 119
US1 = ( CABS(ARG) )**EXU 120
US1 = Z1*US1 121
IF(US1.GT.AL) GC TO 60 122
50 U1 = AL 123
CUW 3 = 0.0 124
CUW 5 = 0.0 125
CUW7 = C.O 126
CAM 1 = 0.0 127
GC TC 100 128
60 IF(USl.GE.AU) GC TO 70 129
U1 = US1 130
CUW3 = US1/(3.»W(3 ) ) 131
CUW5 = US1/(3.*W(5 ) ) 132
CUW7 = -EXU*USl«KM/(KM*h(7) - MU) 133
GAM 1 = C.O 134
GC TC 100 135
70 CONTINUE 136
LI = AU 137
240

CUW3 =
0.0
CUW5 =
0.0
CUW7 =
0.0
GAÑI =
0 .0
100
CONTINUE
U2
US2
U3
US3
GAM 2 =
0.0
GAM 3 =
0.0
G ( 1 ) =
W (2 )
G ( 2 ) =
W(3)/(U1*U1«U2)
G ( 3 ) *
W(2)*W(4 )
G (4 ) =
KM*(ML 4 U 1 * U 3 )
G { 5 ) =
~ W ( 6 ) * W ( 4 )
G ( 6 ) =
-W(5 ) /{U1«U1«U2)
C ( 7 ) =
— to { 6 ) #W { 2 )
200
CONTINUE
CO 220
1 = 1,
7
CC 2 20
J = l,
7
220
CG( I , J)
=
0.0
CG(1,2)
=
1.0
CGU
= -
2.*G(2)/U1
C G ( 2 » 3 )
=
CGU» CUW3 +
CG(2,5)
=
DGU*CUW5
CG(2,7 )
=
C G U * C U W 7
CG(3 » 2 )
=
W ( 4 )
CG{3,4 )
=
W ( 2 )
CGU
=
KM*U3
CG(4 t 3 )
=
DGU*CUW3
CG(4 » 5 )
=
DGU-CUW5
CG{4i7 )
=
DGU*CUW7
CG(5,4)
= -
W ( 6 )
CG{5,6 )
= -
W ( 4 )
CGU
= -
2 .*G(6}/U1
1./(U1*U1*U2)
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
241

300
CGl6 »3 )
CG C 6 * 5 )
CG Í 6,7)
CG(7,2 )
C G ( 7 » 6 )
CONTINUE
RETURN
ENC
DGU * CUW 3
DCU»CUW5
DGU*CUW7
W ( 6 )
W ( 2 )
i. / (U1*U1*U2)
172
173
174
175
176
177
173
179
242

SUBROUTINE CCRRC (N,M,NC,NP,NPAR,ITER,T,P,BL,BR,Y) 180
IMPLICIT REAL *8( A-H , 0“Z ) 181
R E A L * 8 MU,KM,ML,LAM 182
C0MMGN/QUC0MM/X,ERROR,ÃœFFDG,H,E1MAX,E2MAX,BL0 183
COMMCN/DATA/ AU , AL, EU , EL,RU,RL,MU,KM,DU,SU,LAM 184
DIMENSION T(NP),P(NPAR),BL(NC,NC ) ,BR(NC,NC),Y(NP»NC) 135
Y<1, 1 ) = 0.0 186
Y(1 ,2) = 0.0 187
Y{N P , 3 ) = 0.0 188
Y ( N P »4 ) =-LAM 189
Y(NP,5 ) =0.0 190
Y í1,6 ) = 0.0 191
Y(l, 7) =, 0.0 192
RETURN 193
ENC 194

BIBLIOGRAPHY
Ashley, H. and McIntosh, S. C., Jr. 1968 "Application of Aeroelastic
Constraints in Structural Optimization," Proc. 12th Inter.
Congress Appl. Mech., 100-113.
Ashley, H., McIntosh, S. C., Jr., and Weatherill, W. H. 1970 "Optimi¬
zation under Aeroelastic Constraints," [in: Proc, AGARJD Sympo¬
sium on Structural Optimization, (AD-715 483)], 5.1-5.28.
Asimov, I. 1960 Asimov's Biographical Encyclopedia of Science and
Technology, Doubleday.
Barnett, R. L. 1961 "Minimum Weight Design of Beams for Deflection,"
Proc. ASCE J. Eng. Mech. 87, 75-109.
Barnett, R. L. 1963a "Minimum Weight Design of Beams for Deflection,"
Proc. ASCE J. Eng. Mech. 128, 221-255.
Barnett, R. L. 1963b "Minimum Deflection Design of a Uniformly Accel¬
erating Cantilever Beam," Trans. ASME J. Appl. Mech. E30, 466-467.
Barnett, R. L. 1966 "Survey of Optimal Structural Design," Exper.
Mech. 6^, 19A-26A.
Barnes, E. R. 1971 "Necessary and Sufficient Conditions for a Class
of Distributed Parameter Control Systems," SIAM J. Control 9,
62-82.
Barston, E. M. 1974 "A Minimax Principle for Nonoverdamped Systems,"
Inter. J. Eng. Sci. 12, 413-422.
Bellamy, N. W. and West, M. J. 1969 "Methods of Profile Optimization
by Iterative Analog Computation," The Computer J. 12, 132-138.
Berg, P. W. 1962 "Calculus of Variations," [in: Handbook of Engineer¬
ing Mechanics, ed. W. Flugge, McGraw-Hill], 16.1-16.15.
Bergamini, D. 1963 Mathematics, Life Science Library, Time-Life Books.
Bhargava, S. and Duffin, R. J. 1973 "Dual Extremum Principles Relating
to Optimum Beam Design," Arch. Rati. Mech. Anal. 50, 314-330.
Blasius, H. 1913 "Trager kleinster Durchbeigung und Stabe grosster
Knickfestikeit bei gegebenem Materialverbrauch," Zeit. fur Math.
und Phys. 62, 182-197.
244

245
Bliss, G. A. 1925 Calculus of Variations, Open Court Publishing Company.
Bliss, G. A. 1946 Lectures on the Calculus of Variations, University
of Chicago Press.
Bolotin, V. V. 1963 Nonconservative Problems of the Theory of Elastic
Stability, Pergamon Press.
Bolza, 0. 1904 Lectures on the Calculus of Variations, Chelsea Pub¬
lishing Company.
Boykin, W. H., Jr. and Sierakowski, R. L. 1972 "Remarks on Pontryagin's
Maximum Principle Applied to a Structural Optimization Problem,"
J. Royal Aero. Soc. 76, 175-176.
Brach, R. M. 1968 "On the Extremal Fundamental Frequencies of Vibrat¬
ing Beams," Inter. J. Solids Struct. b_, 667-674.
Brach, R. M. and Walters, J. J. 1970 Maximum Frequency of Beams Includ¬
ing Shear Effects, Notre Dame Report 70-02-04 (AD-715 021).
Breakwell, J. V. 1959 "The Optimization of Trajectories," J. SIAM 7,
215-247.
Bryson, A. E., Jr. and Denham, W. F. 1962 "A Steepest Ascent Method
for Solving Optimum Programming Problems," Trans. ASME J. Appl.
Mech. E29, 247-257.
Bryson, A. E., Jr. and Denham, W. F. 1964 "Optimal Programming Problems
with Inequality Constraints II: Solution by Steepest-Ascent,"
AIAA J. 2, 25-34.
Bryson, A. E., Jr., Denham, W. F., and Dreyfus, S. E. 1963 "Optimal
Programming Problems with Inequality Constraints I: Necessary
Conditions for Extremal Solutions," AIAA J. 1, 2544-2550.
Bryson, A. E., Jr. and Ho, Yu-chi. 1969 Applied Optimal Control, Ginn.
Bullock, T. E. 1966 Computation of Optimal Controls by a Method Based
on Second Variations, Stanford University Report SUDAAR No. 297.
Carll, L. B. 1881 A Treatise on the Calculus of Variations, John Wiley
and Sons.
Chem, J. M. 1971a "Optimal Structural Design for Given Deflection in
the Presence of Body Forces," Inter. J. Solids Struct. J7,
373-382.
Chern, J. M. 1971b "Optimal Thermoelastic Design for Given Deflection,"
Trans. ASME J. Appl. Mech. E38, 538-540.

246
Chern, J. M. and Prager, W. 1970 "Optimal Design of Beams for Pre¬
scribed Compliance under Alternative Loads," J. Opt. Theory
Applns. J5, 424-431.
Citron, S. J. 1969 Elements of Optimal Control, Holt, Rinehart, and
Winston, 154-166.
Clausen, T. 1851 "líber die Form architektonisher SMulen,"
Bull, phys.-math. de l'Académie 9_, 368-379.
Cunningham, W. J. 1958 Introduction to Nonlinear Analysis, McGraw-
Hill.
Denn, M. M. 1969 Optimization by Variational Methods, McGraw-Hill.
De Silva, B. M. E. 1969 "Minimum Weight Design of Disks Using a Fre¬
quency Constraint," Trans. ASME J. Eng. Industry B91, 1091-
1099.
De Silva, B. M. E. 1972 "Optimal Vibrational Modes of a Disk,"
J. Sound Vib. 21, 19-34.
Dixon, L. 1967 "Pontryagin's Maximum Principle Applied to the Profile
of a Beam," J. Royal Aero. Soc. 71, 513-515.
Dixon, L. 1968 "Further Comments on Pontryagin's Maximum Principle
Applied to the Profile of a Beam," J. Royal Aero. Soc. 72,
518-519.
Dorn, W. S., Gomory, R. E., and Greerberg, H. S. 1964 "Automatic
Design of Optimal Structures," J. de Mecanique 3, 25-52.
Dreyfus, S. E. 1965 Dynamic Programming and the Calculus of Variations,
Academic Press.
Drucker, D. C. and Shield, R. T. 1957a "Bounds on Minimum Weight
Design," Quart. Appl. Math. 15, 269-281.
Drucker, D. C. and Shield, R. T. 1957b "Design for Minimum Weight,"
Proc. 9th Inter. Congress Appl. Mech., 212-222.
Dubey, R. 1970 "Variational Methods for Nonconservative Problems,"
Trans. ASME J. Appl. Mech. E37, 133-136.
Dupuis, G. 1971 "A Finite Element Approach to Structural Optimiza¬
tion," [in: Optimality Criteria in Structural Design, ed.
W. Prager and P. V. Marcal, AFFDL-TR-70-166], 100-112.
Elsgolc, L. E. 1961 Calculus of Variations, Addison-Wesley.

247
Feigen, M. 1952 "Minimum Weight of Tapered Round Thin-walled Columns,"
Trans. ASME J. Appl. Mech. E19, 375-380.
Fox, R. L. 1971 Optimization Methods for Engineering Design, Addison-
Wesley.
Fox, R. L. and Kapoor, M. P. 1970 "Structural Optimization in the
Dynamic Response Regime: A Computational Approach," AIAA J.
8, 1798-1804.
Fung, Y. C. 1965 Foundations of Solid Mechanics, Prentice-Hall.
Gajewski, A. and Zyczkowski, M. 1970 "Optimal Design of Elastic Col¬
umns Subject to the General Conservative Behavior of Loading,"
ZAMP 21, 806-818.
Gallagher, R. H. and Zienkiewicz, 0. C., eds. 1973 Optimal Structural
Design, John Wiley and Sons.
Gelfand, I. M. and Fomin, S. V. 1963 Calculus of Variations, Prentice-
Hall.
Gellatly, R. A., Ed. 1970 Proc. AGARD Symposium on Structural Optimi¬
zation (AD-715 483).
Gellatly, R. A. and Berke, L. 1971 Optimal Structural Design
(AFFDL-TR-70-166).
Gerard, G. 1966 "Optimum Structural Design Concepts for Aerospace
Vehicles," AIAA J. Spacecraft 3, 5-18.
Ghosh, T. K. 1973 Dynamics and Optimization of a Human Motion Problem,
Ph.D. Dissertation, University of Florida.
Gjelsvik, A. 1971 "Minimum Weight Design of Continuous Beams,"
Inter. J. Solids Struct. _7, 1411-1425.
Gould, S. H. 1957 Variational Methods for Eigenvalue Problems, Univer¬
sity of Toronto Press.
Greenhill, A. G. 1881 "Title Not Known," Proc. Cambridge Phil. Soc. 4-,
Halkin, H. 1963 "The Principle of Optimal Evolution,” [in: Nonlinear
Differential Equations and Nonlinear Mechanics, eds. J. P.
Lasalle and S. Lefschetz, Academic Press], 284-302.
Hanson, J. N. 1968 "Comparison of Two Optimizations of a Nonlinear
Boundary-value Problem," AIAA J. 6^, 1979-1985.

248
Haug, E. J. and Kirmser, P. G. 1967 "Minimum Weight Design of Beams
with Inequality Constraints on Stress and Deflection," Trans.
ASMEJ. Appl. Mech. E34, 999-1004.
Haug, E. J., Streeter, T. D., and Newell, R. S. 1969 Optimal Design of
Elastic Tructural Elements (AD-688 139).
Hennig, G. R. and Miele, A. 1972 Sequential Gradient-Restoration Algo¬
rithm for Optimal Control Problems with Bounded State Variables,
Rice University Report AAR-101.
Hestenes, M. R. 1966 Calculus of Variations and Optimal Control Theory,
John Wiley and Sons.
Hornbuckle, J. E., Nevill, G. E., Jr., and Boykin, W. H., Jr. 1974
"Finite Element Method Applied to a Structural Optimization
Problem," submitted to: J. Royal Aero. Soc.
Hu, T. C. and Shield, T. T. 1961 "Uniqueness in the Optimum Design of
Structures," Trans. ASME J. Appl. Mech. E28, 284-287.
Huang, H. C. 1968 "Optimal Design of Elastic Structure for Maximum
Stiffness," Inter. J. Solids Struct. 4^, 689-700.
Huang, N. C. and Sheu, S. Y. 1968 "Optimal Design of an Elastic Column
of Thin-walled Cross Section," Trans. ASME J. Appl. Mech. E35,
285-288.
Huang, N. C. and Tang, H. T. 1969 "Minimum Weight Design of Elastic
Sandwich Beams with Deflection Constraints," J. Opt. Theory
Applns. 4, 277-298.
Icerman, L. J. 1969 "Optimal Structural Design for Given Dynamic
Deflection," Inter. J. Solids Struct. J5, 473-490.
Ireland, N. 0. 1962 Index to Scientists of the World from Ancient to
Modern Times, F. W. Faxon Company.
Irving, J. and Mullineaux, N. 1959 Mathematics in Physics and Engi¬
neering , Academic Press.
Keller, J. B. 1960 "The Slope of the Strongest Column," Arch. Rati.
Mech. Anal. _5, 275-285.
Keller, J. B. and Niordson, F. I. 1966 "The Tallest Column," J. Math.
Mech. 16, 433-446.
Komkov, V. 1972 Optimal Control Theory for the Damping of Variations
of Simple Elastic Systems, Lecture Notes in Mathematics 253
Springer-Verlag.

249
Kopp, R. E. 1962 "Pontryagin Maximum Principle," [in: Optimization
Techniques, ed. G. Leitmann, Academic Press], 255-279.
Kopp, R. E. 1963 "On the Pontryagin Maximum Principle," [in: Non-
linear Differential Equations and Nonlinear Mechanics, eds.
J. P. Lasalle and S. Lefschetz, Academic Press], 395-402.
Kowalik, J. S. 1970 "Feasible Direction Methods," [in: Proc. AGARD
Symposium on Structural Optimization (AD-715 483)], 2.1-2.16.
Lanczos, C. 1962 Linear Differential Operators, D. Van Nostrand
Company.
Leipholz, H. H. E. 1972 "On the Sufficiency of the Energy Criterion
for the Stability of Certain Nonconservative Systems of the
Follower-load Type," ASME Paper 72-APM-E.
Lovitt, W. V. 1924 Linear Integral Equations, Dover Publications.
Luenberger, D. G. 1969 Optimization by Vector Space Methods, John
Wiley and Sons.
Maday, C. J. 1973 "A Class of Minimum Weight Shafts," ASME Paper
73-DET-10.
Martin, J. B. 1970 "Optimal Design of Elastic Structures for Multi¬
purpose Loading," J. Opt. Theory Applns. b_, 22-40.
Martin, J. B. 1971 "The Optimal Design of Beams and Frames with
Compliance Constraints," Inter. J. Solids Struct. 7_, 63-81.
Masur, E. F. 1970 "Optimum Stiffness and Strength of Elastic Struc¬
tures," Proc. ASCE J. Eng. Mech. 96, 621-640.
Mayeda, R. and Prager, W. 1967 "Minimum Weight Design of Beams for
Multiple Loading," Inter. J. Solids Struct. _3, 1001-1011.
McCart, B. R., Haug, E. J., and Streeter, T. D. 1970 "Optimal Design
of Structures with Constraints on Natural Frequency," AIAA J.
8, 1012-1019.
McIntosh, S. C. and Eastep, F. E. 1968 "Design of Minimum-Mass Struc¬
tures with Specified Stiffness," AIAA J. 6^, 962-964.
McIntosh, S. C., Weishaar, T. A., and Ashley, H. 1969 Progress in
Aeroelasticity Optimization—Analytical versus Numerical
Approaches, Stanford University Report SUDAAR No. 383
(AD-695 057).

250
McNeill, W. A. 1971 "Structural Weight Minimization Using Necessary
and Sufficient Conditions," J. Opt. Theory Applns. J3, 454-466.
Melosh, R. J. and Luik, R. 1967 Approximate Analysis and Allocation
for Least Weight Structural Design (AFFDL-TR-67-59).
Miele, A., Well, K. H., and Tietze, J. L. 1972 Modified Quasilinear¬
ization Algorithm for Optimal Control Problems with Bounded
State Variables, Rice University Report AAR-103.
Mikhlin, S. G. and Smolitskiy, K. L. 1967 Approximate Methods for
Solution of Differential and Integral Equations, American
Elsevier Publishing Company.
Moses, F. and Kinser, D. E. 1967 "Optimum Structural Design with
Failure Probability Constraints," AIAA J. _5, 1152-1158.
Niordson, F. I. 1964 "On the Optimum Design of a Vibrating Beam,"
Quart. Appl. Math. 23, 47-53.
Oden, J. T. and Reddy, J. N. 1974 "On Dual-Complementary Variational
Principles in Mathematical Physics," Inter. J. Eng. Sci. 29,
1-29.
Opatowski, I. 1944 "Cantilever Beams of Uniform Strength," Quart..
Appl. Math. _3, 76-81.
Plaut, R. H. 1970 "On Minimizing the Response of Structures to Dynamic
Loading," ZAMP 21, 1004-1010.
Plaut, R. H. 1971a "Structural Optimization of a Panel Flutter Prob¬
lem," AIAA J. j), 182-184.
Plaut, R. H. 1971b "On the Optimal Structural Design for a Noncon¬
servative Elastic Stability Problem," J. Opt. Theory Applns. ]_,
52-60.
Plaut, R. H. 1971c "Optimal Structural Design for Given Deflec¬
tion Under Periodic Loading," Quart. Appl. Math. 29, 315-318.
Pontryagin, L. S., Boltyanskii, V. G., Gamkrelidze, R. V., and
Mishchenko, E. F. 1962 The Mathematical Theory of Optimal
Processes, John Wiley and Sons.
Pope, C. G. and Schmidt, L. A., eds. 1971 Structural Design Applica¬
tions of Mathematical Programming Techniques, AGARDograph 149
(AD-719 710).
Prager, W. 1969 "Optimality Criteria Derived from Classical Extremum
Principles," [in: An Introduction to Structural Optimization,
University of Waterloo Press], 165-178.

251
Prager, W. 1970 "Optimization of Structural Design," J. Opt. Theory
Applns. 6^, 1-21.
Prager, W. and Shield, R. T. 1967 "A General Theory of Optimal
Plastic Design," Trans. ASME J. Appl. Mech. E34, 184-186.
Prager, W. and Shield, R. T. 1968 "Optimal Design of Multi-Purpose
Structures," Inter. J. Solids Struct. 4_, 469-475.
Prager, W. and Taylor, J. E. 1968 "Problems in Optimal Structural
Design," Trans. ASME J. Appl. Mech. E35, 102-106.
Prasad, S. and Herrmann, G. 1969 "Usefulness of Adjoint Systems in
Solving Nonconservative Stability Problems in Elastic Continua,"
Inter. J. Solids Struct. j5, 727-735.
Przemieniecki J- S. 1968 Theory of Matrix Structural Analysis,
McGraw-Hill.
Rosen, J. B. 1960 "The Gradient Projection Method for Nonlinear Pro¬
gramming: Part I," SIAM J. Appl. Math. 181-217; "Part II,"
9, 414-432.
Roxin, E. 1963 "A Geometric Interpretation of Pontryagin's Maximum
Principle," [in: Nonlinear Differential Equations and Non¬
linear Mechanics, eds. J. P. Lasalle and S. Lefschetz, Academic
Press], 303-324.
Rozonoér, L. I. 1959 "Pontryagin's Maximum Principle in the Theory of
Optimum Systems," Avtomat. i Telemekh 20 [English translation:
Automation and Remote Control 1960, 1288-1302, 1405-1421, 1517-
1532].
Rozvany, G. I. N. 1966 "Analysis versus Synthesis in Structural
Engineering," Civil Eng. Trans., Institution of Engineering,
Australia CE8, 158-166.
Rubin, C. P. 1970 "Minimum Weight Design of Complex Structures Subject
to a Frequency Constraint," AIAA J. ^8, 923-927.
Rudisill, C. S. and Bhatia, K. G. 1971 "Optimization of Complex
Structures to Satisfy Flutter Requirements," AIAA J. 9_, 1487-
1491; Errata, 9, 2479.
Salinas, D. 1968 On Variational Formulations for Optimal Structural
Design, Ph.D. Dissertation, University of California, Los Angeles.
Salukvadze, M. E. 1971 "Optimization of Vector Functionals: Program¬
ming of Optimal Trajectories," English translation: Automa¬
tion and Remote Control J5, 1169-1178.

252
Schmidt, L. A. Jr. 1966 "Automated Design," Inter. Sci. Tech. 54,
63-78 and 115-117.
Schmidt, L. A., Jr. 1968 "Structural Synthesis," [in: Composite Mate¬
rials Workshop, eds. S. W. Tsai, J. C. Halpin, and N. J. Pagano,
Technomic Publishing Company], 309-343.
Sheu, C. Y. 1968 "Elastic Minimum-Weight Design for Specified Funda¬
mental Frequency," Inter. J. Solids Struct. _4, 953-958.
Sheu, C. Y. and Prager, W. 1968a "Minimum Weight Design with Piecewise
Constant Specific Stiffness," J. Opt. Theory Applns. J2,
179-186.
Sheu, C. Y. and Prager, W. 1968b "Recent Developments in Optimal
Structural Design," AMR 21, 985-992.
Shield, R. T. 1963 "Optimum Design Methods for Multiple Loading,"
ZAMP 14, 38-45.
Shield, R. T. and Prager, W. 1970 "Optimal Structural Design for
Given Deflection," ZAMP 21, 513-523.
Simitses, G. J. 1973 "Optimal Vs. the Stiffened Circular Plate,"
AIAA J. 11, 1409-1412.
Simitses, G. J., Kamat, M. P., And Smith, C. V., Jr. 1973 "Strongest
Column by the Finite Element Displacement Method," AIAA J. 11,
1231-1232.
Speyer, J. L. and Bryson, A. E., Jr. 1968 "Optimal Programming Prob¬
lems with a Bounded State Space," AIAA J. 6^, 1488-1491.
Stroud, W. J., Dexter, C. B., and Stein, M. 1971 Automated Prelim¬
inary Design of Simplified Wing Structures to Satisfy Strength
and Flutter Requirements, NASA TN-D-6534.
Tadjbakhsh, I. and Keller, J. B. 1962 "The Strongest Column and Iso-
perimetric Inequalities for Eigenvalues," Trans. ASME J. Appl.
Mech. E29, 159-164.
Taylor, J. E. 1967 "The Strongest Column: An Energy Approach,"
Trans. ASME J. Appl. Mech. E34, 486-487.
Taylor, J. E. 1969 "Maximum Strength of Elastic Structural Design,"
Proc. ASCE J. Eng. Mech. 95, 653-663.
Taylor, J. E. and Liu, C. Y. 1968 "Optimal Design of Columns,"
AIAA J. 6, 1497-1502.

253
Timoshenko, S. P. and Gere, J. M. 1961 Theory of Elastic Stability,
McGraw-Hill.
Todhunter, I. and Pearson, K. 1893 A History of the Theory of Elastic¬
ity and the Strength of Materials, Vol. 2, Part 1, Cambridge
University Press.
Troitskii, V. A. 1971 "Variational Problems in the Theory of Optimum
Processes," J. Opt. Theory Applns. j3, 1-14.
Truesdell, C. 1968 Essays in the History of Mechanics, Springer-Verlag.
Turner, M. J. 1967 "Design of Minimum-Mass Structures with Specified
Natural Frequencies," AIAA J. 406-412.
Turner, M. J. 1969 "Optimization of Structures to Satisfy Flutter
Requirements," AIAA J. _7, 945-951.
Valentine, F. A. 1937 The Problem of Lagrange with Differential
Inequalities as Added Side Conditions, Ph.D. Dissertation,
University of Chicago.
Wang, P. K. C. 1968 "Theory of Stability and Control for Distributed
Parameter Systems (A Bibliography)," Inter. J. Control 7,
101-116.
Wasiutynski, Z. and Brandt, A. 1963 "The Present State of Knowledge
in the Field of Optimum Design of Structures," AMR 16, 341-350.
Weisshar, T. A. 1970 An Application of Control Theory Methods to the
Optimization of Structures Having Dynamic or Aeroelastic Con¬
straints, Stanford University Report SUDAAR No. 412 (AD-722 439).
Wu, J. J. 1973 "Column Instability and Nonconservative Forces, with
Internal and External Damping—Finite Element Using Adjoint
Variational Principles," Developments in Mechanics, Vol. 7,
Proceedings of the 13th Midwestern Mechanics Conference, Uni¬
versity of Pittsburgh, 5-1-514.
Zarghamee, M. S. 1968 "Optimum Frequency of Structures," AIAA J. Jj,
749-750.
Zienkiewicz, 0. C. 1971 The Finite Element Method in Engineering
Science, McGraw-Hill.

BIOGRAPHICAL SKETCH
Jerry C. Hornbuckle was born in Clearwater, Florida, in 1942.
On graduating from Stranahan High School in Fort Lauderdale, he attended
the Aerospace Engineering program of the University of Florida. Studies
there were interrupted by a year's employment at Honeywell's inertial
guidance division in St. Petersburg, Florida. He received his BSAE
from the University of Florida in 1965.
Following graduation he was employed by the Boeing Company in
Huntsville, Alabama. While there he worked on error analyses, the
Apollo-Saturn V on-board computer flight program, and trajectory simu¬
lation. During the evenings he was a part-time student at the Univer¬
sity of Alabama in Huntsville, receiving an MSE in Engineering Mechan¬
ics in 1971.
In 1969 he began doctoral studies in the Engineering Sciences
department of the University of Florida. Although specializing in
dynamics and vibrations, his dissertation research was in the field of
structural optimization. He expects to graduate in August, 1974, with
a Ph.D. in Engineering Mechanics.
Mr. Hornbuckle is married to the former Carolyn A. Smith of
Fort Lauderdale and has two children. He enjoys golf, tennis, racquet-
ball, music, and reading. In the area of reading, he is interested in
psychology and history, especially military and aviation history.
254

I certify that I have read this study and that in my opinion it
conforms to acceptable standards of scholarly presentation and is fully
adequate, in scope and quality, as a dissertation for the degree of
Doctor of Philosophy.
/ /
William H. Boykin, Jr., Chairman
Associate Professor of Engineering Sciences
I certify that I have read this study and that in my opinion it
conforms to acceptable standards of scholarly presentation and is fully
adequate, in scope and quality, as a dissertation for the degree of
Doctor of Philosophy.
Robert L. Sierakowski, Co-Chairman
Professor of Engineering Sciences
I certify that I have read this study and that in my opinion it
conforms to acceptable standards of scholarly presentation and is fully
adequate, in scope and quality, as a dissertation for the degree of
Doctor of Philosophy.
I certify that I have read this study and that in my opinion it
conforms to acceptable standards of scholarly presentation and is fully
adequate, in scope and quality, as a dissertation for the degree of
Doctor of Philosophy.
j/l/
Gene W. Hemp
Associate Professor of Engineering Sciences

I certify that I have read this study and that in my opinion it
conforms to acceptable standards of scholarly presentation and is fully
adequate, in scope and quality, as a dissertation for the degree of
Doctor of Philosophy.
Associate Professor of Mechanical Engineering
This dissertation was submitted to the Graduate Faculty of the College
of Engineering and to the Graduate Council, and was accepted in partial
fulfillment of the requirements for the degree of Doctor of Philosophy.
August, 1974
Dean, Graduate School






PAGE 2

21 7+( $8720$7(' 237,0$/ '(6,*1 2) &21675$,1(' 6758&785(6 %< -(55< & +251%8&./( $ ',66(57$7,21 35(6(17(' 72 7+( *5$'8$7( &281&,/ 2) 7+( 81,9(56,7< 2) )/25,'$ ,1 3$57,$/ )8/),//0(17 2) 7+( 5(48,5(0(176 )25 7+( '(*5(( 2) '2&725 2) 3+,/2623+< 81,9(56,7< 2) )/25,'$

PAGE 3

&RS\ULJKW E\ -HUU\ & +RUQEXFNOH

PAGE 4

'(',&$7,21 7R P\ JUDQGPRWKHU 0UV )ORUHQFH +RUQEXFNOH DQG P\ ZLIH &DURO\Q :LWKRXW WKH ORYH FRQILGHQFH DQG SDWLHQW XQGHUVWDQGLQJ RI *UDQQ\ +RUQEXFNOH DQG &DURO\Q P\ JUDGXDWH VWXGLHV ZRXOG QHYHU KDYH EHHQ DWWHPSWHG

PAGE 5

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

PAGE 6

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nV 0D[LPXP 3ULQFLSOH 0HWKRG RI 6WHHSHVW $VFHQW'HVFHQW 7UDQVLWLRQ 0DWUL[ $HURHODVWLFLW\ 3UREOHPV 2WKHU 0LVFHOODQHRXV 0HWKRGV 0HWKRGV 'LVFUHWH 6\VWHPV 0DWKHPDWLFDO 3URJUDPPLQJ 'LVFUHWH 6ROXWLRQ $SSUR[LPDWLRQV 6HJPHQWZLVH&RQVWDQW $SSUR[LPDWLRQV &RPSOH[ 6WUXFWXUHV ZLWK )UHTXHQF\ &RQVWUDLQWV )LQLWH (OHPHQW $SSUR[LPDWLRQV &ORVXUH Y

PAGE 7

7$%/( 2) &217(176 &RQWLQXHGf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

PAGE 8

7$%/( 2) &217(176 &RQWLQXHGf &+$37(5 3DJH 9,,, &21&/86,216 $1' 5(&200(1'$7,216 6XPPDU\ DQG &RQFOXVLRQV 5HFRPPHQGDWLRQV $33(1',;(6 $ +,6725,&$/ '(9(/230(176 % $ 6,03/( 3522) 2) 7+( .8+178&.(5 7+(25(0 & &20387(5 68%5287,1( /,67,1*6 %,%/,2*5$3+< %,2*5$3+,&$/ 6.(7&+ YLL

PAGE 9

$EVWUDFW RI 'LVVHUWDWLRQ 3UHVHQWHG WR WKH *UDGXDWH &RXQFLO RI WKH 8QLYHUVLW\ RI )ORULGD LQ 3DUWLDO )XOILOOPHQW RI WKH 5HTXLUHPHQWV IRU WKH 'HJUHH RI 'RFWRU RI 3KLORVRSK\ 21 7+( $8720$7(' 237,0$/ '(6,*1 2) &21675$,1(' 6758&785(6 %\ -HUU\ & +RUQEXFNOH $XJXVW &KDLUPDQ 'U :LOOLDP + %R\NLQ -U &R&KDLUPDQ 'U 5REHUW / 6LHUDNRZVNL 0DMRU 'HSDUWPHQW (QJLQHHULQJ 6FLHQFHV 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH LV DSSOLHG WR WKH RSWLPDO GHVLJQ RI HODVWLF VWUXFWXUHV VXEMHFW WR ERWK KDUG LQHTXDOLW\ FRQVWUDLQWV DQG VXEVLGLDU\ FRQGLWLRQV %\ DQDO\]LQJ WKH PD[LPXP SULQFLSOH DV D QRQn OLQHDU SURJUDPPLQJ SUREOHP DQ H[SOLFLW IRUPXODWLRQ LV GHULYHG IRU WKH /DJUDQJLDQ PXOWLSOLHU IXQFWLRQV WKDW DGMRLQ WKH FRQVWUDLQWV WR WKH FRVW IXQFWLRQDO :LWK WKLV UHVXOW WKH XVXDO QHFHVVDU\ FRQGLWLRQV IRU RSWLn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

PAGE 10

IRUPXODWLRQ PHQWLRQHG DERYH 6LQFH WKH SUREOHP FRQVLGHUHG LV FRQVHUYDn WLYH LW LV WKHUHIRUH GHVFULEHG PDWKHPDWLFDOO\ E\ D VHOIDGMRLQW V\VWHP XQGHU WKLV FRQGLWLRQ LW LV VKRZQ WKDW WKH PLQLPXP ZHLJKW SUREOHP LV LGHQWLFDO WR WKH PD[LPXP EXFNOLQJ ORDG SUREOHP ,Q RUGHU WR GHPRQVWUDWH WKH WKHRU\ IRU WKH SURJUDPPLQJ WHFKQLTXHV XVHG WKH EHDP SUREOHP LV DOVR VROYHG E\ XVLQJ D ILQLWH HOHPHQW UHSUHn VHQWDWLRQ RI WKH VWUXFWXUH )URP D FRPSDULVRQ WR WKH PD[LPXP SULQFLSOH VROXWLRQ LW LV VKRZQ WKDW WKH IRUP RI WKH RSWLPDO VROXWLRQ REWDLQHG LV GHSHQGHQW XSRQ WKH PDJQLWXGH RI WKH WROHUDQFH XVHG ZLWK WKH QXPHULFDO VROXWLRQ VFKHPH )XUWKHUPRUH LW LV VKRZQ WKDW FRQYHUJHQFH E\ WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP LV UHODWHG WR WKH UHVSHFWLYH FXUYDWXUHV RI WKH LQLWLDO JXHVV DQG WKH VROXWLRQ 5HFRPPHQGDWLRQV IRU DGGLWLRQDO LQYHVWLJDWLRQV SHUWLQHQW WR WKLV VWXG\ DUH DOVR LQFOXGHG L[

PAGE 11

&+$37(5 ,1752'8&7,21 6XUYH\ 3DSHUV ,W LV H[FHHGLQJO\ GLIILFXOW WR ZULWH D JHQHUDO LQWURGXFWLRQ WR WKH ILHOG RI VWUXFWXUDO RSWLPL]DWLRQ IRU WZR EDVLF UHDVRQV Lf WKHUH LV QR FRQYHQWLRQDOO\ DFFHSWHG QRPHQFODWXUH DQG LLf WKHUH LV DOVR QR FRQYHQWLRQDOO\ DFFHSWHG FODVVLILFDWLRQ RI SUREOHP W\SHV RU FKDUDFWHUn LVWLFV ,Q PDUNHG FRQWUDVW ZKHQ RQH FRQVLGHUV WKH FDOFXOXV RI YDULDn WLRQV FRVW IXQFWLRQDO V\VWHP HTXDWLRQV NLQHPDWLF DQG QDWXUDO ERXQGn DU\ FRQGLWLRQV DGMRLQW YDULDEOHV +DPLOWRQLDQ HWF DOO KDYH ZHOO GHILQHG XQLYHUVDOO\ DFFHSWHG PHDQLQJV $GGLWLRQDOO\ WKHUH LV QR FRQn IXVLRQ ZKHQ VSHDNLQJ RI WKH SUREOHP W\SHV RI 0D\HU /DJUDQJH DQG %RO]D 7KLV FRPPRQ ODQJXDJH DQG FDWHJRUL]DWLRQ RI SUREOHPV GRHV QRW H[LVW LQ VWUXFWXUDO RSWLPL]DWLRQ ,QVWHDG WKH ILHOG WHQGV WR EUDQFK DQG IUDJPHQW LQWR YHU\ VSHFLDOn L]HG VXEGLVFLSOLQHV WKDW DUH RULHQWHG WRZDUGV DSSOLFDWLRQV :KLOH WKHVH EUDQFKHV DUH UHODWHG WR WKH JHQHUDO ILHOG WKH WHFKQLTXHV DQG PHWKRGV RI RQH EUDQFK FDQ VHOGRP EH DSSOLHG WR DQRWKHU 0RUHRYHU DV D UHVXOW RI WKH WHQGHQF\ WR DQ DSSOLFDWLRQV RULHQWDWLRQ VROXWLRQV DUH JHQHUDOO\ DG KRF DQG QRW XVHIXO IRU RWKHU SUREOHPV HYHQ ZLWKLQ WKH VDPH EUDQFK 7KH ODFN RI DQ\ GHILQLWLYH XQLILFDWLRQ RI WKH VXEMHFW FDQQRW EH EODPHG HLWKHU RQ EHLQJ UHFHQWO\ GHYHORSHG RU LQ UHFHLYLQJ WRR OLWWOH DWWHQWLRQ

PAGE 12

7KLV LV UHDGLO\ VHHQ E\ FRQVLGHULQJ WKH VXUYH\ SDSHUV GHVFULEHG LQ WKH IROORZLQJ SDUDJUDSKV 7KH HDUOLHVW FRPSUHKHQVLYH VXUYH\ SDSHU LV :DVLXW\QVNL DQG %UDQGW f $OWKRXJK WKHLU H[FHOOHQW KLVWRULFDO GHYHORSPHQW LV GRPLQDWHG E\ 5XVVLDQ DQG (DVWHUQ (XURSHDQ UHIHUHQFHV WKH DXWKRUV GR LQFOXGH D KLJKHU SHUFHQWDJH RI SDSHUV E\ :HVWHUQ DXWKRUV WKDQ LV HQFRXQWHUHG LQ WKH W\SLFDO SDSHU IURP (DVWHUQ (XURSH $ PRUH IXQGDn PHQWDO FULWLFLVP LV WKDW WRR OLWWOH LV VDLG UHJDUGLQJ SUREOHP W\SHV RU VROXWLRQ PHWKRGV &KURQRORJLFDOO\ WKH QH[W VXUYH\ SDSHU LV *HUDUG f 7KH WKHPH RI WKLV UHYLHZ LV DHURVSDFH DSSOLFDWLRQV ZLWK D SDUWLFXODU RULHQWDWLRQ WR WKH GHVLJQPDQDJHPHQW GHFLVLRQPDNLQJ SURFHVV 0RVW RI WKH SDSHUV FLWHG WUHDW VSHFLDOL]HG DHURVSDFH VWUXFWXUHV DQG DSSOLFDn WLRQV KRZHYHU WKH DXWKRU GRHV WU\ WR JHQHUDOL]H E\ LQWURGXFLQJ D GHVLJQ LQGH[ D PDWHULDO HIILFLHQF\ SDUDPHWHU 0 DQG D VWUXFWXUDO HIILFLHQF\ SDUDPHWHU 6 $IWHU GHILQLQJ WKH H[SUHVVLRQV IRU 0 DQG 6 FRUUHVSRQGLQJ WR VHYHUDO VWUXFWXUDO HOHPHQWV GHVLJQ FKDUWV DUH SUHn VHQWHG ZKLFK VKRZ UHJLRQV RI SRVVLEOH DSSOLFDWLRQ IRU YDULRXV PDWHULDOV 8QIRUWXQDWHO\ WKH GHVLJQ FKDUWV GR QRW VDWLVI\ H[SHFWDWLRQV DURXVHG E\ WKH LQWURGXFWLRQ RI WKH WKUHH JHQHUDO SDUDPHWHUV 5R]YDQ\ f SUHVHQWV D VLPLODU SDSHU SHUWDLQLQJ WR VWUXFWXUHV LQ FLYLO HQJLQHHULQJ 7KLV SDSHU LV OHVV FRPSUHKHQVLYH DQG PRUH RULn HQWHG WRZDUGV VSHFLILF VWUXFWXUDO DSSOLFDWLRQV WKDQ HLWKHU RI WKH SUHn FHGLQJ VXUYH\ SDSHUV 7KH DXWKRU SRVWXODWHV VHYHUDO LQWHUUHODWHG TXDQWLWLHV RU SDUDPHWHUVf ZKLFK FRXOG SHUKDSV EH XVHG WR JHQHUDOL]H

PAGE 13

VWUXFWXUDO RSWLPL]DWLRQ LQWR D PRUH UDWLRQDO PHWKRGRORJ\ +RZHYHU WKHVH TXDQWLWLHVf§ORDGLQJ /f PDWHULDO 0f JHRPHWU\ *f LQLWLDO EHKDYLRU ,%f DQG GHVLJQ EHKDYLRU 6%ff§DUH RQO\ DSSOLHG WR DQ DEVWUDFW GLVFXVVLRQ RI FRQFHSWV %DUQHWW f KDV D UHDGDEOH VKRUW VXUYH\ RI WKH ILHOG WKDW GZHOOV PRUH XSRQ WKHRUHWLFDO DVSHFWV +H SRVWXODWHV D JHQHUDO SUREOHP LQ ZKLFK WKH FRVW LV PLQLPL]HG VXEMHFW WR D V\VWHP LQ HTXLOLEULXP ZLWK LWV ORDGV ZKLOH EHKDYLRU FRQVWUDLQWV RQ VWUHQJWK VWLIIQHVV DQG VWDELOLW\ DUH VDWLVILHG 8QLIRUP VWUHQJWK GHVLJQ LQWURGXFHV WKH GLVn FXVVLRQ RI RSWLPDO WUXVVHV YLUWXDO ZRUN WKHRUHPV WKDW ZHUH GHULYHG RULJLQDOO\ IRU WUXVVHV DUH WKHQ DSSOLHG WR VLPXOWDQHRXV SODVWLF FROODSVH SUREOHPV )ROORZLQJ D EULHI GLVFXVVLRQ RI WKH SODVWLF FROODSVH ORDG ERXQGLQJ WKHRUHPV WKHUH LV D VKRUW WUHDWPHQW RI HODVWLF VWDELOLW\ SUREOHPV DQG PDWHULDO PHULW LQGLFHV %DUQHWWnV VWLIIQHVV GHVLJQ H[DPSOH H[KLELWV VHYHUDO LPSRUWDQW IHDWXUHV ZRUWK QRWLQJ 6SHFLILFDOO\ WKH H[DPSOH LV WR PLQLPL]H WKH ZHLJKW RI D EHDP VXEMHFW WR VRPH JLYHQ ORDG ZKHUH WKH GHIOHFWLRQ DW D FHUWDLQ SRLQW LV VSHFLILHG 9LUWXDO ZRUN LV XVHG WR KDQGOH WKH VXEVLGLDU\ GHIOHFWLRQ FRQGLWLRQ 1HFHVVDU\ FRQGLn WLRQV DUH REWDLQHG IURP WKH FDOFXOXV RI YDULDWLRQV EXW PRUH VLJQLIn LFDQWO\ WKH 6FKZDU] LQHTXDOLW\ SURYLGHV D VXIILFLHQW FRQGLWLRQ IRU JOREDO RSWLPDOLW\ %DUQHWW FRQFOXGHV KLV VXUYH\ ZLWK D VHFWLRQ VWDWLQJ WKDW PXOWLORDG GHVLJQV VDWLVI\LQJ DOO WKUHH EHKDYLRU FULWHULD DUH PRUH HDVLO\ VROYHG LQ GHVLJQ VSDFH E\ PDWKHPDWLFDO SURJUDPPLQJ WHFKQLTXHV GHVFULEHG E\ 6FKPLGW f $V D VHTXHO WR WKH FRPSUHKHQVLYH VXUYH\ WR f E\ 8DVLXW\QVNL DQG %UDQGW 6KHX DQG 3UDJHU Ef SUHVHQW D FRPSOHWH UHYLHZ

PAGE 14

RI GHYHORSPHQWV IURP WR 7KLV SDSHU FRQWDLQV WKUHH PDMRU VHFWLRQV JHQHUDO EDFNJURXQG PHWKRGRORJ\ DQG VSHFLILF SUREOHPV ,Q WKH ILUVW VHFWLRQ WKH\ VWDWH WKDW WKH ZHOOSRVHG SUREOHP RI RSWLPDO VWUXFWXUDO GHVLJQ UHTXLUHV VSHFLILFDWLRQ RI WKH f SXUSRVH RI WKH VWUXFWXUH ORDG DQG HQYLURQPHQWf f JHRPHWULF GHVLJQ FRQVWUDLQWV OLPLWV WR GHVLJQ SDUDPHWHUVf f EHKDYLRUDO GHVLJQ FRQVWUDLQWV OLPLWV WR WKH VWDWH RI WKH VWUXFWXUHf f GHVLJQ REMHFWLYH FRVW IXQFWLRQDOf 7KH PHWKRGRORJ\ VHFWLRQ LV QRW QRWHZRUWK\ EXW WKH WKLUG VHFWLRQ OLVWV ZKDW LV LQ WKHLU RSLQLRQ WKH VSHFLILF SUREOHP W\SHV VWDWLF FRPSOLDQFH G\QDPLF FRPSOLDQFH EXFNOLQJ ORDG SODVWLF FROODSVH ORDG PXOWLSXUSRVH PXOWLFRQVWUDLQW VWUXFWXUHV RSWLPDO OD\RXW HJ WUXVVHVf UHLQIRUFHG SUHVWUHVVHG VWUXFWXUHV DQG IURP WKH EDFNJURXQG VHFWLRQ SUREDELOLVWLF SUREOHPV 7KHLU FRQFOXGLQJ UHPDUNV VXFFLQFWO\ VXPPDUL]H WKH SDUDPRXQW GLIILFXOW\ RI WKH VXEMHFW UHDOLVWLF SUREOHPV DUH WRR FRPSOLFDWHG IRU SUHFLVH DQDO\WLFDO WUHDWPHQW :KLOH SURJUHVV LV EHLQJ PDGH LQ DQDO\Wn LFDO WUHDWPHQW RI VLPSOH VWUXFWXUHV WKH DXWKRUV RSLQH WKDW UHDOLVWLF SUREOHPV UHTXLUH PDWKHPDWLFDO SURJUDPPLQJ WHFKQLTXHV +RZHYHU WKH\ GR IHHO WKDW DQDO\WLFDO WUHDWPHQW LV GHVLUDEOH WR SURYLGH D GHHSHU LQVLJKW LQWR WKH DQDO\WLFDO QDWXUH RI RSWLPDOLW\ $ UHODWHG VXUYH\ E\ :DQJ f RQ GLVWULEXWHG SDUDPHWHU V\VWHPV ZKRVH G\QDPLFDO EHKDYLRUV DUH GHVFULEDEOH E\ SDUWLDO GLIIHUHQWLDO HTXDWLRQV LQWHJUDO HTXDWLRQV RU IXQFWLRQDO GLIIHUHQWLDO HTXDWLRQV FRQVLVWV HQWLUHO\ RI D ELEOLRJUDSK\ :KLOH QRW SHUWLQHQW WR WKH GLVVHUn WDWLRQ LW LV PHQWLRQHG KHUH IRU FRPSOHWHQHVV

PAGE 15

3UDJHU f SURYLGHV DQRWKHU VXUYH\ ZKLFK LV QRW FRPSUHKHQVLYH QRU GRHV KH SUHVHQW DQ\ QHZ UHVXOWV DV FODLPHG )RU H[DPSOH 3UDJHU DQG 7D\ORU f XVHG WKH SULQFLSOH RI PLQLPXP SRWHQWLDO HQHUJ\ DQG WKH DVVXPSWLRQ WKDW VWLIIQHVV LV SURSRUWLRQDO WR VSHFLILF PDVV WR SURYH JOREDO RSWLPDOLW\ ,W FRXOG KDUGO\ EH FDOOHG D QHZ GHYHORSPHQW LQ $W WKH VDPH WLPH WKH DXWKRU GRHV SUHVHQW DQ H[FHOOHQW H[DPSOH RI D PXOWLSXUSRVH RSWLPDO GHVLJQ SUREOHP 3UDJHU DOVR WUHDWV VHJPHQW ZLVH FRQVWDQW DSSUR[LPDWLRQV DQG WKH RSWLPDO OD\RXW RI WUXVVHV 7KH ILQDO VXUYH\ SDSHU 7URLWVNLL f LV DQ XQXVXDO UHYLHZ RI PHWKRGV LQ WKH FDOFXOXV RI YDULDWLRQV :KHUHDV VRPH RI WKH HDUOLHU VXUYH\V SUHVHQW OHQJWK\ OLVWV RI UHIHUHQFHV EXW FRQWDLQ OLWWOH PHWKRGn RORJ\ RU WKHRU\ WKLV VXUYH\ LV MXVW WKH RSSRVLWH 3DUW RI ZKDW PDNHV LW XQLTXH LV WKDW WKH DXWKRU EHOLHYHG RQO\ HLJKW DUWLFOHV PHULWHG FLWDn WLRQf§DOO RI WKHP E\ 7URLWVNLL 7KLV VKRUWFRPLQJ LV PRUH WKDQ RYHUFRPH E\ D WKRURXJK FODVVLILFDWLRQ RI RSWLPDO FRQWURO SUREOHPV LQ WKH FDOFXOXV RI YDULDWLRQV 7URLWVNLL EDVHV WKH FODVVLILFDWLRQ RQ FHUWDLQ FKDUDFWHUn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

PAGE 16

+LVWRULFDO 'HYHORSPHQW 2SWLPDO &ROXPQV 7KH EHJLQQLQJ RI VWUXFWXUDO RSWLPL]DWLRQ LV JHQHUDOO\ DWWULEXWHG WR *DOLOHRnV VWXGLHV LQ RI WKH EHQGLQJ VWUHQJWK RI EHDPV $FFRUGn LQJ WR %DUQHWW f *DOLOHR FRQVLGHUHG D FRQVWDQWZLGWK FDQWLOHYHU EHDP XQGHU D WLS ORDG DV SDUW RI D VWXG\ RI VROLGV RI HTXDO UHVLVWDQFH ,Q UHTXLULQJ WKH PD[LPXP VWUHVV LQ HDFK FURVV VHFWLRQ WR EH FRQVWDQW WKURXJKRXW WKH EHDP WKH KHLJKW PXVW EH D SDUDEROLF IXQFWLRQ RI SRVLWLRQ DORQJ WKH EHDP :KLOH WKLV DSSHDUV WR EH WKH RULJLQ RI WKH ILHOG D SUREn OHP WKDW UHFHLYHG PRUH DWWHQWLRQ LV WKH EXFNOLQJ RI D FROXPQ 8VLQJ WKH QHZO\ GHYHORSHG FDOFXOXV RI YDULDWLRQV LQ /DJUDQJH DWWHPSWHG WR DSSO\ YDULDWLRQDO WHFKQLTXHV WR WKH SUREOHP RI ILQGLQJ WKDW GLVWULEXWLRQ RI D KRPRJHQHRXV PDWHULDO DORQJ WKH OHQJWK RI D FROXPQ ZKLFK PD[LPL]HV WKH EXFNOLQJ ORDG 7UXHVGHOO f UHODWHV WKDW WKURXJK DQ LQVXIILFLHQW PDWKHPDWLFDO IRUPXODWLRQ /DJUDQJH VKRZHG WKH RSWLPDO IRUP WR EH D FLUFXODU F\OLQGHU &ODXVHQ f SURYLGHV WKH HDUOLHVW NQRZQ VROXWLRQ WR WKLV SUREOHP IRU WKH VLPSO\ VXSSRUWHG FDVH $V GHVFULEHG LQ 7RGKXQWHU DQG 3HDUVRQ SS f &ODXVHQ PLQLPL]HG WKH YROXPH RI WKH FROXPQ ZLWK WKH GLIIHUHQWLDO HTXDWLRQ IRU EXFNOHG GHIOHFWLRQ WUHDWHG DV D VXEVLGLDU\ FRQGLWLRQ $VVXPLQJ DOO FURVV VHFWLRQV WR EH VLPLODU DIWHU VHYHUDO YDULDEOH WUDQVIRUPDWLRQV DQG FRPSOLFDWHG PDQLSn XODWLRQ KH REWDLQHG DQ LPSOLFLW DQDO\WLFDO VROXWLRQ 7KH QH[W GHYHORSPHQW ZDV *UHHQKLOO f DFFRUGLQJ WR .HOOHU DQG 1LRUGVRQ f *UHHQKLOO GHWHUPLQHG WKH KHLJKW RI D XQLIRUP SULVPDWLF FROXPQ EH\RQG ZKLFK WKH FROXPQ EXFNOHG GXH VROHO\ WR LWV RZQ ZHLJKW 7LPRVKHQNR DQG *HUH f UHSURGXFH WKH VROXWLRQ LQ ZKLFK WKH

PAGE 17

GHIOHFWLRQ LV H[SUHVVHG DV WKH LQWHJUDO RI D %HVVHO IXQFWLRQ RI WKH ILUVW NLQG RI WKH QHJDWLYH RQHWKLUG RUGHUf %ODVLXV f LQWURGXFHV KLV SDSHU ZLWK D XQLIRUP VWUHQJWK DQG D PLQLPXP GHIOHFWLRQ EHDP SUREOHP )RU D JLYHQ ORDG DQG DPRXQW RI PDWHn ULDO WKH FURVVVHFWLRQDO DUHD GLVWULEXWLRQ LV GHWHUPLQHG ZKLFK PD[Ln PL]HV WKH EXFNOLQJ ORDG RI D FLUFXODU FROXPQ 7KH VROXWLRQ LV LGHQWLFDO WR WKDW REWDLQHG E\ &ODXVHQ ,Q DGGLWLRQ %ODVLXV DOVR REWDLQHG VROXn WLRQV IRU FROXPQV KDYLQJ UHFWDQJXODU FURVV VHFWLRQV DQG GLVFXVVHG WKH HIIHFW RI GLIIHUHQW ERXQGDU\ FRQGLWLRQV RQ WKH UHVXOWV )RU WKH QH[W IHZ GHFDGHV VWUXFWXUDO RSWLPL]DWLRQ DSSHDUV WR KDYH EHHQ GLUHFWHG WRZDUGV DSSOLFDWLRQV LQ WKH DLUFUDIW LQGXVWU\ ZKHUH DLUFUDIW VWUXFWXUDO SUREOHPV DQG UHVXOWV DUH SUHVHQWHG LQ WKH IRUPDW RI D GHVLJQ KDQGERRN )HLJHQ f LV D JRRG H[DPSOH RI WKLV FRQVLGHUn LQJ WKH EXFNOLQJ RI D WKLQZDOO FROXPQ *LYHQ D FRQVWDQW ORDG DQG ZDOO WKLFNQHVV KH UHTXLUHG WKH YDULDEOH LQVLGH GLDPHWHU WR EH FKRVHQ VXFK WKDW WKH EXFNOLQJ ORDG LV PD[LPL]HG :DOO WKLFNQHVV LV VHOHFWHG WR PDNH ORFDO EXFNOLQJ DQG (XOHU EXFNOLQJ RFFXU DW WKH VDPH ORDG 6ROLG WDSHUHG FROXPQV KDYLQJ EOXQW HQGV DUH DOVR WUHDWHG IRU DVVXPHG VWLIIQHVV GLVn WULEXWLRQV 5HQHZHG LQWHUHVW ZDV DURXVHG E\ .HOOHU f ZKR H[DPLQHG WKH SUREOHP IURP WKH SRLQW RI YLHZ RI WKH WKHRU\ RI HODVWLFLW\ DQG LQ FKRRVLQJ WKH FURVVVHFWLRQDO VKDSH WR JLYH WKH PD[LPXP VWLIIQHVV 1HJOHFWLQJ WKH ZHLJKW RI WKH FROXPQ KH REWDLQHG YLD WKH IRUPHU WKDW WZLVWLQJ WKH FROXPQ GRHV QRW DIIHFW WKH EXFNOLQJ 2I DOO FRQYH[ FURVV VHFWLRQV WKH HTXLODWHUDO WULDQJOH LV VKRZQ WR KDYH WKH ODUJHVW VHFRQG

PAGE 18

PRPHQW RI DUHD UHODWLYH WR D FHQWURLGDO D[LV +HQFH IURP WKH GHILQn LWLRQ RI EXFNOLQJ ORDG WKH EHVW FURVVVHFWLRQDO VKDSH LV WKH HTXLn ODWHUDO WULDQJOH .HOOHU DOVR REWDLQHG &ODXVHQnV LPSOLFLW DQDO\WLFDO VROXWLRQ 6XEVHTXHQWO\ 7DGMEDNKVK DQG .HOOHU f JHQHUDOL]HG WKH SUREOHP WR D JHQHUDO HLJHQYDOXH SUREOHP DQG ERXQGDU\ FRQGLWLRQV VXEMHFW WR D VXEVLGLDU\ HTXDOLW\ FRQVWUDLQW 7KH ODWWHU FRUUHVSRQGV WR VSHFn LI\LQJ WKH YROXPH RU ZHLJKWf RI PDWHULDO WR EH GLVWULEXWHG LQ DQ RSWLn PDO PDQQHU 8VLQJ WKH +ROGHU LQHTXDOLW\ WKH\ GHPRQVWUDWH JOREDO RSWLn PDOLW\ RI WKH HLJHQYDOXH IRU WKH KLQJHGKLQJHG FROXPQ .HOOHU DQG 1LRUGVRQ f H[DPLQH WKH FDVH RI D YHUWLFDO FROXPQ IL[HG DW WKH EDVH VXEMHFW WR D YHUWLFDO ORDG DW WKH WLS DQG WKH FROXPQnV RZQ ZHLJKW ,W LV DOVR DVVXPHG WKDW DOO FURVV VHFWLRQV DUH VLPLODU 7KHLU DSSURDFK LV WR VWDWH WKH SUREOHP DV D VLPXOWDQHRXV GXDO RSWLPL]Dn WLRQ RI WKH 5D\OHLJK TXRWLHQW 7KH HLJHQYDOXH LV PLQLPL]HG ZLWK UHVSHFW WR WKH HLJHQIXQFWLRQ DQG PD[LPL]HG ZLWK UHVSHFW WR FURVVVHFWLRQDO DUHD GLVWULEXWLRQ 6SHFLI\LQJ WKH YROXPH RI PDWHULDO DYDLODEOH LV WUHDWHG DV D VXEVLGLDU\ HTXDOLW\ FRQVWUDLQW )URP WKH PD[LPXP ORZHVW HLJHQYDOXH WKH PD[LPXP KHLJKW DW EXFNOLQJ LV GHWHUPLQHG 6ROXWLRQV DUH REWDLQHG E\ DQ LWHUDWLYH WHFKQLTXH HPSOR\HG ZLWK LQWHJUDO HTXDWLRQV ,Q D EULHI QRWH 7D\ORU f VXJJHVWV WKDW HQHUJ\ PHWKRGV PD\ OLQN RSWLPXP FROXPQ SUREOHPV WR WKH WUDGLWLRQDO HLJHQYDOXH SUREOHPV RI PHFKDQLFV 3UDJHU DQG 7D\ORU f FODVVLI\ SUREOHPV LQ RSWLPDO VWUXFn WXUDO GHVLJQ DQG GHPRQVWUDWH JOREDO RSWLPDOLW\ XVLQJ HQHUJ\ SULQFLSOHV 8QIRUWXQDWHO\ WKHLU DVVXPSWLRQ RI WKLQZDOO FRQVWUXFWLRQ OLPLWV WKH UHVXOWV WR VWUXFWXUHV ZKHUH WKH VWLIIQHVV LV SURSRUWLRQDO WR WKH VSHFLILF

PAGE 19

PDVV GHQVLW\ 7KH FRQVHTXHQFH RI WKLV DVVXPSWLRQ LV WKDW LQ WKH HQHUJ\ IRUPXODWLRQ WKH UHVXOWLQJ FRQWURO ODZ DQG JRYHUQLQJ HTXDWLRQV DUH GHFRXSOHG DQG KHQFH HDVLO\ VROYHG +XDQJ DQG 6KHX f DSSO\ WKLV VDPH WKLQZDOO DVVXPSWLRQ WR WKH SUREOHP WUHDWHG E\ .HOOHU DQG 1LRUGVRQ +RZHYHU WKH IRUPHU VHHN WKH PD[LPXP HQG ORDG LQVWHDG RI WKH PD[LPXP KHLJKW 7KH DXWKRUV DOVR DWWHPSW WR OLPLW WKH PD[LPXP DOORZDEOH VWUHVV DQG REWDLQ VROXWLRQV E\ D ILQLWHGLIIHUHQFH PHWKRG )XUWKHU GLVFXVVLRQ RI VDQGZLFK WKLQZDOO FRQVWUXFWLRQf FROXPQV LV JLYHQ E\ 7D\ORU DQG /LX f %DVLFDOO\ WKLV SDSHU LV DQ HODERUDWLRQ RI WKH WHFKQLTXHV GHVFULEHG E\ 3UDJHU DQG 7D\ORU ZKHQ DSSOLHG WR FROXPQV ([WHQVLYH UHVXOWV DUH SURYLGHG IRU YDULRXV FDVHV 3RVWEXFNOLQJ EHKDYLRU IRU FROXPQV VXEMHFW WR FRQVHUYDWLYH ORDGV LV FRQVLGHUHG E\ *DMHZVNL DQG =\F]NRZVNL f $ QRQFRQFHUYD WLYH SUREOHP LV WUHDWHG E\ 3ODXW Ef 7KH ILUVW RI WKHVH WZR SDSHUV LV OHQJWK\ EXW LV PXFK WRR QDUURZ LQ VFRSH WR EH SDUWLFXODUO\ XVHIXO ,Q WKH VHFRQG SDSHU WKH 5LW] PHWKRG LV DSSOLHG WR DQ HQHUJ\ IXQFWLRQDO REWDLQLQJ WKH EHVW IRUP RI WKH DVVXPHG DSSUR[LPDWLRQ WR WKH RSWLPDO VROXWLRQ +LVWRULFDO 'HYHORSPHQW 2SWLPDO 6WDWLF %HDPV 7KDW EHDP SUREOHPV SOD\HG D UROH LQ WKH HDUO\ GHYHORSPHQWV RI VWUXFWXUDO RSWLPL]DWLRQ KDV DOUHDG\ EHHQ LQGLFDWHG LQ WKH SUHFHGLQJ VHFn WLRQ 1R DWWHPSW LV PDGH LQ ZKDW IROORZV WR SUHVHQW D FRPSOHWH KLVWRU\ EXW PHUHO\ WR RXWOLQH WKH W\SH RI SUREOHPV WKDW KDYH EHHQ FRQVLGHUHG

PAGE 20

2SDWRZVNL f KDV DQ RXWVWDQGLQJ SDSHU WKDW GHDOV ZLWK FDQWLOHYHU EHDPV RI XQLIRUP VWUHQJWK %HVLGHV SURYLGLQJ QXPHURXV UHIHUn HQFHV WR HDUOLHU VWXGLHV WKH DXWKRU WUHDWV WKH SUREOHP ZLWK LPSUHVn VLYH PDWKHPDWLFDO ULJRU 7KH EHDP LV FRQVLGHUHG WR GHIOHFW XQGHU LWV RZQ ZHLJKW DQG D WUDQVYHUVH WLS ORDG EHQGLQJ GHIOHFWLRQ LV GHVFULEHG E\ D 9ROWHUUD LQWHJUDO HTXDWLRQ ZKLFK LV VROYHG H[DFWO\ IRU YDULRXV DVVXPHG FODVVHV RI YDULDEOH FURVVVHFWLRQDO JHRPHWU\ 7KLV SDSHU LV UHSUHVHQWDWLYH RI HDUOLHU SDSHUV LQ WKDW LW FRQWDLQV H[WHQVLYH DQDO\VLV DQG DQDO\WLFDO UHVXOWV EXW OLWWOH QXPHULFDO GDWD %DUQHWWnV ZRUN f DQG LWV VHTXDO Df DSSO\ WKH FDOFXOXV RI YDULDWLRQV WR PRUH UHDOn LVWLF ,EHDPV 2QH SUREOHP FRQVLGHUHG LV PD[LPL]LQJ WKH ZHLJKW VXEn MHFW WR JHQHUDO XQVSHFLILHG ORDGV IRU D VSHFLILHG GHIOHFWLRQ DW D JLYHQ SRLQW 7KH 6FKZDU] LQHTXDOLW\ LV XVHG WR GHULYH D VXIILFLHQW FRQGLWLRQ IRU JOREDO RSWLPDOLW\ $OVR LQFOXGHG LV D FRPSDULVRQ RI XQLIRUP VWUHQJWK EHDPV WR WKH PLQLPXP GHIOHFWLRQ EHDP IRU VHYHUDO GLIIHUHQW FDVHV RI DSSOLHG ORDG DQG JHRPHWU\ 7KH SDSHU LV FRQFOXGHG ZLWK YDULRXV PLQLPXP ZHLJKW GHVLJQ H[DPSOHV LQ ZKLFK ERWK EHQGLQJ DQG VKHDU VWLIIQHVV DUH FRQn VLGHUHG +DXJ DQG .LUPVHUnV f ZRUN LV RQH RI WKH PRVW FRPSUHKHQVLYH VWXGLHV RI PLQLPXP ZHLJKW EHDP SUREOHP SXEOLVKHG :KLOH LW PD\ VXFFHHG LQ KDQGOLQJ DQ\ FRQFHLYDEOH SUREOHP DQG LQ HPSOR\LQJ WKH PRVW UHDOLVWLF VWUHVV FRQVWUDLQWV WKLV YHU\ JHQHUDOLW\n UHTXLUHV VR PDQ\ YDULDEOHV DQG FRQGLWLRQV WKDW WKH PDWKHPDWLFV LV FRPSOLFDWHG DOPRVW EH\RQG UHDVRQ $QRWKHU VWXG\ RI PLQLPXP ZHLJKW EHDPV +XDQJ DQG 7DQJ f LV LPSRUWDQW IRU VHYHUDO UHDVRQV %\ GLYLGLQJ WKH EHDP LQWR PDQ\ VHJPHQWV

PAGE 21

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fV 0D[LPXP 3ULQFLSOH :KLOH QR QHZ UHVXOWV RU WHFKQLTXHV DUH FRQWDLQHG LQ &LWURQnV SS f WKH DXWKRU JLYHV D YHU\ UHDGDEOH PLQLPXP ZHLJKW EHDP H[DPSOH 7KH SUREOHP LV VLPSOH GHVFULEHG LQ GHWDLO FRPSOHWHO\ DQG SURYLGHV DQ H[FHOOHQW H[DPSOH RI KRZ FRQWURO WKHRU\ LV DSSOLHG WR DQ RSWLPDO SUREOHP LQ VWUXFWXUDO GHVLJQ $Q DQDO\VLV RI LQWHUPHGLDWH EHDPV ZKLFK PD\ IRUP SODVWLF KLQJHV LV SURYLGHG E\ *MHOVYLN f ,W VKRZV WKDW LI KLQJHV DUH SODFHG DW DOO SRLQWV RI WKH EHDP ZKHUH WKH EHQGLQJ PRPHQW LV ]HUR WKLV PDNHV WKH SODVWLF RU HODVWLF PLQLPXP ZHLJKW EHDP VWDWLFDOO\ GHWHUPLQDWH %RWK WKH HODVWLF DQG SODVWLF EHDP GHVLJQV DUH VKRZQ WR EH IXOO\ VWUHVVHG LH DUH XQLIRUP VWUHQJWK EHDPV $SSOLFDWLRQ RI JHQHUDOL]HG YHFWRU VSDFH WHFKQLTXHV LV FKDUDFWHUn LVWLF RI PRUH UHFHQW SDSHUV %KDUJDYD DQG 'XIILQnV f LV VXFK D VWXG\ ,W WUHDWV WKH PD[LPXP VWUHQJWK RI D FDQWLOHYHU EHDP RQ DQ HODVWLF 6LQFH WKHVH HDUO\ SDSHUV DUH QRW UHDGLO\ DYDLODEOH WKLV VWDWHPHQW LV EDVHG XSRQ GHVFULSWLRQV RI WKHP JLYHQ E\ KLVWRULFDO UHIHUHQFHV FLWHG LQ $SSHQGL[ $

PAGE 22

IRXQGDWLRQ VXEMHFW WR DQ XSSHU ERXQG RQ ZHLJKW $OWKRXJK LQYROYLQJ PRUH DGYDQFHG PDWKHPDWLFV WKDQ QRUPDOO\ UHTXLUHG E\ YDULDWLRQDO WHFKn QLTXHV YHFWRU VSDFH PHWKRGV PD\ DOVR SURYLGH PRUH SRZHUIXO DQDO\Wn LFDO WRROV +LVWRULFDO 'HYHORSPHQW 2SWLPDO '\QDPLFDO %HDPV 7KH WLWOH RI WKLV VHFWLRQ LV D PLVQRPHU 2Q WKH EDVLV RI WKH UHODWHG OLWHUDWXUH D PRUH UHDOLVWLF WHUPLQRORJ\ ZRXOG EH 2SWLPDO 2XDVLVWDWLF %HDPV 3DSHUV GHDOLQJ ZLWK WKH RSWLPL]DWLRQ RI G\QDPLFDO EHDPV XOWLPDWHO\ FRQWDLQ VRPH DVVXPSWLRQ RU JLYHQ FRQGLWLRQ WKDW HIIHFn WLYHO\ WUDQVIRUPV WKH SUREOHP WR DQ HTXLYDOHQW VWDWLF FDVH 6LPSOH KDUPRQLF PRWLRQ LV IUHTXHQWO\ DVVXPHG WR UHPRYH WLPH GHSHQGHQFH IURP WKH JRYHUQLQJ HTXDWLRQ )RU H[DPSOH %DUQHWW Ef PLQLPL]HV WKH WLS RI GHIOHFWLRQ RI D FDQWLOHYHU EHDP DFFHOHUDWLQJ XQLIRUPO\ XSZDUGV ZLWK WKH WRWDO ZHLJKW RI WKH EHDP VSHFLILHG 7KH RSWLPDO VROXWLRQ IRU FURVV VHFWLRQDO DUHD GLVWULEXWLRQ LV VSHFLILHG E\ D QRQOLQHDU LQWHJUDO HTXDn WLRQ VROYHG E\ VXFFHVVLYH DSSUR[LPDWLRQV '\QDPLFV HQWHUV WKH SUREOHP RQO\ DV D WLPH LQYDULDQW LQHUWLD ORDG ZKLFK FRQYHUWV WKH SUREOHP WR D VWDWLF EHDP VXEMHFW WR D ERG\ IRUFH 1LRUGVRQ f ILQGV WKH WDSHULQJ RI D VLPSO\ VXSSRUWHG EHDP RI JLYHQ YROXPH DQG OHQJWK ZKLFK PD[LPL]HV WKH IXQGDPHQWDO IUHTXHQF\ RI IUHH YLEUDWLRQ $VVXPLQJ WKDW DOO FURVV VHFWLRQV DUH VLPLODU 1LRUGVRQ H[SUHVVHV WKH GHVLUHG IUHTXHQF\ DV WKH 5D\OHLJK TXRWLHQW 7KLV LV REWDLQHG E\ WKH HTXDWLRQ IRU VSDWLDO GHSHQGHQFH DVVRFLDWHG ZLWK WKH XVXDO VHSDUDWLRQ RI YDULDEOHV WHFKQLTXH ,Q WKLV FDVH LW LV DVVXPHG

PAGE 23

WKDW GHIOHFWLRQ VKHDU DQG URWDWLRQDO LQHUWLD HIIHFWV DUH VPDOO 6ROXWLRQ RI WKH FRQGLWLRQV IRU RSWLPDOLW\ ZHUH REWDLQHG E\ VXFFHVVLYH DSSUR[LPDWLRQV 7KLV DSSURDFK UHVXOWV LQ D SUREOHP LGHQWLFDO LQ IRUP WR WKH HLJHQYDOXH SUREOHP DVVRFLDWHG ZLWK PD[LPL]LQJ WKH (XOHU EXFNOLQJ ORDG RI D FROXPQ $ YHU\ VSHFLDOL]HG SUREOHP LV WUHDWHG LQ %UDFK f &URVV VHFWLRQDO DUHD DQG VWLIIQHVV DUH SURSRUWLRQDO DQG KDYH ERWK XSSHU DQG ORZHG ERXQGV PDWHULDO SURSHUWLHV DQG OHQJWK RI WKH EHDP DUH VSHFLILHG FRQVWDQWV 7KH REMHFW LV WR PDNH WKH IXQGDPHQWDO IUHTXHQF\ RI IUHH YLEUDWLRQ VWDWLRQDU\ ZLWKRXW DQ\ FRQVWUDLQWV RQ WKH ZHLJKW ,QVWHDG RI XVLQJ WKH 5D\OHLJK TXRWLHQW %UDFK XVHV WKH WRWDO SRWHQWLDO HQHUJ\ +LV VROXWLRQ PHWKRG LV DG KRF DQG QRW JHQHUDOO\ XVHIXO $ PRUH XVHIXO DSSURDFK LV GHVFULEHG E\ ,FHUPDQ f IRU VWUXFWXUHV VXEMHFW WR SHUn LRGLF ORDGV 1HFHVVDU\ DQG VXIILFLHQW FRQGLWLRQV IRU PLQLPXP ZHLJKW DUH REWDLQHG IURP WKH SULQFLSOH RI PLQLPXP SRWHQWLDO HQHUJ\ $PSOLWXGH DQG IUHTXHQF\ RI WKH DSSOLHG ORDG LV VSHFLILHG DV ZHOO DV G\QDPLF UHVSRQVH GHILQHG DV WKH SRWHQWLDO HQHUJ\ DVVRFLDWHG ZLWK WKH ORDG DPSOLWXGH GLVSODFHG D GLVWDQFHVTXDO WR WKH GLVSODFHPHQW DPSOLWXGH DW WKH SRLQW RI DSSOLFDWLRQ ,W LV DOVR UHTXLUHG WKDW WKH ORDGnV IUHTXHQF\ EH OHVV WKDQ WKH IXQGDPHQWDO QDWXUDO IUHTXHQF\ 6XEMHFW WR WKHVH FRQn VWUDLQWV WKH VWUXFWXUHnV ZHLJKW LV WR EH PLQLPL]HG 7UXVVHV DQG VHJPHQWZLVH FRQVWDQW DSSUR[LPDWLRQV DUH DOVR WUHDWHG 2QFH DJDLQ D G\QDPLFDO SUREOHP LV HIIHFWLYHO\ WUDQVIRUPHG LQWR D TXDVLVWDWLF SUREOHP 5HDOLVWLF WUHDWPHQW RI WKH G\QDPLF UHVSRQVH RI EHDPV VXEMHFW WR WLPH GHSHQGHQW ORDGV LV JLYHQ LQ 3ODXW f $Q XSSHU ERXQG RQ WKH

PAGE 24

UHVSRQVH RI WKH EHDP LV REWDLQHG IURP WKH ODUJHVW SRVVLEOH GLVSODFHn PHQW RI WKH EHDP XQGHU D VWDWLF FRQFHQWUDWHG XQLW ORDG 7KLV LQHTXDOn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f 7KH\ PD[LPL]H WKH IXQGDPHQWDO QDWXUDO IUHTXHQF\ ZKLFK LV H[SUHVVHG DV D 5D\OHLJK TXRWLHQW WKDW LQFOXGHV WKH HIIHFW RI VKHDU 6WDQGDUG YDULDn WLRQDO PHWKRGV DUH HPSOR\HG WR GHULYH QHFHVVDU\ FRQGLWLRQV EXW QR VROXWLRQV QRU H[DPSOHV DUH JLYHQ 7KH DXWKRUV GR KRZHYHU VXJJHVW XVLQJ WKH PHWKRG RI TXDVLOLQHDUL]DWLRQ 7KLV SDSHU LV DQRWKHU H[DPSOH RI D TXDVLVWDWLF G\QDPLFDO EHDP SUREOHP 6FRSH RI WKH 'LVVHUWDWLRQ 7KLV GLVVHUWDWLRQ LV SULPDULO\ FRQFHUQHG ZLWK WKH DSSOLFDWLRQ RI 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH WR SUREOHPV LQ VWUXFWXUDO RSWLPL]DWLRQ 2QO\ HODVWLF PDWHULDOV DUH FRQVLGHUHG KRZHYHU YDULRXV W\SHV RI FRQn VWUDLQWV DUH WUHDWHG 7KH WKHRUHWLFDO GHYHORSPHQW RI WKH GLVVHUWDWLRQ SHUWDLQV WR SUREOHPV GHVFULEHG E\ DQ RUGLQDU\ GLIIHUHQWLDO HTXDWLRQ EXW LV EDVHG XSRQ D QXPHULFDO WHFKQLTXH QRUPDOO\ XVHG ZLWK V\VWHPV GHVFULEHG E\

PAGE 25

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n ULWKP LV EDVHG WR WKH WKHRUHWLFDO GHYHORSPHQW DVVRFLDWHG ZLWK FRQWLQn XRXV V\VWHPV )XUWKHUPRUH LW SURYLGHV D FRPSDULVRQ EHWZHHQ WKH VROXWLRQ WR D SUREOHP GHVFULEHG DV D FRQWLQXRXV V\VWHP DQG DOWHUQDWHO\ E\ D GLVFUHWH DSSUR[LPDWLRQ $GGLWLRQDOO\ LW LV WKH LQWHQW RI WKLV GLVVHUWDWLRQ WR UHSODFH VRPH RI WKH FRQIXVLRQ LQ FODVVLILFDWLRQ RI SUREOHP W\SHV FRQWDLQHG LQ WKH YDULRXV VXUYH\ SDSHUV ZLWK DQ RUJDQL]DWLRQ EDVHG XSRQ PDWKHPDWLFDO DWWULEXWHV 7KH UHVXOW LV D ORJLFDO DSSURDFK WR WKH IRUPXODWLRQ RI RSWLPL]DWLRQ SUREOHPV IRU HODVWLF VWUXFWXUHV

PAGE 26

&+$37(5 ,, *(1(5$/ 352%/(06 $1' 0(7+2'6 ,1 6758&785$/ 237,0,=$7,21 ,QWURGXFWLRQ 7KH ILUVW FKDSWHU SUHVHQWV D EURDG YLHZ RI VWUXFWXUDO RSWLPL]Dn WLRQ DQG WKH KLVWRULFDO GHYHORSPHQW RI WZR JHQHUDO W\SHV RI SUREOHPV WKDW DUH XVHG DV H[DPSOHV LQ ODWHU FKDSWHUV %HIRUH WKH PDWKHPDWLFDO WKHRU\ LV GHYHORSHG LQ WKH QH[W FKDSWHUV JHQHUDO SUREOHPV DQG PHWKRGV LQ VWUXFWXUDO RSWLPL]DWLRQ LWVHOI DUH EULHIO\ RXWOLQHG LQ WKLV FKDSWHU 7KH FODVVLILFDWLRQ RI SUREOHP W\SHV YLVKYLV PDWKHPDWLFDO DWWULEXWHV LV GLVFXVVHG ILUVW 7KLV LV IROORZHG E\ VKRUW GHVFULSWLRQV RI WKH PDMRU PHWKRGV RI VWUXFWXUDO RSWLPL]DWLRQ IRU ERWK FRQWLQXRXV DQG GLVn FUHWH V\VWHPV $SSURSULDWH UHIHUHQFHV DUH FLWHG IRU HDFK VHFWLRQ 3UREOHP &ODVVLILFDWLRQ &ULWHULD 3HUKDSV WKH PDMRU VRXUFH RI GLIILFXOW\ LQ FODVVLI\LQJ VWUXFn WXUDO RSWLPL]DWLRQ SUREOHPV OLHV LQ WKH WUDQVODWLRQ IURP WKH SK\VLFV LQYROYHG WR D PDWKHPDWLFDO UHSUHVHQWDWLRQ $ VLQJOH SK\VLFDO FRQFHSW ZKHQ WUDQVIRUPHG WR PDWKHPDWLFV PD\ EHFRPH PRUH WKDQ RQH PDWKHPDWLFDO DWWULEXWH )RU H[DPSOH FRQVLGHU WKH FODVV RI FRQVHUYDWLYH SUREOHPV 6LQFH HQHUJ\ LV FRQVHUYHG WKLV LPPHGLDWHO\ SURKLELWV GLVVLSDWLYH PDWHn ULDOV WLPH YDULHQW FRQVWUDLQWV DQG QRQKRORQRPLF FRQVWDQWV 0RUH LPSRUWDQW FRQVLGHU WKH IROORZLQJ VWDWHPHQWV IURP /DQF]RV 3 f

PAGE 27

DOO WKH HTXDWLRQV RI PDWKHPDWLFDO SK\VLFV ZKLFK GR QRW LQYROYH DQ\ HQHUJ\ ORVVHV DUH GHGXFLEOH IURP D SULQFLSOH RI OHDVW DFWLRQ WKDW LV WKH SULQFLSOH RI PDNLQJ D FHUWDLQ VFDODU TXDQWLW\ D PLQLPXP RU D PD[LPXP DOO WKH GLIIHUHQWLDO HTXDWLRQV ZKLFK DUH VHOIDGMRLQW DUH GHGXFLEOH IURP D PLQLPXPPD[LPXP SULQFLSOH DQG YLFH YHUVD +RZHYHU LW LV VKRZQ LQ &KDSWHU 9 WKDW WR EH VHOIDGMRLQW V\VWHPV SODFHV UHTXLUHPHQWV RQ ERWK WKH GLIIHUHQWLDO HTXDWLRQ DQG WKH ERXQGDU\ FRQGLn WLRQ 7KXV WKH VLQJOH SK\VLFV DWWULEXWH RI EHLQJ D FRQVHUYDWLYH V\Vn WHP LV GHVFULEHG PDWKHPDWLFDOO\ E\ H[SUHVVLRQV LQYROYLQJ WKH FRVW IXQFn WLRQDO WKH JRYHUQLQJ V\VWHP RI GLIIHUHQWLDO HTXDWLRQV ERXQGDU\ FRQGLn WLRQV DQG FRQVWUDLQWV $V D UHVXOW RI WKLV ODFN RI VLPLODULW\ LQ GHVFULSWLRQV D FKRLFH PXVW EH PDGH DV WR ZKLFK UHDOP ZLOO EH XVHG IRU FODVVLILFDWLRQ RI SUREn OHPV 6LQFH DOO SUREOHPV DUH XOWLPDWHO\ WUDQVIRUPHG WR PDWKHPDWLFV PDWKHPDWLFDO FKDUDFWHULVWLFV DUH VHOHFWHG DV WKH FULWHULD 2Q WKH EDVLV RI VXUYH\ SDSHU FRQWHQWV DQG WKH PDQ\ UHODWHG SDSHUV LW LV IHOW WKDW WKH SURSHU QRW QHFHVVDULO\ WKH EHVW QRU DOO LQFOXVLYHf FKDUDFWHULVWLFV IRU FODVVLILFDWLRQ RI SUREOHP W\SHV DUH LfFRVW IXQFWLRQDO LLfV\VWHP HTXDWLRQ DQG ERXQGDU\ FRQGLWLRQV LLLfFRQWURO FRQVWUDLQWV LYfEHKDYLRUDO FRQVWUDLQWV VWDWH DQGRU FRQWURO FRQVWUDLQWVf 6XEVHTXHQW GLVFXVVLRQ LV LQ WHUPV RI WKHVH IRXU FKDUDFWHULVWLFV ([FHSWLRQV WR WKHVH GHVFULSWRUV DUH UHDGLO\ DFNQRZOHGJHG HJ ZKHWKHU WKH SUREOHP LV GHWHUPLQLVWLF RU SUREDELOLVWLF $Q H[DPSOH RI WKH ODWWHU PD\ EH VHHQ LQ 0RVHV DQG .LQVHU f 7KHVH H[FHSWLRQV GR QRW VHUYH

PAGE 28

DV QHJDWLQJ FRXQWHUH[DPSOHV EXW LQVWHDG LQGLFDWH WKH UHTXLUHPHQW IRU DGGLWLRQDO GHVFULSWRUV DQG YHULI\ WKH GLIILFXOW\ RI WKH WDVN VXJJHVWn LQJ WKH QHHG IRU IXUWKHU FRPSUHKHQVLYH VWXG\ 3UREOHP &ODVVLILFDWLRQ *XLGHOLQHV ,Q WKH IROORZLQJ VHFWLRQV HDFK FULWHULRQ LV EULHIO\ GLVFXVVHG 6RPH RI WKH YDULRXV FKDUDFWHULVWLFV RI HDFK DUH PHQWLRQHG DQG ZKHUH DSSURSULDWH UHIHUHQFHV H[LVW WKH FLWDWLRQ LV JLYHQ ,W PXVW DJDLQ EH HPSKDVL]HG WKDW WKH IROORZLQJ LV QRW DOO LQFOXVLYH LW LV DQ DWWHPSW WR FDWHJRUL]H WKH W\SHV RI SUREOHPV H[LVWLQJ LQ WKH OLWHUDWXUH DFFRUGLQJ WR WKH IRXU PDWKHPDWLFDO GHVFULSWRUV SRVWXODWHG 0RUHRYHU WKH GHVFULSn WRUV DUH QRW GLVFXVVHG LQ WKH RUGHU JLYHQ EXW LQVWHDG DUH WUHDWHG LQ WKH RUGHU QRUPDOO\ HQFRXQWHUHG GXULQJ D SUREOHP VROXWLRQ *RYHUQLQJ (TXDWLRQV RI WKH 6\VWHP 7KH LPPHGLDWH TXHVWLRQ WR EH DQVZHUHG LV ZKHWKHU WKH VWUXFWXUDO V\VWHP LV GHVFULEHG E\ D VHW RI FRQWLQXRXV IXQFWLRQV RU D VHW RI GLVn FUHWH FRQVWDQWV %HQGLQJ GHIOHFWLRQ RI VLPSOH VWUXFWXUDO HOHPHQWV LV DQ H[DPSOH RI WKH IRUPHU GHVLJQ RI WUXVVHV LV D JRRG H[DPSOH RI WKH ODWWHU ,Q JHQHUDO YDULDWLRQDO WHFKQLTXHV DUH HPSOR\HG ZLWK FRQWLQn XRXV V\VWHPV ZKLOH PDWKHPDWLFDO SURJUDPPLQJ WHFKQLTXHV DUH PRVW IUHn TXHQWO\ DSSOLHG WR WKH GLVFUHWH V\VWHPV +RZHYHU YDULDWLRQDO PHWKRGV FDQ EH DSSOLHG WR WKH DSSUR[LPDWLRQ RI FRQWLQXRXV V\VWHPV E\ GLVFUHWH HOHPHQWV 7KLV XVXDOO\ WDNHV WKH IRUP RI HLWKHU D ILQLWH HOHPHQW RU VHJPHQWZLVH FRQVWDQW DSSUR[LPDWLRQ RI WKH FRQWLQXRXV VWUXFWXUH 5HIHUHQFHV WUHDWLQJ WKH GLIIHUHQW V\VWHPV GHVFULEHG DERYH DUH LQFOXGHG LQ WKH VHFWLRQ RQ PHWKRGV

PAGE 29

&RQVWUDLQWV 7ZR RI WKH GHVFULSWRUV DUH SRVWXODWHG WR EH FRQWURO FRQVWUDLQWV DQG EHKDYLRUDO FRQVWUDLQWV $ IXUWKHU FRQVLGHUDWLRQ LV ZKHWKHU WKH FRQVWUDLQW LV GHILQHG E\ DQ HTXDOLW\ UHODWLRQVKLS RU E\ DQ LQHTXDOLW\ H[SUHVVLRQ (TXDOLW\ FRQVWUDLQWV DUH KDQGOHG E\ D ORQJNQRZQ WHFKQLTXH HQWLWOHG LVRSHULPHWULF FRQVWUDLQWV 9DOHQWLQHnV f ZRUN LV NQRZQ WR FRQWDLQ WKH LQLWLDO GHYHORSPHQW RI D WHFKQLTXH IRU FRQYHUWLQJ LQHTXDOn LW\ FRQVWUDLQWV WR HTXDOLW\ FRQVWUDLQWV 7KH LQWURGXFWLRQ RI VODFN YDULDEOHV LQFUHDVHV WKH QXPEHU RI YDULDEOHV LQ WKH SUREOHP WR EH VROYHG EXW DW WKH VDPH WLPH SHUPLWV DOO RI WKH LVRSHULPHWULF WHFKQLTXHV WR EH XVHG $ GHWDLOHG DSSOLFDWLRQ RI WKLV DSSURDFK LV SUHVHQWHG LQ $SSHQGL[ % ,W VKRXOG DOVR EH QRWHG WKDW LVRSHULPHWULF FRQVWUDLQWV DUH VRPHWLPHV UHIHUUHG WR DV DFFHVVRU\ RU VXEVLGLDU\ FRQGLWLRQV 0RVW UHDO VWUXFWXUDO RSWLPL]DWLRQ SUREOHPV SRVVHVV DQ LVRSHULn PHWULF FRQVWUDLQW DV ZHOO DV LQHTXDOLW\ FRQVWUDLQWV GHSHQGHQW XSRQ WKH FRQWURO X DQGRU WKH VWDWH [ RI WKH VWUXFWXUDO V\VWHP 7\SLFDOO\ WKH FRQWURO FRQVWUDLQWV DUH WKH UHVXOW RI JHRPHWULFDO OLPLWDWLRQV RU UHVWULFWLRQV WR WKH W\SHV RI DYDLODEOH PDWHULDOV %HKDYLRUDO FRQVWUDLQWV DUH UHODWHG WR WKH VWDWH RI WKH V\VWHP DQG PD\ GHSHQG VROHO\ XSRQ WKH VWDWH RI GHIRUPDWLRQf RU LQ WKH FDVH RI PRVW VWUHVV FRQVWUDLQWV MRLQWO\ XSRQ WKH VWDWH DQG FRQWURO YDULDEOHV :LWK WKLV GLVWLQFWLRQ WKH FRQVWUDLQWV PD\ EH FODVVLILHG YLV£YLV WKH WZR FULWHULD DQG RSWLPDO FRQWURO FKDUDFWHULVWLFV DV IROORZV %\ FRQYHQWLRQ DOO YHFWRUV DUH FROXPQ YHFWRUV XQOHVV LQGLFDWHG RWKHUZLVH

PAGE 30

Lf XQFRQV WUDLQHG LLf R 9, f§V LLLf [ Xf LYf [f FRQWURO FRQVWUDLQW EHKDYLRUDO FRQVWUDLQWV $OO RI WKLV GLVFXVVLRQ SHUWDLQV WR ERWK FRQWLQXRXV DQG GLVFUHWH V\VWHPV 1R UHIHUHQFHV IRU XQFRQVWUDLQHG RSWLPL]DWLRQ SUREOHPV DUH JLYHQ ,W PD\ EH WKDW LQ VRPH SUREOHPV WKH XQFRQVWUDLQHG VWUXFWXUDO RSWLPL]Dn WLRQ VROXWLRQV KDYH HLWKHU LQILQLWH VWLIIQHVV DQG ILQLWH ZHLJKW RU ILQLWH VWLIIQHVV DQG ]HUR ZHLJKW $ GLVFXVVLRQ RI WKLV FDQ EH IRXQG LQ 6DOLQDV SS f ,QYHVWLJDWLRQ RI FRQWURO FRQVWUDLQWV OHG WR WKH GHYHORSPHQW RI WKH PD[LPXP SULQFLSOH $OWKRXJK UHVHUYLQJ GLVFXVVLRQ RI WKH PHWKRG IRU D ODWHU VHFWLRQ WKH FODVVLFDO UHIHUHQFH GHWDLOLQJ WKH GHULYDWLRQ RI WKH SULQFLSOH LV JLYHQ KHUH IRU FRPSOHWHQHVV 5R]RQRU f WUHDWV FRQn WURO FRQVWUDLQWV EXW RQO\ DV UHODWHG WR WKH GHYHORSPHQW RI WKH PD[LPXP SULQFLSOH 0RVW RI WKH OLWHUDWXUH FRQFHUQV HLWKHU ERXQGHG FRQWURO SUREOHPV RU D PRUH JHQHUDO IRUP RI FRQVWUDLQW ZKLFK FDQ EH FODVVLILHG DV D EHKDYn LRUDO FRQVWUDLQW 7KH ODWWHU LV D PL[HG FRQVWUDLQW ZKLFK GHSHQGV XSRQ ERWK WKH VWDWH DQG FRQWURO YDULDEOHV %UHDNZHOO f LV D OXFLG SDSHU GHDOLQJ ZLWK WKLV W\SH RI FRQVWUDLQW 5HIHUHQFHV ZKLFK WUHDW VWDWH DQGRU PL[HG FRQVWUDLQWV DUH %U\VRQ HW DO f DQG 6SH\HU DQG %U\VRQ f &RQVWUDLQWV ZKLFK GHSHQG XSRQ RQO\ WKH VWDWH DUH QRW WUHDWHG LQ WKLV GLVVHUWDWLRQ DQ H[DPSOH RI VXFK D FRQVWUDLQW LV WR GHWHUPLQH WKH RSWLPDO VROXWLRQ IRU VRPH SUREOHP VXEMHFW WR DQ XSSHU ERXQG RQ GHIOHFWLRQ RI WKH VWUXFWXUH DW DQ\ SRLQW

PAGE 31

&RVW )XQFWLRQDOV 7KHUH DUH WZR EDVLF W\SHV RI FRVW IXQFWLRQDOV WKDW RFFXU LQ WKH ILHOG RI VWUXFWXUDO RSWLPL]DWLRQ 7KH\ ZHUH ILUVW LGHQWLILHG E\ 3UDJHU f EXW QRW IRU WKH SURSHU UHDVRQV 8VLQJ 3UDJHUnV QRWDWLRQ WKH\ DUH 0LQ )LSf G9 9 *: G9 9 -4 0LQ +: G9 9 ZKHUH ) DQG + DUH VFDODU IXQFWLRQDOV RI LIL 7KH ODWWHU IXQFWLRQDO UHSUHVHQWV D 5D\OHLJK TXRWLHQW DVVRFLDWHG ZLWK DQ HLJHQYDOXH SUREOHP ,W FDQ EH UHGXFHG WR WKH ILUVW W\SH RI IXQFWLRQDO VKRZQ DERYH E\ FKRRVn LQJ D QRUPDOL]DWLRQ RI WKH HLJHQIXQFWLRQ VXFK WKDW WKH QXPHUDWRU HTXDOV XQLW\ IRU DOO DGPLVVLEOH YDULDWLRQV VHH VHFWLRQ f 7KLV QRUPDOL]DWLRQ LV WKHUHDIWHU WUHDWHG DV D VXEVLGLDU\f FRQVWUDLQW :KDW DFWXDOO\ GLVWLQJXLVKHV WKH VHFRQG IXQFWLRQDO IURP WKH ILUVW LV QRW WKDW WKH IXQFWLRQDO LV D TXRWLHQW EXW WKDW WKH H[WUHP L]DWLRQ RI DQ HLJHQYDOXH UHTXLUHV D GXDO H[WUHPL]DWLRQ VHH VHFWLRQ f ,Q WHUPV RI D VWDWH [ DQG FRQWURO X WKH IXQGDPHQWDO HLJHQYDOXH LV JLYHQ E\ PLQLPL]DWLRQ RI WKH 5D\OHLJK TXRWLHQW ZLWK UHVSHFW WR WKH HLJHQIXQFWLRQ RU 0LQ *[BXf G9 9 A +[Xf G9 9 ZKHUH LL UHSUHVHQWV VRPH VSHFLILHG GHVLJQ SDUDPHWHU ,I WKH GHVLUHG UHVXOW LV WR PD[LPL]H WKH FRVW ZLWK UHVSHFW WR DOO DGPLVVLEOH BX LW LV REVHUYHG WKDW D VHFRQG H[WUHPL]DWLRQ LV UHTXLUHG IRU H[DPSOH VHH

PAGE 32

.HOOHU DQG 1LRUGVRQ f 7KXV D PRUH DSSURSULDWH PDQQHU RI FODVVLn I\LQJ FRVW IXQFWLRQDOV LV RQ WKH EDVLV RI ZKHWKHU WKH SUREOHP VWDWHPHQW LPSOLHV D VLQJOH RU D GRXEOH H[WUHPL]DWLRQ +HQFH WKH WZR EDVLF W\SHV RI FRVW IXQFWLRQDOV HQFRXQWHUHG LQ VWUXFWXUDO RSWLPL]DWLRQ DUH 0LQ )[!-Mf G9 X 9 *[BXf G9 9 0D[ 0LQ f X [ +[Xf G9 9 ZKHUH B[ PXVW VDWLVI\ DQ HTXLOLEULXP FRQGLWLRQ RI WKH VWDWH DQG X LV VXEMHFW WR VRPH DGPLVVLELOLW\ UHTXLUHPHQWV 7KHUH LV D VSHFLDO FDVH UHODWHG WR WKHVH WZR LQ ZKLFK WKH ZHLJKW LV WR EH PLQLPL]HG IRU D VSHFLILHG HLJHQYDOXH 7KLV SUREOHP LV WUHDWHG LQ ,FHUPDQ f ZLWK D PDWKHPDWLFDO GLVFXVVLRQ RI VXFK D YDULDWLRQDO SUREOHP SUHVHQWHG LQ ,UYLQJ DQG 0XOOLQHX[ S f ,Q WHUPV RI WKH WZR FRVW IXQFWLRQDOV WKH VSHFLDO FDVH LV 0LQ *[Xf GAO.[MXf` G9 X 9 A ZKHUH LV D VSHFLILHG FRQVWDQW 7KLV DSSURDFK LV IUHTXHQWO\ HPSOR\HG LQ HLJHQYDOXH SUREOHPV WR DYRLG WKH LQKHUHQW GLIILFXOWLHV DVVRFLDWHG ZLWK WKH GXDO H[WUHPL]DWLRQ SUREOHP 7KHUH KDYH DOVR EHHQ PDQ\ SDSHUV SXEOLVKHG WKDW FRQVLGHU PXOWLSXUSRVH VWUXFWXUHV HJ 3UDJHU f 0DUWLQ f DQG 3UDJHU DQG 6KLHOG f 7KH FRVW IXQFWLRQ IRU VXFK SUREOHPV LV GHILQHG DV

PAGE 33

N = DB[!L2 L O ZKHUH DBA DUH SRVLWLYH FRQVWDQWV VHUYLQJ DV ZHLJKWLQJ SDUDPHWHUV :KLOH SHUKDSV GHPRQVWUDWLQJ PXFK SRWHQWLDO QR VLJQLILFDQW UHVXOWV REWDLQHG ZLWK WKLV DSSURDFK KDYH VR IDU EHHQ SXEOLVKHG :KDW SUREOHPV KDYH EHHQ VROYHG DUH WRR VLPSOH LQGHHG WKH DXWKRUV LQGLFDWH WKH QHHG IRU XVLQJ D GLVFUHWH DSSUR[LPDWLRQ DQG PDWKHPDWLFDO SURJUDPPLQJ WHFKn QLTXHV LQ UHDOLVWLF DSSOLFDWLRQV $ VXEMHFW FORVHO\ UHODWHG WR PXOWLSXUSRVH VWUXFWXUHV LV WKDW RI PXOWLSOH FRQVWUDLQWV ,W LV PHQWLRQHG KHUH RQO\ EHFDXVH PRVW SDSHUV RQ WKH ODWWHU DOVR LQFOXGH WKH IRUPHUf§VHH 0DUWLQ f 7KH LGHD RI PXOWLSOH FRQVWUDLQWV LV QRW QHZ LQ ERWK YDULDWLRQDO DQG PDWKn HPDWLFDO SURJUDPPLQJ ILHOGV WKHUH H[LVW VWDQGDUG WHFKQLTXHV IRU KDQn GOLQJ PXOWLSOH FRQVWUDLQWV $ UHFHQW 5XVVLDQ SDSHU 6DOXNYDG]H f VXJJHVWV DQ DOWHUn QDWH WR WKH PXOWLSXUSRVH FRVW IXQFWLRQDO ,QVWHDG RI WUHDWLQJ D YHFWRU IXQFWLRQDO WKDW UHTXLUHV WKH FKRRVLQJ RI ZHLJKWLQJ FRHIILFLHQWV LW LV VXJJHVWHG WKDW WKH VHYHUDO IXQFWLRQDOV EH FRPELQHG LQWR RQH *LYHQ D V\VWHP DQG YHFWRU FRVW IXQFWLRQDOV [ IB[X Wf >X@ -AM[XWf L ON /HW XL3 GHQRWH WKH RSWLPDO VROXWLRQ IRU ZKLFK DVVXPHV WKH RSWLPDO f§237 A [ ) YDOXH RQ WKH WUDMHFWRU\ RI WKH V\VWHP )RU HDFK RI WKH -BA WKHUH LV D GLIIHUHQW 7KHVH N YDOXHV FDQ EH WKRXJKW RI DV FRPSRQHQWV RI f§237 L

PAGE 34

r W D YHFWRU U ZKHUH )RU DQ\ DUELWUDU\ XL WKH UHVXOW MU>X@ ^ >X@ -N>X@`7 LV MXVW VRPH YHFWRU IXQFWLRQDO N 9HFWRU BU UHSUHVHQWV D FRQVWDQW SRLQW LQ WKH VSDFH RI f 6LQFH QR FKRLFH RI X FDQ RSWLPL]H DOO RI WKH VLPXO N f§ L r WDQHRXVO\ WKDW LV WR DWWDLQ WKH SRLQW BU LQ -AVSDFH WKH EHVW DOWHU N QDWLYH LV WR PLQLPL]H WKH GLVWDQFH EHWZHHQ BU>X@ DQG UB 7KDW GLVWDQFH LV GHILQHG E\ WKH (XFOLGHDQ QRUP 7R DYRLG WKH TXHVWLRQ RI LQFRQn VLVWHQW GLPHQVLRQV WKH IXQFWLRQDOV DUH UHGXFHG WR GLPHQVLRQOHVV IRUP 7KXV ->X@ N = L O L>+O n Lf 237 DQG XAfP LV WKDW IXQFWLRQ X ZKLFK PLQLPL]HV WKH IXQFWLRQDO ->XM f§237 f§ f§ 0DWKHPDWLFDOO\ VSHDNLQJ '37 $5*0,1 7KLV W\SH RI YHFWRU FRVW IXQFWLRQDO LV PXFK PRUH DSSHDOLQJ WKDQ WKH W\SH WUHDWHG LQ WKH SDSHUV RQ PXOWLSXUSRVH VWUXFWXUHV ,W DOVR VXJJHVWV DQ HQWLUHO\ QHZ ILHOG RI VWXG\ WKH PRUH UHDOLVWLF I 7 6XSHUVFULSW 7 HJ MX GHQRWHV WKH WUDQVSRVH RI WKH YHFWRU WX

PAGE 35

FKRLFH RI FRVW IXQFWLRQDOV 7KH PDWKHPDWLFV RI D SUREOHP VHOGRP DFFRPPRGDWHV ILQDQFLDO FRQVLGHUDWLRQV )RU H[DPSOH WKH GHVLJQ ZKLFK UHTXLUHV WKH OHDVW PDWHULDO PD\ UHGXFH WKH FRVW RI PDWHULDOV DW DQ RYHUZKHOPLQJ H[SHQVH LQ PDQXIDFWXULQJ RU IDEULFDWLRQ :KHQ DHVWKHWLF DSSHDO DQG HQYLURQPHQWDO LPSDFW DUH LQFOXGHGf§DV PXVW EH GRQH LQ DQ\ UHDO FRPPHUFLDO DSSOLFDWLRQf§WKH VHOHFWLRQ RI DQ DSSURSULDWH FRVW IXQFWLRQDO LV DQ DOPRVW LQVXUPRXQWDEOH WDVN +RZHYHU D VLPSOH H[WHQn VLRQ RI 6DOXNYDG]HnV FRPSRVLWH FRVW IXQFWLRQDO PD\ UHGXFH WKH GLIILFXOn WLHV WR RSHUDWLRQV UHVHDUFK FRQVLGHUDWLRQV ,Q SUREOHPV ZKHUH LW LV GHVLUHG WR RSWLPL]H VLPXOWDQHRXVO\ VHYHUDO GLIIHUHQW IXQFWLRQDOV QRW DOO KDYLQJ WKH VDPH GLPHQVLRQV WKH FRQFHSW RI D JHQHUDOL]HG LQQHU SURGXFW PD\ SURYH XVHIXO ,W LV GHILQHG LQ WHUPV RI D PHWULF RSHUDWRU $ IRU VRPH JHQHUDO YHFWRU B] __B] __ ]B]f$ B=7$] DQG V\PERO PHDQV LV GHILQHG E\ :LWK UHIHUHQFH WR WKH YHFWRU FRVW IXQFWLRQDO $ UHSUHVHQWV D VHW RI VFDOH IDFWRUV ZKLFK FRQYHUWV DOO RI WKH VHSDUDWH FRVW IXQFWLRQDOV WR D FRPPRQ GLPHQVLRQ 7KLV LV ZKHUH WKH RSHUDWLRQV UHVHDUFK HQWHUVf§UHODWLQJ PDWHULDO H[SHQVH WR IDEULFDWLRQ WR VRFLRORJLFDO FRQVLGHUDWLRQV DQG VR IRUWKf§WR GHWHUPLQH WKH PHWULF $ )RU D YHFWRU HB ZKRVH HOHPHQWV DUH IXQFWLRQDOV -/ Q>M@ f / ZKHUH U>X@ DQG U DUH GHILQHG DERYH WKH FRPSRVLWH FRVW IXQFWLRQDO LV W X@ F$Ff

PAGE 36

7KH ZHDNQHVV LQ WKLV PHWKRG LV WKDW DQ RSWLPDO VROXWLRQ PXVW EH REWDLQHG IRU HDFK LQGLYLGXDO FRVW IXQFWLRQDO SULRU WR DWWHPSWLQJ D VROXWLRQ WR WKH FRPSRVLWH SUREOHP $GGLWLRQDOO\ VRPH RI WKH PRUH DEVWUDFW FRVW REMHFWLYHV PD\ EH GLIILFXOW WR TXDQWLI\ LQ D PHDQLQJIXO PDQQHU 'HVSLWH WKHVH VKRUWFRPLQJV WKLV DSSURDFK GRHV VXJJHVW LQWHUn HVWLQJ DSSOLFDWLRQV 0HWKRGV &RQWLQXRXV 6\VWHPV 7KH SUREOHPV FKDUDFWHUL]HG E\ PDWKHPDWLFDO IXQFWLRQV LQ FRQn WUDVW WR WKRVH UHSUHVHQWHG E\ D VHW RI GLVFUHWH FRQVWDQWV DUH QRUPDOO\ WUHDWHG E\ YDULDWLRQDO WHFKQLTXHV 0DQ\ ERRNV RQ WKLV VXEMHFW KDYH EHHQ SXEOLVKHG WKH EHWWHU DXWKRUV LQFOXGH (OVJROF f *HOIDQG DQG )RPLQ f 'UH\IXV f +HVWHQHV f 'HQQ f /XHQEHUJHU f DQG %U\VRQ DQG +R f $Q H[FHOOHQW VXPPDU\ SDSHU LV DYDLOn DEOH LQ %HUJ f 7R VHH KRZ WKHVH WHFKQLTXHV DUH DSSOLHG WKUHH SDSHUV DUH UHFRPPHQGHG 7KH ILUVW LV %ODVLXV f ZKLFK SURYLGHV VXIILFLHQW GHWDLO DQG H[SODQDWLRQ WR PDNH LW TXLWH ZRUWKZKLOH $OWKRXJK LW GRHV LQFOXGH VHYHUDO H[DPSOHV WKDW LQYROYH VXEVLGLDU\ FRQGLWLRQV QR LQHTXDOn LW\ FRQVWUDLQWV DUH WUHDWHG $Q H[DPSOH WKDW LQFOXGHV LQHTXDOLW\ FRQn VWUDLQWV WR WKH FRQWURO YDULDEOH LV FRRQWDLQHG LQ %UDFK f $ PXFK PRUH JHQHUDO DSSOLFDWLRQ RI YDULDWLRQDO SULQFLSOHV LV SUHVHQWHG LQ 2GHQ DQG 5HGG\ f ,Q WKLV SDSHU D GXDOFRPSOHPHQWDU\ YDULDWLRQDO SULQFLSOH LV GHYHORSHG IRU D SDUWLFXODU FODVV RI SUREOHPV ,W LV VKRZQ WKDW WKH FDQRQLFDO HTXDWLRQV REWDLQHG DUH WKH (XOHU/DJUDQJH HTXDWLRQV IRU D FHUWDLQ IXQFWLRQDO

PAGE 37

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n WLRQDO $ JRRG GLVFXVVLRQ RI WKH IRUPHU LV DYDLODEOH LQ )XQJ f RU 3U]HPLHQLHFNL f WKH EHVW JHQHUDO WUHDWPHQW RI WKH BDWWHU LV DYDLOn DEOH LQ HLWKHU *RXOG &KDSWHU f RU LQ 0LNKOLQ DQG 6PROLWVNL\ &KDSWHU f 7KH SULQFLSOH RI PLQLPXP SRWHQWLDO HQHUJ\ LV IUHTXHQWO\ XVHG ZLWK VLPSOH SUREOHPV WR SURYH WKDW D QHFHVVDU\ FRQGLWLRQ IRU RSWLPDOLW\ LV DOVR VXIILFLHQW 3UDJHU DQG 7D\ORU f FRQWDLQV VXFK D SURRI IRU WKH JOREDO PD[LPXP VWLIIQHVV GHVLJQ RI DQ HODVWLF VWUXFWXUH RI VDQGZLFK FRQVWUXFWLRQ WZR SDSHUV WKDW DOVR FRQVLGHU WKLV SUREOHP DUH +XDQJ f DQG 7D\ORU f 6SHFLILF DSSOLFDWLRQ RI WKH HQHUJ\ PHWKRG WR DQ HLJHQYDOXH SUREOHP LV GHPRQVWUDWHG LQ 7D\ORU DQG /LX f $ PXFK PRUH JHQHUDO GLVFXVVLRQ RI WKH HQHUJ\ PHWKRG LV SURYLGHG LQ 6DOLQDV f )XUWKHU H[WHQVLRQV RI WKH PHWKRG DUH SUHVHQWHG LQ 0DVXU f LQ ZKLFK WKH SULQFLSOH RI PLQLPXP FRPSOHPHQWDU\ HQHUJ\ LV DSSOLHG WR SUREOHPV RI WKH RSWLPXP VWLIIQHVV DQG VWUHQJWK RI HODVWLF

PAGE 38

VWUXFWXUHV ,Q WKHVH SUREOHPV D QHFHVVDU\ FRQGLWLRQ IRU RSWLPDOLW\ LV WKDW WKH VWUDLQ HQHUJ\ GHQVLW\ EH FRQVWDQW WKURXJKRXW WKH VWUXFWXUH 7KLV FRQGLWLRQ LV DOVR VXIILFLHQW IRU RSWLPDOLW\ LQ FHUWDLQ FODVVHV RI VWUXFWXUHV WKDW VDWLVI\ D VSHFLILF UHODWLRQVKLS EHWZHHQ WKH VWUDLQ HQHUJ\ GHQVLW\ DQG GHVLJQ YDULDEOHV 0DQ\ RI WKH HQHUJ\ SUREOHPV EHORQJ WR WKH FODVV RI SUREOHPV KDYLQJ TXDGUDWLF FRVW IXQFWLRQDOV 7KH VLJQLILFDQFH RI WKLV FKDUDFWHUn LVWLF LV WKDW WKH (XOHU/DJUDQJH HTXDWLRQV GHULYHG IURP VXFK IXQFWLRQDOV DUH OLQHDU $ PRUH UHFHQW HQHUJ\ PHWKRG GHYHORSPHQW LV WKH FRQFHSW RI PXWXDO SRWHQWLDO HQHUJ\ 0XWXDO SRWHQWLDO HQHUJ\ WHFKQLTXHV UHVHPEOH WKRVH RI WKH SULQFLSOH RI PLQLPXP SRWHQWLDO HQHUJ\ ,Q ERWK PHWKRGV D FRVW IXQFWLRQDO LV GHILQHG RYHU WKH HQWLUH GRPDLQ RFFXSLHG E\ WKH VWUXFWXUH DQG RSWLPL]HG ZLWK UHVSHFW WR WKH FRQWURO YDULDEOH ,I LW LV GHVLUHG WKDW WKH RSWLPDO VROXWLRQ EH UHTXLUHG WR KDYH D VSHFLILHG GHIOHFWLRQ DW D FHUWDLQ SRLQW WKLV FRQGLWLRQ FRUUHVSRQGV WR D VXEn VLGLDU\ VWDWH FRQVWUDLQW ZKHQ XVLQJ WKH SULQFLSOH RI PLQLPXP SRWHQWLDO HQHUJ\ 7KH PXWXDO SRWHQWLDO HQHUJ\ PHWKRG LQFRUSRUDWHV WKLV W\SH RI ORFDOL]HG FRQVWUDLQW LQWR WKH FRVW IXQFWLRQDO ZKLFK LV GHILQHG RYHU WKH HQWLUH GRPDLQ RI WKH VWUXFWXUH %\ LWVHOI WKLV DORQH LV DGYDQWDn JHRXV KRZHYHU IRU FHUWDLQ W\SHV RI SUREOHPV WKH PXWXDO SRWHQWLDO HQHUJ\ PHWKRG DOVR SURYLGHV ERWK D QHFHVVDU\ DQG VXIILFLHQW FRQGLWLRQ IRU JOREDO RSWLPDOLW\ ,Q WKH ZD\ RI D FULWLFDO FRPPHQW HLWKHU WKH PHWKRG KDV UHFHLYHG WRR OLWWOH DWWHQWLRQ RU HOVH LW GRHV QRW HIILn FLHQWO\ KDQGOH SUREOHPV PRUH GLIILFXOW WKDQ WKH VLPSOH H[DPSOHV SUHVHQWHG

PAGE 39

)RXU SDSHUV WKDW DUH UHSUHVHQWDWLYH RI WKH OLWHUDWXUH DVVRFLDWHG ZLWK WKLV PHWKRG DUH 6KLHOG DQG 3UDJHU f &KHUQ D Ef DQG 3ODXW Ff $QRWKHU UHFHQW GHYHORSPHQW LV WKDW WKH FODVV RI SUREOHPV IRU ZKLFK HQHUJ\ PHWKRGV LV DSSOLFDEOH KDV EHHQ H[SDQGHG WR LQFOXGH FHUn WDLQ W\SHV RI QRQFRQVHUYDWLYH V\VWHPV 7RJHWKHU ZLWK WKH PXWXDO SRWHQn WLDO HQHUJ\ FRQFHSWV WKLV VXJJHVWV WKDW SHUKDSV WKH FODVVLFDO HQHUJ\ PHWKRG LV D VSHFLDO FDVH RI D PRUH JHQHUDOL]HG PHWKRG ,I D WHFKQLTXH FDQ EH GHYHORSHG ZKLFK XVHV WKH DGMRLQW YDULDEOHV WR WUDQVIRUP D JHQn HUDO QRQFRQVHUYDWLYH V\VWHP LQWR DQ HTXLYDOHQW VHOIDGMRLQW IRUP WKH PHWKRG PLJKW EH GHGXFHG 6RPH SDSHUV SHUWDLQLQJ WR WKH VXEMHFW DUH 3UDVDG DQG +HUUPDQQ f :X f DQG %DUVWRQ f 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH 7KHUH DUH PDQ\ WH[WERRNV ZKLFK GHULYH H[SODLQ DQG JLYH H[DPSOHV IRU WKH PD[LPXP SULQFLSOH 7KH RULJLQDO 3RQWU\DJLQ HW DO f UHTXLUHV D NQRZOHGJH RI IXQFWLRQDO DQDO\VLV $ FRQGHQVHG IRUP RI WKLV VDPH PDWHULDO LV DYDLODEOH LQ 5R]RQRU f 'HQQ f SURYLGHV DQRWKHU SRLQW RI YLHZ LQ ZKLFK WKH SULQFLSOH LV GHULYHG IURP *UHHQnV IXQFWLRQV ,Q WKLV PDQQHU WKH VHQVLWLYLW\ WR YDULDWLRQV LV UHDGLO\ REVHUYHG 7R XQGHUVWDQG 'HQQnV WUHDWPHQW UHTXLUHV RQO\ D NQRZOHGJH RI WKH VROXWLRQ RI GLIIHUHQWLDO HTXDWLRQV 6KRUWO\ DIWHU 3RQWU\DJLQnV ERRN ZDV SXEOLVKHG PDQ\ SDSHUV GHYRWHG WR WKH WKHRUHWLFDO DVSHFWV RI WKH PD[LPXP SULQFLSOH ZHUH SXEn OLVKHG 6RPH RI WKH PRUH UHDGDEOH RQHV DUH .RSS f 5R[LQ f DQG +DONLQ f $QRWKHU HDUO\ SDSHU %UHDNZHOO f

PAGE 40

DSSHDUV WR EH D FRPSOHWHO\ LQGHSHQGHQW GHULYDWLRQ RI WKH PD[LPXP SULQn FLSOH $OWKRXJK TXLWH JHQHUDO LQ WKH PDWKHPDWLFDO VHQVH WKH H[DPSOHV SUHVHQWHG DUH WUDMHFWRU\ RSWLPL]DWLRQ SUREOHPV DQG QRW D JHQHUDO W\SH RI PDWKHPDWLFDO SUREOHP 7KLV PD\ EH DQ H[SODQDWLRQ IRU ZKDW VHHPV WR EH D ODFN RI UHFRJQLWLRQ IRU D VLJQLILFDQW DFKLHYHPHQW 7KH DSSOLFDWLRQ RI 303 WR SUREOHPV LQ VWUXFWXUDO RSWLPL]DWLRQ LV UHODWLYHO\ UHFHQW :KHQ WKH PHWKRG LV XVHG RQH RI WZR GLIILFXOWLHV LV RIWHQ HQFRXQWHUHG 7KH ILUVW LV HUURUV LQ WKH IRUPXODWLRQ RI WKH RSWLPDO FRQWURO SUREOHP WKH VHFRQG LV WKDW RQFH D ZHOOSRVHG QRQn OLQHDU WZRSRLQW ERXQGDU\ YDOXH SUREOHP 73%93f LV REWDLQHG LW LV GLIn ILFXOW WR VROYH $Q H[DPSOH RI WKH ILUVW LV SURYLGHG E\ 'L[RQ ff§ WKH FRUUHFWLRQ ZDV JLYHQ LQ %R\NLQ DQG 6LHUDNRZVNL f 'H 6LOYD f SURYLGHV D FOHDU SUHVHQWDWLRQ RQ WKH DSSOLFDWLRQ RI 303 WR D VSHFLILF SUREOHP EXW LQFOXGHV QR GDWD EHFDXVH D VROXWLRQ FRXOG QRW EH REWDLQHG 'HVSLWH WKH IDLOXUH WR GHWHUPLQH WKH VROXWLRQ WKLV SDSHU LV ZRUWKZKLOH IRU LWV OXFLG GLVFXVVLRQ RI WKH 303 DSSOLFDWLRQ $QRWKHU SDSHU WKDW JLYHV D JRRG VSHFLILF DSSOLFDWLRQ RI 303 LV 0DGD\ f $OWKRXJK PXFK DQDO\VLV LV SUHVHQWHG YHU\ OLWWOH LV VDLG UHJDUGLQJ WKH VROXWLRQ WHFKQLTXHV $OO RI WKH DERYH UHIHUHQFHV DUH DSSOLFDEOH RQO\ WR V\VWHPV WKDW DUH GHVFULEHG E\ RUGLQDU\ GLIIHUHQWLDO HTXDWLRQV LQ FRQWUDVW ZLWK WKH FDOFXOXV RI YDULDWLRQV ZKLFK DOVR KDQGOHV SUREOHPV GHVFULEHG E\ SDUWLDO GLIIHUHQWLDO HTXDWLRQV 6LQFH PDQ\ RI WKH SUREOHPV RI PDWKHn PDWLFDO SK\VLFV LQYROYH SDUWLDO GLIIHUHQWLDO HTXDWLRQV DQ H[WHQVLRQ RI 303 WR LQFOXGH WKLV FODVV RI SUREOHPV LV WKH QH[W ORJLFDO GHYHORSPHQW

PAGE 41

6RPH ZRUN KDV DOUHDG\ EHHQ GRQH IRU H[DPSOH %DUQHV f DQG .RPNRY f $ VXUYH\ RI WKHVH GLVWULEXWHG SDUDPHWHU V\VWHPVf§ VHH 6HFWLRQ f§LV SUHVHQWHG LQ :DQJ f 0HWKRG RI 6WHHSHVW $VFHQW'HVFHQW 7KLV PHWKRG LV IUHTXHQWO\ FLWHG LQ WKH OLWHUDWXUH IRU WUDMHFn WRU\ RSWLPL]DWLRQ DQG RFFDVLRQDOO\ LQ UHIHUHQFHV UHODWHG WR RSWLPDO VWUXFWXUHV :KHQ WKH PHWKRG RI TXDVLOLQHDUL]DWLRQ FRQYHUJHG IRU WKH GLVVHUWDWLRQ H[DPSOH SUREOHPV WKHUH ZDV QR QHHG WR LQYHVWLJDWH RWKHU PHWKRGV VXFK DV WKH PHWKRG RI VWHHSHVW DVFHQW &RQVHTXHQWO\ OLWWOH LV VDLG DERXW LW $FFRUGLQJ WR WKH UHIHUHQFHV LW LV DSSOLHG LQ D VWUDLJKWn IRUZDUG PDQQHU )XUWKHUPRUH WKH H[DPSOH SUREOHP VROXWLRQV SUHVHQWHG VHHP WR EH UHDO SUREOHPV DQG QRW DFDGHPLFDOO\ VLPSOH 7KH IROORZLQJ IRXU SDSHUV WUHDW WKH PHWKRG LQ JHQHUDO ZLWK WUDMHFWRU\ RSWLPL]DWLRQ DSSOLFDWLRQV %U\VRQ DQG 'HQKDP f %U\VRQ HW DO f DQG +DQVRQ f ,Q +DXJ HW DO f WKH PHWKRG RI VWHHSHVW DVFHQW LV GHULYHG LQ GHWDLO FRPSOHWHO\ GLVFXVVHG DQG FRPSDUHG WR WKH PD[LPXP SULQFLSOH 6HYHUDO VWUXFWXUDO RSWLPL]DWLRQ SUREOHPV DUH WKHQ VROYHG E\ WKH PHWKRG RI VWHHSHVW DVFHQW $OWKRXJK QR H[FLWLQJ UHVXOWV DUH REWDLQHG XVH RI WKH PHWKRG LV FOHDUO\ LOOXVWUDWHG E\ WKH DSSOLFDWLRQV WR UHDOLVWLF VWUXFWXUDO SUREOHPV 7UDQVLWLRQ 0DWUL[ $HURHODVWLFLW\ 3UREOHPV )RU WKH SDVW IHZ \HDUV D JURXS DW 6WDQIRUG 8QLYHUVLW\ KDV VWXGLHG WKH RSWLPL]DWLRQ RI VWUXFWXUHV VXEMHFW WR G\QDPLF RU DHURG\QDPLF FRQn VWUDLQWV 7KH JHQHUDO SUREOHP RI WKHLU LQWHUHVW LV WKDW RI PLQLPL]LQJ

PAGE 42

WKH ZHLJKW RI D JLYHQ VWUXFWXUH IRU VSHFLILHG HLJHQYDOXH VXEMHFW WR LQHTXDOLW\ FRQVWUDLQWV RQ FRQWURO 7KUHH W\SHV RI VROXWLRQ WHFKQLTXHV DUH XVHG RQFH WKH QHFHVVDU\ FRQGLWLRQV IRU PLQLPXP ZHLJKW DUH GHWHUPLQHG ([DFW VROXWLRQV DUH REWDLQHG IRU PRVW RI WKH SUREOHPV EHFDXVH WKH\ DUH VR VLPSOH WKDW DQDO\WLFDO PHWKRGV DUH DSSOLFDEOH 0RUH FRPSOLFDWHG SUREOHPV DUH VROYHG E\ D WUDQVLWLRQ PDWUL[ PHWKRG GHVFULEHG LQ %U\VRQ DQG +R f 2Q WKH EDVLV RI FRQYHUJHQFH GLIILFXOWLHV UHSRUWHG LQ WKH UHIHUHQFHV WKLV PHWKRG VKRXOG EH XVHG ZLWK FDXWLRQ 5HVXOWV KDYH EHHQ REWDLQHG RQO\ IRU YHU\ VLPSOH SUREOHPV +RZHYHU WKHVH UHVXOWV DUH FRUURERUDWHG E\ GDWD REWDLQHG IURP D GLVFUHWH DSSUR[LPDWLRQ PHWKRG )LYH SDSHUV WKDW DUH UHSUHVHQWDWLYH RI WKLV ZRUN DUH 0F,QWRVK DQG (DVWHS f $VKOH\ DQG 0F,QWRVK f 0F,QWRVK HW DO f $VKOH\ HW DO f DQG :HLVVKDDU f 0LVFHOODQHRXV 0HWKRGV 7KH SUHFHGLQJ VHFWLRQV KDYH EULHIO\ RXWOLQHG WKH PHWKRGV RI VWUXFWXUDO RSWLPL]DWLRQ PRVW IUHTXHQWO\ HQFRXQWHUHG LQ WKH OLWHUDWXUH $SSURSULDWH UHSUHVHQWDWLYH UHIHUHQFHV KDYH DOVR EHHQ JLYHQ 1RW DOO PHWKRGV DUH OLVWHG ZKLOH VRPH DUH RPLWWHG IRU QRW EHLQJ JHQHUDOO\ XVHn IXO RWKHUV DUH RPLWWHG IRU QRW EHLQJ JHQHUDOO\ XVHG 7ZR H[DPSOHV RI WKH ODWWHU DUH WKH PRGLILHG TXDVLOLQHDUL]DWLRQ DQG VHTXHQWLDO JUDGLHQW UHVWRUDWLRQ DOJRULWKPV GHVFULEHG LQ 0LHOH HW DO f DQG LQ +HQQLJ DQG 0LHOH f $W VRPH ODWHU WLPH WKHVH PHWKRGV PD\ EH DFNQRZOHGJHG DV PDMRU PHWKRGV WKDW DUH DSSOLFDEOH WR PDQ\ GLIIHUHQW RU LPSRUWDQW SUREn OHPV EXW IRU QRZ WKH\ DUH PHQWLRQHG RQO\ LQ SDVVLQJ

PAGE 43

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f *HOODWO\ DQG %HUNH f 3RSH DQG 6FKPLGW f DQG *DOODJKHU DQG =LHQNLHZLF] f $QRWKHU UHSRUW 0HORVK DQG /XLN f SURYLGHV D JRRG H[SRVLWLRQ RI WKH GLIILFXOWLHV DVVRFLDWHG ZLWK WKH DQDO\VLV SRUWLRQ RI OHDVW ZHLJKW VWUXFWXUDO GHVLJQ ,W DOVR FRQn WDLQV D EULHI FRPSDULVRQ RI YDULRXV PDWKHPDWLFDO SURJUDPPLQJ PHWKRGV 0F1HLOO f LV WKH ODVW UHIHUHQFH WR EH FLWHG LQ WKH VHFWLRQ RQ JHQHUDO PHWKRGV IRU WKH RSWLPL]DWLRQ RI GLVFUHWH V\VWHPV 0LQLPXP ZHLJKW GHVLJQ RI JHQHUDO VWUXFWXUHV LV WUHDWHG LQ D PDWKHPDWLFDOO\ SUHFLVH IRUPXODWLRQ /HJHQGUHnV QHFHVVDU\ FRQGLWLRQ LV FRPELQHG ZLWK WKH FRQFHSWV RI FRQYH[ IXQFWLRQV DQG VHWV WR GHULYH WKH QHFHVVDU\ DQG VXIILFLHQW FRQGLWLRQV IRU JOREDO RSWLPDOLW\ )XOO\ VWUHVVHG GHVLJQV DQG FRQVWUDLQWV WR HLJHQYDOXHV DUH DOVR GLVFXVVHG ,Q VXPPDU\ WKLV SDSHU RURYLGHV D JRRG H[DPSOH RI WKH JHQHUDO PDWKHPDWLFDO SUREOHP WKDW PXVW EH VROYHG LQ WKH RSWLPL]DWLRQ RI GLVFUHWH V\VWHPV

PAGE 44

:KLOH FHUWDLQ YDULDWLRQDO PHWKRGV PD\ EH DSSOLHG WR GLVFUHWH V\VWHPV WKH PRVW IUHTXHQWO\ XVHG WHFKQLTXH LV PDWKHPDWLFDO SURJUDPPLQJ ,Q WKH IROORZLQJ VHFWLRQV WKLV PHWKRG DQG RWKHU PDMRU PHWKRGV DUH GLVn FXVVHG DQG UHSUHVHQWDWLYH UHIHUHQFHV FLWHG 0DWKHPDWLFDO 3URJUDPPLQJ 7KH JHQHUDO PHWKRG RI PDWKHPDWLFDO SURJUDPPLQJ LV GLVFXVVHG LQ VHFWLRQ RI WKH GLVVHUWDWLRQ DQG WKH VROXWLRQ RI DQ H[DPSOH SUREOHP XVLQJ WKLV PHWKRG LV GHWDLOHG LQ &KDSWHU 9/ ,Q WKH OLWHUDWXUH UHODWHG WR WKLV VXEMHFW D YHU\ UHDGDEOH WH[WERRN LV DYDLODEOHf§)R[ f 7KLV ERRN FRPSOHPHQWV WKH WKHRU\ ZLWK QXPHURXV GLVFXVVLRQV SHUWDLQLQJ WR QXPHULFDO WHFKQLTXHV DQG PHWKRGV WKDW FDQ EH HPSOR\HG WR RYHUFRPH FHUWDLQ GLIILFXOWLHV WKDW PD\ DULVH $OWKRXJK LW GRHV FRQWDLQ IORZFKDUWV RI VHYHUDO DOJRULWKPV WKHUH DUH IHZ VSHFLILF H[DPSOHV JLYHQ )RU D GLVFXVVLRQ RI WKH JHQHUDO WKHRU\ WZR DOWHUQDWLYHV WR WKLV ERRN H[LVW LQ WKH IRUP RI SDSHUV 6FKPLGW @f 7KH ILUVW LV ZULWWHQ LQ D FRQYHUVDWLRQDO VW\OH FRQWDLQV QR PDWKHPDWLFV DQG LV LQWHQGHG WR SURYLGH RQO\ D JHQHUDO GHVFULSWLRQ RI WKH VXEMHFW 7KH ODWWHU SDSHU LV WKHRUHWLFDO LQ FRQWHQW $Q H[FHOOHQW DSSOLFDWLRQ WR D UHDOLVWLF SUREOHP LV WR EH IRXQG LQ 6WURXG HW DO f 7KV SDSHU FRQWDLQV OLWWOH GLVFXVVLRQ RI WKH PHWKRG LWVHOI EXW GRHV GHPRQVWUDWH DQ DSSOLFDWLRQ WKDW DOORZV D FRQFHS WXDO YLVXDOL]DWLRQ RI WKH VROXWLRQ 7KH DSSURDFK LV WR DVVXPH WKH VROXWLRQ WR EH D OLQHDU FRPELQDWLRQ RI VSHFLILHG IXQFWLRQV DQG WR FKRRVH WKH ZHLJKWLQJ FRHIILFLHQWV WR PLQLPL]H WKH FRVW 0DWKHPDWLFDO SURJUDPPLQJ LV HPSOR\HG WR GHWHUPLQH WKH RSWLPDO VHW RI FRHIILFLHQWV

PAGE 45

7KLV DSSURDFK UHVHPEOHV *DOHUNLQnV PHWKRG DQG WKRXJK QRW PDWKHPDWn LFDOO\ ULJRURXV LW PD\ SURYLGH D XVHIXO DSSUR[LPDWLRQ WR ODUJH XQZLHOG\ SUREOHPV 'LVFUHWH 6ROXWLRQ $SSUR[LPDWLRQV ,Q WKH SUHYLRXV VHFWLRQ D SDSHU LV FLWHG WKDW FRQWDLQV DQ DSSUR[LPDWH VROXWLRQ REWDLQHG E\ *DOHUNLQnV PHWKRG 7KH XVH RI WKH *DOHUNLQ RU 5D\OHLJK5LW] DSSUR[LPDWH VROXWLRQ WHFKQLTXHV LV VXIILn FLHQWO\ ZLGHVSUHDG WR EH FRQVLGHUHG D JHQHUDO PHWKRG )RU ERWK PHWKRGV WKH VROXWLRQ LV DVVXPHG WR EH D OLQHDU FRPELQDWLRQ RI WKH VROXWLRQ WR WKH OLQHDU SDUW RI WKH JRYHUQLQJ HTXDWLRQ DQG D VHW RI SUHVFULEHG IXQFn WLRQV 7KLV DSSUR[LPDWH VROXWLRQ GRHV QRW VDWLVI\ WKH JLYHQ HTXDWLRQ H[DFWO\ EXW SURGXFHV VRPH UHVLGXDO IXQFWLRQ $ FRVW IXQFWLRQ WKDW GHSHQGV XSRQ WKH UHVLGXDO LV WKHQ PLQLPL]HG ZLWK UHVSHFW WR WKH XQNQRZQ FRHIILFLHQWV 7KH WZR ZHLJKWHG UHVLGXDO PHWKRGV PHQWLRQHG DERYH KDYH GLIIHUHQW FRVW IXQFWLRQV EXW WKH PHWKRGV DUH LGHQWLFDO IRU OLQHDU HTXDWLRQVf§VHH &XQQLQJKDP S f 7KH DGYDQWDJH WR XVLQJ WKHVH PHWKRGV LV WKDW DIWHU DVVXPLQJ WKH SDUWLFXODU IRUP RI VROXWLRQ WKH SUREOHP RI VROYLQJ IRU WKH ZHLJKWn LQJ FRHIILFLHQWV PD\ EH PXFK VLPSOHU WKDQ WKH RULJLQDO SUREOHP ,Q WKH FDVH RI 6WURXG HW DO f WKH FRHIILFLHQWV ZHUH REWDLQHG E\ PDWKHn PDWLFDO SURJUDPPLQJ WHFKQLTXHV +RZHYHU WKH ZHDNQHVV RI WKH PHWKRG LV WKH UHVWULFWHG IXQFWLRQ VSDFH RI SRVVLEOH VROXWLRQV :LWK WKH FRHI ILFHQWV REWDLQHG E\ WKHVH PHWKRGV WKH UHVXOWLQJ VROXWLRQ LV WKH EHVW DSSUR[LPDWLRQ WKDW LV SRVVLEOH IURP WKH VHW RI VROXWLRQ IXQFWLRQV

PAGE 46

SUHVFULEHG 7KHUH LV QR JXDUDQWHH WKDW WKH DSSUR[LPDWLRQ HYHQ UHVHPEOHV WKH WUXH VROXWLRQ 7KH IOXWWHU RI D SDQHO LV VROYHG XVLQJ *DOHUNLQnV PHWKRG LQ 3ODXW Df 1R JHQHUDO GHYHORSPHQWV DUH SUHVHQWHG DQG WKH DVVXPHG VROXWLRQ IXQFWLRQV DUH WULYLDOO\ VLPSOH +RZHYHU WKLV SDSHU GRHV SURYLGH DQ DSSOLFDWLRQ RI WKH PHWKRG WR REWDLQ DQ DSSUR[LPDWH VROXWLRQ WR D YHU\ GLIILFXOW RSWLPL]DWLRQ SUREOHP LQYROYLQJ WKH VWDELOLW\ RI D QRQFRQVHUYDWLYH V\VWHP $ VLPLODU SUREOHP LV WUHDWHG LQ D PRUH WKHRn UHWLFDO PDQQHU LQ 3ODXW Ef XVLQJ D PRGLILHG 5D\OHLJfQ5LW] PHWKRG 6HJPHQWZLVHFRQVWDQW FRQWURO IXQFWLRQV DUH DVVXPHG DOVR WKLV SDUWLFn XODU DSSUR[LPDWLRQ LV GLVFXVVHG ZLWK PRUH GHWDLO LQ WKH IROORZLQJ VHFn WLRQ $GGLWLRQDO QRQFRQVHUYDWLYH SUREOHPV DUH WUHDWHG LQ /HLSKRO] f DSSO\LQJ *DOHUNLQnV DSSUR[LPDWH VROXWLRQ WR WKH HQHUJ\ PHWKRG 6HYHUDO VLPSOH H[DPSOHV DUH LQFOXGHG 1RQFRQVHUYDWLYH HODVWLF VWDELOLW\ SUREOHPV RI HODVWLF FRQWLQXD DUH WUHDWHG LQ 3UDVDG DQG +HUUPDQQ f XVLQJ DGMRLQW V\VWHPV 7KLV DSSURDFK LV PRUH UHDOLVWLF WKDQ WKH VHJPHQWZLVHFRQVWDQW FRQWURO DVVXPSn WLRQ GHVFULEHG LQ WKH IROORZLQJ VHFWLRQ 6ROXWLRQV IRU WKH VWDWH DQG DGMRLQW V\VWHP DUH DVVXPHG VXFK WKDW DSSUR[LPDWLRQ SURFHVV UHVHPEOHV WKH 5D\OHLJK5LW] PHWKRG +RZHYHU RQO\ D VLQJOH W\SH RI QRQFRQVHUYDn WLYH V\VWHP LV FRQVLGHUHG ([WHQVLRQ WR VHYHUDO RWKHU W\SHV RI QRQFRQn VHUYDWLYH HODVWLF FRQWLQXD SUREOHPV LV JLYHQ LQ 'XEH\ f 9DULDn WLRQDO HTXDWLRQV FRUUHVSRQGLQJ WR ERWK WKH *DOHUNLQ DQG 5D\OHLJK5LW] PHWKRGV DUH GHULYHG )XUWKHUPRUH WKH FRQGLWLRQ IRU HTXLYDOHQFH RI WKH WZR PHWKRGV LV VKRZQ WR EH WKDW WKH DGPLVVLEOH YHORFLW\ ILHOG PXVW

PAGE 47

VDWLVI\ D QDWXUDO ERXQGDU\ FRQGLWLRQ RYHU WKDW SRUWLRQ RI WKH ERG\fV VXUIDFH ZKHUH WUDFWLRQV DUH SUHVFULEHG 6HJPHQWZLVH&RQVWDQW $SSUR[LPDWLRQV 7KH GHILQLWLYH FKDUDFWHULVWLF RI WKLV PHWKRG LV DSSUR[LPDWLQJ WKH VWUXFWXUDO V\VWHP E\ D QXPEHU RI GLVFUHWH VHJPHQWV ZKHUH ZLWKLQ HDFK VHJPHQW WKH FRQWURO IXQFWLRQ KDV D FRQVWDQW YDOXH ,Q JHQHUDO WKH FRQVWDQW YDOXH RI WKH FRQWURO GLIIHUV IURP VHJPHQW WR VHJPHQW )RU WKH PDQ\ SDSHUV RQ WKLV PHWKRG WKDW KDYH EHHQ SXEOLVKHG WKH SURFHGXUH LV WKH VDPH $Q RSWLPDOLW\ FRQGLWLRQ QHFHVVDU\ LQ DOO FDVHV EXW DOVR VXIILFLHQW LQ VRPHf RU FRVW IXQFWLRQDO LV GHULYHG IRU WKH FRQWLQXRXV V\VWHP $IWHU GHILQLQJ WKH VHJPHQWZLVHFRQVWDQW DSSUR[LPDWLRQ WKH FRQGLWLRQ RU IXQFWLRQDO LV UHIRUPXODWHG LQ WHUPV RI WKH GLVFUHWH SDUDPn HWHUV 0RVW RI WKH SDSHUV XVH VR IHZ HOHPHQWV WKDW VROYLQJ IRU WKH GLVFUHWH YDOXHV RI WKH FRQWURO SDUDPHWHU SRVHV QR GLIILFXOWLHV $OWKRXJK WKLV PHWKRG GRHV VLPSOLI\ WKH PDWKHPDWLFDO SUREOHP WR EH VROYHG WKH FUXGHQHVV RI WKH DSSUR[LPDWLRQ LV QRW DSSHDOLQJ )LYH SDSHUV ZKLFK WUHDW D YDULHW\ RI SUREOHPV XVLQJ WKLV DSSUR[LPDWLRQ DUH FLWHG EHORZ 0LQLPXP ZHLJKW RI VDQGZLFK VWUXFWXUHV VXEMHFW WR VWDWLF ORDGV LV GLVFXVVHG LQ 6KHX DQG 3UDJHU Df ,Q 6KHX f WKH VDPH W\SH RI VWUXFWXUH LV FRQVLGHUHG ,W GLIIHUV IURP WKH ILUVW SUREOHP E\ UHTXLULQJ SRLQW PDVVHV WR EH VXSSRUWHG VXFK WKDW WKH WRWDO VWUXFWXUH KDV D SUHVFULEHG IXQGDPHQWDO IUHTXHQF\ RI IUHH YLEUDWLRQ ,FHUPDQ f WUHDWV WKH SUREOHP RI HODVWLF VWUXFWXUHV VXEMHFW WR D FRQFHQWUDWHG ORDG RI KDUPRQLFDOO\ YDU\LQJ DPSOLWXGH 7KH PLQLPXP ZHLJKW GHVLJQ LV REWDLQHG

PAGE 48

VXEMHFW WR D FRPSOLDQFH FRQVWUDLQW UHODWHG WR WKH DSSOLHG ORDG DQG ZKLFK LV HIIHFWLYHO\ D ERXQGDU\ FRQGLWLRQ RQ GLVSODFHPHQW DW WKH SRLQW RI DSSOLFDWLRQ $ WUXVV SUREOHP LV DOVR LQFOXGHG 7KH FRQFHSW RI D FRPSOLDQFH FRQVWUDLQW LV SXUVXHG IXUWKHU LQ &KHP DQG 3UDJHU f 7KH PLQLPXP ZHLJKW GHVLJQ IRU VDQGZLFK FRQn VWUXFWLRQ EHDPV XQGHU DOWHUQDWLYH ORDGV LV IRXQG VXEMHFW WR WKLV W\SH RI FRQVWUDLQW 7KH SDSHU XVHV XS WR HLJKW VHJPHQWV WKHUHE\ REWDLQLQJ D PRUH UHDOLVWLF DSSUR[LPDWLRQ WR WKH FRQWLQXRXV SUREOHP 0LQLPXP ZHLJKW GHVLJQ RI HODVWLF VWUXFWXUHV VXEMHFW WR ERG\ IRUFHV DQG D SUHVFULEHG GHIOHFWLRQ LV GLVFXVVHG LQ &KHUQ Df 7KLV LQYHVWLJDWLRQ LV QRn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n WXUH )XUWKHUPRUH PDQ\ VWUXFWXUHV PXVW EH GHVLJQHG WR DYRLG FHUWDLQ QDWXUDO IUHTXHQFLHV EHFDXVH RI UHVRQDQFH RU VHOILQGXFHG RVFLOODWLRQV WKLV VLWXDWLRQ LQGLFDWHV WKDW WKH QDWXUDO IUHTXHQF\ FRQVWUDLQW LV DOVR DSSURSULDWH 0DQ\ GLIIHUHQW VROXWLRQ VFKHPHV KDYH EHHQ GHYHORSHG ZKLFK DUH XVXDOO\ EDVHG XSRQ JHQHUDO PDWKHPDWLFDO SURJUDPPLQJ WHFKQLTXHV

PAGE 49

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n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n LFDO RUGHU ZLWKRXW DGGLWLRQDO FRPPHQWV ,QWHUHVWHG UHDGHUV DUH UHIHUUHG WR 7XUQHU f =DUJKDPHH f 7XUQHU f 'H 6LOYD f 5XELQ f )R[ DQG .DSRRU f 0F&DUW HW DO f DQG 5XGLVLOO DQG %KDWLD f

PAGE 50

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n PHQWV RYHU ZKLFK D SDUWLFXODU IRUP RI DSSUR[LPDWLRQ RI HLWKHU WKH GLVn SODFHPHQW DQGRU IRUFH LV DVVXPHG $ QXPEHU RI QRGHV FRPPRQ WR RQH RU PRUH HOHPHQW LV SUHVFULEHG FRQWLQXLW\ LV UHTXLUHG WR H[LVW DW WKHVH QRGHV EXW QRW QHFHVVDULO\ HOVHZKHUH $Q HTXLOLEULXP HTXDWLRQ LV GHULYHG IRU HDFK HOHPHQW DQG WKHQ DOO RI WKH LQGLYLGXDO HTXDWLRQV DUH FRPELQHG LQWR D VLQJOH HTXLOLEULXP HTXDWLRQ IRU WKH HQWLUH V\VWHP 7KH UHVXOWn LQJ HTXDWLRQ LV D OLQHDU DOJHEUDLF HTXDWLRQ ZKRVH XQNQRZQV DUH GLVSODFHn PHQWV DQGRU IRUFHV DW WKH QRGHV 2QFH WKH PDWUL[ HTXDWLRQ LV LQYHUWHG WKH QRGDO GLVSODFHPHQWV DQGRU IRUFHV DUH XVHG ZLWK WKH DVVXPHG DSSUR[n LPDWLRQ IRUP WR GHVFULEH WKH VWDWH RI WKH VWUXFWXUH WKURXJKRXW HDFK DQG HYHU\ HOHPHQW DQG KHQFH WKH V\VWHP +HUHDIWHU WKLV PHWKRG LV UHIHUUHG WR DV WKH ILQLWH HOHPHQW PHWKRG 7KH PRVW IUHTXHQW DSSOLFDWLRQ RI WKH ILQLWH HOHPHQW PHWKRG LV WR SUREOHPV KDYLQJ FRPSOLFDWHG ORDGV JHRPHWU\ DQG UHVSRQVH *HQHUDOO\ VSHDNLQJ WKH PHWKRG LV HPSOR\HG ZKHUHYHU WKH SK\VLFDO V\VWHP LV WRR

PAGE 51

FRPSOH[ WR EH GHVFULEHG DGHTXDWHO\ E\ D VLQJOH GLIIHUHQWLDO HTXDWLRQ DQG ERXQGDU\ FRQGLWLRQV )RU D FRPSOHWH WKHRUHWLFDO GHYHORSPHQW RI WKH ILQLWH HOHPHQW PHWKRG DQG QXPHURXV H[DPSOHV VHH =LHQNLHZLF] f :LWK UHVSHFW WR VWUXFWXUDO RSWLPL]DWLRQ WKH PHWKRG LV HPSOR\HG WR VLPSOLI\ WKH SUREOHP WR EH VROYHG 9HU\ OLWWOH KDV EHHQ SXEOLVKHG RQ WKLV VXEMHFW EXW WKH SDSHUV DYDLODEOH FRYHU D ZLGH VSHFWUXP RI WHFK LTXHV )RU H[DPSOH 'XSXLV f FRPELQHV WKH ILQLWH HOHPHQW DQG VHJPHQWZLVHFRQVWDQW PHWKRGV DV DSSOLHG WR PLQLPXP ZHLJKW EHDP GHVLJQ $ VLPLODU DSSOLFDWLRQ WR FROXPQ EXFNOLQJ LV FRQWDLQHG LQ 6LPLWVHV HW DO f $QRWKHU SDSHU :X f LV D VWXG\ RI WZR FODVVLFDO QRQFRQn VHUYDWLYH VWDELOLW\ SUREOHPV $OWKRXJK DGDSWHG WR VWDELOLW\ FRQVLGHUn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f WUHDWV WKH RSWLPDO OD\RXW RI WUXVVHVf§DQ LPSRUWDQW VXEMHFW EXW QRW UHODWHG WR WKH JHQHUDO SUREOHP WR EH FRQVLGHUHG LQ WKLV GLVVHUWDWLRQ ,Q DGGLWLRQ RQO\ HODVWLF VWUXFWXUHV KDYH EHHQ FRQVLGHUHG DOWKRXJK WKHUH DUH QXPHURXV

PAGE 52

SXEOLFDWLRQV RQ RSWLPDO GHVLJQ RI LQHODVWLF VWUXFWXUHV 5HIHUHQFHV WKDW DUH UHSUHVHQWDWLYH RI WKLV VXEMHFW DUH 'UXFNHU DQG 6KLHOG D Ef +X DQG 6KLHOG f 6KLHOG f 3UDJHU DQG 6KLHOG f DQG 0D\HGD DQG 3UDJHU f 2Q FRQVLGHULQJ WKH YDULRXV UHIHUHQFHV PHQWLRQHG DERYH LW ZRXOG DSSHDU WKDW WKHUH DUH WZR SRVVLEOH SLWIDOOV LQ VWUXFWXUDO RSWLPL]DWLRQ WKDW VKRXOG EH DYRLGHG 7KH ILUVW LV WKH FRQIXVLQJ RI PHWKRG RI RSWLn PL]DWLRQ ZLWK WKH VROXWLRQ WHFKQLTXHV HPSOR\HG WR REWDLQ D VROXWLRQ WR WKH UHVXOWLQJ 73%93 ,Q RUGHU WR DYRLG SRVVLEOH HUURUV WKH WZR VKRXOG EH GHDOW ZLWK LQGHSHQGHQWO\ XQOHVV LW LV FOHDUO\ DGYDQWDJHRXV WR UHODWH RQH WR WKH RWKHU %HVLGHV WKLV LW PXVW EH UHFRJQL]HG WKDW DQ\ VROXWLRQ REWDLQHG LV RSWLPDO RQO\ ZLWK UHVSHFW WR WKH JLYHQ FRQGLn WLRQV RI WKH SDUWLFXODU SUREOHP $Q\ FKDQJH LQ WKH SUREOHP VWDWHPHQW LQYDOLGDWHV WKH DSSOLFDELOLW\ RI WKDW VROXWLRQ 7KH FKDQJH PD\ OHDG WR D PRUH GHVLUDEOH VROXWLRQ EXW WKH RULJLQDO VROXWLRQ LV QR OHVV YDOLG 6LPLWVHV f LV DQ H[DPSOH ZKHUH WKLV VLWXDWLRQ LV QRW DFNQRZOHGJHG ,Q WKLV SDSHU WKH WKLFNQHVV RI D WKLQ UHLQIRUFHG FLUFXODU SODWH RI VSHFn LILHG ZHLJKW DQG GLDPHWHU LV GHWHUPLQHG VXFK WKDW WKH DYHUDJH GHIOHFn WLRQ GXH WR D XQLIRUP ORDG LV PLQLPL]HG $Q HDUOLHU SDSHU ZKLFK GLG QRW LQFOXGH VWLIIHQLQJ LV FLWHG ZLWK WKH LPSOLFDWLRQ WKDW WKH RSWLPXP VROXWLRQ IRU WKH XQVWLIIHQHG SODWH LV QRW FRUUHFW 7KH SRLQW PDGH DERYH LV WKDW ERWK RI WKHVH VROXWLRQV DUH RSWLPXP XQGHU WKH UHVSHFWLYH FRQGLn WLRQV RI WKH WZR SUREOHPV 1HLWKHU VROXWLRQ LV PRUH RU OHVV YDOLG WKDQ WKH RWKHU

PAGE 53

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n LFDO PHWKRG WR WKH VROXWLRQ RI WKH QHFHVVDU\ FRQGLWLRQV IURP 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH 303f LV GHWDLOHG LQ 6HFWLRQ 5HVXOWV RI WKLV DSSURDFK DUH VKRZQ WR EH FRQVLVWHQW ZLWK WKH QHFHVVDU\ FRQGLWLRQV JLYHQ LQ 6HFWLRQ WKHVH UHVXOWV SURYLGH D FODULI\LQJ LQVLJKW WR WKH PDWKn HPDWLFDO SURFHVVHV HQWDLOHG LQ WKH PD[LPXP SULQFLSOH DQG DQ H[SOLFLW IRUPXODWLRQ IRU WKH /DJUDQJLDQ PXOWLSOLHU IXQFWLRQV 7KLV H[SOLFLW IRUn PXODWLRQ LV XVHG LQ WKH QH[W VHFWLRQ WR VKRZ WKH QHFHVVDU\ FRQGLWLRQV PD\ WKHQ EH UHJDUGHG DV DQ DOJRULWKP 7KH ILQDO VHFWLRQ FRQWDLQV D EULHI VXPPDU\ RI VROXWLRQ PHWKRGV 7KH PDLQ WKHRUHWLFDO GHYHORSPHQW RI WKH GLVVHUWDWLRQ LV FRQWDLQHG LQ WKH ILUVW WKUHH VHFWLRQV ,W LV ZHOO NQRZQ WKDW WKH SUREOHPV HQFRXQn WHUHG LQ WKH FDOFXOXV RI YDULDWLRQV DUH HTXLYDOHQW WR WKH RSWLPL]DWLRQ RI

PAGE 54

D IXQFWLRQDO LQ WKH VHQVH RI PDWKHPDWLFDO SURJUDPPLQJ SUREOHPVf XQGHU FHUWDLQ UHVWULFWLRQV XSRQ WKH YDULDWLRQV $ JRRG H[SRVLWLRQ RI WKLV LV DYDLODEOH LQ /XHQEHUJHU f :LWK WKLV HTXLYDOHQFH LQ PLQG LW LV QRWHG WKDW WKH 303 LV LWVHOI ZRUGHG DV D FRQVWUDLQHG RSWLPL]DWLRQ SUREn OHP :KHQ WUHDWHG ZLWK ZKDW LV QRUPDOO\ UHJDUGHG DV D QXPHULFDO PHWKRG WKH JUDGLHQW SURMHFWLRQ PHWKRG DQ H[SOLFLW IRUPXODWLRQ RI WKH DWWHQn GDQW /DJUDQJLDQ PXOWLSOLHUV LV REWDLQHG 7KLV IRUP VDWLVILHV DOO RI WKH FDOFXOXV RI YDULDWLRQ QHFHVVDU\ FRQGLWLRQV DQG DOORZV RQH WR XVH WKHP LQ D PRVW VWUDLJKWIRUZDUG IDVKLRQ $V D UHVXOW WKHVH QHFHVVDU\ FRQGLn WLRQV PD\ EH GLUHFWO\ XVHG LQ WKH IRUP RI DQ DOJRULWKP WR REWDLQ D VROXn WLRQ )XUWKHUPRUH LW LV EHOLHYHG WKDW WUHDWLQJ WKH 303 DV D PDWKHPDWn LFDO SURJUDPPLQJ SUREOHP LQ FRQMXQFWLRQ ZLWK WKH JUDGLHQW SURMHFWLRQ PHWKRG KHOSV WR H[SODLQ WKH HIIHFW RI FRPELQHG FRQWUROVWDWH FRQVWUDLQWV XSRQ WKH PD[LPXP SULQFLSOH 3UREOHP 6WDWHPHQW DQG 1HFHVVDU\ &RQGLWLRQV $ JHQHUDO SUREOHP ZKLFK UHSUHVHQWV D ODUJH FODVV RI VWUXFWXUDO RSWLPL]DWLRQ SUREOHPV LV WUHDWHG LQ WKH VHTXHO 7KH IXQFWLRQDO W) / [Xf GW f LV WR EH PLQLPL]HG ZLWK UHVSHFW WR WKH FRQWURO XWf ZKHUH WKH VWDWH [Wf PXVW VDWLVI\ FHUWDLQ ERXQGDU\ FRQGLWLRQV DQG D GLIIHUHQWLDO FRQn VWUDLQW LQ DGGLWLRQ DQ LQHTXDOLW\ FRQVWUDLQW LQYROYLQJ ERWK WKH VWDWH DQG FRQWURO PXVW EH VDWLVILHG )RU

PAGE 55

X7Wf >X Wf XfWf X Wf@ f f§ ] P [7Wf >[A&Wf [Wf AQWf@ f WKH VXEVLGLDU\ FRQGLWLRQV WR PLQLPL]LQJ WKH FRVW IXQFWLRQ DUH [ IB[BXf f 6SHFLILHG %RXQGDU\ &RQGLWLRQV RQ [W]f f I!" [Xf  T f 7HUPLQDO WLPH W LV FRQVLGHUHG WR EH FRQVWDQW DOORZLQJ LW WR EH U XQVSHFLILHG UHTXLUHV RQO\ D VOLJKW PRGLILFDWLRQ WR WKH IROORZLQJ GHULYDWLRQ 7KLV SUREOHP LV D SDUWLFXODU IRUP RI D YHU\ JHQHUDO RQH WUHDWHG E\ +HVWHQHV f +LV UHVXOWV DUH D VHW RI QHFHVVDU\ FRQGLWLRQV ZKLFK PXVW EH VDWLVILHG E\ WKH RSWLPDO VROXWLRQ DQG LQFOXGH WKH PD[LPXP SULQn FLSOH 7R REWDLQ WKH QHFHVVDU\ FRQGLWLRQV WKH LQHTXDOLW\ FRQVWUDLQWV DUH FRQYHUWHG WR HTXDOLW\ FRQVWUDLQWV LQ WKH PDQQHU RI 9DOHQWLQH f 7KHVH FRQVWUDLQWV DQG WKH GLIIHUHQWLDO FRQVWUDLQWV DUH WKHQ DGMRLQHG WR WKH FRVW IXQFWLRQ YLD /DJUDQJLDQ PXOWLSOLHU IXQFWLRQV S Wf DQG SWf $ UHVSHFWLYHO\ A [}+f VaWf ZKHUH WKH VODFN YDULDEOHV DUH GHILQHG VXFK WKDW 7KH V\PERO GHQRWHV LV GHILQHG E\

PAGE 56

VeWf >I!e[Xf @ WS WS W /Q[XfGW S Wf >[I[Xf@ 8 IF) 3eWf >Se[Xf V_Wf@ GW GW ,PSOLHG VXPPDWLRQ FRQYHQWLRQ LV XVHG ZKHQHYHU D YHFWRU IRUPXODWLRQ OHDGV WR SRVVLEOH DPELJXLWLHV LQ ODWHU GHYHORSPHQWV ,QWHJUDWLQJ WKH VHFRQG LQWHJUDO E\ SDUWV JLYHV D UHVXOW WKDW OHDGV WR WKH YDULDWLRQDO +DPLOWRQLDQ >,Y 7 (O GW 'HILQH +[Xef /T[!Kf 3 Wf I[Xf f WKH YDULDWLRQDO +DPLOWRQLDQ DQG + ZKLFK ZLOO LQFOXGH WHUPV DULVLQJ IURP WKH LQHTXDOLW\ FRQVWUDLQW + + RU + e7Wf IB[BXf /T[Xf 3eWf M!L [Xf +HQFH 7 e ) (+ [7S \-O6@ GW :LWK WKH H[FHSWLRQ RI WKH PD[LPXP SULQFLSOH DOO RI +HVWHQHVf QHFHVVDU\ FRQGLWLRQV DUH REWDLQHG IURP WKH UHTXLUHPHQW WKDW WKH ILUVW YDULDWLRQ RI WKH FRVW IXQFWLRQ YDQLVK ,Q WKH IROORZLQJ MF GHVLJQDWHV

PAGE 57

WKH YDULDWLRQ RI B[ D VXEVFULSW YHFWRU GHVLJQDWHV WKH SDUWLDO GHULYDn WLYH ZLWK UHVSHFW WR WKDW YHFWRU ZLWK WKH UHVXOW LWVHOI D FROXPQ YHFWRU 7KXV •V WS WS 7 r >[ + f 7 r X + f§ X \$6e6e@ GW r 7R GHULYH WKH 303 UHTXLUHV DQ H[WHQVLYH PDWKHPDWLFDO GHYHORSPHQW DQG LV QRW LQFOXGHG VLQFH LW FRQWULEXWHV QRWKLQJ WR WKH SUHVHQW GLVFXVVLRQ +RZHYHU WKH QHFHVVDU\ FRQGLWLRQV DUH OLVWHG LQ RUGHU WR EH DYDLODEOH IRU ODWHU UHIHUHQFH [ + I[Xf BSB r S [ W A SM[L W WS 6SHFLILHG %RXQGDU\ &RQGLWLRQV RQ [AWf + MM XA&Wf 9LeWf +A237f A237f f§A +A237f f§f A 7KH RSWLPDO VROXWLRQ PXVW VDWLVW\ WKHVH VL[ FRQGLWLRQV WRJHWKHU ZLWK WKH LQHTXDOLW\ FRQVWUDLQW f

PAGE 58

7KH 303 VWDWHV WKDW DORQJ WKH RSWLPDO WUDMHFWRU\ HDFK LQVWDQW RI WLPH W VWDWH [A Wf DQG DGMRLQW VWDWH M"Wf WUHDWHG DV IL[HG WKH RSWLPDO FRQWURO XASA&Wf LV WKDW DGPLVVLEOH FRQWURO ZKLFK PLQLPL]HV WKH YDULDWLRQDO +DPLOWRQLDQ ,Q WKH SUHVHQW FRQWH[W DGPLVVLELOLW\ UHTXLUHV WKDW BXWf EH SLHFHZLVH FRQWLQXRXV WKH VHW RI DGPLVVLEOH FRQWUROV EHLQJ GHQRWHG E\ 4 +HQFH WKH 303 LQGLFDWHV WKDW 8437Wf $5*0,1 >+[4S7 X Sf@ f MX H IW 1RWLFH WKDW WKH QHFHVVDU\ FRQGLWLRQV VXJJHVW QRWKLQJ DERXW KRZ D VROXWLRQ LV REWDLQHG EXW PHUHO\ LQGLFDWH FHUWDLQ IXQFWLRQDO UHODWLRQn VKLSV WKDW PXVW EH VDWLVILHG +RZHYHU HTXDWLRQ f VHHPV WR LQWLn PDWH WKDW VROXWLRQ RI WKH QHFHVVDU\ FRQGLWLRQ RI 303 LQYROYHV D PDWKHn PDWLFDO SURJUDPPLQJ SUREOHP 0DWKHPDWLFDO 3URJUDPPLQJ *UDGLHQW 3URMHFWLRQ 0HWKRG +DYLQJ VKRZQ WKDW WKH 303 IURP WKH FDOFXOXV RI YDULDWLRQV DSSURDFK WR DQ RSWLPL]DWLRQ SUREOHP PD\ SHUKDSV EH UHODWHG WR D PDWKHn PDWLFDO SURJUDPPLQJ SUREOHP WKH ODWWHU ZLOO EH GLVFXVVHG LQ JHQHUDO WHUPV &RQVLGHU D QRQOLQHDUO\ FRQVWUDLQHG RSWLPL]DWLRQ SUREOHP [RS7 $5*0,1 >)[f@ [ H 4 VXEMHFW WR JM[f M P ZKHUH  GHQRWHV WKH VHW RI DGPLVVLEOH VWDWH FRPSRQHQWV [S L Q DQG WR EH DGPLVVLEOH UHTXLUHV RQO\ WKH VDWLVIDFWLRQ RI WKH P LQHTXDOLWLHV

PAGE 59

1HFHVVDU\ FRQGLWLRQV ZKLFK [ASA PXVW VDWLVI\ DUH JLYHQ LQ WKH .XKQ7XFNHU WKHRUHP Lf FRQVWUDLQWV DUH VDWLVILHG LLf PXOWLSOLHUV H[LVW VXFK WKDW DQG IRU DOO M P LLLf DQG P 9 )[ f = [ Y37 a $ 2EVHUYH WKDW LI GHQRWHV WKH VHW $ RI FRQVWUDLQWV WKH ILUVW WZR FRQGLWLRQV M e ;$ r JM [f DQG $ M =$ JM [f DQG $ JM A237A e r $ $MMb37f r )R[ SS f SUHVHQWV D YHU\ UHDGDEOH SURRI RI WKLV WKHRUHP D PRUH PDWKHPDWLFDO SURRI XVLQJ YHFWRU VSDFH FRQFHSWV LV DYDLODEOH LQ /XHQEHUJHU f 0DQ\ PHWKRGV IRU REWDLQLQJ D QXPHULFDO VROXWLRQ WR WKH QRQOLQHDU SURJUDPPLQJ SUREOHP GHVFULEHG E\ WKH ILUVW WZR HTXDWLRQV RI WKLV VHFWLRQ KDYH EHHQ GHYHORSHG 7KH JUDGLHQW SURMHFWLRQ PHWKRG E\ 5RVHQ f LV XVHG IUHTXHQWO\ LQ VWUXFWXUDO RSWLPL]DWLRQ %DVLF WR WKH PHWKRG LV WKH RUWKRJRQDO SURMHFWLRQ RI WKH FRVW IXQFWLRQ JUDGLHQW RQWR D VXEVSDFH GHILQHG E\ WKH QRUPDO YHFWRUV RI WKH DFWLYH FRQVWUDLQWV $Q LQKHUHQW SDUW RI WKH DOJRULWKP LV WKH FRQFHSW RI D IHDVLEOH XVDEOH GLUHFWLRQ $Q\ GLUHFWLRQ G LV IHDVLEOH LI DQ LQFUHPHQW [ LQ WKDW GLUHFWLRQ LPSURYHV WKH FRVW IXQFWLRQ LH GHFUHDVHV )[f 'LUHFWLRQ G LV VDLG WR EH XVDEOH LI LW DOVR VDWLVILHV WKH FRQVWUDLQWV $V ORQJ DV D IHDVLEOH

PAGE 60

XVDEOH GLUHFWLRQ H[LVWV WKH FRVW IXQFWLRQ PD\ EH LPSURYHG $ FRQVWUDLQHG RSWLPDO VROXWLRQ [ASS RFFXUV DW WKDW SRLQW ZKHUH QR IHDVLEOH GLUHFWLRQ LV DOVR XVDEOH LH DQ\ DWWHPSW WR LPSURYH WKH FRVW YLRODWHV D FRQn VWUDLQW ,Q $SSHQGL[ % WKHVH FRQFHSWV DUH XVHG LQ D FRQFLVH SURRI RI WKH .XKQ7XFNHU FRQGLWLRQV )R[ f GHULYHV WKH PDWUL[ 3 ZKLFK SURMHFWV WKH FRVW IXQFWLRQ JUDGLHQW LQWR WKH VXEVSDFH GHILQHG E\ YHFWRUV QRUPDO WR WKH DFWLYH FRQn VWUDLQWV 7KLV LV HTXLYDOHQW WR VXEWUDFWLQJ DOO FRPSRQHQWV SDUDOOHO WR YHFWRUV WKDW DUH QRUPDO WR VXUIDFHV RI DFWLYH FRQVWUDLQWV IURP WKH QHJDn WLYH JUDGLHQW RI WKH FRVW IXQFWLRQ 5HFDOOLQJ WKH GHILQLWLRQ RI VHW FRQVLGHU U FRQVWUDLQWV WR EH DFWLYH VXFK WKDW A f§ f f QAB` 'HILQH D YHFWRU ZKRVH HOHPHQWV DUH WKH FRUUHVSRQGLQJ QRQ]HUR /DJUDQJLDQ PXOWLSOLHUV DQG DQRWKHU YHFWRU ZKRVH HOHPHQWV DUH WKH DFWLYH FRQVWUDLQW IXQFWLRQV $ >$ $ $ f§ D Df D n U 1 > J J %D tRW D U )URP WKH 1 YHFWRU D PDWUL[ 1 LV LQWURGXFHG HDFK FROXPQ RI ZKLeK LV WKH JUDGLHQW RI DQ DFWLYH FRQVWUDLQW +HQFH 1 LV DQ Q [ Uf PDWUL[ ZKHUH 7 1 1 f§[ DQG 1 1 LM [ >1 @ LM JDL [ L L Q M }f f f f

PAGE 61

:LWK WKHVH GHILQLWLRQV RI $B DQG 1 WKH WKLUG .XKQ7XFNHU FRQGLWLRQ FDQ EH ZULWWHQ DV 9 )[f 1 $B $W DQ\ IHDVLEOH SRLQW [ ZKHUH J[f WKH GLUHFWLRQ ZKLFK EHVW LPSURYHV WKH FRVW IXQFWLRQ LV WKH QHJDWLYH JUDGLHQW RI WKH FRVW ,I WKRVH GLUHFWLRQV ZKLFK OHDG WR FRQVWUDLQW YLRODWLRQV DUH VXEWUDFWHG IURP 9A)[f WKH SURMHFWLRQ PDWUL[ 3 LV REWDLQHG 'LUHFWLRQV FDXVLQJ D FRQVWUDLQW YLRODWLRQ DUH VSHFLILHG E\ WKH JUDGLHQWV RI DFWLYH FRQVWUDLQWV 7 LH WKH FROXPQV RI 1A :KDW LV UHTXLUHG RI 6 WKH SURMHFWLRQ RI WKH JUDGLHQW LV WKDW 6 9 )[ff 17 $ f B[ f§ f§[ f§ ZKHUH $B DUH VFDODU FRHIILFLHQWV WR EH GHWHUPLQHG VXFK WKDW 6 LV RUWKR 7 JRQDO WR HDFK FROXPQ RI 1 RU 7 7 r 1 f 6 a[ 7 :KHQ WKH PDWUL[ HTXLYDOHQW WR LV XVHG WRJHWKHU ZLWK WKH 6 H[SUHVn VLRQ f WKH UHVXOW LV 17Y[)[f 1 ;f VXFK WKDW WKH ZKLFK VDWLVILHV WKLV RUWKRJRQDOLW\ FRQGLWLRQ LV $ 171f 19 )[ff f§ [ f§ f 8QOHVV GHQW WKH DFWLYH ERXQGDU\ VXUIDFH QRUPDOV [f 7 WKH PDWUL[ 1 1f LV QRQVLQJXODU &RQYHUVHO\ DUH OLQHDUO\ GHSHQ LI WKLV PDWUL[ LV

PAGE 62

VLQJXODU WKH DFWLYH FRQVWUDLQWV DUH QRW OLQHDUO\ LQGHSHQGHQW KRZHYHU WKLV LV QRW D FRQGLWLRQ HQFRXQWHUHG LQ PRVW UHDO FDVHV 6XEVWLWXWLRQ RI WKH B; H[SUHVVLRQ LQWR WKH 6 HTXDWLRQ OHDGV GLUHFWO\ WR WKH SURMHFWLRQ PDWUL[ 3 6 3 9 )[f [ f§ ZKHUH 3 1171fB 17 f ZKHUH LV WKH LGHQWLW\ PDWUL[ 7KH GLUHFWLRQ 6 ZKLFK EHVW LPSURYHV WKH FRVW LV JLYHQ LQ WHUPV RI 3 ZKHUH 3 DQG 1 DUH JLYHQ E\ f DQG f ,I QR FRQVWUDLQWV DUH DFWLYH DW D SRLQW WKHQ 1 LV D QXOO PDWUL[ 3 UHGXFHV WR DQ LGHQWLW\ PDWUL[ DQG WKH GLUHFWLRQ RI EHVW LPSURYHPHQW LV FRLQFLGHQW ZLWK WKH GLUHFWLRQ RI VWHHSHVW GHVFHQW ,Q WKH DOJRULWKP DVVRFLDWHG ZLWK WKLV PHWKRG WKH VWDUWLQJ SRLQW PXVW EH D IHDVLEOH SRLQW ZKHUH JA [f IRU DOO M OP 7KH GHVLJQ WKHQ SURFHHGV LQ WKH 6 GLUHFWLRQ XQWLO WKH VROXWLRQ LV VDWLVILHG WR ZLWKLQ D VSHFLILHG SRVLWLRQ WROHUDQFH H 1HFHVVDU\ FRQGLWLRQV JHQHUDOO\ SURJUDPPHG LQ D FRPSXWHU SURJUDP DUH n9 r ( L f $ R M e ;$ $ R M L r$ ,W LV UHDGLO\ VHHQ WKDW IRU 6 3 DQG B$ GHILQHG DV DERYH WKHVH DUH FRPSOHWHO\ HTXLYDOHQW WR WKH .XKQ7XFNHU FRQGLWLRQV

PAGE 63

*UDGLHQW 3UR MHFWLRQ 0HWKRG $SSOLHG WR WKH 0D[LPXP 3ULQFLSOH %DVHG XSRQ WKH SUHFHGLQJ GLVFXVVLRQ WKH VLPLODULW\ EHWZHHQ 303 DQG WKHnPDWKHPDWLFDO SURJUDPPLQJ SUREOHP FDQ EH GLVFXVVHG 7KH PD[LPXP SULQFLSOH VWDWHV WKDW WKH RSWLPDO FRQWURO PLQLPL]HV WKH YDULDWLRQDO +DPLOWRQLDQ ZLWK UHVSHFW WR DOO DGPLVVLEOH BX 2U DW HDFK WLPH W WS PLQLPL]HV +[Xf ZLWK UHVSHFW WR X IRU JLYHQ B[ DQG Mf DQG ZKHUH BWL LV VXEMHFW WR FRQVWUDLQWV S e OT 7UHDWLQJ WKLV DV D PDWKHPDWLFDO SURJUDPPLQJ SUREOHP WKH IROORZLQJ FRUUHVSRQGHQFHV DUH UHFRJQL]HG \ nY BX )[f A +[XMSf JM [f A I!A [Xf $ A S Wf 9 ) [f + f§ 8 6 A + SOXV FRQVWUDLQWV BX &RQWLQXLQJ WR LGHQWLI\ FRUUHVSRQGLQJ TXDQWLWLHV DW HDFK WLPH W OHW GHQRWH WKH VHW RI DFWLYH FRQVWUDLQWV WDNHQ WR EH U LQ QXPEHU $ A DA f f f M 7KHQ 1 ? $7 >$ $ L @ D Df D U DW 37 >S Wf S Wf S Wf@ f§ D Df D U A .P >ALL@ A DM [ / M Q M OP

PAGE 64

)XUWKHUPRUH GHILQH + fS WR EH WKH JUDGLHQW RI + ZLWK UHVSHFW WR X ZKHUH DOO FRPSRQHQWV WKDW FDXVH D FRQVWUDLQW YLRODWLRQ KDYH EHHQ UHPRYHG 6LQFH SURMHFWLRQ PDWUL[ 3 UHPRYHV FRVW IXQFWLRQ JUDGLHQW FRPSRQHQWV WKDW OHDG WR FRQVWUDLQW YLRODWLRQV FRQVLGHU LWV XVH LQ WKH PD[LPXP SULQFLSOH 7 :LWK WKH FRUUHVSRQGHQW WR 1 LGHQWLILHG DV S WKHQ DQG 3 eL 7AfB f§ 8 f§ X f + f 3 + f X 3 ,, )URP WKH .XKQ7XFNHU FRQGLWLRQV WKLV LPSOLHV WKDW DORQJ WKH RSWLPDO WUDMHFWRU\ WAA MM37f i +XfS f 6LPLODUO\ DW HDFK WLPH \Wf 7 L9 L 7 +X f X f§ X f§ IURP ZKLFK LW IROORZV L 7 LK 9 $ 7 +X X f§ X f§ r +X r + S7F^!f X 2U + f Knf X f f

PAGE 65

+HQFH WKH FRQWURO ODZ IURP +HVWHQHVn QHFHVVDU\ FRQGLWLRQV FDQ EH GHULYHG IURP WKH 303 FRQGLWLRQ E\ WUHDWLQJ LW DV D QRQOLQHDUO\ FRQVWUDLQHG PDWKn HPDWLFDO SURJUDPPLQJ SUREOHP :KLOH XVLQJ WKH JUDGLHQW SURMHFWLRQ PHWKRG LQ WKH GHULYDWLRQ LW LV VHHQ WKDW HTXDWLRQ f LV HTXLYDOHQW WR WKH WKLUG .XKQ7XFNHU FRQGLWLRQ 7KH VHFRQG .XKQ7XFNHU FRQGLWLRQ LV LGHQn WLFDO WR +HVWHQHVf QHFHVVDU\ FRQGLWLRQ RQ WKH /DJUDQJLDQ PXOWLSOLHUV XVHG WR DGMRLQ WKH LQHTXDOLW\ FRQVWUDLQWV WR WKH FRVW IXQFWLRQ 6DWLVIDFWLRQ RI WKH LQHTXDOLW\ LV LPSOLHG E\ UHTXLULQJ WKH ILUVW .XKQ7XFNHU FRQGLWLRQ WR EH IXOILOOHG ZKHUH !r B,9 R R f ;MJ&[f S\ Wf-!e[Xf f J[f R f 7KXV E\ WUHDWLQJ WKH VROXWLRQ RI WKH QHFHVVDU\ FRQGLWLRQV RI WKH PD[n LPXP SULQFLSOH DV D SURJUDPPLQJ SUREOHP ZLWK LQHTXDOLW\ FRQVWUDLQWV XVLQJ WKH JUDGLHQW SURMHFWLRQ PDWUL[ DQG E\ UHTXLULQJ VDWLVIDFWLRQ RI WKH .XKQ7XFNHU FRQGLWLRQV DQ H[SOLFLW IRUPXOD IRU +HVWHQHVn /DJUDQJLDQ PXOWLSOLHU IXQFWLRQV KDV EHHQ GHULYHG ,W LV IXUWKHU GHPRQn VWUDWHG WKDW ZLWK WKH \eWf VR GHILQHG VDWLVIDFWLRQ RI WKH H[WUHPXP FRQn WURO ODZ FRQGLWLRQ LV LPSOLHG +RZHYHU EHIRUH WKLV WUHDWPHQW FDQ EH DFFHSWHG DV YDOLG LW PXVW DOVR EH VKRZQ WKDW WKH V\VWHP RI FDQRQLFDO GLIIHUHQWLDO HTXDWLRQV LV XQFKDQJHG &RQVLGHU WKH VWDWH V\VWHP HTXDWLRQV N [ + I[Xf

PAGE 66

ZKHUH + M!7W.[Xf /T[Xf BS7Wf[Xf ,W LV REYLRXV WKDW WKH H[SOLFLW IRUP RI \Wf KDV DEVROXWHO\ QR HIIHFW XSRQ WKH VWDWH V\VWHP HTXDWLRQ H[SUHVVHG LQ FDQRQLFDO IRUP 'HPRQVWUDWLQJ WKDW WKH DGMRLQW V\VWHP HTXDWLRQ LV XQFKDQJHG UHTXLUHV WKH PHWKRG GHVFULEHG E\ %U\VRQ HW DO f &RQVLGHU WKH JHQHUDO SUREOHP RI 6HFWLRQ DJDLQ EXW ZLWK RQO\ WKH GLIIHUHQWLDO FRQVWUDLQWV DGMRLQHG WR WKH FRVW IXQFWLRQ LH 0LQ ^X + [7SfGW` f VXEMHFW WR ^Le[Xf e OT f ZKHUH 7 +[Xf / [Xf e WfIB[Xf f DQG [ I [ Xf f $JDLQ OHW GHQRWH WKH VHW RI LQGLFHV DVVRFLDWHG ZLWK U DFWLYH $ FRQVWUDLQWV DW DQ\ WLPH W WD a ^DL f D ` U  7D e e ,$ \eWf DQG LV GHILQHG DV EHIRUH e >`!f f L! @ U fO f 7KH SUREOHP FDQ WKHQ EH WKRXJKW RI DV PLQLPL]LQJ f VXEMHFW WR M![Xf

PAGE 67

:KLOH RQ WKH FRQVWUDLQW VXUIDFHV GHILQHG E\ WKLV HTXDWLRQ WKH YDULDWLRQV LQ FRQWURO BXWf DQG VWDWH [Wf DUH QRW LQGHSHQGHQW EXW LQVWHDG DUH UHODWHG WKURXJK WKH VXEVLGLDU\ UHTXLUHPHQW WKDW RU [7-!7 [Xf X7I!7[Xf ; f§ f8 f§ f§ f 7KLV LPSRVHV D UHVWULFWLRQ WR WKH DGPLVVLEOH YDULDWLRQV )RU FRVW IXQFn WLRQ f WR EH D PLQLPXP LW LV QHFHVVDU\ WKDW LWV ILUVW YDULDWLRQ YDQLVK LH ) ) f 7 f 7 7 f 7 [ + X + [ e f§ ; f§ X MR [ ,W KDV DOUHDG\ EHHQ VKRZQ WKDW GW f +r + \7L!f X X ZKLFK ZLOO EH XVHG WR DGYDQWDJH VKRUWO\ DIWHU KDYLQJ DGGHG DQG VXE 7 7 WUDFWHG WKH WHUP X B8 -2 IURP WKH LQWHJUDQG RI f M! [ W) &)  [A+ f§ Sf X + f§ [ f§ X •X7S7f X7S7Af f§ FX ,BX GW 5HDUUDQJLQJ WHUPV JLYHV Be R !F IF) IF) [7+ f X7+ \7Mf f 6X7Y7f f§ X A X f§ / X GW

PAGE 68

%XW DQG +HQFH + \7W!f + Y7sf +ff X f§ X X O/7LfX A e [ L ) ) f 7 f f 7 7 R R [ + f X \ GW ,W LV KHUH WKDW WKH UHVWULFWLRQV LPSRVHG E\ WKH DFWLYH FRQVWUDLQWV DUH DSSOLHG IURP f 7 7 7 7 RX E R[ f§ +L f§ f§[ VXFK WKDW [7 + f [7 A / GW W) I I [7+ S !7 \f f§ [ [ f§ GW / $SSO\LQJ (XOHUnV OHPPD IRU DUELWUDU\ YDULDWLRQV LQ WKH VWDWH ZKLFK VDWLVILHV WKH FRQVWUDLQWV +[ f 3 A Yf ZKLFK E\ WKH IROORZLQJ PDQLSXODWLRQV LV VKRZQ WR EH WKH DGMRLQW V\VWHP HTXDWLRQ RI +HVWHQHV  +[ \ f[ + \A + ;

PAGE 69

7KXV WKH H[SOLFLW IRUPXODWLRQ IRU \BWf REWDLQHG E\ DSSO\LQJ WKH JUDGLHQW SURMHFWLRQ PHWKRG WR WKH 303 VDWLVILHV DOO WKH QHFHVVDU\ FRQGLWLRQV RI +HVWHQHV ,W PD\ KDSSHQ WKDW LQ VRPH FDVHV WKH FRQVWUDLQW XSRQ FRQWURO GRHV QRW GHSHQG XSRQ WKH VWDWH ,W FDQ EH VKRZQ WKDW WKH BSWf H[SOLFLW IRUPXODWLRQ LV HTXDOO\ YDOLG LQ WKLV LQVWDQFH %\ H[DPLQDWLRQ RI HTXDWLRQV f WKURXJK f LW FDQ EH YHULILHG WKDW DOO WKH QHFHVn VDU\ FRQGLWLRQV H[FHSW WKH DGMRLQW V\VWHP HTXDWLRQ DUH VDWLVILHG 7R GHPRQVWUDWH WKH ODWWHU UHFDOO WKDW ZKHQ RQ D FRQVWUDLQW ERXQGDU\ WKH ILUVW YDULDWLRQ RI ERWK WKH FRVW IXQFWLRQDO DQG WKH FRQVWUDLQW IXQFWLRQ PXVW YDQLVK 7KDW LV IRU sXf ERWK DQG Mf XA Mf f f§ f§ f§X 7R GHULYH WKH GHVLUHG HTXLYDOHQFH WKH VDPH WHUP PXVW EH DGGHG DQG VXEWUDFWHG IURP WKH LQWHJUDQG RI DV EHIRUH DJDLQ DUULYLQJ DW WKH UHVXOW :KHQ WKH FRQVWUDLQW YDULDWLRQ f LV LQWURGXFHG LQWR WKLV ODVW HTXDWLRQ WKHQ E\ (XOHUnV OHPPD A [7+ 3f [ 7 r7 X r S Sf

PAGE 70

6LQFH LW ZDV VWLSXODWHG WKDW MeXf LV QRW D IXQFWLRQ RI [B WKH HTXDWLRQ PD\ EH ZULWWHQ + \9![  +r f[  7KXV WKH H[SUHVVLRQ IRU BSWf LV YDOLG ZKHQ WKH FRQVWUDLQW LQHTXDOLW\ GHSHQGV RQO\ XSRQ WKH FRQWURO BXWf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f DOVR VDWLVILHV DOO RI WKH RWKHU QHFHVVDU\ FRQGLWLRQV IRU DQ RSWLPDO VROXWLRQ 7KH DELOLW\ WR GHWHUPLQH SAWf H[SOLFLWO\ LQ WHUPV RI SDUDPHWHUV DQG IXQFWLRQV WKDW GHVFULEH WKH SUREOHP VXJJHVWV WKH SRVVLELOLW\ RI FRQYHUWLQJ WKH QHFHVVDU\ FRQGWLRQV RI DQ RSWLPDO VROXWLRQ LQWR DQ DOJRULWKP IRU REWDLQLQJ LW

PAGE 71

(QVXLQJ GLVFXVVLRQ RI WKH DOJRULWKPLF IRUP RI WKH QHFHVVDU\ FRQGLWLRQV FRQWDLQV WKH LPSOLFLW DVVXPSWLRQ WKDW DOO HTXDWLRQV DUH YDOLG DORQJ WKH RSWLPDO WUDMHFWRU\ ,W LV IXUWKHU DVVXPHG WKDW WKH SUREOHP XQGHU FRQVLGHUDWLRQ LV WKDW RQH GHVFULEHG LQ HTXDWLRQV f 7KH DOJRULWKP UHTXLUHV WKDW [Wf DQG eWf EH NQRZQ DW HDFK WLPH W Wf IRU ZKLFK WKH VROXWLRQ SURFHGXUH LV DV IROORZV Lf 8VH 303 RQ WKH YDULDWLRQDO +DPLOWRQLDQ WR GHWHUPLQH DQ RSWLPDO -/ FRQWURO MX Wf LQGHSHQGHQW RI FRQVWUDLQWV XWf $5*0,1 >+[Xef@ (YDOXDWLQJ WKH LQHTXDOLW\ FRQVWUDLQWV ZLWK MD MM UHYHDOV ZKLFK RI WKH e T FRQVWUDLQWV DUH DFWLYH /HW U GHQRWH WKH QXPEHU RI DFWLYH FRQVWUDLQWV DQG ,$ WKH VHW RI LQGLFHV DVVRFLDWHG ZLWK $ WKHP D D D ` $ f U O [!-f e e ,$ I!e[Xf O L D )URP WKLV WKH YHFWRUV ZKRVH HOHPHQWV DUH WKH QRQ]HUR /DJUDQJLDQ PXOWLSOLHUV DQG FRUUHVSRQGLQJ FRQVWUDLQW IXQFWLRQV DUH GHILQHG UHVSHFWLYHO\ DW WKH LQVWDQW RI WLPH W S Wf >X S S @ D D D U [Xf 8 Mf LLf +DYLQJ LGHQWLILHG ZKLFK RI WKH T FRQVWUDLQWV DUH DFWLYH U FRPSRQHQWV RI DUH VSHFLILHG E\ 7KH\ PD\ EH VROYHG E\ XVLQJ WKH LPSOLFLW IXQFWLRQ WKHRUHP ZKLFK UHTXLUHV

PAGE 72

7 Mf WR EH RI UDQN U 7KLV LQ WXUQ UHTXLUHV WKH U FRQVWUDLQWV f§7ZKLFK DUH DFWLYH DW SRLQW B[Wf WR EH OLQHDUO\ LQGHSHQGHQW 7R GHWHUPLQH WKH UHPDLQLQJ PUf FRPSRQHQWV RI XA UHTXLUHV WKDW S EH NQRZQ DW WLPH W EXW K Wf 7 W! 7 + f§ f§ X f§ L X X f§ X f§ 7KLV YDOXH RI BS LV XVHG WR GHWHUPLQH WKH FRQVWUDLQHG +DPLOn WRQLDQ +r + X7$f N LLLf :LWK WKH QRQ]HUR /DJUDQJLDQ PXOWLSOLHUV MM NQRZQ DQG + FRQVHn TXHQWO\ GHILQHG WKH UHPDLQLQJ PUf XQNQRZQ FRPSRQHQWV RI DUH GHWHUPLQHG IURP WKH FRQWURO ODZ IRU WKH FRQVWUDLQHG V\VWHP LH IURP r + LM 2QFH LV FRPSOHWHO\ NQRZQ WKH DGMRLQW V\VWHP HTXDWLRQV DUH GHWHUPLQHG E\ -/ f R e a + [ 7KH SURFHVV RXWOLQHG DERYH WKHQ DOORZV 8T33 WR EH ZULWWHQ DV X4S7 $5*0,1 >+[RS7Xef@ X H6VLQFH WKH BX REWDLQHG LQ WKLV IDVKLRQ VDWLVILHV A [BXf  ZKLFK LV WKH RQO\ UHTXLUHPHQW IRU EHLQJ DGPLVVLEOH +RZHYHU LW PXVW EH UHFDOOHG WKDW WKHVH HTXDWLRQV DUH YDOLG DORQJ WKH RSWLPDO WUDMHFWRU\ LW UHPDLQV WR EH VKRZQ WKDW WKLV DOJRULWKP PD\ EH HPSOR\HG LQ VRPH PDQQHU WR REWDLQ WKDW RSWLPDO WUDMHFWRU\ DQG WR GHPRQVWUDWH WKHLU VDWLVIDFWLRQ DORQJ LW

PAGE 73

6ROXWLRQ 0HWKRGV 1HFHVVDU\ FRQGLWLRQV IURP WKH FDOFXOXV RI YDULDWLRQV SURYLGHV D 7ZR 3RLQW %RXQGDU\ 9DOXH 3UREOHP 73%93f WR EH VROYHG ZKLFK LV LQ JHQHUDO QRQOLQHDU )RU DOO EXW WKH PRVW VLPSOH SUREOHPV QR DQDO\WLFDO VROXWLRQ LV SRVVLEOH DQG LI DQ\ VROXWLRQ LV WR EH REWDLQHG D FRPSXWHU PXVW EH XVHG ZLWK VRPH QXPHULFDO PHWKRG $ GLVFXVVLRQ RI WKH DYDLODEOH PHWKRGV DQG WKHLU UHODWLYH DGYDQWDJHVGLVDGYDQWDJHV LV QRW LQFOXGHG KHUH GXH WR WKH DYDLODELOLW\ RI VXFK GLVFXVVLRQV LQ WKH OLWHUDWXUH HJ %XOORFN f $OO RI WKH PHWKRGV LQYROYH VRPH LWHUDWLYH VFKHPH DQG IRU RSWLPDO FRQWURO FDQ EH VHSDUDWHG LQWR WZR JHQHUDO FDWHJRULHV Lf ,QGLUHFW PHWKRGV 6FKHPHV ZKLFK UHTXLUH DQ LQWLDO JXHVV RI WKH VWDWHnV VROXWLRQ ,Q WKHVH PHWKRGV WKH VWDUWLQJ SRLQW LV DQ LQLWLDO JXHVV RI WKH WLPH KLVWRU\ RI WKH VROXWLRQ 7KH FRQn WURO DVVRFLDWHG ZLWK WKH VROXWLRQ LV D VXEVHTXHQW FDOFXODWLRQ ,WHUDWLRQ FRQWLQXHV XQWLO WKH VWDWH VDWLVILHV VRPH FULWHULRQ FRQQRWLQJ FRQYHUJHQFH WKH ILQDO FRQWURO KLVWRU\ DW FRQYHUn JHQFH LV WKH RSWLPDO FRQWURO LLf 'LUHFW PHWKRGV 6FKHPHV ZKLFK UHTXLUH DQ LQLWLDO JXHVV RI WKH FRQWURO IXQFWLRQ 7KH VWDUWLQJ SRLQW IRU WKHVH PHWKRGV LV DQ LQLWLDO JXHVV RI WKH FRQWURO WLPH KLVWRU\ )RU WKLV FODVV RI PHWKRGV WKH VWDWH DVVRFLDWHG ZLWK WKH FRQWURO LV D VXEVHTXHQW FDOFXODWLRQ ,WHUDWLRQ FRQWLQXHV XQWLO WKH FRQWURO VDWLVILHV VRPH FRQYHUJHQFH FULWHULRQ 7KH PHWKRG RI TXDVLOLQHDUL]DWLRQ ZDV VHOHFWHG EDVHG XSRQ WKH VXFFHVV RI %R\NLQ DQG 6LHUDNRZVNL f LQ DSSO\LQJ LW WR D FRQVWUDLQHG

PAGE 74

VWUXFWXUDO RSWLPL]DWLRQ SUREOHP ([FHOOHQW FRQYHUJHQFH IRU WKHLU SUREOHP WKH FDSDELOLW\ WR KDQGOH QRQOLQHDU V\VWHPV DQG WKH DYDLOn DELOLW\ DV DQ ,%0 6+$5( SURJUDP $%6 48$6, GLFWDWHG LWV VHOHFWLRQ ,Q WKH DSSOLFDWLRQ WR WKH H[DPSOHV LQ &KDSWHUV ,9 DQG 9 WKH SURJUDP UHTXLUHG QR PRGLILFDWLRQ $V D UHVXOW D GHWDLOHG GLVFXVVLRQ RI WKH PHWKRG RI TXDVLOLQHDUL]DWLRQ LV QRW LQFOXGHG 7KH SUREOHP GLVFXVVHG LQ SUHFHGLQJ VHFWLRQV RI WKLV FKDSWHU IDOOV LQWR WKH JHQHUDO FODVV RI SUREOHPV WKDW 48$6, KDQGOHV WKDW LV e,Wf ZLWK WKH ERXQGDU\ FRQGLWLRQ RI WKH IRUP %e
PAGE 75

IRUPXODWLRQ IRU WKH GHVFULSWLRQ RI WKH VWUXFWXUDO V\VWHP 7KLV LV D PHWKRG RI WKH VHFRQG NLQG PHQWLRQHG DERYH 5HVXOWV RI WKH WZR PHWKRGV DUH FRPSDUHG

PAGE 76

&+$37(5 ,9 &21675$,1(' '(6,*1 2) $ &$17,/(9(5 %($0 %(1',1* '8( 72 ,76 2:1 :(,*+7 ,QWURGXFWLRQ $ VWUXFWXUDO RSWLPL]DWLRQ SUREOHP KDV EHHQ VHOHFWHG IRU LWV VLPSOLFLW\ DQG VWDWHG DV DQ RSWLPDO FRQWURO SUREOHP 7KH PD[LPXP SULQn FLSOH LV DSSOLHG JLYLQJ D QRQOLQHDU 73%93 RI WKH 0D\HU W\SH $PRQJ WKH HDUOLHVW H[SRVLWRU\ SDSHUV RQ WKH PD[LPXP SULQFLSOH 5R]RQRU f JLYHV DQ H[FHOOHQW WUHDWPHQW WR D VLPLODU W\SH RI SUREOHP KLV WHFKQLTXH LV XVHG WR REWDLQ ERWK WKH YDULDWLRQDO +DPLOWRQLDQ DQG DGMRLQW YDULDEOH ERXQGDU\ FRQGLWLRQV ,W LV VKRZQ WKDW QR ILQLWH VROXWLRQ H[LVWV IRU WKH VLWXDWLRQ RI XQFRQVWUDLQHG FRQWURO 1XPHULFDO VROXWLRQV IRU FRQVWUDLQHG FRQWURO DUH REWDLQHG E\ WKH PHWKRG RI TXDVLOLQHDUL]DWLRQ &RQVWUDLQWV LQFOXGH ERWK JHRPHWULF OLPLWDWLRQV WR FRQWURO DV ZHOO DV PD[LPXP VWUHVV OLPLWV WKDW EHFRPH PL[HG FRQVWUDLQWV GHSHQGLQJ XSRQ ERWK VWDWH DQG FRQn WURO YDULDEOHV 3UREOHP 6WDWHPHQW $ FDQWLOHYHU EHDP RI YDULDEOH UHFWDQJXODU FURVV VHFWLRQ LV WR EH GHVLJQHG IRU PLQLPXP WLS GHIOHFWLRQ GXH VROHO\ WR LWV RZQ ZHLJKW 7KH PDWHULDO LV VSHFLILHG WR WKH H[WHQW WKDW WKH PRGXOXV ( DQG GHQVLW\ S DUH FRQVWDQWV /HQJWK / LV VSHFLILHG EXW WKH GHVLJQ YDULDEOHV KHLJKW K[f DQG ZLGWK Z[f PD\ EH FKRVHQ LQGHSHQGHQWO\ RI HDFK RWKHU VXEMHFW WR

PAGE 77

KDUG FRQVWUDLQWV XSRQ WKH DOORZDEOH GLPHQVLRQV 7KDW LV D Z[f F f E K[f G ,I <[f GHQRWHV WKH GHIOHFWLRQ RI WKH FHQWHUOLQH WKH SUREOHP LV JLYHQ ( S / DQG WKH FRQVWUDLQWV ILQG K[f DQG Z[f WR PLQLPL]H

PAGE 78

1RWD <[f LV FHQWHUOLQH GHIOHFWLRQ SRVLWLYH GRZQZDUG )LJXUH 6WUXFWXUDO &RQYHQWLRQV

PAGE 79

'HVLJQ YDULDEOHV DQG WKH UHODWHG FRQVWUDLQWV f DUH SXW LQWR GLPHQVLRQOHVV IRUP VXFK WKDW YL EG Y L LG DQG Z[f f§f§ DF Y F 9;f n &G 99 5HSODFLQJ WKH LQGHSHQGHQW YDULDEOH ZLWK D GLPHQVLRQOHVV HTXLYDOHQW DQG XVLQJ WKH FRQWURO FRPSRQHQWV DOORZV WKH JRYHUQLQJ HTXDWLRQ WR EH SXW LQWR D GLPHQVLRQOHVV IRUP )RU [ / aAf X_X"< [Wf X7fX[fG7 f ZKHUH 8Wf Y[Wff :KHQ FRQVWDQW & LV GHILQHG WKH XVXDO NLQHPDWLFDO UHODWLRQVKLSV IRU % D EHDP PD\ EH ZULWWHQ LQ D VLPSOH GLPHQVLRQOHVV IRUP WKDW LV OHW F -/-/rLf % 9 8QLWV /HQJWK % [L FE\ [ f &%< 'HIOHFWLRQ 6ORSH ; &88< [ f &% GW O8
PAGE 80

7KHVH VWDWH FRPSRQHQW GHILQLWLRQV DUH XVHG ZLWK WKH QDWXUDO ERXQGDU\ FRQGLWLRQV WR REWDLQ [A&Of [' )URP f WKH VWDWH FRPSRQHQW GHILQLWLRQV DQG WKH DERYH ERXQGDU\ FRQGLWLRQV LW IROORZV WKDW [ [f [f rX8 ; [ XOX [f [f [Of [A f 7KHVH HTXDWLRQV DQG ERXQGDU\ FRQGLWLRQV DUH XVHG LQ WKH IROORZLQJ VHFWLRQ WR SUHFLVHO\ VWDWH WKH SUREOHP 7KH VROXWLRQ DQG UHVXOWV DUH JLYHQ ODWHU 8QPRGLILHG $SSOLFDWLRQ RI WKH 0D[LPXP 3ULQFLSOH ,Q WHUPV RI WKH VWDWH YDULDEOHV GHILQHG LQ WKH SUHFHGLQJ VHFWLRQ WKH SUREOHP FDQ EH VWDWHG ZLWK PRUH PDWKHPDWLFDO SUHFLVLRQ )LQG XA&Wf $5*0,1 >[Of@ XHLO VXEMHFW WR LfGLIIHUHQWLDO FRQVWUDLQWV LLfNLQHPDWLF ERXQGDU\ FRQGLWLRQV QDWXUDO ERXQGDU\ FRQGLWLRQV [ BI [BXf [A&2f [f [A&Of [A&Of LLLfKDUG JHRPHWULF FRQVWUDLQWV EG 8OWf DF XWf

PAGE 81

$FFRUGLQJ WR WHUPLQRORJ\ LQ WKH FDOFXOXV RI YDULDWLRQV WKLV LV D 0D\HU W\SH SUREOHP $PRQJ WKH HDUO\ SDSHUV FRQFHUQLQJ WKH 303 5R]RQRU f DSSOLHV WKH 303 WR D VLPLODU SUREOHP JLYLQJ D JHRPHWULF LQWHUSUHWDWLRQ WR WKH IXQFWLRQ RI WKH DGMRLQW YDULDEOHV ,Q 5R]RQRUnV SUREOHP WKH FRVW LV D JHQHUDOL]DWLRQ RI WKH RUGLQDU\ 0D\HU SUREOHP LQ WKH VHQVH WKDW WKH FRVW IXQFWLRQ LV D OLQHDU FRPELQDWLRQ RI WKH WHUPLQDO VWDWH FRPSRQHQWV ,W FDQ EH VKRZQ YLD WKH FDOFXOXV RI YDULDWLRQV WKDW WR PLQLPL]H ZKHUH F LV D YHFWRU RI SUHVFULEHG FRQVWDQWV WKH QHFHVVDU\ FRQGLWLRQV DUH + e7WfIB[X Wf [ + I e [ + X e7f [f e WS0OOM eAWSf @ 7 QA237f A237 SWf +[QS7 X eWf 7KDW LV 0LQ>-@ 0D[ >+@ X X

PAGE 82

,I ZH FRQVLGHU WKH FDQWLOHYHU EHDP SUREOHP WKH IRUPV LQ WKH QHFHVVDU\ FRQGLWLRQV DUH F > @7 f§U + SA 3[X_X S[ 3AXA IURP ZKLFK ? [XOX 988 3 R 3OOf 3 3L 3Of S 3XX 3f 3 3 3rf $GMRLQW YDULDEOHV SAWf DQG SWf FDQ EH LQWHJUDWHG E\ LQVSHFWLRQ S Wf 3Wf Wf W  f VXFK WKDW + [ OWf[X_X S[ 3XO8

PAGE 83

:LWK WKLV UHVXOW WKH QHFHVVDU\ FRQGLWLRQ IRU FRQWURO WR PLQLPL]H WLS GHIOHFWLRQ LV +X[ OWf[XAX SX +X OWf[X_8 SX[ $W ILUVW WKLV DSSHDUV WR EH D FRQWUDGLFWLRQ VLQFH WKH WZR HTXDWLRQV FDQ EH VDWLVILHG RQO\ E\ WKH WULYLDO VROXWLRQ EHFDXVH WKH\ KDYH HTXLYDOHQW IRUPV 388 Wf[ 388 OWf[ f )XUWKHU H[DPLQDWLRQ KRZHYHU OHDGV WR WKH FRQFOXVLRQ WKDW ZKHQ WKH FRQWURO LV FRPSOHWHO\ XQFRQVWUDLQHG WKHUH LV QR KRUL]RQWDO WDQJHQW SODQH WR WKH VXUIDFH + +XX"f :KHQ WKH JHRPHWULF FRQVWUDLQWV WR WKH FRQWURO DUH LQFOXGHG D FRQVWUDLQHG PLQLPXP PD\ H[LVW ,I VXFK LV WKH FDVH WKH PD[LPXP YDOXH RI + RFFXUV RQ WKH ERXQGDU\ RI DGPLVVLEOH FRQWURO VSDFH 7R WKDW HQG 303 LV HPSOR\HG DORQJ WKH FRQWURO VSDFH ERXQGDU\ WR GHWHUPLQH 8R37Wf DW HDFK WLPH W %HIRUH GHWDLOLQJ WKLV SURFHGXUH LW LV QHFHVn VDU\ WR ILUVW FRQVLGHU VRPH VWUXFWXUDO DVSHFWV RI WKH SUREOHP %\ GHILQLWLRQ WKH FRQWURO FRPSRQHQWV DUH SRVLWLYH ZKLFK LQ WXUQ LPSOLHV /RDG f§ m r 88 6KHDU ; r r ; ‘3 f§1 0 V ,, R 0RPHQW ; R ; [Of

PAGE 84

)XUWKHUPRUH VLQFH SWf IURP f 3 H S 3f S A 3 3rf 7KLV H[HUFLVH PDNHV LW SRVVLEOH WR XVH WKH LQIRUPDWLRQ RI WKH VHQVH IRU [ DQG SA WR VLPSOLI\ WKH VHDUFK IRU XAA RQ WKH FRQWURO ERXQGDU\ %\ DUUDQJLQJ WKH +DPLOWRQLDQ LQ WKH IROORZLQJ IDVKLRQ + [ SA S>8OX OWf[SX_X@ LW LV REVHUYHG WKDW ERWK WHUPV LQ WKH EUDFNHWHG H[SUHVVLRQ DUH SRVLWLYH 7KLV DQG WKH SA RXWVLGH WKH OHDGLQJ EUDFNHW DOORZV WKH IROORZLQJ HTXLYDOHQFH 0D[ >+@ X H 8 RU A237 ZKHUH rXf +LU >8OX OWf[ SXMX@ XH 8 $5*0,1 >Xf@ X H 8 >8S8S )WfXX" @ DQG )Wf OWf[S f $W HDFK SRVLWLRQ W DORQJ WKH EHDP WKH VWDWH DQG DGMRLQW YDULDEOHV PXVW VDWLVI\ WKH DSSURSULDWH GLIIHUHQWLDO HTXDWLRQV DQG BX LV VSHFLILHG E\ WKH SUHFHGLQJ WKUHH HTXDWLRQV &RQWURO VSDFH ERXQGDU\ 68 LV LOOXVWUDWHG LQ )LJXUH ZKHUH WKH 8S D[LV LV WUHDWHG DV WKH RUGLQDWH VLQFH XAWf DQG XWf FRUUHVSRQG

PAGE 85

EG DF X EG DF )LJXUH $GPLVVLEOH &RQWURO 6SDFH

PAGE 86

WR WKH KHLJKW DQG ZLGWK UHVSHFWLYHO\ RI WKH FURVV VHFWLRQ RI WKH EHDP DW SRVLWLRQ W $ORQJ WKH FRQVWDQW X HGJHV RI 8 OHW X X ZKHUH X KDV WKH F F YDOXH RI HLWKHU EG RU XQLW\ ,I XFXf KDV D PLQLPXP SRLQW G GXf DQG G GX ZKHUH !X X f X X )WfXX F F F X ) WfXX GX F F Ga" GX )WfXX F ] 7KH YDOXH RI ZKLFK VDWLVILHV WKH ILUVW FRQGLWLRQ LV X )WfX F )XUWKHUPRUH LW LV REVHUYHG WKDW RQO\ RQH H[WUHPXP RI Xf H[LVWV DORQJ 8 X DQG WKDW LW LV D PLQLPXP +HQFH HLWKHU X Xff KDV D PLQLPXP F F  RQ WKH FRQVWDQW XA HGJH RU LV PRQRWRQLFDOO\ GHFUHDVLQJLQFUHDVLQJ ,I HLWKHU r r XA DF RU XA WKHQ DORQJ WKH FRQVWDQW X HGJH + KDV LWV PD[LPXP YDOXH DW D FRUQHU RI WKH UHFWDQJXODU 8 2Q WKH RWKHU KDQG LI DF X WKHQ + KDV LWV PD[LPXP YDOXH RQ WKH OLQH X X LQWHULRU WR WKH F HQGSRLQWV

PAGE 87

6LPLODUO\ DORQJ WKH HGJHV RI 8 GHQRWH KDV WKH YDOXH RI HLWKHU DF RU XQLW\ ,I XX f KDV F X ZKHUH X F F D PLQLPXP SRLQW ZKHUH .XUXFf G DQG f§ GX XX )WfXX F F X )WfXMX F F )WfXAX F ,W LV REVHUYHG WKDW XOXFf KDV RQO\ RQH H[WUHPXP DORQJ X XF LW LV D PLQLPXP DQG RFFXUV DW WKH SRLQW ZKHUH 8/ ^ ) Wf X ` 7KXV E\ WKH VDPH DUJXPHQW SRVHG LQ WKH SUHFHGLQJ SDUDJUDSK LI HLWKHU N N XA EG RU XA WKHQ DORQJ WKH FRQVWDQW X" HGJH + KDV LWV PD[LPXP YDOXH DW D FRUQHU RI WKH UHFWDQJXODU 8 :KHUHYHU EG XA + KDV LWV PD[LPXP YDOXH RQ WKH OLQH Xf X LQWHULRU WR WKH HQGSRLQWV F

PAGE 88

2Q WKH EDVLV RI WKHVH DUJXPHQWV WKH IROORZLQJ V\VWHP ZDV VROYHG E\ WKH PHWKRG RI TXDVLOLQHDUL]DWLRQ K ‘  ; &0 ; ,, ;S&2f [X_X [f ,, ; [Of XOX [Of 3XLX 3f a3 3f 237 $5*01 AXLX a OBWf[3AX_XA X H 8 7KH EHDP LV UHSUHVHQWHG E\ LQWHUYDOV FRPSRVLQJ WKH UDQJH W ZKLFK LV VHSDUDWHG E\ PHVK SRLQWV $Q LQLWLDO JXHVV RI WKH VROXWLRQ [Wf DQG SWf LV FKRVHQ LW LV VHOHFWHG WR VDWLVI\ WKH ERXQGn DU\ FRQGLWLRQV 7KLV JXHVV LV QRW D VROXWLRQ DQG GRHV QRW VDWLVI\ WKH GLIIHUHQWLDO HTXDWLRQV 7KH [ DQG  HTXDWLRQV DUH OLQHDUL]HG DERXW WKH LQLWLDO JXHVV WKHQ WKH UHVXOWLQJ OLQHDU 73%93 LV VROYHG WR REWDLQ QHZ [Wf DQG MG Wf IXQFWLRQV ZKLFK PRUH FORVHO\ VDWLVI\ WKH GLIIHUHQWLDO HTXDWLRQV $W HDFK WLPH W FRUUHVSRQGLQJ WR D PHVK SRLQW ,, LV QXPHUn LFDOO\ HYDOXDWHG DORQJ HDFK RI WKH IRXU VWUDLJKW OLQH VHJPHQWV FRPn SRVLQJ 8 WR GHWHUPLQH 7KH SRLQW XAXAf RQ 8 ZKLFK JLYHV +X[SWf LWV PD[LPXP YDOXH LV 7KLV SURFHVV LV UHSHDWHG XQWLO WKH &Wf DQG MfWf LWHUDWH VDWLVILHV WKH GLIIHUHQWLDO HTXDWLRQV WR ZLWKLQ D VSHFLILHG WROHUDQFH 7KH HTXDWLRQV QHFHVVDU\ WR XVH WKH ,%0 SURJUDP DYDLODEOH DUH JLYHQ LQ $SSHQGL[ & LQ WKH IRUP RI D VXEURXWLQH OLVWLQJ

PAGE 89

5HVXOWV *HRPHWULF &RQWURO &RQVWUDLQWV )RU WKH PRVW SDUW QR PDMRU GLIILFXOWLHV ZHUH HQFRXQWHUHG LQ XVLQJ TXDVLOLQHDUL]DWLRQ WR REWDLQ D VROXWLRQ WR WKH VL[WK RUGHU V\Vn WHP GHULYHG LQ WKH SUHYLRXV VHFWLRQ &HUWDLQ SDUDPHWHU YDOXHV GLG HQJHQGHU QXPHULFDO LQVWDELOLW\ 7KHVH FDVHV WKH VRXUFH RI WKH GLIILn FXOW\ DQG LWV FLUFXPYHQWLRQ DUH GLVFXVVHG LQ &KDSWHU 9,, 0RUHRYHU DOO FDOFXODWLRQV ZHUH GRQH LQ GRXEOH SUHFLVLRQ DV QHFHVVLWDWHG E\ PDWUL[ LQYHUVLRQ DFFXUDF\ UHTXLUHPHQWV 3DUDPHWHU YDOXHV VHOHFWHG WR LOOXVWUDWH WKH VROXWLRQ PHWKRG DUH X EG f X DF 7KH PHDVXUH RI HUURU RI VDWLVIDFWLRQ RI WKH GLIIHUHQWLDO HTXDWLRQV LQ WKH 73%93 LV LQ WHUPV RI WKH JHQHUDO V\VWHP G< f§A I<[f < <[f G[ L f§ L Q (5525 0D[ G< I<[fG[_ O O f§ ; f 'HIOHFWLRQ RI D XQLIRUP EHDP GXH WR LWV RZQ ZHLJKW ZDV XVHG WR LQIHU DQ LQLWLDO JXHVV [LKLFK VDWLVILHV DOO ERXQGDU\ FRQGLWLRQV [[Wf W [Wf W [Wf W W f [AWf W 3Wf W 3Wf aW

PAGE 90

:LWK WKHVH VSHFLILHG SDUDPHWHU YDOXHV DQG LQLWLDO JXHVV RI WKH VROXn WLRQ WKH SURJUDP FRQYHUJHG WR D VROXWLRQ LQ ILYH LWHUDWLRQV )URP WKLV UXQ D WROHUDQFH ZDV VHOHFWHG IRU DOO VXEVHTXHQW FDVHV WKH IROORZn LQJ WDEXODWLRQ SURYLGHV WKH GDWD XVHG LQ LWV VHOHFWLRQ ,WHUDWLRQ (5525 7LS 'HIOHFWLRQ &RVWf ; f ; f [ LRf ; f [ f ; B [ f ; ,W LV VHHQ IURP WKHVH WDEXODU GDWD WKDW WKHUH LV OLWWOH LPSURYHPHQW LQ FRVW WLS GHIOHFWLRQf DV D UHVXOW RI WKH ILIWK LWHUDWLRQ )RU WKLV UHDVRQ D YDOXH IRU WKH WROHUDQFH ZDV VHOHFWHG DV [ r ZKLFK FRUUHn VSRQGV WR DERXW VL[ VLJQLILFDQW GLJLWV LQ WKH FRVW IXQFWLRQDO 5HFDOO IURP WKH SUHYLRXV VHFWLRQ WKDW QR XQFRQVWUDLQHG PLQLPXP H[LVWV :LWK WKH FRQWURO ERXQGV LQFOXGHG WKH LQWXLWLYH VROXWLRQ LV RQH LQ ZKLFK WKH FURVVVHFWLRQDO DUHD LV PD[LPXP QHDU WKH URRW DQG UHGXFHV WR D PLQLPXP DW WKH WLS 5HFDOOLQJ WKDW IRU WKH RSWLPDO FRQWURO 8 0D[ >+@ 0LQ >k@ f§237 Q Q BX H  LL H  $ VHTXHQFH RI LOOXVWUDWLRQV LQ )LJXUH GHPRQVWUDWHV WKH ORFDWLRQ RI 8T3A RQ 68 IRU VHYHUDO VWDWLRQV DORQJ WKH EHDP &RQVWDQW FRQWRXUV RI BXf DUH SORWWHG RQ WKH DGPLVVLEOH FRQWURO VSDFH DW ILYH GLVWLQFW SRVLn WLRQV ,I DQ H[WUHPDO SRLQW H[LVWV LQWHULRU WR 8 VRPH OLQHV RI FRQVWDQW rXf FRQWRXUV PXVW EH FORVHG FXUYHV LQ MMVSDFH 7KLV LV LPSRVVLEOH IRU WKLV H[DPSOH

PAGE 91

2 0LQLPXP iXf XRS7 ’ 0D[LPXP £!Xf 'LUHFWLRQ RI ,QFUHDVLQJ Xf )LJXUH &RQWRXU 3ORWV RI Xf DW 9DULRXV 6WDWLRQV $ORQJ WKH %HDP

PAGE 92

7KH ILUVW LOOXVWUDWLRQ LV IRU WKH W FURVV VHFWLRQ QHDU WKH URRW RI WKH EHDP 6LQFH BXAS RFFXUV DW WKH SRLQW RI PLQLPXP FLf WKH RSWLPXP YDOXH IRU ERWK 8S DQG XA LV XQLW\ WKH PD[LPXP DOORZDEOH GLPHQVLRQV IRU ERWK KHLJKW DQG ZLGWK &RQVWDQW FRQWRXU OLQHV LQGLFDWH WKDW Xf LV PDWKHPDWLFDOO\ GHFUHDVLQJ LQ HLWKHU GLUHFWLRQ RI BXVSDFH /LQHV RI FRQVWDQW X DUH DOVR SORWWHG IRU WKH FURVV VHFWLRQ RI W 7KH RSWLPXP FRQWURO KDV WKH PD[LPXP DGPLVVLEOH YDOXH IRU KHLJKW 8S EXW KDV D YDOXH VRPHZKDW OHVV WKDQ XQLW\ +RZHYHU WKHUH DUH VWLOO QR FRQWRXU OLQHV ZKLFK DUH FORVHG FXUYHV $W WKH PLGSRLQW FURVV VHFWLRQ WKH PLQLPXP Xf SRLQW RFFXUV DW 8S DQG XA DF $OWKRXJK WKH VXUIDFH BXf IRUPV D VFRRSOLNH VKDSH WKHUH DUH VWLOO QR FORVHG FXUYH FRQWRXUV DQG KHQFH QR H[WUHPDO LQWHULRU WR DGPLVVLEOH BXVSDFH 7KH QH[W FURVV VHFWLRQ DW ZKLFK +BXf LV GLVSOD\HG RFFXUV DW W 2Q WKLV VHFWLRQ XA LV VWLOO DW LWV ORZHU ERXQG EXW 8S LV QR ORQJHU DW WKH PD[LPXP DOORZDEOH YDOXH RI XQLW\ DV VKRZQ LQ )LJXUH ,Q WKH ODVW RI WKH VHTXHQFH Xf FRQWRXUV IRU WKH FURVV VHFWLRQ DW WKH WLS RI WKH EHDP DUH GLVSOD\HG 7KH SRLQW RI PLQLPXP L!Xf RFFXUV ZKHUH ERWK FRPSRQHQWV RI FRQWURO KDYH WKHLU PLQLPXP DOORZDEOH YDOXHV $JDLQ QR FRQWRXU OLQHV RI FRQVWDQW BXf IRUP D FORVHG FXUYH LQGLFDWLQJ WKH H[LVWHQFH RI DQ LQWHULRU H[WUHPDO SRLQW 7KLV VHTXHQFH RI LOOXVWUDWLRQV LQGLFDWHV WZR WKLQJV )LUVW WKH ODFN RI FORVHG FXUYH FRQWRXU OLQHV RI Xf YHULILHV WKDW H[LVWV RQ 8 :LWK IXUWKHU VWXG\ LW PD\ EH SRVVLEOH WR REWDLQ VRPH FRQGLWLRQ RQ ZKLFK LPSOLHV WKH HTXDWLRQV FRUUHVSRQGLQJ WR f FDQ QHYHU \LHOG D ILQLWH XQFRQVWUDLQHG RSWLPXP 6XFK D FRQGLWLRQ ZRXOG GHILQH WKH FODVV RI VWUXFWXUHV ZKRVH XQFRQVWUDLQHG VROXWLRQ LV WKH

PAGE 93

]HUR YROXPH VROXWLRQ IUHTXHQWO\ GHVFULEHG LQ WKH OLWHUDWXUH RQ VWUXFn WXUDO RSWLPL]DWLRQ 6HFRQGO\ DW W WKH SRLQW RFFXUV DW 7 X >@ WKH SRLQW RI PD[LPXP FURVVVHFWLRQDO DUHD DV W LQFUHDVHV IURP ]HUR WR RQH WKH SRLQW XAA PRYHV DORQJ WKH XA ERXQGDU\ RI 8 WR WKH XA ORZHU ERXQG DQG WKHQ GRZQ WKH XA DF ERXQGDU\ RI 8 WR XA ORZHU ERXQG %\ WKH WLPH W WKH RSWLPDO FURVVVHFWLRQDO DUHD LV WKH PLQLPXP DOORZDEOH DUHD $V D UHVXOW RI WKH SUHVFULEHG IRUP RI 8 LI 84S7Wf IROORZV WKLV SDUWLFXODU SDWK DV W LQFUHDVHV IURP ]HUR WR RQH HDFK FRPSRQHQW RI XAA&Wf KDV LWV RZQ GLVWLQFW UHJLRQ RI WUDQVLWLRQ 7KDW LV DW DQ\ YDOXH RI W LI EG XA WKHQ XA PXVW EH RQ HLWKHU LWV XSSHU RU ORZHU ERXQG &RQYHUVHO\ LI X LV LQ WUDQVLWLRQ ZKHUH DF X WKHQ XA PXVW EH RQ RQH RI LWV ERXQGV 7KLV HIIHFW LV VHHQ PRVW FOHDUO\ LQ )LJXUH ZKHUH \ASS LV GLVSOD\HG IRU WKH H[DPSOH FDVH SDUDPHWHU YDOXHV VSHFLILHG E\ f 7KH SURILOHV DUH GLVSOD\HG RQ D WZRYLHZ GUDZLQJ DV D SODQIRUP RI WKH EHDP PLJKW DSSHDU 6WDWH FRPSRQHQWV FRUUHVSRQGLQJ WR WKLV EHDP DUH VKRZQ LQ )LJXUH UHSUHVHQWLQJ GLPHQVLRQOHVV GHIOHFWLRQ VORSH PRPHQW DQG VKHDU UHVSHFWLYHO\ $V REVHUYHG LQ )LJXUH WKHUH DUH ILYH GLVWLQFW UHJLRQV RI WKH EHDP Lf W 8I L FRQWUROV RQXSSHU ERXQGV 8 f LLf W DF X X WUDQVLWLRQ

PAGE 94

723 9, (OL 6,'( 9,(: )LJXUH 3ODQIRUP 9LHZV RI 2SWLPDO 6ROXWLRQ IRU EG DQG DF

PAGE 95

)LJXUH 6WDWH &RPSRQHQWV RI 2SWLPDO 6ROXWLRQ IRU EG DQG DF

PAGE 96

LLLf W 8 X DF LYf W EG X DF Yf W XA EG X DF RQ XSSHU ERXQG RQ ORZHU ERXQG XA WUDQVLWLRQ FRQWUROV RQ ORZHU ERXQGV 7KH FXUYHV WKDW VKRZ WKH LQWHUFHSW ORFDWLRQV DV D IXQFWLRQ RI SDUDPHWHU YDOXHV EG KDYH EHHQ FDOOHG FRUUHODWLRQ FXUYHV LQ HDUOLHU VWXGLHV OLEHQ WKH ZLGWK LV DOORZHG WR YDU\ DOVR WKH VHFRQG SDUDPHWHU DF LV LQWURGXFHG )RU WKH VDNH RI FRPSDULVRQ WR SUHYLRXV VWXGLHV WKH LQWHUFHSWFRUUHODWLRQ FXUYHV DUH SORWWHG DV GHSHQGHQW XSRQ EG DQG SDUDPHWULF LQ DF +RZHYHU LW ZRXOG EH MXVW DV FRUUHFW WR GR WKH RSSRVLWH ,QWHUFHSW ORFDWLRQ FXUYHV GHVFULEHG DERYH DUH VKRZQ LQ )LJXUH 7KH KHDY\ EODFN FXUYH LV WKH FDVH ZKHUH DF D EHDP RI FRQVWDQW XQLIRUP ZLGWKf§WKH FDVH FLWHG IURP HDUOLHU OLWHUDWXUH $QRWKHU VSHFLDO FDVH LV UHSUHVHQWHG E\ GDVKHG OLQHV FRUUHVSRQGLQJ WR DF ZKLFK LV WKH FDVH FRUUHVSRQGLQJ WR D PLQLPXP DOORZDEOH WKLFNn QHVV HTXDO WR ]HUR 'DVKHG OLQHV DUH XVHG EHFDXVH WKHVH GDWD DUH DQ H[WUDSRODWLRQ FRQYHUJHQFH SUREOHPV HQFRXQWHUHG IRU SDUDPHWHU YDOXHV OHVV WKDQ SUHYHQWHG REWDLQLQJ QXPHULFDO UHVXOWV

PAGE 98

$ GLVFXVVLRQ RQ WKH FRQYHUJHQFH GLIILFXOWLHV H[SHULHQFHG E\ WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP IRU SDUDPHWHU YDOXHV DSSURDFKLQJ ]HUR LV SUHVHQWHG LQ WKH FKDSWHU RQ QXPHULFDO LQVWDELOLWLHV ,Q WKDW GLVn FXVVLRQ LVRODWLRQ RI WKH VRXUFH RI GLIILFXOW\ LV UHSRUWHG LW LV SRVn VLEOH WKDW WKLV GLIILFXOW\ PD\ EH D JHQHUDO UHVXOW DSSOLFDEOH WR DOO SUREOHPV WR EH VROYHG E\ WKH PHWKRG RI TXDVLOLQHDUL]DWLRQ $ VROXn WLRQ IRU WKLV FDVH LV ODWHU REWDLQHG E\ ILQLWH HOHPHQW WHFKQLTXHV 1RWH WKDW VLQFH DF WKHVH WZR FDVHV UHSUHVHQW OLPLWV WR WKH VROXWLRQV RI WKH SUREOHP ,Q DGGLWLRQ LI WKH IRXU LQWHUFHSW ORFDWLRQV DUH SORWWHG YHUVXV DF DQG SDUDPHWULF LQ EG FXUYHV UA DQG UA DSSHDU DV KRUL]RQWDO YHHV ZLWK UA DQG UA OLQHV WKDW DUH QHDUO\ SDUDOOHO ,W LV LQWHUHVWLQJ WR QRWH IURP WKH ILJXUH GHSLFWLQJ WKH VROXn WLRQ RI WKLV FDVH DV D SODQIRUP WKDW LQ WKH FHQWUDO UHJLRQ RI WKH EHDP WKH KHLJKW LV JUHDWHU WKDQ WKH ZLGWK 7KLV UHVXOW FDQ EH DQWLFn LSDWHG VLQFH VXFK D FRQILJXUDWLRQ JLYHV D JUHDWHU EHQGLQJ UHVLVWDQFH SHU XQLW ZHLJKW :LWK IXUWKHU UHIHUHQFH WR )LJXUH WKH WUDQVLWLRQ RI XB Wf LV VHHQ WR EH DOPRVW D OLQHDU WDSHU ZKHUHDV WKH XAWf WUDQVLWLRQ H[KLELWV D PXFK PRUH SURQRXQFHG FXUYDWXUH 7R JHQHUDOL]H IURP WKLV VSHFLILF FDVH RI JLYHQ YDOXHV RI EG DQG DF WR DUELWUDU\ YDOXHV UHTXLUHV WKH LQWURGXFWLRQ RI IRXU TXDQWLWLHV FKDUDFWHUL]LQJ WKH VROXn WLRQ 7KHVH TXDQWLWLHV DUH WKH YDOXHV RI W DW WKH SRLQWV ZKHUH WKH WUDQVLWLRQV LQWHUFHSW WKH ERXQGV RQ XA DQG XA VLQFH W UHSUHVHQWV D QRUPDOL]HG SRVLWLRQ [/ WKHVH TXDQWLWLHV FDQ EH WKRXJKW RI DV DQ LQWHUFHSW ORFDWLRQ H[SUHVVHG DV SHUFHQW RI WKH EHDPnV OHQJWK 7KH\

PAGE 99

DUH GHILQHG ZLWK UHIHUHQFH WR WKH ILYH GLVWLQFW UHJLRQV RI WKH EHDP SUHYLRXVO\ JLYHQ ZKHUH UA GHVLJQDWHV XAWf LQWHUFHSW ZLWK ORZHU ERXQG GHVLJQDWHV XAWf LQWHUFHSW ZLWK XSSHU ERXQG UA GHVLJQDWHV XA&Wf LQWHUFHSW ZLWK ORZHU ERXQG UA GHVLJQDWHV XAWf LQWHUFHSW ZLWK XSSHU ERXQG VXFK WKDW WKH ILYH UHJLRQV DUH FRQWURO RQ XSSHU ERXQGV XA WUDQVLWLRQ Lf  W U F LLf U F W U D LLLf U D W UG FRQWURO RQ XSSHUORZHU ERXQGV LYf UG W UID XA WUDQVLWLRQ Yf U W E FRQWURO RQ ORZHU ERXQGV ,QHTXDOLW\ 6WUHVV &RQVWUDLQWV 7KLV VHFWLRQ WUHDWV DQ LQHTXDOLW\ OLPLW WR DOORZDEOH QRUPDO DQG VKHDU VWUHVVHV DVVRFLDWHG ZLWK EHQGLQJ 8VLQJ WKH RUGLQDU\ VWUHQJWK RI PDWHULDOV IRUPXODWLRQV LW FDQ EH VKRZQ WKDW IRU WKH UHFn WDQJXODU FURVV VHFWLRQ WKHVH FRQVWUDLQWV WDNH WKH IRUP 9[f [f K[f r0$; DQG L 9[f 62UI f [f 70$; -G :KHQ GLPHQVLRQOHVV TXDQWLWLHV DUH LQWURGXFHG A ;88 f0$;
PAGE 100

WKH LQHTXDOLWLHV PD\ EH ZULWWHQ LQ WKH UHTXLUHG IRUP IRU PL[HG FRQn VWUDLQWV LH DV D IXQFWLRQ RI ERWK FRQWURO DQG VWDWH FRPSRQHQWV A[MXf [XAX f ff f ;A88 WT f ZKHUH J0$; G \/ 70$; \/ 7KH WZR VWUHVV FRQVWUDLQWV SODFH UHVWULFWLRQV XSRQ WKH PLQLPXP FURVVVHFWLRQDO GLPHQVLRQV WR NHHS WKH QRUPDO DQG VKHDU VWUHVVHV OHVV WKDQ SUHVFULEHG YDOXHV 6SHFLILFDOO\ IURP WKH FRQVWUDLQWV f DQG f WZR FRQWURO LQHTXDOLWLHV PXVW EH VDWLVILHG DW HDFK VWDWLRQ W DQG WKHVH LQHTXDOLWLHV GHSHQG XSRQ WKH VWDWH RI WKH VWUXFWXUDO V\Vn WHP 7KH LQHTXDOLWLHV DUH XOX ;R 88 97 [Wf IURP ZKLFK FDQ EH GHULYHG ERXQGDU\ DUFV LQ XVSDFH b XODX` [28f 8WXf ;78 r $f %RWK RI WKHVH ERXQGDU\ DUFV DUH K\SHUERODV UHVWULFWHG WR WKH ILUVW TXDGUDQW RI XVSDFH 'HSHQGLQJ XSRQ WKH ORFDWLRQ RI WKH DUFV YLVDYLV WKH UHFWDQJXODU 8 LQFOXVLRQ RI VWUHVV FRQVWUDLQWV KDV RQH RI WKUHH

PAGE 101

HIIHFWV LQ GHWHUPLQLQJ ZKDW BX LV DGPLVVLEOH $W DQ\ VWDWLRQ W LQ VRPH VWUXFWXUDO VWDWH [Wf LfLI rTWT LV WRR VPDOO WKH VWUHVV ERXQGDU\ DUF OLHV HQWLUHO\ DERYH UHFWDQJXODU 8 DOO JHRPHWULFDOO\ DGPLVVLEOH BX YLRODWH WKH VWUHVV FRQVWUDLQWV LLfLI *T7T LV WRR ODUJH WKH VWUHVV ERXQGDU\ DUF OLHV HQWLUHO\ EHORZ UHFWDQJXODU 8 DOO JHRPHWULFDOO\ DGPLVVLEOH MX VDWLVI\ WKH VWUHVV FRQVWUDLQWV LLLf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f WR EH D SDUW RI •8 7KHQ WR ILQG XAS LQ WKH PDQQHU RXWOLQHG LQ 6HFWLRQ &Xf PXVW EH HYDOXDWHG DORQJ XA XA&Af ,I D PLQLPXP H[LVWV DORQJ WKH RUWKRJRQDO SURMHFWLRQ RI XA&XSf RQ WADH pQf VXUIDFH WKHQ 0LQ X .Xf Xf s!Xf A X

PAGE 102

$ORQJ WKH QRUPDO VWUHVV ERXQGDU\ DUF 8DAXAf 88 ;2 8 ;8 6XEVWLWXWLQJ IRU 8 LQnBXf LW IROORZV IURP f WKDW Xf [D )WfDffX/ 6LQFH WKH H[SUHVVLRQ LQ SDUHQWKHVHV LV SRVLWLYH LW LV UHDGLO\ REVHUYHG WKDW WKH YDOXH XA f PLQLPL]HV Xf DORQJ XAALXAf )URP f WKH YDOXH RI 8 DW WKLV SRLQW LQ MDVSDFH LV ]HUR 7KH VDPH UHVXOW H[LVWV IRU !BXf HYDOXDWHG DORQJ XA RQ WKH VKHDU VWUHVV ERXQGDU\ DUF XOX f 97 a f f 99O 6XEVWLWXWLQJ IRU 8 LQ f LW IROORZV IURP f Xf 8O B;Wf B[WfB SWfXL O[ WKDW 6LQFH ERWK WKH H[SUHVVLRQ LQ SDUHQWKHVHV DQG )]Wf DUH SRVLWLYH LW FDQ EH DUJXHG DV DERYH WKDW WKH SRLQW RI PLQLPXP pBXf DORQJ RFFXUV DW WKH SRLQW X Xf f $W DQ\ FURVV VHFWLRQ WKHUH H[LVWV D VLQJOH PLQLPXP RI Xf DORQJ HLWKHU VWUHVV FRQVWUDLQW ERXQGDU\ DUF 6LQFH WKHVH PLQLPD RFFXU DW WKH SRLQW X m!f DV RQH SURFHHGV DORQJ HLWKHU ERXQGDU\ LQ WKH GLUHFWLRQ RI LQFUHDVLQJ XA .BXf LV D PRQRWRQLFDOO\ GHFUHDVLQJ IXQFWLRQ )RU WKLV UHDVRQ DORQJ WKH VWUHVV FRQVWUDLQW ERXQGDU\ DUF LW LV QHFHVVDU\ WR HYDOXDWH Xf DW RQO\ RQH SRLQW 7KDW SRLQW LV WKH LQWHUVHFWLRQ RI WKH VWUHVV ERXQGDU\ FRQVWUDLQW DUF ZLWK WKH UHFWDQJXODU ERXQGDU\ KDYLQJ WKH

PAGE 103

ODUJHU YDOXH RI XA ,W LV SRVVLEOH WKDW WKH WZR VWUHVV ERXQGDULHV LQWHUVHFW ZLWK WKH FRRUGLQDWHV RI WKDW SRLQW JLYHQ E\ (YHQ WKRXJK WKLV DGGV D FRPSOLFDWLRQ WR WKH ERXQGDU\ RI ZKDW LV DGPLVn VLEOH MM WKH PRQRWRQLFDOO\ GHFUHDVLQJ SURSHUW\ RI ERWK VWUHVV ERXQGDU\ DUFV VWLOO UHVXOWV LQ WKH QHFHVVLW\ RI HYDOXDWLQJ Xf DW RQO\ D VLQJOH SRLQW DORQJ WKH VWUHVV ERXQGDU\ DUFV 7R LOOXVWUDWH WKLV )LJXUH GHSLFWV DQ DGPLVVLEOH FRQWURO UHJLRQ GHWHUPLQHG E\ D FRPELQDWLRQ RI JHRPHWULF FRQVWUDLQWV DQG D FRPSRVLWH VWUHVV FRQVWUDLQW ERXQGDU\ WRJHWKHU ZLWK WKH SRLQW DW ZKLFK L!XBf LV D PLQLPXP DORQJ WKH FRPSRVLWH VWUHVV ERXQGDU\ :LWK FRQWLQXHG UHIHUHQFH WR WKH ILJXUH XA LV WKDW SRLQW DORQJ OLQH $%& ZKLFK PLQLPL]HV XBf 5HVXOWV 6WUHVV &RQVWUDLQWV ,QFOXGHG )RU WKH SXUSRVH RI FRPSDULVRQ WKH SDUDPHWHU YDOXHV XVHG LQ WKH FDVH ZLWK RQO\ JHRPHWULF FRQVWUDLQWV ZHUH DOVR XVHG LQ WKH FDVHV ZKHUH VWUHVV LQHTXDOLW\ FRQVWUDLQWV DUH SUHVHQW 7KH UHVXOWV DUH TXLWH VLPSO\ VWDWHG QHLWKHU VWUHVV FRQVWUDLQW LV HYHU DFWLYH ,Q JHQHUDO LW ZDV IRXQG WKDW WKH FULWLFDO VHFWLRQ RFFXUV DW WKH URRW $W WKDW VHFWLRQ WKH JUHDWHVW SRUWLRQ RI JHRPHWULFDOO\ DGPLVVLEOH FRQWURO VSDFH LV SURn KLELWHG E\ WKH VWUHVV FRQVWUDLQWV 7KH IXUWKHU DORQJ WKH EHDP WRZDUGV WKH WLS WKH OHVV WKH SURKLELWHG JHRPHWULFDOO\ DGPLVVLEOH FRQWURO VSDFH

PAGE 104

DF 8 )LJXUH %HKDYLRU RI MMf $ORQJ &RPSRVLWH 6WUHVV %RXQGDU\ 92

PAGE 105

%RWK DQG [A GHFUHDVH PRQRWRQLFDOO\ WR ]HUR DW W DQG LV WKH UHDVRQ IRU WKLV SKHQRPHQRQ (DFK RI WKH SDUDPHWHUV DQG WKXV KDV WKUHH LPSRUWDQW FULWLFDO YDOXHV )RU WKH ILUVW SDLU RI YDOXHV WKH PD[LPXP DOORZDEOH VWUHVVHV DUH VXIILFLHQWO\ ODUJH WKDW DOO JHRPHWULFDOO\ DGPLVVLEOH FRQWURO FRQWDLQHG E\ WKH UHFWDQJXODU 8 VDWLVI\ WKH VWUHVV FRQVWUDLQWV ,Q WKLV FDVH D WT DQG WKH VWUHVV FRQVWUDLQWV LQWHUVHFW WKH ORZHU OHIWKDQG FRUQHU RI 8 DW W DQG OLH EHORZ 8 IRU DOO W )RU WKH UDQJH RI YDOXHV D4 W WKH VWUHVV ERXQGDU\ DUFV LQWHUFHSW WKH XA SRUWLRQ RI UHFWDQJXODU 8 ,I WKH\ DUH H[DFWO\ HTXDO WR WKH ORZHU YDOXHV UHVSHFWLYHO\ ERWK DUFV SDVV WKURXJK WKH SRLQW f DW W DQG OLH EHORZ WKDW SRLQW IRU DOO W +RZHYHU WKHVH YDOXHV DUH QRW DV LPSRUWDQW DV HLWKHU WKH ILUVW SDLU RU WKH QH[W )RU WKH UDQJH RI YDOXHV R4 WT ERWK DUFV LQWHUFHSW WKH XA SRUWLRQ RI UHFWDQJXODU 8 :KHQ HTXDO WR WKH ORZHU YDOXHV ERWK DUFV SDVV WKURXJK WKH XSSHU ULJKWKDQG FRUQHU RI 8 DW W ,I WKH YDOXHV RI

PAGE 106

R4 WT WKHQ DW W WKHUH LV QR FRQWURO ZKLFK VDWLVILHV ERWK JHRPHWULF FRQVWUDLQWV DQG VWUHVV FRQVWUDLQWV 7KH SUREOHP KDV LQFRQVLVWHQW FRQVWUDLQWV VXFK WKDW QR VROXWLRQ LV SRVVLEOH $ VHTXHQFH RI SORWV VKRZLQJ WKH DGPLVVLEOH FRQWURO VSDFH LV VKRZQ LQ )LJXUH IRU VL[ VHSDUDWH VWDWLRQV DORQJ WKH EHDP 7KH YDOXH RI 2T LV IRU WKLV FDVH VXFK WKDW LQ VHFWLRQ W WKH VHW RI DGPLVVLEOH FRQWUROV LV WKH VLQJOH SRLQW MM f 7KDW SRUWLRQ RI WKH JHRPHWULFDOO\ DGPLVVLEOH FRQWURO VSDFH GLVDOORZHG E\ WKH VWUHVV FRQVWUDLQW LV LQGLFDWHG E\ GDVKHG OLQHV $GGLWLRQDOO\ WKH SRLQW FRUn UHVSRQGLQJ WR LV LQFOXGHG IRU HDFK VHFWLRQ %\ REVHUYLQJ WKH VHTXHQFH RI SORWV DV W LQFUHDVHV ERWK WKH VWUHVV ERXQGDU\ DQG DWH VHHQ WR PRYH FORVHU WR WKH RULJLQ RI BXVSDFH +RZHYHU QHYHU RYHUn WDNHV WKH VWUHVV ERXQGDU\ WKXV WKH QRUPDO VWUHVV ERXQGDU\ QHYHU EHFRPHV DFWLYH )LJXUH GHSLFWV WKH VDPH FDVH IRU WKH VKHDULQJ VWUHVV FRQVWUDLQW 7ZR DGGLWLRQDO ILJXUHV DUH LQFOXGHG ZLWK WKH LQWHUPHGLDWH FULWLFDO YDOXHV RI DQG W 7KH QRUPDO VWUHVV FRQVWUDLQW FDVH 2T LV VKRZQ LQ )LJXUH VKHDU VWUHVV FRQVWUDLQW FDVH 7T LQ )LJXUH 1R LOOXVWUDWLRQV DUH SURYLGHG IRU WKH FDVHV DVVRFLDWHG ZLWK WKH ODUJHVW FULWLFDO YDOXHV VLQFH IRU W DOO JHRPHWULFDOO\ DGPLVVLEOH FRQWUROV VDWLVI\ WKH VWUHVV FRQVWUDLQWV

PAGE 107

W W )LJXUH $GPLVVLEOH &RQWURO R R fA237

PAGE 108

W /B W W

PAGE 109

)LJXUH $GPLVVLEOH &RQWURO D R 9' 92

PAGE 110

)LJXUH $GPLVVLEOH &RQWURO WT

PAGE 111

&+$37(5 9 &21675$,1(' '(6,*1 )25 $1 237,0$/ (,*(19$/8( 352%/(0 ,QWURGXFWLRQ $ FROXPQ LV WR EH GHVLJQHG IRU PD[LPXP EXFNOLQJ ORDG LQ RUGHU WR GHPRQVWUDWH Lf DQ DSSOLFDWLRQ RI WKH WKHRU\ GHYHORSHG LQ &KDSWHU ,,, DQG LLf DQ RSWLPDO HLJHQYDOXH SUREOHP 'LIILFXOWLHV SHUWLQHQW WR WKLV FODVV RI SUREOHPV DUH GLVFXVVHG DV ZHOO DV D PHWKRG WR REWDLQ D VROXn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f PDWHULDO PRGXOXV ([f DQG GHQVLW\ S[f XSRQ

PAGE 112

SRVLWLRQ [ DORQJ WKH FROXPQ LV WR EH VXFK WKDW WKH YHUWLFDO HQG ORDG 3 DW EXFNOLQJ LV PD[LPL]HG :HLJKW LV QRW WR EH QHJOHFWHG DQG DOO FURVV VHFWLRQDO VKDSHV DUH VLPLODU ,Q PRUH VSHFLILF WHUPV D JLYHQ ZHLJKW RI PDWHULDO LV WR EH DUUDQJHG ZLWK YDULDEOH SURSHUWLHV DQG JHRPHWU\ VXFK WKDW WKH UHVXOWn LQJ PDVV GLVWULEXWLRQ 0[f DQG VWLIIQHVV GLVWULEXWLRQ 6[f PD[LPL]HV WKH YHUWLFDO WLS ORDG WKH FROXPQ FDQ VXSSRUW ZKHQ WKH HIIHFW RI ZHLJKW LV LQFOXGHG 8SSHU DQG ORZHU ERXQGV WR DOO GHVLJQ YDULDEOHV $[f $\ (O ([f (X f ? 3[f 58 DUH DOVR VSHFLILHG *RYHUQLQJ HTXDWLRQV IRU WKH VWUXFWXUDO V\VWHP DUH GHULYHG LQ WKH QH[W VHFWLRQ 6WUXFWXUDO 6\VWHP ,Q WKH VXEVHTXHQW DQDO\VLV VPDOO GHIOHFWLRQ EHQGLQJ WKHRU\ LV DVVXPHG WKH UHVXOW LV D OLQHDU GLIIHUHQWLDO HTXDWLRQ ZLWK YDULDEOH FRHIILFLHQWV :LWKRXW WKLV DVVXPSWLRQ WKH JRYHUQLQJ HTXDWLRQ LV WKDW RI WKH QRQOLQHDU HO£VWLFD SUREOHP $GGLWLRQDOO\ D[LDO FRPSUHVVLRQ HIIHFWV DUH QHJOHFWHG DV VHFRQG RUGHU /HW <[f GHQRWH WKH GHIOHFWLRQ RI WKH FHQWHUOLQH DW SRVLWLRQ [ DORQJ WKH EHDP DV GHSLFWHG LQ )LJXUH 6LQFH WKH V\VWHP LV FRQVHUYDWLYH WKH JRYHUQLQJ HTXDWLRQV FDQ EH GHULYHG TXLWH VLPSO\ E\ HQHUJ\ WHFKQLTXHV ,Q WHUPV RI VWLIIQHVV 6[f DQG PDVV SHU XQLW OHQJWK WKH HTXDWLRQ DQG ERXQGDU\ FRQGLWLRQV DUH

PAGE 113

9[f )LJXUH 6WUXFWXUDO &RQYHQWLRQV

PAGE 114

6[f
PAGE 115

W [/ Q
PAGE 116

GW VWfQf $LO N Q P&fGFM R Q VWfQf 77 &V &WfQf ;I_ DW W W ZKHUH f ; 3, 9R N 9 (R,R ,QWHJUDWLQJ E\ SDUWV DQG XVLQJ WKH JLYHQ ERXQGDU\ FRQGLWLRQV HTXLYDOHQW IRUPXODWLRQV WR f DUH REWDLQHG ZKLFK DUH XVHG WR FRQYHUW WKH JRYHUQLQJ HTXDWLRQ WR D VWDWH FRPSRQHQW UHSUHVHQWDWLRQ (TXLYDOHQW IRUPXODWLRQV DUH VWfKf ;>QOf QWf@ N P>Q QWf@Ge W VWfKf f§ ;I_ f§ NIL P2Ge $f GW HW GA n ,fGW /HW [ Q 'HIOHFWLRQ • 6ORSH VWfUL 0RPHQW A VWfKf 6KHDU r 6ORSH

PAGE 117

WKHQ IURP WKLV FKRLFH RI VWDWH FRPSRQHQWV DQG WKH ODVW RI WKH HTXDWLRQV f LW IROORZV WKDW  [VWf ; ;; [ NPWf 7R FRPSOHWH WKH WUDQVIRUPDWLRQ RI WKH VWUXFWXUDO SUREOHP WR DQ RSWLPDO FRQWURO SUREOHP WKH GHVLJQ SDUDPHWHUV DUH WUHDWHG DV WKH FRPSRQHQWV RI FRQWURO XWf ZKHUH XWf D[Wff XWf H[Wff XWf U[Wff +HQFH ZLWK WKH GHILQLWLRQ RI FRQWURO \LWf DQG WKH UHSUHVHQWDWLRQ RI WKH V\VWHPfV VWDWH E\ D IRXUGLPHQVLRQDO YHFWRU WKH JRYHUQLQJ HTXDWLRQV DUH ; ; [f [M 8"8 [B [f[ [ NXXf [A&2f [f [Of [AOf $ f 7R FRPSOHWH WKH FRQYHUVLRQ RI WKH VWUXFWXUDO SUREOHP WR DQ RSWLPDO FRQWURO SUREOHP WKH VSHFLILHG ZHLJKW FRQGLWLRQ PXVW EH FRQn VLGHUHG ,Q WHUPV RI WKH VSHFLILF PDVV GLVWULEXWLRQ 0[f WKDW FRQGLn WLRQ LV

PAGE 118

RU / : J 0[fG[ Z S4$ J/ PWfGW 7KLV LQWURGXFHV \HW DQRWKHU GLPHQVLRQOHVV SDUDPHWHU WR WKH V\VWHP LQ WKDW \ X WfX WfGW ZKHUH \ SR$R6/ Z :LWK WKLV ODVW GHYHORSPHQW WKH EXFNOLQJ SUREOHP FDQ EH VWDWHG ZLWK PRUH PDWKHPDWLFDO SUHFLVLRQ )LQG XRS7 $5*0$; >[Of@ [Of $ 8 (  VXEMHFW WR Lf GLIIHUHQWLDO FRQVWUDLQWV [ I[Xf LLf NLQHPDWLF ERXQGDU\ FRQGLWLRQV [f [f QDWXUDO ERXQGDU\ FRQGLWLRQV [f [Gf ` LLLf KDUG JHRPHWULF FRQVWUDLQW D/ 8S&Wf D 8 LYf KDUG PDWHULDO FRQVWUDLQWV H/ XWf H 8 U/  8Wf U 8 Yf VXEVLGLDU\ FRQVWUDLQW \ XO8GW

PAGE 119

$ FXUVRU\ H[DPLQDWLRQ RI WKLV SUREOHP VWDWHPHQW UHYHDOV WKH LQKHUHQW GLIILFXOW\ RI RSWLPDO HLJHQYDOXH SUREOHPV ,I WKH HLJHQYDOXH ; LV WKH FRVW IXQFWLRQDO WKH FRVW IXQFWLRQDO DSSHDUV LQ WKH JRYHUQLQJ HTXDWLRQ DQGRU WKH ERXQGDU\ FRQGLWLRQV )RU WKH VWDWH FRPSRQHQWV VHOHFWHG WKH HLJHQYDOXH ; DSSHDUV RQO\ DV D ERXQGDU\ FRQGLWLRQ +RZHYHU DV VWDWHG LQ f WKH HLJHQYDOXH DSSHDUV LQ ERWK WKH HTXDWLRQ DQG ERXQGDU\ FRQGLWLRQV $QDO\VLV RI WKH 3UREOHP ,Q WKH SUHFHGLQJ VHFWLRQ WKH JRYHUQLQJ HTXDWLRQ IRU FROXPQ EXFNOLQJ LV H[SUHVVHG DV HLWKHU D VHFRQG WKLUG RU IRXUWK RUGHU HTXDn WLRQ ZLWK DSSURSULDWH ERXQGDU\ FRQGLWLRQV 6WDWH FRPSRQHQWV DUH WKHQ GHILQHG IURP VWUXFWXUDO TXDQWLWLHV RQH RI ZKLFK LV WKH VKHDU GLYLGHG E\ WKH VORSH 6LQFH WKH VORSH YDQLVKHV DW W LW PXVW EH VKRZQ WKDW WKLV VWDWH YDULDEOH [AWf LV QRW LQGHWHUPLQDWH DW WKH SRLQW LQ TXHVn WLRQ 0RUHRYHU LQ RUGHU WR XVH WKH WKHRUHWLFDO WHFKQLTXHV GHYHORSHG LQ SUHFHGLQJ FKDSWHUV WKH ILUVW HLJHQYDOXH FRVWf PXVW EH H[SUHVVHG DV VRPH IXQFWLRQDO RI WKH VWDWH DQG GHVLJQ SDUDPHWHUV $FFRUGLQJ WR %RORWLQ S f FRQVHUYDWLYH V\VWHPV DUH GHVFULEHG E\ VHOIDGMRLQW ERXQGDU\ YDOXH SUREOHPV &ODVVLFDO HODVWLF VWDELOLW\ WKHRU\ LV FRQYHQWLRQDOO\ UHVWULFWHG WR FRQVHUYDWLYH SUREOHPV LQ ZKLFK WKH EXFNOLQJ ORDG LV WKH IXQGDPHQWDO HLJHQYDOXH ,W LV VKRZQ LQ ZKDW IROORZV WKDW WKH FROXPQ EXFNOLQJ SUREOHP LV VHOIDGMRLQW DQG WKDW WKH HLJHQYDOXH PD\ EH REWDLQHG YLD WKH WRWDO SRWHQWLDO HQHUJ\ RI WKH V\VWHP IURP WKH 5D\OHLJK TXRWLHQW 6XIILFLHQF\ IRU [AWf WR EH GHWHUn PLQDWH DW W LV WKDW WKH SUREOHP EH D VHOIDGMRLQW HLJHQYDOXH SUREOHP

PAGE 120

&RQVLGHU D GHSHQGHQW YDULDEOH LQ WKH WKLUGRUGHU UHSUHVHQWDn WLRQ RI WKH JRYHUQLQJ HTXDWLRQ f DQG DSSURSULDWH ERXQGDU\ FRQn GLWLRQ ,I ] Q DQRWKHU YDOLG UHSUHVHQWDWLRQ RI WKH V\VWHP LV BGB GW V]f $N I PfGef] ] VWf] W W 8VLQJ WKH QRWDWLRQ RI /RYLWW f ZKR WUHDWV WKH JHQHUDO SUREOHP SXnf T $UfX SXXnf WKH FRUUHVSRQGLQJ TXDQWLWLHV DUH X Q} ] [ A W S[f Q V Wf T[f A N PefG" W U[f A :KHQ ZULWWHQ DV D OLQHDU GLIIHUHQWLDO RSHUDWRU / f DQG LWV DGMRLQW X -8 $ $ / f WKH RSHUDWRU / f LV VDLG WR EH VHOIDGMRLQW LI / f + / f ,I WKH 73%93 LV WR EH VHOIDGMRLQW IRU DQ\ DGPLVVLEOH IXQFWLRQ I! DQG LS WKHQ E\ *UHHQnV )RUPXOD

PAGE 121

,OO E M E ^I!f S/ LSf ` G[ % IfMMf D D ZKHUH %I!L_Mf LV WKH ELOLQHDU FRQFRPLWDQW RI / f ,Q WHUPV RI D JHQHUDO VHFRQG RUGHU HTXDWLRQ /c!f 3MA 33n T $UfM! O } SAf &SAfn T $UfA 7KH QHFHVVDU\ DQG VXIILFLHQW FRQGLWLRQV WKDW / f EH VHOIDGMRLQW LV WKDW S_ ,I / f LV D VHOIDGMRLQW RSHUDWRU DQG WKH ELOLQHDU FRQFRPLWDQW YDQLVKHV WKH SUREOHP LV VDLG WR EH VHOIDGMRLQW )RU WKH JHQHUDO RSHUDWRU E % ! LM2 3[ M! SLSn f D RU LQ WHUPV RI /RYLWWnV HTXDWLRQ RSHUDWRU ZKLFK REYLRXVO\ KDV D VHOIDGMRLQW %! SSn LS WSLSff LW LV VHHQ WKDW WKH SUREOHP LV VHOIDGMRLQW LI WKH ERXQGDU\ FRQGLWLRQV RI WKH DGMRLQW DQG JLYHQ V\VWHPV DUH LGHQWLFDO 7KHUH LV D ZHOOGHYHORSHG WKHRU\ DVVRFLDWHG ZLWK VHOIDGMRLQW HLJHQYDOXH SUREOHPV ZKHUH WKH HLJHQYDOXH SUREOHP LWVHOI LV WKH UHVXOW RI PLQLPL]LQJ DQ HQHUJ\ IXQFWLRQDO $ORQJ WKLV OLQH /RYLWW VKRZV WKDW XQGHU WKH UHTXLUHPHQWV WKDW S[f DQG LV SLHFHZLVH VPRRWK LQ f DQG WKDW T[f LV D ZHOOEHKDYHG IXQFWLRQ ZKLFK LV ILQLWH HYHU\ZKHUH RQ f WKHUH DUH DQ LQILQLWH QXPEHU RI UHDO SRVLWLYH HLJHQYDOXHV ZKLFK PD\ EH DUUDQJHG DV

PAGE 122

$ $ $ Q ,I S[f DQG T[f DUH VSHFLILHG IXQFWLRQV ZKLFK VDWLVI\ WKH DERYH FRQn GLWLRQV WKH IXQGDPHQWDO HLJHQYDOXH LV JLYHQ E\ $ 0LQ >'Xf @ X[f 'Xf ^S[fXnf T[fX` G[ ZKHUH X H & SXXn DQG U[fXG[ /RYLWW VKRZV IRU X H & DQG S[f T[f U[f ERXQGHG WKDW $A LV ILQLWH +LJKHU RUGHUHG HLJHQYDOXHV DUH REWDLQHG E\ XVLQJ WKH RUWKRn JRQDOLW\ SURSHUW\ RI HLJHQIXQFWLRQV DV DGGLWLRQDO VXEVLGLDU\ FRQGL )RU H[DPSOH $ Q 0LQ fQ[! >'X f@ Q 'L2 S[fXAf T [f X ` G[ WLRQV

PAGE 123

IRU X H & Q LO SX Xn Q QOR U[fX G[ Q DQG U[fX XG[ L QO Q 7KH LPPHGLDWH XVH RI WKH VHOIDGMRLQW SURSHUW\ RI WKH FROXPQ EXFNOLQJ SUREOHP LV WKDW WKH IXQGDPHQWDO HLJHQYDOXH LV ILQLWH 7KLV LV XVHG WR SURYH WKDW [Af LV GHWHUPLQDWH 5HFDOO WKH GHILQLWLRQ [ Q )URP WKH WKLUGRUGHU IRUPXODWLRQ RI WKH JRYHUQLQJ HTXDWLRQ f ,G fMM VQf $ N PefG W VXFK WKDW [ f $ N PefGe ZKHUH J LV D SRVLWLYH FRQVWDQW WR EH GHWHUPLQHG %\ VXEVWLWXWLQJ WKH VSHFLILHG ZHLJKW VXEVLGLDU\ FRQGLWLRQ IRU WKH LQWHJUDO [Af $ NSf %RWK N DQG S DUH VSHFLILHG ILQLWH SDUDPHWHUV RI WKH V\VWHP ,I $ LV ILQLWH WKHQ [Af LV DOVR ,Q IDFW WKH EXFNOLQJ ORDG ? LV WKH

PAGE 124

IXQGDPHQWDO HLJHQYDOXH $A ZKLFK LV VKRZQ E\ /RYLWW WR EH ILQLWH IRU WKH FRQGLWLRQV SUHVFULEHG +HQFH [Af LV GHWHUPLQDWH ,Q WKH OLWHUDWXUH RI FODVVLFDO HODVWLF VWDELOLW\ SUREOHPV WKH HLJHQYDOXH WR EH RSWLPL]HG LV H[SUHVVHG DV D 5D\OHLJK TXRWLHQW 7KLV TXRWLHQW FDQ EH IRXQG IURP WKH WRWDO SRWHQWLDO HQHUJ\ )RU WKH JHQHUDO HTXDWLRQ RI /RYLWW WKH TXRWLHQW LV LSXnf TX` G[ UXG[ RU LQ WHUPV RI WKH FROXPQ EXFNOLQJ SDUDPHWHUV DQG FRPSRQHQWV f [VWf N PefGe [` GW W $BBBB B [ GW 2Q UHYHUVLQJ WKH RUGHU RI LQWHJUDLRQ IRU WKH GRXEOH LQWHJUDO FRQWDLQHG LQ WKH QXPHUDWRU DQG LQWURGXFLQJ GHVLJQ SDUDPHWHUV WKH 5D\OHLJK TXRWLHQW EHFRPHV W [XX NX X [efGe` GW $ 4 L GW 7KH QRUPDOL]DWLRQ FRQGLWLRQ RI /RYLWW UHTXLUHV WKDW WKH TXRWLHQWnV GHQRPLQDWRU HTXDOV XQLW\ IRU DOO DGPLVVLEOH VROXWLRQV RU [WfGW

PAGE 125

7KLV SDUWLFXODU QRUPDOL]DWLRQ LV VHOHFWHG LQ RUGHU WR REWDLQ WKH VLPSOHVW H[SUHVVLRQ IRU WKH IXQFWLRQDO $ $OVR UHFDOO WKDW [Af ; NSf DQG [AOf $ ,W LV VHHQ IURP WKHVH ERXQGDU\ FRQGLWLRQV DQG WKH GHILQLWLRQ [A NXAXA WKDW WKH VXEVLGLDU\ ZHLJKW FRQGLWLRQ FDQ EH ZULWWHQ DV D PL[HG ERXQGDU\ FRQGLWLRQ RQ [AWf [AOf [Af NS 7KLV WRJHWKHU ZLWK FRQYHUWLQJ WKH QRUPDOL]DWLRQ FRQGLWLRQ WR DQRWKHU VWDWH FRPSRQHQW ZLWK ERXQGDU\ FRQGLWLRQV VSHFLILHG DW ERWK HQGV WUDQVIRUPV WKH SUREOHP WR A0D[ 0D[ ^->X@` X H 4 >X@ ^[XX N8/X[` GW VXEMHFW WR  ; [f  [8X [f ; ;; [f r NXO [Lf [f ; r [f [Of NS

PAGE 126

ZKHUH BX H IW PHDQV WKDW MM e & DQG VDWLVILHV WKH JHRPHWULF FRQWURO LQHTXDOLWLHV )RU UHDVRQV DV \HW XQGHWHUPLQHG WKH TXDVLOLQHDUL]Dn WLRQ DOJRULWKP GLG QRW FRQYHUJH ZKHQ WKH ODWWHU WZR VWDWH ERXQGDU\ FRQGLWLRQV ZHUH XVHG HYHQ WKRXJK WKH\ EHORQJ WR WKH FODVV RI SUREOHPV WR ZKLFK WKH DOJRULWKP LV WKHRUHWLFDOO\ DSSOLFDEOH 7KLV PLJKW EH GXH VROHO\ WR WKH PL[HG ERXQGDU\ FRQGLWLRQ EHLQJ GHSHQGHQW RQ WKH YDOXH RI WKH FRVW IXQFWLRQDO EHLQJ PLQLPL]HG 2Q IDLOLQJ WR REWDLQ D VROXWLRQ E\ PLQLPL]DWLRQ RI WKH 5D\OHLJK TXRWLHQW WKH QH[W DWWHPSW ZDV VXJJHVWHG E\ WKH VWDWH ERXQGDU\ FRQGLn WLRQV )URP WKH FRQGLWLRQ [Of $ DQG GHILQLWLRQ RI $ ZKHUHE\ $ LW IROORZV WKDW 0LQ >[Of@ 0D[ >$@ X X 7KXV WKH SUREOHP FDQ EH FDVW DV D 0D\HU W\SH RI RSWLPDO FRQWURO SUREOHP 9[ >[f XHLO ; ; [ ;A88 ; ;; ; NXO8 [f [f [Of ; f e ZKHUH LV D SRVLWLYH FRQVWDQW WR EH GHWHUPLQHG 7KH TXDVLOLQHDUL]Dn WLRQ DOJRULWKP GLG QRW FRQYHUJH IRU WKLV IRUPXODWLRQ RI WKH SUREOHP DOVR DJDLQ IRU XQGHWHUPLQHG UHDVRQV

PAGE 127

&RQVLGHU IURP WKH FKRLFH RI VWDWH YDULDEOHV WKDW [ NPWf DQG [f $ ,Q DQ HDUOLHU GHULYDWLRQ J ZDV GHILQHG VXFK WKDW [f $ N PWf GW J ,Q DQRWKHU IRUP % N M PWf GW $ &RQVWDQWf :HLJKWf /RDGf RU LQ WHUPV RI WKH ZHLJKW $ % N: 7KHUHIRUH LI WKH ZHLJKW : LV PLQLPL]HG VXEMHFW WR FRQVWUDLQWV WKLV LV WKH VDPH DV PD[LPL]LQJ $ WKH ORDG VXEMHFW WR WKH VDPH FRQVWUDLQWV 1RWLFH WKDW FRQVWDQW % LV DQ XQGHWHUPLQHG ERXQGDU\ FRQGLWLRQ LH D IUHH FRQGLWLRQ VR WKDW WKH FRUUHVSRQGLQJ DGMRLQW YDULDEOHnV ERXQGDU\ FRQGLWLRQ LV NQRZQ $W WKLV SRLQW WKH UHVXOWV RI D GRFWRUDO GLVVHUWDWLRQf§6DOLQDV ff§ZHUH XVHG LQ DQRWKHU IRUPXODWLRQ RI WKH VDPH SUREOHP 6DOLQDV FRQVLGHUHG WKH DSSOLFDWLRQ RI WKH HQHUJ\ PHWKRG WR VHOIDGMRLQW SUREn OHPV +H VKRZHG WKDW WKH IROORZLQJ SUREOHPV DUH LGHQWLFDO Lf 0D[LPXP ORDG VSHFLILHG ZHLJKW XA $5*0$; ^ -S > X @ ` 8 e IW -S WKH EXFNOLQJ ORDG FRVW IXQFWLRQDO D FRQVWDQW WKH ZHLJKW Z

PAGE 128

LLf 0LQLPXP ZHLJKW VSHFLILHG ORDG 8T37 $5*0,1 ^-Z >X@` OL H  -Z WKH ZHLJKW FRVW IXQFWLRQDO -S D FRQVWDQW WKH ORDG %RWK SUREOHPV DUH VXEMHFW WR WKH VDPH GLIIHUHQWLDO FRQVWUDLQWV ERXQGDU\ FRQGLWLRQV DQG FRQWURO DGPLVVLELOLW\ FRQVWUDLQWV $ FRQFLVH ZD\ RI VWDWLQJ WKH HTXLYDOHQFH LV X $5*0$; ^>X@ VSHFLILHG` f§237 B U f§ : X H IW $5*0,1 >MM@ -S VSHFLILHG` BX H 7KH HTXLYDOHQFH RI WKHVH WZR SUREOHPV LV XVHG WR VWDWH WKH FROXPQ EXFNOLQJ SUREOHP DV XA $5*0,1 >-Z@ XH% -: 88 GW ; n [ ;Of [ [n fOf XLUXU [f ; ;; f [ NXX [f ; V ZKHUH WKH VXEVFULSW V RQ WKH HLJHQYDOXH LQGLFDWHV D VSHFLILHG ORDG :LWK UHIHUHQFH WR WKH VWDWHPHQW RI WKH SUREOHP DV D 0D\HU W\SH QRWLFH WKDW

PAGE 129

-: N [AA 0LQ W[AOf@ X e IL ZLWK [Af D IUHH ERXQGDU\ FRQGLWLRQ )XUWKHUPRUH VLQFH [AOf ; DQG ; REVHUYH WKDW I -Z -S $GPLVVLELOLW\ RI FRQWURO X H 6f UHTXLUHV WKDW BXWf LV SLHFHZLVH FRQWLQXRXV WKURXJK WKH VHFRQG GHULYDWLYH DQG WKDW WKH FRQWURO LQHTXDOn LW\ FRQVWUDLQWV DUH QRW YLRODWHG )RU HYHU\ YDOXH ;J DVVLJQHG D SULRUL WR WKH HLJHQYDOXH ; WKHUH LV D VROXWLRQ YO ;WM f WKDW PLQLPL]HV WKH ZHLJKW ZLWK UHVSHFW WR X H IW DQG VDWLVILHV WKH HLJHQYDOXH SUREOHP +RZHYHU DQ DUELWUDU\ ;J QHHG QRW VDWLVI\ WKH VSHFLILHG ZHLJKW FRQGLn WLRQ DQG LQ JHQHUDO GRHV QRW 7R IXOILOO WKH JLYHQ WRWDO ZHLJKW UHTXLUHn PHQW WKH YDOXH RI ;A ZDV DGMXVWHG VXFK WKDW -Z b X X GW 237 :KHQ WKH FRVW LV QRW RQO\ D PLQLPXP EXW LV DOVR HTXDO WR XQLW\ DOO FRQGLWLRQV RI WKH EXFNOLQJ SUREOHP DV VWDWHG LQ VHFWLRQ DUH VDWLVn ILHG %\ 6DOLQDVn WKHRUHP WKH ;J FDXVLQJ WKLV VDWLVIDFWLRQ LV WKH PD[LPXP ORZHVW HLJHQYDOXH LH WKH PD[LPXP EXFNOLQJ ORDG 7KLV HIIHFW LV LOOXVWUDWHG LQ )LJXUH DQG LV PDWKHPDWLFDOO\ VWDWHG E\ 0D[ ^9AZ 237 : >AZ ` 237 $5*0,1 ^-77>X@ ; ; ` : f§ V 8eLO ZKHUH

PAGE 130

)LJXUH 5HODWLRQVKLS %HWZHHQ 2SWLPDO (LJHQYDOXH DQG 0LQLPXP :HLJKW &RVW )XQFWLRQDO

PAGE 131

$SSOLFDWLRQ RI WKH 0D[LPXP 3ULQFLSOH ,Q WKH SUHFHGLQJ DQDO\VLV GLIILFXOWLHV DVVRFLDWHG ZLWK WKH RSWLPDO HLJHQYDOXH SUREOHP DUH GLVFXVVHG &LUFXPYHQWLRQ RI VXFK GLIn ILFXOWLHV LV GHVFULEHG IRU WKRVH SUREOHPV WKDW DUH VHOIDGMRLQW ,W LV VKRZQ WKDW WKH FROXPQ EXFNOLQJ EHORQJV WR WKDW FODVV RI SUREn OHPV +RZHYHU EHIRUH WKH WKHRUHWLFDO GHYHORSPHQWV SUHVHQWHG LQ WKH WKLUG FKDSWHU PD\ EH DSSOLHG WKH SDUWLFXODU IRUPV RI SDUDPHWHUV DQG IXQFWLRQV PXVW EH LGHQWLILHG $FFRUGLQJO\ IRU WKH RSWLPL]DWLRQ RI WKH FROXPQ EXFNOLQJ ORDG [ A [ [f [f [f f§ M 7 A -: 9r!XfGW /[Xf SXLX bf [f f %RXQGDU\ &RQGLWLRQV V

PAGE 132

DAf XMAf I!e[!Xf b a AH8a8A8OBH/f U8a8f 9U/f ZKHUH WKH PD[LPXP YDOXH RI WKH $V LV WKDW YDOXH VXFK WKDW -: S X X GW f %DVHG XSRQ WKHVH SDUWLFXODU IRUPV IRU WKH JHQHUDO SDUDPHWHUV WKH YDULDWLRQDO +DPLOWRQLDQ LV + 3;A88 3;; 3NXOXA SXOX N $GMRLQW YDULDEOHV eWf DUH GHILQHG IURP WKH +DPLOWRQLDQ + ZKLFK FRQ WDLQV WKH FRQVWUDLQW IXQFWLRQV 6LQFH WKH FRQVWUDLQW IXQFWLRQV M!A DUH LQGHSHQGHQW RI [ IRU WKH FDVH RI RQO\ JHRPHWULF FRQVWUDLQWV S f§ +r + [ [ )URP WKLV UHVXOW DQG IURP WKH VWDWH ERXQGDU\ FRQGLWLRQV LW IROORZV WKDW 3 R 3 BSL[ f 3; Sf 3 S[_X 3f 3A a3; 3f R 9DULDEOH SAWf PD\ EH LQWHJUDWHG E\ LQVSHFWLRQ SAWf IRU DOO W ZKLFK LPPHGLDWHO\ UHGXFHV WKH QXPEHU RI DGMRLQW YDULDEOHV WR EH GHWHUn PLQHG DQG VLPSOLILHV WKH +DPLOWRQLDQ DV ZHOO 7KXV

PAGE 133

+ ^S[[ S[X8/ NSAW2XAX ` ZKHUH S S[ 3Of S SXX 3f 3 3[ 3rf 2SWLPDO FRQWURO DV GHWHUPLQHG E\ WKH 303 VWDWHG LQ f LV X4S7 $5*0,1 >+[S7XSf@ f /O e 4 ZKHUH WKH DGPLVVLELOLW\ UHVWULFWLRQ LQ 3RQWU\DJLQnV GHULYDWLRQ UHTXLUHV RQO\ WKDW BXWf EH SLHFHZLVH FRQWLQXRXV $SSO\LQJ 6DOLQDVn WKHRUHP IXUWKHU UHVWULFWV WKH FRQWURO WR BX H & &RQVLGHULQJ IRU WKH PRPHQW WKDW QR IXUWKHU FRQVWUDLQWV RQ WKH FRQWURO H[LVW WKH XQFRQn VWUDLQHG 843S LV WKDW FKRLFH RI XSWf XWf XWff ZKLFK PLQLPL]HV WKH YDULDWLRQDO +DPLOWRQLDQ + 5HFRJQL]LQJ WKDW WKH SK\VLFDO TXDQWLWLHV UHSUHVHQWHG E\ X KDYH PHDQLQJ RQO\ LI WKH\ DUH SRVLWLYH DVVXPH WKLV WR EH WKH FDVH :LWK QRQQHJDWLYH FRQWUROV WKH VWDWH HTXDWLRQV DQG ERXQGDU\ FRQGLWLRQV MRLQWO\ LPSO\ WKDW [Wf L [Wf )XUWKHUPRUH E\ PHDQV RI f DQG WKH H[SUHVVLRQ IRU + 3;f ! X S[f r X m!

PAGE 134

DQG NSA\f r NSA\f &RQWURO X UHPDLQV WR EH GHWHUPLQHG $VVXPLQJ WKDW X!X DUH NQRZQ TXDQWLWLHV WKHQ XA PXVW EH FKRVHQ LQGHSHQGHQWO\ WR PLQLPL]H WKH YDULDWLRQDO +DPLOWRQLDQ ,Q WKDW FRQWH[W +X ^S[X_X NSA\fXA` + ^SB[BXX ` XAXA fN 7KH XQFRQVWUDLQHG RSWLPDO FRQWURO YDULDEOH LV GHILQHG WR VDWLVI\ WKH ILUVW FRQGLWLRQ WKH VHFRQG FRQGLWLRQ LQGLFDWHV WKDW SWf PXVW EH QHJDn WLYH IRU +&;TSA8TSA3f WR EH D PLQLPXP 6R r X S[ D ANSBSA88 DQG 3Wf ZKLFK OHDGV WR WKH IROORZLQJ DUJXPHQW RQ WKH VLJQV RI SWf S ‘r 3Wf DQG 3Of 3Wf S S [ DQG [AWf 3 A 3 3; DQG [Wf 3AWf S DQG SAf

PAGE 135

,W LV QRWHG WKDW ZLWK WKHVH VLJQV LI NSA\f WKH RSWLPDO XQFRQn VWUDLQHG FRQWURO LV U XL ‘A237 A 9 \ N ZKHUH WKH H[LVWHQFH RI XA LV GLVFXVVHG LQ WKH QH[W VHFWLRQ :KDW LV LPSRUWDQW LV WKDW WKH VROXWLRQ UHTXLUHV D PDWHULDO RI LQILQLWH VWUHQJWK DQG ]HUR GHQVLW\ 6RPH VWUXFWXUDO DQDO\VWV FRPPRQO\ UHIHU WR WKLV K\SRWKHWLFDO PDWHULDO DV %RORJQLXP ,Q WKH FRXUVH RI DQ DFWXDO GHVLJQ SURFHVV FRQWURO ERXQGV FRUUHn VSRQGLQJ WR WKH FRQVWUDLQW IXQFWLRQV DUH VSHFLILHG 2Q WKH EDVLV RI WKH SUHFHGLQJ DQDO\VLV ZLWK JLYHQ ERXQGV RQ FRQWUROV RSWLPDO XAWf DQG XAWf DUH EDQJEDQJ FRQWUROV GHILQHG E\ XWf bH-HXf bH/HXf 6*1 S[f f XWf bU,UXf bU/aUX! 6*1 NSA\f f ZKHUH IRU DQ DUELWUDU\ DUJXPHQW ] RI IXQFWLRQ 6*1 f UO 6*1 ]f ] ] ] N 'HSHQGLQJ XSRQ WKH PDJQLWXGH RI X YLV£YLV D DQG D WKHUH DUH WKUHH /L 8 SRVVLELOLWLHV IRU P DW HDFK SRLQW WM[Wf eWff

PAGE 136

A237 ‘ f LLf D/ Xn DX ^X 7KDW FODVV RI SUREOHPV KDYLQJ QR PL[HG FRQVWUDLQWV UHTXLUHV FRPSXWLQJ WKH /DJUDQJLDQ PXOWLSOLHU IXQFWLRQV MMWf RI f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f [Xf

PAGE 137

(DFK FRQVWUDLQW !e[!B\f ZKLFK LV DFWLYH VXFK WKDW I!e [Xf LV VROYHG IRU D FRPSRQHQW RI FRQWURO 7KH UHVXOWLQJ QRQ]HUR PXOWLn SOLHU IXQFWLRQ S" Wf PXVW EH GHWHUPLQHG IURP WKH QHFHVVDU\ FRQGLWLRQ r + BX 7KLV SAWf LV XVHG LQ WXUQ WR GHWHUPLQH WKH HIIHFW XSRQ WKH V\VWHP IURP WKH VROXWLRQnV O\LQJ RQ D PL[HG FRQVWUDLQW VXUIDFH ZKLFK E\ GHILQLWLRQ GHSHQGV XSRQ WKH VWDWH 7KLV HIIHFW LV PDQLIHVWHG E\ WKH -/ $ DGMRLQW GLIIHUHQWLDO HTXDWLRQ  ZKLFK FDQ EH ZULWWHQ DV L ? ,I FRQVWUDLQWV M! GHSHQG RQO\ XSRQ WKH FRQWURO WKH VHFRQG WHUP LV 7 XQQHFHVVDU\ EHFDXVH A 6XFK LV WKH FDVH RI WKH FROXPQ EXFNOLQJ SUREOHP SRVVHVVLQJ RQO\ FRQWURO ERXQGV )RU WKLV SUREOHP ZKHQHYHU WKH FRQVWUDLQWV DUH DFWLYH WKH PXOWLSOLHU IXQFWLRQV GHWHUPLQHG E\ WKH DERYH SURFHGXUH DUH 3Wf 3[X_X NS\fX f &D/ 2 3Wf 3;88 X HXH/f 3Wf NSAXfX X U U f X / f f f

PAGE 138

7KHVH VDPH SDUDPHWHUV ZLOO QRZ EH GHWHUPLQHG IURP WKH H[SOLFLW IRUPXODn WLRQ RI MMWf GHULYHG LQ &KDSWHU ,,, 5HFDOO WKDW ZKHUH GHQRWHV WKH VHW RI DFWLYH FRQVWUDLQWV ZKLFK $ DUH U LQ QXPEHU ,$ DA RA A` I! ^L! ` D Df D U >Y X L L P M ffU DQG WKDW LQ WHUPV RI M! DQG + MMWf LV GHILQHG DV \Wf L L +X ? 8 f§ X f§ f )RU WKH FROXPQ EXFNOLQJ SUREOHP VXEMHFW WR FRQVWUDLQWV f LQ RUGHU WR FDOFXODWH BS WKH IROORZLQJ TXDQWLWLHV PXVW EH GHILQHG U ‘! U U r rf nrr AX r  N A A U K XL +8 ,, O -fXY S[XX NS_fX ‘3;88 9 NSaSfX-

PAGE 139

ZLWK X D D7f 8 / Xf H8H/f f Xf &RQVLGHU ILUVW WKH FDVH ZKHUH XA LV QRW RQ D FRQWURO ERXQGDU\ 2QO\ WKH ODWWHU WZR FRQVWUDLQWV DUH DFWLYH VLQFH DQG DUH EDQJ EDQJ FRQWUROV WKXV \A DQG ERWK DQG \A DUH GHWHUPLQHG E\ f )URP WKH SUHFHGLQJ SDUDJUDSKnV GHILQLWLRQV U ,$ ^` L 7R 8 f§ r r [2 X f§ A -A I! } U a? + 8 + X + 9X\ + 3R R 9f f\

PAGE 140

+HQFH B a3;88 XO n 9H/f NSA\fX X f U8U/f ZKLFK DUH LGHQWLFDO WR WKH H[SUHVVLRQV f DQG f REWDLQHG E\ WKH XVXDO PHWKRG :KHQHYHU WKH FRQVWUDLQW LV DFWLYH XA LV SUHVFULEHG DQG LV QRQ]HUR ,I WKH GHULYDWLRQ LQ WKH SUHFHGLQJ SDUDJUDSK LV H[SDQGHG WR LQFOXGH WKLV DGGLWLRQDO FRQVWUDLQW WKH UHVXOWLQJ H[SUHVVLRQ IRU LV LGHQWLFDO WR f 7KH UHVXOW RI WKLV DQDO\VLV LV D QRQOLQHDU 73%93 ZKRVH VROXWLRQ LV WKH GHVLUHG RSWLPDO VROXWLRQ  ; [@Bf ; [XOX [f ; ;; [Gf ;r -1 ,, NXOX [Lf $ V S 3; 3f S 388 3f ,, U f D 3; 3rf A237 $3Y*0,1 >+&L6RS;f8f3f XHIL + ^SA[A 3[_IXLX N3KfXX` $ 0D[ A-3A37r -:A237A `

PAGE 141

ZKHUH XL H UHTXLUHV WKDW BX VDWLVI\ FRQVWUDLQWV f DQG /DJUDQJLDQ PXOWLSOLHU IXQFWLRQV JLYHQ E\ f WKURXJK f PXVW EH QRQn QHJDWLYH $ VROXWLRQ WR WKLV SUREOHP LV REWDLQHG E\ WKH PHWKRG RI TXDVLOLQHDUL]DWLRQ D VXEURXWLQH OLVWLQJ RI WKH HTXDWLRQV UHTXLUHG E\ WKH ,%0 SURJUDP LV JLYHQ LQ $SSHQGL[ & 2QH DGGLWLRQDO SUHOLPLQDU\ LWHP PXVW EH DWWHQGHG WR EHIRUH SUHn VHQWLQJ UHVXOWV ,Q RUGHU WR SURYLGH VROXWLRQV VXEMHFW WR FRPSDULVRQ VWDQGDUGV DQ HLJHQIXQFWLRQ QRUPDOL]DWLRQ PXVW EH SUHVFULEHG 0RVW 5D\OHLJK TXRWLHQW SUREOHPV LQ WKH RSHQ OLWHUDWXUH QRUPDOL]H WKH HLJHQn IXQFWLRQ VXFK WKDW WKH GHQRPLQDWRU RI WKH TXRWLHQW LV DOZD\V HTXDO WR XQLW\ )RU WKH FROXPQ EXFNOLQJ SUREOHP WKLV LV HTXLYDOHQW WR ["Wf GW f 7KH XQPRGLILHG UHVXOWV RI VROXWLRQ YLD TXDVLOLQHDUL]DWLRQ GR QRW VDWLVI\ WKLV FRQGLWLRQ LQVWHDG [Wf GW $ ZKHUH $ LV VRPH FRQVWDQW $ WUDQVIRUPDWLRQ ZKLFK VDWLVILHV QRUPDOL]Dn WLRQ f VXFK WKDW WKH GLIIHUHQWLDO HTXDWLRQV ERXQGDU\ FRQGLWLRQV DQG RSWLPDO FRQWURO DUH XQDIIHFWHG E\ WKH WUDQVIRUPDWLRQ LV JLYHQ E\ r [A$ [ [$ [ [$

PAGE 142

:LWK D VSHFLILHG QRUPDOL]DWLRQ RI WKH HLJHQIXQFWLRQ AKH VWDWH VROXn WLRQ ;TSA PD\ EH FRPSDUHG WR DQ\ RWKHU VLPLODU VROXWLRQ IRU ZKLFK WKH QRUPDOL]DWLRQ LV NQRZQ $OO VXEVHTXHQW UHVXOWV DUH QRUPDOL]HG DV DERYH WR VDWLVI\ f 5HVXOWV *HRPHWULF &RQWURO &RQVWUDLQWV 7KUHH GLPHQVLRQOHVV SDUDPHWHUV RI WKH V\VWHP DUH GHULYHG LQ VHFWLRQ 7KH ILUVW LV $ WKH HLJHQYDOXH WR EH PD[LPL]HG 7KH VHFRQG LV N ZKLFK LV D UHODWLYH PHDVXUH RI WKH ZHLJKW RI D XQLIRUP FROXPQ WR LWV VWLIIQHVVf§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n HWHU N LQYROYHV ERWK PDWHULDO DQG JHRPHWULF SURSHUWLHV )RU FRQYHQLHQFH WKH PDWHULDO VHOHFWHG LV VWUXFWXUDO VWHHO DQG WKH (XOHU EXFNOLQJ VOLPn QHVV UDWLR /NAf YDOXH RI LV DVVXPHG :LWK WKHVH DVVXPSWLRQV WKH YDOXHV XVHG WKURXJKRXW DUH \ N

PAGE 143

$W WKLV SRLQW D FRPPHQW XSRQ WKH HIIHFW WKDW WKH PDJQLWXGH RI N KDV XSRQ WKH QDWXUH RI 8J Wf LV DSSURSULDWH ,Q WKH ODVW VHFWLRQ LW ZDV VKRZQ WKDW XA DQG XA DUH EDQJEDQJ FRQWUROV GHSHQGHQW XSRQ WKH VLJQ RI VZLWFKLQJ IXQFWLRQV ";A DQA NSA\f UHVSHFWLYHO\ ,QWXLn WLYHO\ RQH H[SHFWV WKH VWURQJHVW OHDVW GHQVH PDWHULDO WR EH RSWLPDO FRUUHVSRQGLQJ WR S[f X H 8 NSA\f r XA U/ $OVR VKRZQ LQ WKH ODVW VHFWLRQ LV WKDW IRU +DPLOWRQLDQ + WR EH PLQLPL]HG -/ $ E\ DQ XQFRQVWUDLQHG YDOXH RI XA XA DW VRPH WLPH LW LV QHFHVVDU\ WKDW 3Wf 1RWH WKDW LI WKLV GRHV QRW RFFXU VRPHZKHUH LQ W WKH FRQWURO LV XQLIRUP WKURXJKRXW DQG WKH V\VWHP GHVFULEHG LV D XQLIRUP FROXPQ XQOHVV WW WKH VROXWLRQ RI WKH VWDWH LV WKH WULYLDO VROXn WLRQA $ GLUHFW FRQVHTXHQFH RI S EHLQJ QRQSRVLWLYH LV WKDW SAWf LV DOVR +HQFH DQ\ SRVLWLYH YDOXH RI N LV VXIILFLHQW WR PDNH NSA\f VR WKDW WKH PLQLPXP GHQVLW\ LV UHTXLUHG 7KLV LQVLJQLILFDQW UHTXLUHPHQW WKDW D QRQWULYLDO RSWLPDO VROXn WLRQ WR WKH EXFNOLQJ SUREOHP H[LVW SURYLGHV DQ XQXVXDO UHVXOW :KHQ WKH YDULDEOH VLJQV UHTXLUHG IRU D QRQWULYLDO VROXWLRQ DUH XVHG LQ WKH 303 H[SUHVVLRQV f DQG f WKH LQWXLWLYH VROXWLRQ LV VKRZQ PDWKHPDWLFDOO\ WR EH WKH RSWLPDO VROXWLRQ ,W LV IRU WKLV UHDVRQ WKDW DOO VXEVHTXHQW FDVHV ZHUH UXQ ZLWK FRQWURO ERXQG YDOXHV A&RGLQJ HUURUV SURGXFHG SUHFLVHO\ WKLV FDVH DQG LQGHHGWR ZLWKLQ WKH QXPHULFDO OLPLWV RI WKH FRPSXWHU WKH DOJRULWKP FRQYHUJHG WR WKH WULYLDO VROXWLRQ

PAGE 144

HX r H -/M UX n U -BL 8QLW\ YDOXHV ZHUH VHOHFWHG WR VLPSOLI\ UHVXOWV QRQXQLW\ FRQWURO ERXQG YDOXHV ZHUH LQFOXGHG WR UHYHDO XQIRUHVHHQ VZLWFKLQJ LQ DQG XA 1RQH ZHUH HQFRXQWHUHG (IIHFWLYHO\ WKH SUREOHP DV VWDWHG KDV D VLQJOH FRQWURO FRPSRQHQW XAWf D[/f 7R LOOXVWUDWH WKH RSHUDWLRQ RI WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP GHWDLOHG UHVXOWV DUH SUHVHQWHG IRU RQH FDVH 'DWD IRU WKLV FDVH DUH D LL f -XVW DV LQ WKH FDQWLOHYHU EHDP H[DPSOH WKH LQLWLDO IURP WKH XQLIRUP VROXWLRQ DQG PXVW VDWLVI\ RQO\ WKH ,QLWLDO JXHVVHV DUH LQ WHUPV RI &Wf FRV b7Wf W 6Wf VLQ bUUWf JXHVV LV ERXQGDU\ WDNHQ FRQGLWLRQV VXFK WKDW [Wf [Wf [Wf [Wf [Wf [Wf [Wf a O&Wff 7 6Wf &Wf ; OWfNS 6 6Wf 77

PAGE 145

*LYHQ WKLV LQLWLDO JXHVV WKH YDOXH RI $A ZKLFK VDWLVILHV WKH ZHLJKW FRQGLWLRQ LV $B 7KLV LV GHPRQVWUDWHG E\ WKH IROORZ 0D[ LQJ YDOXHV RI $A DQG ZKLFK UHSUHVHQW VSHFLILF SRLQWV RQ D FXUYH OLNH WKDW RI )LJXUH $ -7 V : 6 )RU $!U $ VXFK 0D[ V WKDW -BB : WKH FRQYHUJHQFH LV PHDVXUHG LQ WHUPV RI (5525 DV GHILQHG LQ 6HFWLRQ DQG VKRZQ LQ WKH IROORZLQJ WDEXODWLRQ ,WHUDWLRQ (5525 &RPSRQHQW 3RVLWLRQf &RV -Z fr -3 ra rRf ra [f r [Af rf [f 'HIOHFWLRQ RI WKH FHQWHUOLQH LV VKRZQ LQ )LJXUH ZKLFK FRQWDLQV WKUHH FXUYHV WKH LQLWLDO JXHVV D SORW RI VXEVHTXHQW LWHUDWHV FRLQn FLGHQW IRU WKH VFDOH XVHGf DQG WKH QRUPDOL]HG VROXWLRQ 7KH RSWLPDO FRQWURO LV GLVSOD\HG LQ )LJXUH &RQWURO IXQFWLRQV IRU LWHUDWLRQV VXEVHTXHQW WR WKH LQLWLDO JXHVV SORW FRLQFLGHQW WR D VLQJOH FXUYH $OO QRQ]HUR FRPSRQHQWV RI WKH VWDWH DQG DGMRLQW YDULDEOHV DUH VKRZQ LQ )LJXUH 6KHDU UHSUHVHQWHG E\ WKH SURGXFW [ [Af rV DAVR LQFOXGHG WR JLYH D FRPSOHWH VHW RI VWUXFWXUDO YDULDEOHV

PAGE 147

)LJXUH 7\SLFDO 2SWLPDO &RQWURO )XQFWLRQ

PAGE 148

)LJXUH 7\SLFDO 6WDWH DQG $GMRLQW 9DULDEOH 6ROWXLRQV

PAGE 149

1RWLFH WKDW IRU WKH FDVH ZKHUH GHQVLW\ DQG PRGXOXV DUH HTXDO WR XQLW\ WKH FRQGLWLRQ RQ VSHFLILHG ZHLJKW VLPSOLILHV WR -77 X WfGW f Z ZKHUH LPSOLFLWO\ \ $V WKH YDOXH RI LQFUHDVHV LW UHDFKHV DQ XSSHU OLPLW RI XQLW\ YDOXHV RI DA ODUJHU WKDQ XQLW\ YLRODWH WKH ZHLJKW FRQGLWLRQ f 7KXV IRU QR XSSHU ERXQG RQ XA WKHUH DUH WZR OLPLWLQJ FDVHV DVVRFLDWHG ZLWK ORZHU ERXQG YDOXHV D DQG D -/L 9DOXHV RI D QRW FRQWDLQHG LQ WKLV UDQJH DXWRPDWLFDOO\ YLRODWH VRPH -M JLYHQ FRQGLWLRQ (DFK YDOXH RI D KDV D FRUUHVSRQGLQJ RSWLPDO HLJHQYDOXH $A DULG GHVLJQ SURILOH X Wf D D[/f 6HYHUDO RI WKHVH SURILOHV DUH VKRZQ LQ )LJXUH ,Q DOO EXW WKH XQLIRUP FROXPQ OLPLWn LQJ FDVH IRU D WKH RSWLPDO SURILOH KDV D WUDQVLWLRQ UHJLRQ /L r r HUH XAA XS 7KH SRLQW ZKHUH GHFUHDVLQJ XAWf LQWHUVHFWV WKH ORZHU ZK ERXQG D RFFXUV DW D WLPH GHVLJQDWHG W 2SWLPDO FRQWURO LV RQ WKH /L LM ORZHU ERXQG IRU W W $ FRPSDUDWLYH PHDVXUH IRU WKH HIIHFWLYHQHVV RI WKH RSWLPL]Dn WLRQ LV WKH UDWLR RI $ WR WKH HLJHQYDOXH IRU D XQLIRUP FROXPQ RI 0D[ HTXDO ZHLJKW 7KLV ODWWHU HLJHQYDOXH LV WKH FODVVLFDO (XOHU EXFNOLQJ ORDG GHQRWHG $ DQG FRUUHVSRQGV WR WKH FDVH D 7KXV Y OL ( 9R 7KH GHSHQGHQFH RI WKH HIIHFWLYHQHVV UDWLR $AA$A DQG WKH ORZHU ERXQG LQWHUFHSW W7 XSRQ D7 LV VKRZQ LQ )LJXUH $V LV W\SLFDO RI VLPSOH ‘/L

PAGE 150

)LJXUH 2SWLPDO $UHD 'LVWULEXWLRQV 3DUDPHWULF LQ /RZHU %RXQG 1R 8SSHU %RXQG

PAGE 151

$ $r 0D[ Q r/ &,17 )LJXUH (IIHFWLYHQHVV 5DWLR DQG /RZHU %RXQG ,QWHUFHSW 1R 8SSHU %RXQG

PAGE 152

SUREOHPV RI WKLV W\SH WKH PRVW LPSURYHPHQW RFFXUV IRU WKH VPDOOHVW H[FXUVLRQV IURP WKH XQLIRUP FDVH D f 7R DOORZ D WR EH /L f / UHGXFHG IURP WR SURYLGHV PXFK OHVV LPSURYHPHQW LQ WKH FRVW 2EVHUYH WKDW WKH PD[LPXP YDOXH IRU D[f RFFXUV DW W IRU DA DQG LV 7KXV IRU DA QR RSWLPDO SURILOH LQ )LJXUH LV FRQVWUDLQHG E\ WKH XSSHU ERXQG RQ XAWf )RU DA DW RQH SRLQW W Df DA DQG DWf DX W 7KHUHIRUH LQVWHDG RI EHLQJ XQERXQGHG LQ f LW LV RQO\ QHFHVVDU\ WKDW DA $Q DA WKLV ODUJH LV HVVHQWLDOO\ XQERXQGHG IRU WKH JLYHQ ZHLJKW FRQVWUDLQW 5HVXOWV VLPLODU WR WKRVH DERYH KDYH EHHQ REWDLQHG IRU VLPLODU SUREOHPV E\ RWKHU PHWKRGV 7KH QH[W H[DPSOHV DUH IRU D FDVH QRW \HW UHSRUWHG LQ WKH RSHQ OLWHUDWXUH 7KHVH FDVHV FRXOG QRW EH REWDLQHG ZLWKRXW VSHFLI\LQJ ERWK XSSHU DQG ORZHU FRQWURO ERXQGV 7KH ILUVW H[DPSOH KDV D ORZHU ERXQG VSHFLILHG ZLWK D UDQJH RI YDULRXV XSSHU ERXQGV GHFUHDVLQJ WR WKH OLPLWLQJ FDVH RI D XQLIRUP VROXWLRQ 7KH ILUVW H[DPSOH RI WKLV FDVH LV DI6LQFH WKH QDWXUH RI WKH VROXWLRQ PHWKRGV GLIIHU RQO\ LQ GHWDLOV QR VROXWLRQ PHWKRGV DUH SUHVHQWHG 2SWLPDO SURILOHV DUH SUHVHQWHG IRU YDULRXV YDOXHV RI DA JLYHQ DA ,Q )LJXUH WKH OLPLWLQJ FDVHV

PAGE 153

)LJXUH 2SWLPDO $UHD 'LVWULEXWLRQV 3DUDPHWULF LQ 8SSHU %RXQG /RZHU %RXQG D LX

PAGE 154

DUH VKRZQ ZLWK WKUHH LQWHUPHGLDWH FDVHV DV ZHOO 2QH FDVH LV GLVSOD\HG DV D GDVKHG OLQH VLQFH LW LV REWDLQHG DV DQ DSSUR[LPDWLRQ $V DAr WKH QXPHULFDO SURFHGXUH EHFRPHV XQVWDEOH IRU UHDVRQV GLVFXVVHG LQ &KDSWHU 9,, WKH FDVH IRU DA PXVW DSSURDFK WKH XQLIRUP F\OLQGHU DVVRFLDWHG ZLWK DA %HFDXVH WKH WUDQVLWLRQ UHJLRQ ZKHUH D[/f DA LV DSSURDFKLQJ D YHUWLFDO VWUDLJKW OLQH LQ WKH OLPLW DV DA WKH DSSUR[LPDWLRQ LV WDNHQ WR EH D[/f DA [/ W\ D[/f D9 M W [/ 7KDW YDOXH RI W ZKLFK VDWLVILHV WKH VSHFLILHG ZHLJKW FRQGLWLRQ XQGHU WKH DVVXPHG IRUP RI WKH DSSUR[LPDWLRQ LV JLYHQ E\ W 8 1R IXUWKHU XVH LV PDGH RI WKLV DSSUR[LPDWLRQ LWV VROH UHDVRQ IRU H[LVWHQFH LV WR SURYLGH DQ DGGLWLRQDO FXUYH LQ )LJXUH WR PRUH HIIHFn WLYHO\ GHPRQVWUDWH WKH WUDQVLWLRQ IURP WKH FDVH ZLWK QR XSSHU ERXQG WR WKH XQLIRUP EHDP FDVH 7KH FKDUDFWHULVWLF IRUP RI VROXWLRQV LV LQGLFDWHG E\ D SORW RI W\ YHUVXV DA VKRZQ LQ )LJXUH WRJHWKHU ZLWK WKH HIIHFWLYHQHVV UDWLR $ $A -XVW DV REVHUYHG LQ WKH FDVH SRVVHVVLQJ RQO\ D ORZHU ERXQG 0D[ ( ZKHQ RQO\ DQ XSSHU ERXQG LV SUHVHQW WKH ILUVW H[FXUVLRQV IURP D XQLn IRUP FROXPQ DA f SURYLGHV WKH JUHDWHVW FRVW LPSURYHPHQW ,Q WKH SUHFHGLQJ FDVH DOWKRXJK RQO\ W LV SUHVHQWHG WKH LPSOLFDWLRQ WKDW W7 GRHV QRW H[LVW LV LQFRUUHFW %\ REVHUYDWLRQ RI /M

PAGE 155

$0D[$(n r8 )LJXUH (IIHFWLYHQHVV 5DWLR DQG 8SSHU %RXQG ,QWHUFHSW /RZHU %RXQG DA

PAGE 156

)LJXUH DQG WKH GHILQLWLRQ RI W LW LV REYLRXV WKDW IRU DOO FDVHV OBL W 7KLV RFFXUV DV D FRQVHTXHQFH RI VHOHFWLQJ D LQ RUGHU /L /M WR GHPRQVWUDWH D PRUH JHQHUDO EHKDYLRU RI WKH VROXWLRQ SURFHVV D QRQn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f 0DWKHPDWLFDOO\ WKH DSSUR[LPDWLRQ LV DWf 8 WX W W8f n Dm ‘ : W W / VXFK WKDW WKH FDVH ZKLFK VDWLVILHV WKH VSHFLILHG ZHLJKW FRQGLWLRQ UHTXLUHV WKH FRQWURO ERXQG LQWHUFHSWV RI WKH DSSUR[LPDWLRQ WR VDWLVI\ 8 IF/ D77 D D / /

PAGE 157

[/ )LJXUH 2SWLPDO $UHD 'LVWULEXWLRQV 3DUDPHWULF LQ 8SSHU %RXQG /RZHU %RXQG D /L

PAGE 158

)URP DQ H[WUDSRODWLRQ RI LQWHUFHSW YDOXHV DVVRFLDWHG ZLWK KLJKHU YDOXHV RI D LQ )LJXUH W LV IRXQG WR EH DERXW 7KH FRUUHVSRQGLQJ 8 LL YDOXH RI W LV WKXV -XVW DV LQ WKH SUHFHGLQJ FDVH LW PXVW EH HPSKDVL]HG WKDW WKLV DSSUR[LPDWLRQ LV LQFOXGHG RQO\ WR PRUH FOHDUO\ LOOXVWUDWH WKH WUDQVLWLRQ RI D[/f IURP RQH OLPLWLQJ FDVH WR WKH RWKHU DV DA GHFUHDVHV IURP WR (IIHFWLYHQHVV RI GHVLJQ RSWLPL]DWLRQ LV UHYHDOHG LQ WKH SORW RI $ $f YHUVXV DB7 VHH )LJXUH f &KDUDFWHULVWLF VROXWLRQ IRUPV 0D[ ( 8 DUH DOVR GHSLFWHG WKHUH LQ WKH SORW RI W DQG WA GHSHQGHQFH XSRQ DA ,W LV WR EH QRWHG WKDW W7 GHYLDWHV OLWWOH IURP WKH YDOXH RI XQLW\ EXW YHU\ GHILQLWHO\ KDV D IXQFWLRQDO GHSHQGHQFH XSRQ DA ,QHTXDOLW\ 6WUHVV &RQVWUDLQWV +DYLQJ REWDLQHG VROXWLRQV IRU YDULRXV FRQWURO FRQVWUDLQWV WKH QH[W SUREOHP FRQILJXUDWLRQ DWWHPSWHG LQYROYHV D PL[HG FRQVWUDLQW ,QLWLDOO\ LW ZDV GHFLGHG WR UHVWULFW WKH PD[LPXP DOORZDEOH QRUPDO VWUHVV GXH WR EHQGLQJ ,r%;f r0D[ [ / 5HFRJQL]LQJ WKDW IRU DQ\ FURVV VHFWLRQ WKH PD[LPXP YDOXH RI 2J[f RFFXUV DW WKH RXWHUPRVW ILEUHV WKH GLVWDQFH G[f RI WKHVH ILEUHV IURP WKH EHQGLQJ D[LV PXVW EH GHWHUPLQHG 2Q WKH EDVLV RI DVVXPLQJ VLPLODU FURVV VHFWLRQV WKH IXQFWLRQ I[f DSSHDULQJ LQ WKH $[f H[SUHVVLRQ f DOVR GHWHUPLQHV G[f LH G [f G4I[f G4D/L[f

PAGE 159

)LJXUH (IIHFWLYHQHVV 5DWLR DQG &RQWURO %RXQG ,QWHUFHSWV /RZHU %RXQG DA

PAGE 160

ZKHUH GA LV VRPH UHIHUHQFH YDOXH ,Q WHUPV RI %HUQRXOOL(XOHU EHQGLQJ HTXDWLRQ WKH LQHTXDOLW\ FRQVWUDLQW RQ EHQGLQJ QRUPDO VWUHVV LV OD%[f [fG[f 9f R 0D[ (,
PAGE 161

,Q WKH SUHVHQW H[DPSOH IRU VPDOO YDOXHV QR VROXWLRQ H[LVWV WKDW FDQ IXOILOO ERWK WKH VWUHVV DQG JHRPHWULF FRQVWUDLQWV /DUJH YDOXHV RI D UHVXOW LQ DOO JHRPHWULFDOO\ DGPLVVLEOH FRQWURO VLPXOWDQHRXVO\ VDWLVI\n LQJ WKH PL[HG FRQVWUDLQW ,QWHUPHGLDWH YDOXHV SURGXFH VROXWLRQ LWHUn DWHV LQ WKH TXDVLOLQHDUL]DWLRQ SURFHVV WKDW ZHUH FRQVWUDLQHG E\ f RYHU VRPH LQWHUYDO +RZHYHU WKH ILQDO LWHUDWH XAA&Wf LV QRW FRQn VWUDLQHG E\ WKH PL[HG FRQVWUDLQW DW DQ\ SRLQW W )RU WKLV UHDn VRQ QR PL[HG FRQVWUDLQW H[DPSOHV DUH SUHVHQWHG 7KHRUHWLFDO FRQVLGHUn DWLRQV RI RWKHU SRVVLEOH VWUHVV FRQVWUDLQWV DUH JLYHQ LQ WKH IROORZLQJ SDUDJUDSKV $ PRUH UHDOLVWLF VWUHVV FRQVWUDLQW LV WR OLPLW WKH VWUHVV FRPSRQHQW R[f QRUPDO WR D FURVV VHFWLRQ FRPSRVHG RI D WHUP DA&[f GXH WR EHQGLQJ DQG D WHUP 2S[f DWWULEXWHG WR WKH HQG ORDG 3 :LWK UHIHUn HQFH WR WKH GHILQLWLRQ RI VKRZQ LQ )LJXUH D[f DQG LWV FRQVWLWXn HQWV DUH R[f R%[f 2S[f D%[f 3/ [fG[f 9U DS[f 3 FRV [f $[f 'LIIHUHQWLDO FDOFXOXV DQG WULJRQRPHWU\ SURYLGH WZR LGHQWLWLHV WKDW DUH XVHG WR HYDOXDWH FRV f DV D IXQFWLRQ RI [ LQ WHUPV RI WKH VORSH RI WKH FHQWHUOLQH 7KH\ DUH G< WDQ G[ WDQ VHF

PAGE 162

)URP WKLV FRV 6f ?aK 7KXV WKH QRUPDO VWUHVV LQHTXDOLW\ EHFRPHV R[f 0IL[fG[f S & DY?!b [f n $[f G[f r0D[ %\ LQWURGXFLQJ GLPHQVLRQOHVV VWDWH DQG FRQWURO YDULDEOHV DIWHU PXFK PDQLSXODWLRQ WKH IROORZLQJ PL[HG FRQVWUDLQW LQHTXDOLW\ LV REWDLQHG -![Xf X! A O f GA [fnX-f [D 7KUHH SDUDPHWHUV H[LVW IRU WKLV GHULYDWLRQ / D 0D[ 9R NR f§ D WKH (XOHU EXFNOLQJ VOLPQHVV UDWLR M f§ D D JHRPHWULFDO VOLPQHVV UDWLR f§L $ VLPSOHU FRQVWUDLQW IXQFWLRQ FDQ EH REWDLQHG E\ WDNLQJ 2S[f DV PHUHO\ rS[f f -f IRU ZKLFK ![Xf Xf >aOf Lcf X 9R

PAGE 163

7KLV IXQFWLRQ FDQ EH DQDO\WLFDOO\ VROYHG IRU ZKHQHYHU b :KDW PXVW EH VROYHG LV D FXELF HTXDWLRQ LQ 8XS $ XS % f % [D4 $ UHGXFHG FXELF HTXDWLRQ ZLWK UHDO FRHIILFLHQWV \ S\ T KDV D GLVFULPLQDQW $ 5 5  Sf Tf WKDW GHWHUPLQHV WKH QDWXUH RI WKH URRWV WR WKH HTXDWLRQ )RU $ RQH UHDO WZR FRPSOH[ URRWV $ !‘ WZR UHDO HTXDO URRWV $ r WKUHH GLVWLQFW UHDO URRWV &RQVLGHU D W\SLFDO SUREOHP ZKHUH WKH FURVV VHFWLRQ LV FLUFXODU DQG KDYLQJ W\SLFDO SDUDPHWHU YDOXHV N Rf LW IROORZV IURP

PAGE 164

DQG $ S T D f [ ; [ 7KXV IRU W\SLFDO SDUDPHWHU YDOXHV WKHUH LV D VLQJOH UHDO XA YDOXH ZKHQHYHU [ ; [ f [ a RU [ ; [afA r [ OI ,Q RWKHU ZRUGV QHDU WKH WLS ZKHUH ERWK W DQG [A DSSURDFK ]HUR WKUHH UHDO GLVWLQFW URRWV WR f H[LVW 8QOHVV WKLV SRVVLELOLW\ LV H[FOXGHG E\ D VXIILFLHQWO\ ODUJH PLQLPXP ERXQG WR XA WKH TXHVWLRQ RI ZKLFK URRW LV DSSURSULDWH PXVW EH DQVZHUHG +DYLQJ GHPRQVWUDWHG KRZ FRPSOLFDWHG WKH DQDO\VLV EHFRPHV IRU WKLV SDUWLFXODU VWUHVV FRQVWUDLQW QR IXUWKHU FRQVLGHUDWLRQV DUH SUHVHQWHG 7KH GHYHORSPHQWV DUH LQFOXGHG WR VKRZ WKDW HYHQ VLPSOH VWUHVV FRQVWUDLQWV FDQ EH WRR FRPSOLFDWHG IRU RUGLQDU\ PHWKRGV RI DQDO\VLV

PAGE 165

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f IRU GHWDLOV RI WKH PHWKRG %HUQRXOOL(XOHU %HQGLQJ 7KHRU\ LV DSSOLHG WR D IL[HG OHQJWK FDQWLOHYHU EHDP RI FRQVWDQW ZLGWK

PAGE 166

DQG PDWHULDO SURSHUWLHV ZKLFK LV V\PPHWULF DERXW LWV FHQWUDO D[LV 7KH SUREOHP LV WR ILQG WKH YDULDEOH KHLJKW K[f WKDW JLYHV WKH PLQLPXP WLS GHIOHFWLRQ GXH WR LWV RZQ KHLJKW VXEMHFW WR KDUG FRQVWUDLQWV RI PD[LPXP DQG PLQLPXP DOORZDEOH KHLJKWV $ KLJKHU RUGHUHG HOHPHQW LV XVHG WR REWDLQ EHWWHU DFFXUDF\ ERWK WKH GLVSODFHPHQW DQG LWV GHULYDWLYH DW WKH QRGHV DUH SUHVFULEHG WR EH WKH LQGHSHQGHQW YDULDEOHV )RU WKLV VLPSOH SUREOHP Q QRGHV GLYLGH WKH EHDP LQWR Qf HOHPHQWV RI HTXDO OHQJWK +HQFH WKH GLVSODFHPHQW DQG LWV GHULYDWLYH DUH WKH GHIOHFWLRQ DQG URWDWLRQ RI WKH EHDPnV FHQWHUOLQH DW WKH ORFDWLRQV RI WKH QRGHV ,W LV DOVR DVVXPHG WKDW HDFK HOHPHQW LV OLQHDUO\ WDSHUHG ZKHUHE\ WKH %HUQRXOOL(XOHU EHQGLQJ HTXDWLRQ IRU HOHPHQW L WDNHV WKH IRUP (, [f 0[f G[ ,[f ZK[f 3RVLWLRQ DORQJ WKH EHDP [ KHLJKW K[f DQG GLVSODFHPHQW RI WKH EHDP \[f DQG VORSH [f RI WKH FHQWHUOLQH DUH DSSUR[LPDWHG E\ YHFWRUV FRPSRVHG RI GLVFUHWH HOHPHQWV 7 [ >[ [ [ @ f§ s ] Q K[f >K[ f K[ f K[ @  Q O[f >\[f \[! \&[A@ [f >[ f [ f [ f@7 ] Q :KHQHYHU D KHLJKW GLVWULEXWLRQ LV VSHFLILHG D ILQLWH HOHPHQW VROXWLRQ IRU WKH EHDPnV VORSH DQG GHIOHFWLRQ FDQ EH REWDLQHG

PAGE 167

0DWKHPDWLFDO 3URJUDPPLQJ *UDGLHQW 3URMHFWLRQ 0HWKRG :LWK WKH VWUXFWXUDO DQDO\VLV JLYHQ LQ WKH IRUP RI D ILQLWH HOHPHQW VROXWLRQ WKH PDWKHPDWLFDO VWDWHPHQW RI WKH RSWLPL]DWLRQ SUREOHP LV *LYHQ Lf PDWHULDO SURSHUWLHV S( LLf GLPHQVLRQV / Z LLLf QXPEHU RI QRGHV Q )LQG 0LQ >\[ f@ K[f Q VXEMHFW WR E K[Af G L OQ 7KLV PD\ EH VWDWHG DV D QRQOLQHDU PDWKHPDWLFDO SURJUDPPLQJ SUREOHP ZKRVH RSWLPDOLW\ FRQGLWLRQV DUH SUHVFULEHG E\ WKH .XKQ7XFNHU 7KHRUHP JLYHQ LQ VHFWLRQ )RU WKLV SDUWLFXODU EHDP SUREOHP FRUUHVSRQGLQJ TXDQWLWLHV RI WKH WKHRUHP DUH [ A KB )[f A Y[f [ / UK[ f G L Q M L J [f A YE K!Af L OQ M LQ 7KH QXPHULFDO DOJRULWKP XVHG WR REWDLQ WKH RSWLPDO VROXWLRQ VDWLVI\LQJ WKH .XKQ7XFNHU FRQGLWLRQV LV D PRGLILHG JUDGLHQW SURMHFn WLRQ PHWKRG GHVFULEHG E\ .RZDOLN f %ULHIO\ VSHDNLQJ LW XVHV SURMHFWLRQV RI WKH FRVW IXQFWLRQ JUDGLHQW LQWR D VXEVSDFH VDWLVI\LQJ FXUUHQWO\ DFWLYH FRQVWUDLQWV 6WDUWLQJ IURP D IHDVLEOH GHVLJQ SRLQW KA D VHTXHQFH RI IHDVLEOH GHVLJQV LV JHQHUDWHG VXFK WKDW

PAGE 168

ZKHUH ‘9 ff f f r f9?Of f f f fA237,0$/ K W K D 6 T aT T 6 39) D T 7 7 3 11 1f 1 $ GHULYDWLRQ RI SURMHFWLRQ PDWUL[ 3 LV IRXQG LQ .RZDOLN f &ROXPQV RI 1 DUH WKH JUDGLHQW YHFWRUV RI WKH DFWLYH FRQVWUDLQWV WKXV 1 LV DQ Q r Uf PDWUL[ ZKHUH U LV WKH QXPEHU RI DFWLYH FRQVWUDLQWV DQG UQ ,I U WKHQ E\ GHILQLWLRQ 1 LV D QXOO PDWUL[ DQG 3 UHGXFHV UHGXFHV WR WKH LGHQWLW\ PDWUL[ :LWK 3 VR GHILQHG WKRVH FRPSRQHQWV RI 9) WKDW OHDG WR D FRQVWUDLQW YLRODWLRQ DUH VXEWUDFWHG IURP WKH GLUHFWLRQ RI VWHHSHVW GHVFHQW ,Q WKLV PDQQHU 6 LV WKH GLUHFWLRQ ZKLFK EHVW LPSURYHV WKH FRVW IXQFWLRQ ZLWKRXW OHDYLQJ WKH UHJLRQ RI IHDVLEOH GHVLJQV ZKHUH JA [f )URP WKH IRUP RI JA [f LQ WKLV SDUWLFXODU SUREOHP WKH W\SLFDO HOHPHQW 1 RI 1 LV GHILQHG DV 1 LN [ L OQ N OU ZKHUH D DUH WKRVH LQGLFHV M Q DVVRFLDWHG ZLWK WKH U DFWLYH N FRQVWUDLQWV IRU ZKLFK JA[f +HQFH 1 [N L A D U L L D 1 LN OO L D DQG DQG D Q N D Q N

PAGE 169

7KHUH LV QR GLIILFXOW\ LQ FDOFXODWLQJ WKH JUDGLHQWV UHTXLUHG E\ 1 VLQFH WKH &ALf DUH VSHFLILHG IXQFWLRQV WKH JUDGLHQWV FDQ EH GHWHUPLQHG H[SOLFLWO\ 'HULYDWLYHV UHTXLUHG WR HYDOXDWH 9) DUH QRW VR HDVLO\ REWDLQHG 7R FLUFXPYHQW WKH ODFN RI DQ H[SOLFLW IRUPXODWLRQ WKH GHULYDWLYHV DUH REWDLQHG DV IROORZV 7KH LrA FRPSRQHQW RI 9) LV IRXQG QXPHULFDOO\ E\ ) [ L )[ $[f )[f $[ ZKHUH WKH RQO\ QRQ]HUR FRPSRQHQW RI $[ LV $[ L -/B [ L 7R DYRLG LQVWDELOLW\ DV WKH VROXWLRQ LV DSSURDFKHG WKH LQFUHPHQW LV PRGLILHG WR N[ [ ZKHUH N LV WKH PD[LPXP SHUFHQW FKDQJH RI DOO FRPSRQHQWV RI [ IURP WKH SUHYLRXV LWHUDWLRQ 1R QXPHULFDO LQVWDELOLWLHV ZHUH HQFRXQWHUHG ZLWK WKLV VFKHPH KRZHYHU LI WRR VPDOO D YDOXH RI QXPHULFDO WROHUDQFH LV FKRVHQ WKH FRQYHUJHQFH FULWHULD DUH QHYHU VDWLVILHG DQG WKH DOJRULWKP EHJLQV WR KXQW LQ WKH QHLJKERUKRRG RI WKH RSWLPDO VROXWLRQ $ VLPn SOLILHG IORZFKDUW LV VKRZQ LQ )LJXUH 7R IXUQLVK D FRPSDULVRQ VWDQGDUG WKH %HUQRXOOL(XOHU HTXDWLRQ IRU D EHDP ZLWK D OLQHDUO\ WDSHUHG FHQWHU VHFWLRQ ZDV LQWHJUDWHG DQDOn \WLFDOO\ 7KLV UHYHDOHG VRPH LQWHUHVWLQJ TXDOLWDWLYH UHVXOWV ZKLFK DUH GLVFXVVHG EHORZ ZLWK TXDQWLWDWLYH UHVXOWV JLYHQ LQ WKH IROORZLQJ VHFn WLRQ )LUVW WKH H[SUHVVLRQ IRU WKH VHFRQG GHULYDWLYH RI \[f LV

PAGE 170

)LJXUH 6LPSOLILHG )ORZFKDUW RI $OJRULWKP

PAGE 171

RQO\ SLHFHZLVH VPRRWK GXH WR ,[f 7R LQWHJUDWH WKH HTXDWLRQV IRU WKH OLQHDUO\ WDSHUHG EHDP UHTXLUHG GLYLGLQJ WKH UDQJH RI [ LQWR WKUHH UHTLRQV ,Q WKH UHJLRQ [ WKH KHLJKW K[f G DQG \[f LV D TXDUWLF SRO\QRPLDO LQ [ 9DULDEOH KHLJKW FRPSOLFDWHV WKH HTXDWLRQ LQ WKH UHJLRQ [A [ [A KRZHYHU WKH FHQWHUOLQH GHIOHFWLRQ LV VWLOO D IXQFWLRQ RI D VLQJOH YDULDEOH X[f GHILQHG DV X[f O / [L [ ZKHUH $ f§ WDQ D D FRQVWDQW G WDQ D /9 G` ,G ; ; VXFK WKDW X[f  2QH RI WKH FRPSOLFDWHG WHUPV RI WKH H[SUHVVLRQ IRU \[f LQ WKLV UHJLRQ RI WKH EHDP LV ,Q X [f f X[f @ ZKLFK LV XQGHILQHG DW [ [A ZKHQ EG :LWKLQ WKH WKLUG UHJLRQ [ / WKH KHLJKW K[f E DQG WKH FHQWHUOLQH GHIOHFWLRQ LV D TXDUWLF SRO\QRPLDO LQ [ [Af f ,W LV VKRZQ LQ +RUQEXFNOH HW DO f WKDW LQ DOO WKUHH UHJLRQV WKH GHIOHFWLRQ FRXOG EH ZULWWHQ LQ QRQGLPHQVLRQDOL]HG IRUP

PAGE 172

A0 IFW E [fG` \ R f 7KLV UHVXOW VXSSRUWV 'L[RQnV VWDWHPHQW WKDW G/ LV QRW D SDUDPHWHU RI WKH GLPHQVLRQOHVV V\VWHP 7KHVH HTXDWLRQV LQGLFDWH WKDW GLPHQVLRQOHVV UHVXOWV LQGHHG GHSHQG RQO\ XSRQ EG +RZHYHU IRU D SDUWLFXODU EHDP RI JLYHQ GLPHQVLRQV WKH DFWXDO WLS GHIOHFWLRQ GRHV LQFUHDVH DV WKH VTXDUH RI WKH VOLPQHVV UDWLR /G 5HVXOWV %HIRUH DQ\ DWWHPSWV ZHUH PDGH WR REWDLQ UHVXOWV FHUWDLQ EDVLF TXHVWLRQV KDG WR EH DQVZHUHG )LUVW WKH LQWHJUDWLRQ LQKHUHQW WR WKH ILQLWH HOHPHQW VROXWLRQ DV GHVFULEHG LQ =LHQNLHZLF] f LV GRQH QXPHULFDOO\ WKHUHE\ UHTXLULQJ GHWHUPLQDWLRQ RI WKH SURSHU VWHS VL]H ,W ZDV IRXQG WKDW ZLWK WKH UDQJH RI SDUDPHWHUV XVHG LQ WKLV SUREOHP WHQ LQWHJUDWLRQ LQWHUYDOV SHU HOHPHQW JDYH WKH GHVLUHG DFFXUDF\ ZLWKRXW H[FHVVLYH FRPSXWDWLRQ 6HFRQGO\ WKHUH ZDV D TXHVWLRQ RI KRZ PDQ\ QRGHV DUH UHTXLUHG WR JLYH D YDOLG UHSUHVHQWDWLRQ RI WKH EHDP %HFDXVH WKH V\VWHP LV VR VLPSOH DV IHZ DV WZR ILQLWH HOHPHQWV SURGXFH JRRG VWUXFn WXUDO UHVXOWV YLVDYLV WLS GHIOHFWLRQ +RZHYHU XQOHVV HLJKW RU VL[n WHHQ HOHPHQWV ZHUH XVHG WKH DSSUR[LPDWLRQ RI K[f E\ K[f ZDV QRW DFFHSWDEOH 7KH UHVXOWV IRU HLJKW RU VL[WHHQ HOHPHQWV ZHUH DOPRVW LGHQn WLFDO )RU WKH ODWWHU WZR FDVHV WKH H[HFXWLRQ WLPHV RQ DQ ,%0 ZHUH VHFRQGV DQG VHFRQGV UHVSHFWLYHO\f

PAGE 173

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f ZHUH REWDLQHG IRU WKLV SUREOHP E\ YDULRXV DXWKRUV ,W ZDV IRXQG ZLWK WKH ILQLWH HOHPHQW PHWKRG WKDW IRU QXPHULFDO SUHFLVLRQ FRUUHVSRQGLQJ WR VOLGH UXOH DFFXUDF\ WKH RSWLPDO SURILOH LV DQ DOPRVW OLQHDUO\ WDSHUHG EHDP *UHDWHU QXPHULFDO DFFXUDF\ SURYLGHV D GLVWLQFW FXUYDWXUH LQ WKH WDSHUHG SRUWLRQ RI WKH EHDP 7KLV LV VKRZQ LQ )LJXUH E\ WKH WZR FXUYHV JHQHUDWHG ZLWK WKH ILQLWH HOHPHQW PHWKRG 7KH SURILOH GHSLFWHG E\ WKH

PAGE 174

,QWHUPHGLDWH )HDVLEOH 'HVLJQ &RQVWUDLQHG 2SWLPDO 'HVLJQ )LJXUH 'HVLJQ ,WHUDWLRQ WR 2SWLPDO 3URILOH

PAGE 175

EG /G K[/f G )LJXUH 2SWLPDO 3URILOH &RPSDULVRQ RI 0HWKRGV DQG 1XPHULFDO 7ROHUDQFH

PAGE 176

GRWV LV WDNHQ IURP WKH ZRUN E\ %R\NLQ DQG 6LHUDNRZVNL f DQG LV SUHVHQWHG IRU FRUURERUDWLRQ )RU WKLV ODWWHU FXUYH WKH DEVROXWH PDJQLn WXGH RI WKH HUURU LQ VDWLVI\LQJ WKH GLIIHUHQWLDO HTXDWLRQV LV OHVV Q WKDQ 6LQFH WLS GHIOHFWLRQ LV WKH FRVW IXQFWLRQ WKH WLS GHIOHFWLRQ IURP WKH ILQLWH HOHPHQW DOJRULWKP ZDV FRPSDUHG WR WKDW IURP WKH OLQHDUO\ WDSHUHG EHDP )LJXUH f 'LIIHUHQFHV ZHUH OHVV WKDQ b H[FHSW QHDU EG 7KLV LV DWWULEXWHG WR WKH QDWXUDO ORJDULWKP WHUP LQ WKH OLQHDU WDSHU VROXWLRQ GLVFXVVHG SUHYLRXVO\ 1R FRPSXWHU UHVXOWV ZHUH REWDLQHG IRU WKH OLQHDU WDSHU VROXWLRQ ZLWK EG DOPRVW ]HUR EHFDXVH WKH QDWXUDO ORJDULWKP WHUP EHFDPH H[FHVVLYHO\ ODUJH 7KLV PD\ H[SODLQ ZK\ DOO HDUOLHU VWXGLHV XVLQJ WKLV H[DPSOH SUREOHP FRQWDLQ QR UHVXOWV IRU EG +RZHYHU WKH ILQLWH HOHPHQW DOJRULWKP FRQYHUJHG WR D VROXn WLRQ IRU EG ZLWK QR GLIILFXOW\ )XUWKHU DQDO\VLV FHQWHUHG RQ D VWDWHPHQW LQ ZKLFK %HOODP\ DQG :HVW f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

PAGE 177

< 7,3 IHHWf )LJXUH 7LS 'HIOHFWLRQ 9HUVXV EG

PAGE 178

)LJXUH $QJOH RI 7DSHU 9HUVXV EG

PAGE 179

K[/f G )LJXUH 2SWLPDO %HDP 3URILOHV YLD )LQLWH (OHPHQWV

PAGE 180

QRGHV EHFDXVH LW VSDQV VXFK D VKRUW VHFWLRQ RI EHDP 7KLV LV EHOLHYHG WR EH LQVXIILFLHQW IRU DFFXUDWH UHSUHVHQWDWLRQ 7KH VDPH VKRUWFRPLQJ LV DOVR DSSDUHQW LQ )LJXUH ZKLFK VKRZV WKH RSWLPDO SURILOH GLPHQn VLRQV UHSRUWHG LQ WKH YDULRXV VWXGLHV )RU EG WKH YDOXH RI U G REWDLQHG E\ WKH ILQLWH HOHPHQW PHWKRG GLYHUJHV IURP WKH RWKHU VROXWLRQ PHWKRG UHVXOWV 7R YHULI\ LQGHSHQGHQFH RI WKH VROXWLRQ WR /G D VHULHV RI UXQV ZHUH PDGH ZLWK IL[HG EG DQG WKH VOLPQHVV UDWLR /G UDQJLQJ IURP ILYH WR HLJKW\ 5HVXOWLQJ SURILOHV DW ILUVW VXJJHVWHG WKDW /G LV D SDUDPn HWHU EXW FORVHU DQDO\VLV UHYHDOHG WKHUH ZDV DQ RYHUVLJKW LQ WKH XVH RI QXPHULFDO WROHUDQFHV ,QLWLDOO\ D VLQJOH WROHUDQFH ZDV DSSOLHG WR ERWK JHRPHWULFDO FRQVWUDLQWV DQG WKH .XKQ7XFNHU WHVW 7KH IRUPHU LV LQGHn SHQGHQW RI UHODWLYH EHDP VKDSH EXW WKH ODWWHU LV QRW WKLV LV LPSOLHG E\ WKH IRUP RI WKH QRQGLPHQVLRQDOL]DWLRQ FRHIILFLHQW JLYHQ E\ f %DVLFDOO\ WKH SURMHFWLRQ PDWUL[ 3 GHOHWHV WKRVH FRPSRQHQWV IURP 9) ZKLFK DUH QRUPDO WR DQ DFWLYH JHRPHWULFDO FRQVWUDLQW ERXQGDU\ 6LQFH DOO QXPHULFDO VFKHPHV FRQWDLQ LQKHUHQW DFFXUDF\ OLPLWV WKH .XKQ 7XFNHU WHVW LV FRQVLGHUHG WR EH VDWLVILHG ZKHQ DW WKRVH SRLQWV QRW RQ D ERXQGDU\ \[Qf [ &* L
PAGE 181

'LJLWDO 'L[RQf ff§ $QDORJ +\EULG 4XDVLOLQHDUL]DWLRQ )LJXUH 'HSHQGHQFH RI ,QWHUFHSW /RFDWLRQ XSRQ EG

PAGE 182

DOJRULWKP ZDV UHTXLUHG DQG WKH V\VWHP EHJDQ KXQWLQJ DURXQG WKH VROXWLRQ IDLOLQJ WR FRQYHUJH %\ QXPHULFDO H[SHULPHQWDWLRQ DQG FXUYH ILWWLQJ LW ZDV IRXQGf§+RUQEXFNOH HW DO ff§WKDW XQLIRUP DFFXUDF\ FRXOG EH REWDLQHG IRU WKLV SDUWLFXODU SUREOHP ZKHQ WKH WROHUDQFH XVHG LQ WKH .XKQ7XFNHU WHVW LV 1 eJ eR NR/ 9 A G 1T N4 [ 6XEVFULSW ]HUR LQGLFDWHV YDOXHV DVVRFLDWHG ZLWK UHIHUHQFH OHQJWK FRUUHn VSRQGLQJ WR /AG WKH YDOXH RI ILYH ZDV FKRVHQ EHFDXVH LW UHSUHVHQWV DQ DSSUR[LPDWH ORZHU ERXQG IRU WKH YDOLGLW\ RI %HUQRXOOL(XOHU EHQGLQJ WKHRU\ ,Q XQLIRUP EHDPV ZKHQHYHU /G WKH HIIHFW RI VKHDU LV QR ORQJHU QHJOLJLEOH $V DQ H[DPSOH RI ZK\ FDUH VKRXOG EH H[HUFLVHG LQ VHOHFWLQJ WKH UDQJH RI SDUDPHWHUV WR JHQHUDWH GDWD 'L[RQ f REWDLQHG UHVXOWV IRU EG KDYLQJ DVVXPHG WKH HIIHFW RI VKHDU QHJOLJLEOH KH FRQVLGHUHG RQO\ WKH UDQJH RI /G ZKHUH LW LV VLJQLILFDQW

PAGE 183

&+$37(5 9,, &200(176 21 180(5,&$/ ,167$%,/,7< ,1 7+( 28$6,/,1($5,=$7,21 $/*25,7+0 ,QWURGXFWLRQ 4XDVLOLQHDUL]DWLRQ LV DQ LQGLUHFW W\SH RI QXPHULFDO VFKHPH LQ ZKLFK D VROXWLRQ LV REWDLQHG IRU WKH GLIIHUHQWLDO HTXDWLRQV WKDW PXVW EH VDWLVILHG E\ WKH RSWLPDO VROXWLRQ 6LQFH D VWDQGDUG ,%0 6+$5( SURn JUDP LV XVHG ZLWKRXW PRGLILFDWLRQ WKH PHWKRG LV QRW GLVFXVVHG ,QWHUn HVWHG UHDGHUV DUH UHIHUUHG WR DQ\ VWDQGDUG WH[WERRN RQ QXPHULFDO WHFKn QLTXHV RU WR *KRVK f 7KLV W\SH RI DOJRULWKP UHTXLUHV DQ LQLWLDO JXHVV RI WKH VWDWH DQG WKHQ LWHUDWHV IURP WKDW JXHVV WR WKH VROXWLRQ &RQWURO IXQFWLRQV DVVRFLDWHG ZLWK WKH ILQDO LWHUDWH DUH WKH RSWLPDO FRQWURO IXQFWLRQV DFWXDOO\ VRXJKW ,I WKH SURFHVV FRQYHUJHV DW DOO WKH FRQYHUJHQFH LV UDSLG DQG W\SLFDOO\ UHTXLUHV IHZHU WKDQ WHQ LWHUDWHV +RZHYHU WKH SDUDPRXQW GLIILFXOW\ RI XVLQJ WKLV PHWKRG LV WR VHOHFW DQ LQLWLDO JXHVV ZKLFK LV JRRG HQRXJK WR \LHOG FRQYHUJHQFH 7KLV FKDSWHU FRQWDLQV VRPH QXPHULFDO GDWD UHODWHG WR VSHFLILF FDVHV RI WKH H[DPSOH SUREOHPV ZKHUH FRQYHUJHQFH GLIILFXOWLHV DUH HQFRXQWHUHG &RPSXWHU 3URJUDP &RQYHUJHQFH )HDWXUHV $V H[SODLQHG LQ VHFWLRQ FRQYHUJHQFH LV PHDVXUHG E\ ZKHWKHU RU QRW WKH LWHUDWH VDWLVILHV WKH GLIIHUHQWLDO HTXDWLRQ ,Q WHUPV RI

PAGE 184

WKH TXDQWLW\ (5525 GHILQHG E\ f LQ VXEURXWLQH 48$6, WKH WHVW IRU QXPHULFDO LQVWDELOLW\ LQYROYHV WKUHH SURJUDP SDUDPHWHUV (0$; WKH YDOXH RI (5525 IRU WKH LQLWLDO JXHVV (0$; WKH YDOXH RI (5525 IRU VRPH VXEVHTXHQW LWHUDWLRQ %/2 D VSHFLILHG LQVWDELOLW\ SDUDPHWHU :KHQHYHU WKH UDWLR RI (0$; WR (0$; H[FHHGV %/2 WKH SURJUDP WHUPLQDWHV RQ DQ HUURU H[LW 7\SLFDO YDOXHV RI %/2 DUH WR 7KH SULPDU\ PRGH RI FRQYHUJHQFH LV IRU (0$; ZKLFK GHFUHDVHV ZLWK HDFK LWHUDWLRQ XQGHU QRUPDO FLUFXPVWDQFHV WR EHFRPH VPDOOHU WKDQ D VSHFLILHG YDOXH DVVXPHG WR EH [ f 2WKHU VSHFLDO PRGHV RI FRQYHUJHQFH FRQWDLQHG LQ 48$6, ZHUH QHYHU XVHG E\ DQ\ RI WKH FDVHV UXQ IRU HLWKHU H[DPSOH SUREOHP 7KH SXUSRVH RI WKLV WHVW LV WZRIROG ,Q VRPH LQVWDQFHV WKH LQLWLDO JXHVV LV QRW JRRG HQRXJK DQG WKH ILUVW IHZ LWHUDWLRQV SRVVHVV (0$; YDOXHV JUHDWHU WKDQ (0$; EXW LV VXIILFLHQWO\ JRRG IRU WKH DOJRn ULWKP WR XOWLPDWHO\ FRQYHUJH 7KH RWKHU SXUSRVH LV WR SUHYHQW H[FHVVLYH FRPSXWDWLRQ IRU WKRVH FDVHV ZKHUH WKH LQLWLDO JXHVV LV VR EDG WKDW WKH QXPHULFDO SURFHVV LV GLYHUJHQW $QRWKHU UHODWHG FRQYHUJHQFH WHVW LV UHTXLUHG E\ WKH VXEURXWLQH 5(&,3 ZKLFK SHUIRUPV UHTXLUHG PDWUL[ LQYHUVLRQ 7KLV WHVW LQYROYHV D SURJUDP SDUDPHWHU 680 ZKLFK LV GHILQHG LQ WHUPV RI D PDWUL[ WR EH LQYHUWHG DQG WKH LQYHUVH REWDLQHG QXPHULFDOO\ E\ 5(&,3 ,I & DQG & GHQRWH DSSURSULDWH HOHPHQWV UHVSHFWLYHO\ WKHQ 680 ( ( ,&7 & n [N NM N LM

PAGE 185

,I WKH PDWUL[ KDV EHHQ LQYHUWHG H[DFWO\ WKH YDOXH RI 680 LV ]HUR :KHQHYHU 680 LV OHVV WKDQ SURJUDP SDUDPHWHU 2))'* WKH LQYHUVLRQ LV FRQn VLGHUHG WR EH VXIILFLHQWO\ DFFXUDWH 7KH YDOXH VHOHFWHG IRU DOO FDVHV IRU ERWK H[DPSOH SUREOHPV LV [ :KHQHYHU WKH LQYHUWHG PDWUL[ IDLOV WR VDWLVI\ WKH WHVW RQ 680 WKH VXESURJUDP LWHUDWHV XSRQ WKH LQYHUVH WR LPSURYH LW ,I WKH LPSURYHG LQYHUVH VDWLVILHV WKH DFFXUDF\ UHTXLUHPHQW WKH UHVXOW LV UHWXUQHG WR 48$6, $IWHU ILYH XQVXFFHVVIXO LPSURYHPHQW DWWHPSWV UHVXOWV RI WKH ILIWK DWWHPSW DUH UHWXUQHG WR 48$6, $ VLQJXODU PDWUL[ UHVXOWV LQ LPPHGLDWH WHUPLQDWLRQ RI DOO FDOFXODWLRQV YLD DQ HUURU H[LW 'XULQJ WKH RSHUDWLRQ RI WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP WZR GLIIHUHQW PDWULFHV PXVW EH LQYHUWHG 2QH RI WKHVH H[LVWV IRU HYHU\ f§L HOHPHQW L DQG HDFK VHSDUDWH (K PXVW EH LQYHUWHG )URP DOO RI WKH ,K DQRWKHU PDWUL[ LV GHILQHG ZKLFK DOVR UHTXLUHV LQYHUVLRQ 7KH VSHFLILF IRUP RI WKHVH PDWULFHV LV QRW LPSRUWDQW KRZHYHU WKHLU GHILQLWLRQV PD\ EH IRXQG LQ *KRVK SS f ,W LV QRWHZRUWK\ WKDW LQ WKH FDVHV UXQ IRU WKH H[DPSOH SUREOHPV QR GLIILFXOWLHV ZHUH H[SHULHQFHG LQ WKH LQYHUVLRQ RI DQ\ ,f PDWUL[ $OO PDWUL[ LQYHUVLRQ SUREOHPV ZHUH DVVRn FLDWHG ZLWK WKH PDWUL[ 1XPHULFDO ,QVWDELOLWLHV IRU &DQWLOHYHU %HDP ([DPSOH $Q LQLWLDO JXHVV IRU WKH EHDP SUREOHP ZDV REWDLQHG IURP WKH VROXWLRQ IRU D XQLIRUP EHDP WKDW LV IRU SDUDPHWHU YDOXHV DF EG &DVHV ZHUH WKHQ UXQ IRU D UDQJH RQ ERWK SDUDPHWHUV IURP WR $V ZRXOG EH H[SHFWHG QR FRQYHUJHQFH GLIILFXOWLHV ZHUH HQFRXQWHUHG ZLWK

PAGE 186

FRPELQDWLRQV RI SDUDPHWHU YDOXHV FRUUHVSRQGLQJ WR D QHDUO\ XQLIRUP EHDP ,Q IDFW ZLWK WKH LQLWLDO JXHVV EDVHG XSRQ WKH XQLIRUP EHDP RQO\ WKUHH FDVHV IDLOHG WR FRQYHUJH )RU WKLV LQLWLDO JXHVV JLYHQ E\ f WKH IROORZLQJ WKUHH FDVHV H[KLELWHG QXPHULFDO LQVWDELOLW\ DF EG f $OO RWKHU FRPELQDWLRQV RI SDUDPHWHU YDOXHV UHVXOWHG LQ FRQYHUJHQFH E\ WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP 6LQFH DOO RI WKRVH FDVHV ZKLFK GLG FRQYHUJH VKRZHG YHU\ OLWWOH FKDQJH LQ VWUXFWXUDO VWDWH FRPSRQHQWV [AWf L WKH XQLIRUP EHDP LQLWLDO JXHVV ZDV UHSODFHG E\ D JXHVV EDVHG XSRQ WKH FDVH ZKHUH DF DQG EG ,W ZDV VXVSHFWHG WKDW WKH DGMRLQW YDULDEOH JXHVVHV ZHUH QRW VXIILFLHQWO\ JRRG +HQFH WKH HTXDWLRQV f ZHUH UHSODFHG E\ ;LWf r W ["Wf r W [ Wf r Wf [Wf r Wf 3Wf r W 3Wf r W W f :LWK WKLV LQLWLDO JXHVV RQO\ RQH FDVH FRUUHVSRQGLQJ WR WKH WKUHH VHWV RI SDUDPHWHU YDOXHV f IDLOHG GXH WR QXPHULFDO LQVWDELOLW\ WKH ILUVW WZR FDVHV RI WKHVH FDVHV FRQYHUJHG

PAGE 187

+DYLQJ VXFFHHGHG LQ IRUFLQJ WZR DGGLWLRQDO FDVHV WR FRQYHUJH E\ LPSURYLQJ WKH LQLWLDO JXHVV WKLV VDPH DSSURDFK ZDV WULHG ZLWK WKH WKLUG FDVH DV ZHOO 7KH LQLWLDO JXHVV ZDV EDVHG XSRQ WKH FRQYHUJHG VROXWLRQ IRU WKH FDVH ZLWK DF DQG EG (TXDWLRQV f ZHUH UHSODFHG E\ [ Wf r W [Wf r W [Wf r Wf [Wf r Wf SWf r W SWf r W W f :LWK WKLV LQLWLDO JXHVV WKH FDVH IRU DF DQG EG FRQYHUJHG DOVR 7KLV ODVW FDVH JDYH D FRPSOHWH GDWD VHW IRU SDUDPHWHU YDOXHV DF EG LQ LQFUHPHQWV RI 1R FDVHV IRU SDUDPHWHU YDOXHV HTXDO WR ]HUR ZHUH UXQ VLQFH WKH SDUDPHWHUV DSSHDU LQ WKH GHQRPLQDWRU RI YDULRXV H[SUHVVLRQV $V PHQWLRQHG HDUOLHU LQGLUHFW PHWKRGV VXFK DV TXDVLOLQHDUL]Dn WLRQ JHQHUDOO\ FRQYHUJH UDSLGO\ LI WKH LQLWLDO JXHVV LV VXIILFLHQWO\ JRRG 7R WU\ WR GHWHUPLQH ZK\ WKH WKUHH FDVHV IDLOHG WR FRQYHUJH DOO RI WKH FDVHV ZHUH FORVHO\ H[DPLQHG 7ZR FKDUDFWHULVWLFV ZHUH REVHUYHG Lf 7KHUH LV OLWWOH FKDQJH LQ WKH PD[LPXP YDOXH RI HDFK VHSDUDWH VWDWH FRPSRQHQW UHJDUGOHVV RI WKH SDUDPHWHU YDOXHV &XUYDn WXUH RI [Wf DQG [AWf GLG FKDQJH IRU ORZHU SDUDPHWHU YDOXHV

PAGE 188

WKLV LV HYLGHQW LQ GLIIHUHQW H[SUHVVLRQV XVHG IRU WKH LQLWLDO JXHVV RI WKHVH WZR YDULDEOHV +RZHYHU WKH YDUn LDEOH WKDW DFWXDOO\ FDXVHG WKH LQVWDELOLW\ WHUPLQDWLRQ ZDV SAWf DQG QRW D VWDWH FRPSRQHQW LLf 7KH PD[LPXP YDOXH RI ERWK DGMRLQW YDULDEOHV LQFUHDVHV ZLWK GHFUHDVLQJ SDUDPHWHU YDOXHV ZLWK DQ DVVRFLDWHG LQFUHDVH LQ FXUYDWXUH :LWK SAWf LVRODWHG DV WKH VRXUFH RI FRQYHUJHQFH GLIILFXOWLHV IRU FDVHV ZLWK VPDOO SDUDPHWHU YDOXHV SAWf ZDV SORWWHG IRU VHYHUDO FRPELQDWLRQV RI VPDOO SDUDPHWHU YDOXHV 7KH LQFUHDVHG FXUYDWXUH PHQWLRQHG DERYH LV VKRZQ LQ )LJXUH $OO FDVHV WKDW IDLOHG WR FRQYHUJH H[KLELW D QXPHULFDO LQVWDELOLW\ WHUPLQDWLRQ IRU ZKLFK SAWf EHFRPHV GLYHUJHQW 0D[LPXP HUURU LQ VDWLVn I\LQJ WKH SA HTXDWLRQ RFFXUV LQ WKH UHJLRQ W IRU WKHVH GLYHUJHQW UXQV )URP WKH ILJXUH LW FDQ EH VHHQ WKDW WKLV LV WKH UHJLRQ RI JUHDWHVW FKDQJH LQ WKH FXUYDWXUH RI WKH VROXWLRQ SAWf 7KLV VXJn JHVWV WKDW WKH OLQHDUL]DWLRQ JUDGLHQWV LQ WKH TXDVLOLQHDUL]DWLRQ DOJRn ULWKP PD\ QRW EH YDOLG ,I WKDW LV WKH FDVH WKHQ WKH OLQHDU H[SDQVLRQ DERXW WKH NQRZQ LWHUDWH FDQQRW EH H[SHFWHG WR UHVXOW LQ DQ LPSURYHPHQW $ PRGLILFDWLRQ GHYHORSHG E\ *KRVK f WR LQKLELW LQVWDELOLWLHV DVVRn FLDWHG ZLWK WRRODUJH LPSURYHPHQWV KDV QR HIIHFW XSRQ WKHVH QXPHULFDOO\ GLYHUJHQW FDVHV 5HPRYDO RI WKLV SRVVLEOH VRXUFH RI QXPHULFDO LQVWDELOn LW\ IXUWKHU LPSOLHV WKH GLIILFXOW\ LV UHODWHG WR WKH OLQHDULW\ DVVXPSWLRQV $Q H[SODQDWLRQ IRU WKH LQFUHDVLQJ VORSH LV DYDLODEOH IURP WKH GLIIHUHQWLDO HTXDWLRQ IRU WKH DGMRLQW YDULDEOHV ZKHUH

PAGE 189

)LJXUH 6ROXWLRQV IRU $GMRLQW 9DULDEOH SAWf IRU &HUWDLQ &DVHV RI ,QWHUHVW

PAGE 190

3 OWfXX 3f f 3f 6LQFH ERWK XA DQG X DUH HTXDO WR WKHLU PLQLPXP DOORZDEOH YDOXHV QHDU W S! LQFUHDVHV ZLWK GHFUHDVLQJ SDUDPHWHU YDOXHV LQ WKLV UHJLRQ +RZHYHU WKH WHUP Wf IRUFHV S WR ]HUR DW W )RU XQNQRZQ UHDn VRQV WKH ODUJH FXUYDWXUH DVVRFLDWHG ZLWK WKLV HIIHFW VHHPHG WR SRVH QR GLIILFXOWLHV IRU WKH QXPHULFDO DOJRULWKP $V VHHQ LQ HTXDWLRQV f LI 3Wf FDQ EH GHWHUPLQHG WKHQ SAWf LV UHDGLO\ REWDLQHG &XUYHV IRU SWf FRUUHVSRQGLQJ WR FDVHV SUHVHQWHG LQ )LJXUH DUH JLYHQ LQ )LJXUH $OO VL[ FDVHV H[KLELW SRVLWLYH FXUYDWXUH DQG SRVVHVV QRQH RI WKH LQIOHFWLRQ SRLQWV VHHQ LQ WKHLU FRXQWHUSDUWV RI )LJXUH 7KH LQFUHDVH LQ DPSOLWXGH RI WKHVH WZR YDULDEOHV IRU GHFUHDVLQJ YDOXHV RI EG LV LOOXVWUDWHG LQ )LJXUH ,Q WKLV ILJXUH LW LV VHHQ WKDW ZLWK UHVSHFW WR LQFUHDVLQJ DPSOLWXGH DV EG } 3Wf LV WKH FULWLFDO YDULDEOH 7KH PDMRU GLVDGYDQWDJH RI WKH LQGLUHFW VROXWLRQ PHWKRG LV WKDW OLWWOH LV NQRZQ DERXW WKH FRQQHFWLRQ EHWZHHQ FRQYHUJHQFH DQG UHTXLUHn PHQWV XSRQ WKH LQLWLDO JXHVV )RU WKLV VSHFLILF H[DPSOH RQO\ RQH RI WKH VHYHQ YDULDEOHV FDXVHV FRQYHUJHQFH GLIILFXOWLHV DQG WKH UHVXOWn LQJ QXPHULFDO LQVWDELOLW\ RFFXUV LQ D UHJLRQ RI ODUJH UDSLGO\ FKDQJLQJ FXUYDWXUH 1R FRQYHUJHQFH SUREOHPV DUH HQFRXQWHUHG ZKHQ WKH FXUYDWXUH RI WKH LQLWLDO JXHVV PRUH FORVHO\ UHVHPEOHV WKH ILQDO VROXWLRQ $OO RI WKLV VXJJHVWV WKDW WKH TXHVWLRQ RI ZKHWKHU RU QRW DQ LQLWLDO JXHVV DOORZV WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP WR FRQYHUJH PD\ GHSHQG XSRQ WKH UHVSHFWLYH FXUYDWXUHV RI WKH LQLWLDO JXHVV DQG WKH VROXWLRQ

PAGE 191

)LJXUH 6ROXWLRQV IRU $GMRLQW 9DULDEOH SAWf IRU &HUWDLQ &DVHV RI ,QWHUHVW

PAGE 192

3Wf SAWf f EG )LJXUH 7HUPLQDO 3RLQW $PSOLWXGH RI SAWf DQG SAWf 'HSHQGHQFH XSRQ 3DUDPHWHU EG 9DOXH

PAGE 193

1XPHULFDO ,QVWDELOLWLHV IRU &ROXPQ %XFNOLQJ ([DPSOH $QRWKHU W\SH RI FRQYHUJHQFH IDLOXUH ZDV H[SHULHQFHG ZLWK WKH FROXPQ EXFNOLQJ SUREOHP ,Q WKHVH FDVHV ZKLFK GLG QRW FRQYHUJH HDFK RI WKH YDULDEOHV ZDV GLYHUJHQW LQFUHDVLQJ ZLWK HDFK LWHUDWLRQ EHJLQn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f ZDV DOVR LQHIIHFWLYH $ SRVVLEOH VRXUFH RI WKH QXPHULFDO GLIILFXOW\ ZDV WKH PDWUL[ LQYHUVLRQ )RU WKRVH FDVHV IDLOLQJ WR FRQYHUJH HDFK RI WKH (K PDWULFHV ZDV LQYHUWHG DFFXUDWHO\ EXW WKH LQYHUVLRQ RI & ZDV TXHVWLRQDEOH DV LQGLFDWHG E\ WKH YDOXH RI 680 6LQFH HDFK LWHUDWH GHSHQGV GLUHFWO\

PAGE 194

XSRQ WKH LQYHUVH RI -& D GHJUDGDWLRQ LQ ZRXOG FDXVH D FRUUHVSRQGn LQJ GHJUDGDWLRQ LQ WKH VROXWLRQ DW HDFK LWHUDWLRQ 7R LOOXVWUDWH ZKDW KDSSHQV LQ WKHVH FDVHV FRQVLGHU D VSHFLILF H[DPSOH D D $ SORW RI YDOXHV RI 680 YHUVXV W WKH -M X X LQWHUFHSW RI WKH XSSHU ERXQG LV VKRZQ LQ )LJXUH 7KHVH GDWD SRLQWV UHSUHVHQW WKH A DFFXUDF\ DW GLIIHUHQW LWHUDWLRQV IRU YDULRXV LQLWLDO JXHVVHVf§DFFHSWDEOH DFFXUDF\ LV LQGLFDWHG E\ WKH VWUDLJKW OLQH %DVLn FDOO\ LI DQ LQLWLDO JXHVV \LHOGHG D W RI DSSUR[LPDWHO\ RQHKDOI HDFK LWHUDWLRQ JLYHV DQ LQFUHDVHG YDOXH RI WA ZLWK D FRUUHVSRQGLQJO\ OHVV DFFXUDWH &f $V WA DSSURDFKHV RQH WKH VROXWLRQ DSSURDFKHV WKH OLPLWn LQJ XQLIRUP FROXPQ 7KH UHVXOW RI WKLV LV WKDW & PD\ EHFRPH LOOFRQGLWLRQHG LI WKH SDUDPHWHUV DUH FKRVHQ VXFK WKDW WKH VROXWLRQ LV VLPLODU WR D XQLIRUP FROXPQ 7KHUH DUH QR H[SOLFLW GDWD RU GHULYDWLRQV WR LQGLFDWH WKLV LV WKH VLWXDWLRQ RQO\ WKH LPSOLFLW VXJJHVWLRQ RI WKH GDWD LQ )LJXUH +RZHYHU DV VWDWHG LQ WKH SUHFHGLQJ VHFWLRQ WKHUH LV OLWWOH NQRZQ DERXW ZK\ FHUWDLQ LQLWLDO JXHVVHV IDLO WR FRQYHUJH LQ WKH LQGLUHFW VROXWLRQ DOJRULWKPV

PAGE 195

680 [ [ D M k k k DX k R k k k ? $FFHSWDEOH 0DWULF ,QYHUVLRQ W 8 )LJXUH 'HJUDGDWLRQ RI & A $FFXUDF\ DV ,WHUDWH $SSURDFKHV D 6ROXWLRQ 6LPLODU WR WKH 8QLIRUP &ROXPQ /LPLWLQJ 6ROXWLRQ

PAGE 196

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n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r ‘S + + Bf ZKLFK LQ XVSDFH GHQRWHV WKH GLUHFWLRQ RI FRQVWUDLQHG

PAGE 197

PD[LPXP GHVFHQW ZKHQHYHU MM LV JLYHQ E\ HTXDWLRQ f :KLOH WKLV DWWULEXWH LV QRW SXUVXHG IXUWKHU LQ WKH GLVVHUWDWLRQ LW PD\ SURYH XVHn IXO LQ VXEVHTXHQW DSSOLFDWLRQV RI WKH PD[LPXP SULQFLSOH 7R IXUWKHU LOOXVWUDWH WKH QRQOLQHDU SURJUDPPLQJ SULQFLSOHV XVHG LQ WKH WKHRUHWn LFDO GHYHORSPHQW WKH H[DPSOH RI &KDSWHU ,9 LV DSSUR[LPDWHG E\ ILQLWH HOHPHQWV DQG VROYHG E\ D JUDGLHQW SURMHFWLRQ DOJRULWKP 2SHUDWLRQ RI WKH DOJRULWKP LV LOOXVWUDWHG DQG WKH UHVXOWV FRPSDUHG WR WKRVH REWDLQHG ZLWK WKH PD[LPXP SULQFLSOH :KHQ D VXIILFLHQW QXPEHU RI ILQLWH HOHn PHQWV DUH XVHG WR DGHTXDWHO\ GHVFULEH WKH VWUXFWXUH WKH UHVXOWV RI WKH WZR PHWKRGV DUH LGHQWLFDO $ SUREOHP FRQFHUQLQJ WKH PLQLPXP GHIOHFWLRQ RI D EHDP LV RSWLn PL]HG E\ XVLQJ FODVVLFDO PD[LPXP SULQFLSOH WHFKQLTXHV ,W LV VKRZQ WKDW QR ILQLWH VROXWLRQ LV REWDLQHG IRU XQFRQVWUDLQHG FRQWURO WKLV DOVR KROGV IRU WKH FROXPQ EXFNOLQJ SUREOHP WUHDWHG LQ &KDSWHU 9 )XUWKHUn PRUH LW LV VKRZQ WKDW IRU WKH FRQVHUYDWLYH V\VWHP FRUUHVSRQGLQJ WR WKH EXFNOLQJ SUREOHP WKH PLQLPXP ZHLJKW DQG PD[LPXP EXFNOLQJ SUREOHPV DUH LGHQWLFDO $QRWKHU UHVXOW RI XVLQJ WKH PD[LPXP SULQFLSOH LV WKDW IRU WKH SUREOHP GHVFULEHG LQ RUGHU WR KDYH D QRQWULYLDO VROXWLRQ WKH PRGXn OXV DQG GHQVLW\ PXVW EH WKH PD[LPXP DQG PLQLPXP DOORZDEOH FRQVWDQW YDOXHV UHVSHFWLYHO\ 7KLV LV D PDWKHPDWLFDO YHULILFDWLRQ RI DQ LQWXLn WLYH UHVXOW 1XPHULFDO LQVWDELOLWLHV H[SHULHQFHG E\ WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP DUH GHWDLOHG LQ &KDSWHU 9,, :KLOH QR FRQFUHWH UHVXOWV DUH REWDLQHG IURP WKH GDWD SUHVHQWHG LW DSSHDUV WKDW WKH TXHVWLRQ RI FRQn YHUJHQFH LV UHODWHG WR WKH UHVSHFWLYH FXUYDWXUHV RI WKH LQLWLDO JXHVV DQG WKH VROXWLRQ

PAGE 198

5HFRPPHQGDWLRQV )XUWKHU ZRUN VXJJHVWHG LQ WKH FRXUVH RI GLVVHUWDWLRQ UHVHDUFK IDOOV LQWR WKUHH FDWHJRULHV VWUXFWXUDO FRQVLGHUDWLRQV PDWKHPDWLFDO WKHRU\ DQG QXPHULFDO VROXWLRQ WHFKQLTXHV 8QGHU WKH ILUVW FDWHJRU\ WKH PRVW REYLRXV H[WHQVLRQ LV WR QRW LQYRNH WKH OLQHDU EHQGLQJ DVVXPSn WLRQ 6LQFH WKH TXDVLOLQHDUL]DWLRQ DOJRULWKP KDQGOHV QRQOLQHDU 73%93 WKHRUHWLFDOO\ WKHUH LV QR QHHG IRU WKH DVVXPSWLRQ ,Q FRQWUDVW WR WKH ILUVW RI HTXDWLRQV f WKH FRPSOHWH QRQOLQHDU EHQGLQJ SUREOHP LV GHVFULEHG E\ VWfKf 0Wf>O œ@ W 0Wf $>QOf XWf@ N P >QQWf @Ge W Q f f $QRWKHU LQWHUHVWLQJ SRVVLELOLW\ LV WR LQFOXGH WKH HIIHFW RI VKHDU 7KHVH VWUXFWXUDO DVSHFWV SHUWDLQ WR VSHFLILF SUREOHPV $ PRUH JHQHUDO UHVXOW DQG KHQFH PRUH LQWHUHVWLQJ LV UHODWHG WR H[SDQGLQJ WKH FODVV RI SUREOHPV WKDW FDQ EH WUHDWHG E\ HQHUJ\ WHFKQLTXHV ,W LV VXJn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

PAGE 199

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f WR HUURU VRXUFHV VXFK DV LQLWLDO JXHVV VWHS VL]H HWF IRU D FDVH ZLWK D NQRZQ VROXWLRQ /LWWOH LV NQRZQ DERXW WKLV VXEMHFW

PAGE 200

$33(1',;(6

PAGE 201

$33(1',; $ +,6725,&$/ '(9(/230(176 7KH GLVVHUWDWLRQ VXEMHFW PDWWHU RULJLQDWHG IURP WZR SDSHUV D VKRUW VXUYH\ RI VWUXFWXUDO RSWLPL]DWLRQ SUREOHPV E\ 3UDJHU DQG 7D\ORU f DQG D WHFKQLFDO QRWH E\ %R\NLQ DQG 6LHUDNRZVNL f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n GL[ LV D VKRUW ELRJUDSKLFDO WDEOH WKDW OLVWV WKH OLIH VSDQ DQG QDWLRQDOn LW\ RI PRVW RI WKH SHRSOH PHQWLRQHG

PAGE 202

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f 0 WKH SDWK $0% GHVFHQGLQJ DORQJ ZKLFK E\ LWV RZQ JUDYLW\ DQG EHJLQQLQJ WR EH XUJHG IURP WKH SRLQW $ LW PD\ LQ WKH VKRUWHVW WLPH UHDFK WKH RWKHU SRLQW % $FFRUGLQJ WR YDULRXV VRXUFHV VROXWLRQV ZHUH RIIHUHG E\ 1HZWRQ /HLEQHW] GH On+RSLWDO DQG WKH %HUQRXOOLV $W /HLEQHW]n UHTXHVW WKH VROXWLRQ ZDV ZLWKKHOG WR HQFRXUDJH RWKHUV WR FRQVLGHU WKH SUREOHP ,Q -DQXDU\ WKH SUREOHP ZDV UHDQQRXQFHG ,Q WKH IROORZLQJ 0D\ LVVXH RI $FWD (UXGLWRUXP WKH VROXWLRQV RI WKH %HUQRXOOLV DQG WKH 0DUTXLV GH On+RSLWDO ZHUH SXEOLVKHG -RKQnV LV DOOHJHG WR EH WKH PRVW UHDGDEOH EXW VLQFH LW LV VROYHG DV DQ DQDORJ WR DQ RSWLFV SUREOHP WKH PHWKRG LV UHVWULFWHG WR D VPDOO FODVV RI SUREOHPV 7KH VROXWLRQ RI -DPHV LV TXLWH JHRPHWULFDO WUHDWLQJ WKH FXUYH DV D SRO\JRQ ZLWK VLGHV RI LQILQLWHVLPDO OHQJWK ,W ZDV DOVR DVVXPHG WKDW ZKDWHYHU RSWLPDO

PAGE 203

SURSHUW\ WKH HQWLUH FXUYH SRVVHVVHG ZDV DOVR SRVVHVVHG E\ HDFK SDUW 7KH VROXWLRQ RI GH On+RSLWDO ZDV JLYHQ EXW ZLWKRXW SURRI $W WKH FRQFOXVLRQ RI KLV SDSHU -DPHV SRVHG WZR PRUH GLIILFXOW SUREOHPV )LUVW ILQG WKH FXUYH RI TXLFNHVW GHVFHQW IURP D IL[HG SRLQW $ WR D JLYHQ YHUWLFDO OLQH 7KLV YHUVLRQ RI WKH EUDFKLVWRFKURQH SUREOHP LQYROYHV FRQGLWLRQV IRU PLQLPL]LQJ D GHILQLWH LQWHJUDO ZKHUH YDULDWLRQV DW WKH OLPLWV DUH SHUPLWWHG 6XFK FRQGLWLRQV OHDG WR QDWXUDO ERXQGDU\ FRQGLWLRQV RU WUDQVYHUVDOLW\ FRQGLn WLRQV DQG ZHUH QRW REWDLQHG XQWLO VRPH \HDUV ODWHU E\ /DJUDQJH 6HFRQG DPRQJ DOO FXUYHV ZLWK JLYHQ OHQJWK DQG EDVH ILQG WKH WKH FXUYH VXFK WKDW D VHFRQG FXUYH ZKRVH FRRUGLQDWHV DUH VRPH IXQFWLRQ RI WKH ILUVW FRQWDLQV D PD[LPXP RU PLQLPXP DUHD 7KLV VXEVLGLDU\ FRQGLWLRQ LV FDOOHG DQ LVRSHULPHWULFDO FRQVWUDLQW %HVLGHV LQYLWLQJ DOO PDWKHPDWLFLDQV WR DWWHPSW VROXWLRQV -DPHV VSHFLIn LFDOO\ RIIHUHG KLV EURWKHU -RKQ D SUL]H RI ILIW\ GXFDWV IRU WKH FRUUHFW VROXWLRQf§DFFRUGLQJ WR %OLVV S f -RKQ FODLPHG WKH DZDUG EXW -DPHV UHIXVHG VKRZLQJ WKDW WKH DVVXPSWLRQ RI D XQLIRUP RSWLPDOLW\ SURSn HUW\ IRU HDFK LQILQLWHVLPDO HOHPHQW ZDV QRW YDOLG -DPHV SXEOLVKHG WKH FRUUHFW VROXWLRQV LQ WKH 0D\ LVVXH RI $FWD (UXGLWRUXP WR LQFOXGH WKH HIIHFW RI WKH FRQVWUDLQW KH XVHG WKUHH DGMDFHQW SRO\JRQ HOHPHQWV RI LQILQLWHVLPDO OHQJWK LQVWHDG RI WZR WKDW VXIILFHG IRU WKH EUDFKLVWRFKURQH DQG QR IHZHU WKDQ IRXU VROXWLRQ WHFKQLTXHV f (TXLOLEULXP RI IRUFHV RQ DQ HOHPHQW f (TXLOLEULXP RI PRPHQWV RQ DQ HOHPHQW f 3ULQFLSOH RI YLUWXDO ZRUN f 3ULQFLSOH RI PLQLPXP SRWHQWLDO HQHUJ\ :KHWKHU RU QRW WKH SUL]H H[LVWHG LQ LPPDWHULDO LW LV WUXH WKH SUREOHP HPEURLOHG WKH EURWKHUV LQ D ELWWHU TXDUUHO DQG WRJHWKHU ZLWK WKH EUDFKn LVWRFKURQH SUREOHP PDUNV WKH EHJLQQLQJ RI WKH FDOFXOXV RI YDULDWLRQV

PAGE 204

$OO RI WKLV ZRUN E\ WKH %HUQRXOOLV ZDV EDVHG XSRQ FRPSOLFDWHG JHRPHWULFDO UHVXOWV $ VWXGHQW RI WKHLUV /HRQKDUG (XOHU H[SDQGHG DQG JHQHUDOL]HG WKHLU LQLWLDO GHYHORSPHQWV +LV ILUVW PDMRU FRQWULEXWLRQ WR WKH FDOFXOXV RI YDULDWLRQV ZDV WKH GLIIHUHQWLDO HTXDWLRQ EHDULQJ KLV QDPH (XOHU VXJJHVWHG WKDW IRU FRQWLQXRXV IXQFWLRQV D GHILQLWH LQWHJUDO YDQLVKHV LGHQWLFDOO\ LI WKH LQWHJUDQG YDQLVKHV :KHQ WKH LQWHJUDQG LV WKH SURGXFW RI VRPH IXQFWLRQ 0[[Wf DQG DQ DUELWUDU\ YDULDWLRQ 6[Wf WKH IXQFWLRQ PXVW YDQLVK LI WKH YDULDWLRQ LV WR EH DUELWUDU\ (XOHUnV HTXDWLRQ LV 0[[Wf XS WR WKH ILUVW SDUW RI WKLV FHQWXU\ WKLV YDQLVKLQJ LQWHJUDQG ZDV UHIHUUHG WR DV (XOHUnV )XQGDPHQWDO /HPPDf§ (XOHUnV HTXDWLRQ LV WKH PRUH UHFHQW DSSHOODWLRQ +H DOVR ZDV WKH ILUVW WR LQGLFDWH WKH WZR JHQHUDO FODVVHV RI RSWLPDO VROXWLRQV REWDLQHG IURP KLV WHFKQLTXH DEVROXWH DQG UHODWLYH $IWHU UHDGLQJ D PHPRLU E\ /DJUDQJH ZKLFK LQWURGXFHG WKH V\PERO (XOHU DGRSWHG LWV XVH IRU KLV RZQ VWXGLHV DQG QDPHG LW LQ DQ DUWLFOH HQWLWOHG (OHPHQWLD &DOFXOL 9DULD WLRQXP 2WKHU LQYHVWLJDWLRQV UHODWHG WR WKLV LQFOXGH FRPELQLQJ WKH FDOn FXOXV WHFKQLTXHV RI 1HZWRQ DQG /HLEQHW] YLD GLIIHUHQWLDO HOHPHQWV DQG ZKDW LV DSSDUHQWO\ WKH HDUOLHVW VWXG\ RI LQWHJUDELOLW\ (XOHUnV OLVW RI ERRNV DQG DUWLFOHV DOVR LQFOXFEV WKH LQWURGXFWLRQ RI e WR GHQRWH VXPPDWLRQ DQG DQ LQWXLWLYH GHVFULSWLRQ RI FRQWLQXRXV IXQFWLRQVf§DOO RI ZKLFK DUH QHFHVVDU\ HOHPHQWV RI WKH FDOFXOXV RI YDULDWLRQV $V PHQWLRQHG LQ WKH SUHYLRXV SDUDJUDSK /DJUDQJH LQWURGXFHG WKH FRQFHSW DQG V\PERO XVLQJ LW HQDEOHG KLP WR GHULYH (XOHUnV HTXDWLRQ

PAGE 205

ZLWKRXW UHVRUWLQJ WR WKH LQILQLWHO\VLGHG SRO\JRQ DQG WKH DFFRPSDQ\LQJ FRPSOLFDWHG JHRPHWU\ ,W DOVR IDFLOLWDWHG WKH HYDOXDWLRQ RI SUREOHPV KDYLQJ GHILQLWH LQWHJUDOV ZLWK XQVSHFLILHG FRQGLWLRQV DW WKH OLPLWVf§ WKHVH ZHUH FDOOHG WKH GHILQLWH HTXDWLRQV DQG ODWHU WHUPV DW WKH OLPLWV EXW DUH FXUUHQWO\ UHIHUUHG WR DV QDWXUDO ERXQGDU\ FRQGLWLRQV 7KLV UHSUHVHQWHG D JHQHUDOL]DWLRQ RQ WKH W\SH RI SUREOHP RULJLQDOO\ SRVHG E\ -DPHV %HUQRXOOL /DJUDQJH LQWURGXFHG \HW DQRWKHU JHQHUDOL]Dn WLRQ E\ XVLQJ LQGHWHUPLQDWH PXOWLSOLHUV WR HQIRUFH LPSOLFLW FRQVWUDLQWV HJ GLIIHUHQWLDO HTXDWLRQV 7KH ILQDO PDMRU FRQWULEXWLRQ E\ /DJUDQJH WR EH FLWHG KHUH ZDV WKH H[WHQVLRQ WR SUREOHPV KDYLQJ GRXEOH LQWHJUDOV ZLWK IL[HG FRQGLWLRQV DW WKH OLPLWV )ROORZLQJ D VWXG\ RI WKH PDMRU ZRUNV RI 1HZWRQ /DSODFH DQG /DJUDQJH :LOOLDP 5RZDQ +DPLOWRQ DSSOLHG WKH DQDO\WLFDO PHWKRG RI /DJUDQJH WR KLV RZQ VWXG\ RI RSWLFV ,Q WKH 7KHRU\ RI 6\VWHPV RI 5D\V f DQG LWV VXSSOHPHQWV +DPLOWRQ GHGXFHG DOO RSWLFDO SURSHUWLHV IURP D VLQJOH &KDUDFWHULVWLF )XQFWLRQ XVLQJ WKH /DZ RI 9DU\LQJ $FWLRQ D JHQHUDOL]DWLRQ RI WKH /DZ RI /HDVW $FWLRQ GXH WR 3 / 0 GH 0DXSHUWLV :KHQ DSSOLHG WR G\QDPLFDO V\VWHPV LQ 2Q *HQHUDO 0HWKRGV RI '\QDPLFV f WKLV PHWKRG LV UHFRJQL]HG DV +DPLOWRQnV 3ULQFLSOH 7KH FDQRQLFDO HTXDWLRQV RI PRWLRQ GHULYHG IURP WKLV PHWKRG DUH FDOOHG +DPLOWRQnV HTXDn WLRQV DQG ZHUH XVHG E\ -DFREL WR ODWHU GHULYH WKH +DPLOWRQ-DFREL HTXDn WLRQ (XOHU GHULYHG WKH ILUVW QHFHVVDU\ FRQGLWLRQ IRU DQ RSWLPDO VROXn WLRQ LQ WKH FDOFXOXV RI YDULDWLRQV LQ WKH PDQQHU GHYHORSHG E\ /DJUDQJH WKH ILUVW YDULDWLRQ PXVW YDQLVK %XW WKLV PD\ EH VDWLVILHG E\ HLWKHU

PAGE 206

D PLQLPXP PD[LPXP RU DQ LQIOHFWLRQ SRLQW /HJHQGUH PDGH WKH ILUVW DWWHPSW WR GLVWLQJXLVK ZKDW WKH FDVH PD\ EH E\ LQYHVWLJDWLQJ WKH VHFRQG YDULDWLRQ LQ 8VLQJ D WUDQVIRUPDWLRQ RI YDULDEOHV KH DUULYHG DW ZKDW ZDV EHOLHYHG WR EH D VXIILFLHQW FRQGLWLRQ IRU HLWKHU D PD[LPXP RU D PLQLPXPf IRU ILIW\ \HDUV +RZHYHU LQ -DFREL GLVFRYHUHG WKDW /HJHQGUHnV FRQGLWLRQ IDLOHG LQ VRPH FDVHV 7KH FRQMXJDWH SRLQWVf DVVRn FLDWHG ZLWK WKHVH FDVHV ZHUH IRXQG IURP GHULYDWLYHV RI WKH VROXWLRQ WR (XOHUnV HTXDWLRQ ZLWK UHVSHFW WR WKH FRQVWDQWV RI LQWHJUDWLRQ $V D UHVXOW /HJHQGUHnV FRQGLWLRQ ZDV GHWHUPLQHG WR EH RQO\ QHFHVVDU\ DORQJ ZLWK WKH QHZHU -DFRELn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f WKH FODVVLFDO PHPRLUV RI WKH %HUQRXOOLV (XOHU /DJUDQJH /HJHQGUH DQG -DFREL DUH FRQWDLQHG LQ 2VWZDOGnV .ODVVLNHU GHU H[DFWHQ :LVVHQVFKDIWHQ

PAGE 207

7$%/( $ 0$-25 '(9(/230(176 ,1 7+( &$/&8/86 2) 9$5,$7,216 1HZWRQ ILUVW WR SRVH D SUREOHP RI D W\SH UHTXLUn LQJ WKH FDOFXOXV RI YDULDWLRQV QRW WUHDWHG DV VLJQLILFDQW DW WKH WLPH -RKQ %HUQRXOOL SRVHG WKH EUDFKLVWRFKURQH SUREOHP 6ROXn WLRQV ZHUH REWDLQHG E\ -DPHV %HUQRXOOL /HLEQHW] DQG 1HZWRQ DV ZHOO DV E\ -RKQ %HUQRXOOL DQG GH On+RSLWDO $FFRPSDQ\n LQJ KLV VROXWLRQ ZHUH WZR DGGLWLRQDO SUREOHPV SURSRVHG E\ -DPHV %HUQRXOOL RQH LQYROYLQJ QDWXUDO ERXQGDU\ FRQGLWLRQV DQRWKHU ZLWK DQ LVRSHULPHWULFDO FRQVWUDLQW -DPHV %HUQRXOOL VROYHG WKH LVRSHULPHWULFDO FRQVWUDLQW SUREOHP SRVHG LQ KLV VROXWLRQ WR WKH EUDFKLVWRFKURQH (XOHU H[WHQGHG WKH ZRUN RI WKH %HUQRXOOL EURWKHUV WR VHYHUDO FODVVHV RI SUREOHPV ZLWK YDUn LRXV FRQVWUDLQWV (XOHU LQWURGXFHG WKH WHFKQLTXH WKDW VHWWLQJ FRHIILFLHQW RI < LQ LQWHJUDQG WR ]HUR VDWLVILHV QHFHVVDU\ FRQGLWLRQ 7KH UHVXOW LV FDOOHG (XOHUnV (TXDWLRQ (XOHU GLYLGHG RSWLPL]DWLRQ SUREOHPV LQWR WZR JHQHUDO FODVVHV DEVROXWH DQG UHODWLYH 7KLV LV WKH ILUVW V\VWHPDWLF H[SRVLWLRQ DOO SUHFHGLQJ ZRUN UHSUHVHQWHG DG KRF VROXWLRQV WR VSHFLILF SUREOHPV (XOHU SXEOLVKHG RQH WUDFW JLYLQJ WKH QDPH FDOFXOXV RI YDULDWLRQV DQG DQRWKHU FRQWDLQLQJ WKH ILUVW VWXG\ RI FRQGLWLRQV RI LQWHJUDELOLW\ r /DJUDQJH LQWURGXFHV WKH V\PERO WR GLVWLQJXLVK EHWZHHQ YDULDWLRQ DQG GHULYDWLYH r /DJUDQJH GHGXFHG (XOHUnV (TXDWLRQ ZLWKRXW XVLQJ DGMDn FHQW HOHPHQWV RI DQ LQILQLWHO\VLGHG SRO\JRQ 1R VSHFLILF GDWHV DUH DYDLODEOH WZR PHPRLUV RQ WKH JHQHUDO VXEMHFW IURP ZKLFK WKHVH PLJKW RULJLQDWH ZHUH SXEOLVKHG LQ DQG

PAGE 208

7$%/( $O &RQWLQXHGf r /DJUDQJH GHWHUPLQHG WKH PHWKRG WR WUHDW XQVSHFn LILHG HQG FRQGLWLRQV WKH UHVXOWV DUH QRZ FDOOHG QDWXUDO ERXQGDU\ FRQGLWLRQV r /DJUDQJH LQWURGXFHG HQIRUFHPHQW RI LPSOLFLW FRQn VWUDLQWV E\ WKH XVH RI LQGHWHUPLQDWH PXOWLSOLHUV r /DJUDQJH H[WHQGHG WKH FDOFXOXV RI YDULDWLRQV WR LQFOXGH GRXEOH LQWHJUDOV ZLWK IL[HG ERXQGDU\ FRQGLWLRQV /HJHQGUH LQYHVWLJDWHG WKH VHFRQG YDULDWLRQ RI WKH FRVW IXQFWLRQ WR ILQG D FULWHULRQ ZLWK ZKLFK PD[LPD DQG PLQLPD FRXOG EH GLVWLQJXLVKHG )LUVW UHVXOWV ZHUH QRW UHDOO\ D VXIILFLHQW FRQGLWLRQ EHFDXVH RI QRQXQLTXH VROXWLRQV GHPRQVWUDWHG E\ /DJUDQJH %UXQDFL H[WHQGHG /HJHQGUHfV ZRUN WR GRXEOH LQWHJUDOV UHWDLQLQJ WKH VDPH JHQHUDO IODZ +DPLOWRQ GHYHORSHG WKH FDQRQLFDO HTXDWLRQV DQG +DPLOWRQnV 3ULQFLSOH EDVHG XSRQ D JHQHUDOn L]DWLRQ RI WKH /DZ RI /HDVW $FWLRQ -DFREL VWXGLHG WKH WUDQVIRUPDWLRQ RI /HJHQGUH GLVFRYHULQJ KRZ WR GHWHUPLQH ZKHQ WKH PHWKRG IDLOHG 7KH UHVXOWLQJ FRQMXJDWH SRLQWV DUH GHWHUPLQHG IURP GHULYDWLYHV RI VROXWLRQV WR (XOHUn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

PAGE 209

7$%/( $O &RQFOXGHGf %ULRVFKL &OHEVFK 7RGKXQWHU (UGPDQQ :HLHUVWUDVV %OLVV 9DOHQWLQH ZDV WKH ILUVW WR DSSO\ GHWHUPLQDQWV WR WKH LQYHVWLJDWLRQ RI VHFRQG RUGHU WHUPV JHQHUDOL]HG -DFRELnV 7KHRUHP WR VHYHUDO GHSHQGHQW YDULDEOHV ZLWK RU ZLWKRXW FRQQHFWLQJ HTXDWLRQV DQG IRU PXOWLSOH LQWHJUDOV VXJJHVWHG WKDW YDULDWLRQV PLJKW EH RI UHVWULFWHG VLJQ DOORZLQJ WKH SRVVLELOLW\ RI GLVFRQWLQXRXV VROXWLRQV GHULYHG WKH FRQGLWLRQV WKDW PXVW EH VDWLVILHG E\ D GLVFRQWLQXRXV VROXWLRQ FDOOHG WKH :HLHUVWUDVV(UGPDQQ FRQGLWLRQV KHOSHG WR PDNH WKH WKHRU\ RI WKH FDOFXOXV RI YDULDWLRQV PRUH SUHFLVH LQ JHQHUDO 6SHFLILFDOO\ KH IRXQG D QHZ QHFHVVDU\ FRQGLWLRQ LQ WHUPV RI KLV IXQFWLRQ (;<3
PAGE 210

1RV DQG +LVWRULFDO GHYHORSPHQWV DUH WUHDWHG E\ &DUOO f 7RGKXQWHU DQG 3HDUVRQ f %RO]D f DQG %OLVV f %LRJUDSKLFDO GDWD DUH IRXQG LQ 7UXHVGHOO f %HUJDPLQL f ,UHODQG f DQG $VLPRY f $QDO\WLFDO PHFKDQLFV EHFDPH D YLDEOH PHWKRG ZLWK WKH LQWURGXFn WLRQ RI FDOFXOXV E\ /HLEQHW] DQG 1HZWRQ )RU H[DPSOH *DOLOHR *DOLOHL REVHUYHG LQ WKDW WKH IUHH IDOO RI YDULRXV REMHFWV RI GLIIHUHQW VL]H DQG ZHLJKW LV JLYHQ E\ WKH IRUPXOD < W 1HZWRQ LV UHSRUWHG WR KDYH GLIIHUHQWLDWHG WZLFH WR GLVFRYHU WKDW DOO REMHFWV IDOO ZLWK DQ DFFHOHUn DWLRQ RI WKLUW\WZR IHHW SHU VHFRQG VTXDUHG
PAGE 211

$ PRUH GHWDLOHG OLVW RI WKH EDVLF GHYHORSPHQWV RI PHFKDQLFV LV JLYHQ LQ 7DEOH $ D IHZ ELRJUDSKLFDO GDWD DUH SUHVHQWHG LQ 7DEOH $ 7$%/( $ 0$-25 '(9(/230(176 ,1 $1$/<7,&$/ 0(&+$1,&6 -DPHV %HUQRXOOL -DPHV %HUQRXOOL -DPHV %HUQRXOOL -DPHV %HUQRXOOL /HLEQHW] (XOHU SXEOLVKHG WKH YHU\ ILUVW SDSHU RQ WKH PDWKHPDWLFDO WKHRU\ RI HODVWLFLW\ FRQFOXGHG D WKLUWHHQ \HDU VWXG\ ZKLFK LQFOXGHG DQ LQYHVWLJDWLRQ RI WKH FDWHQDU\ UHVXOWV ZHUH SXEOLVKHG LQ f 3UREOHP ZDV DOVR VROYHG E\ -RKQ %HUQRXOOL XVLQJ D WKHRUHP FRQFHUQLQJ WKH FHQWHU RI JUDYLW\ RI DQ DUF GHFODUHG HODVWLFLW\ LV JHQHUDOO\ QRQn OLQHDU ZLWK OLQHDU HODVWLFLW\ D VSHFLDO FDVH +H DOVR GHULYHG WKH GLIIHUHQWLDO HTXDWLRQ IRU WKH HO£VWLFD ILUVW FKDUDFWHUL]HG D PDWHULDO ZLWK VWUHVV DV D IXQFWLRQ RI VWUDLQ LQVWHDG RI WKH WKHQ DFFHSWHG ORDG YHUVXV GLVSODFHPHQW IRU D SDUWLFXODU VSHFLPHQ +H DYRLGHG WKH ( RI OLQHDU HODVWLFLW\ VXJJHVWLQJ (A E\ LPSOLFDWLRQ SHUIRUPHG WKH ILUVW DQDO\VLV RI D ORDGHG EHDP DVVXPLQJ WKDW VRPH ILEUHV ZHUH LQ WHQVLRQ 7KDW WKH WHQVLRQ YDULHG OLQn HDUO\ JDYH WKH UHVXOW WKDW WKH PRPHQW ZDV SURSRUWLRQDO WR WKH VHFRQG DUHD PRn PHQW RI WKH FURVV VHFWLRQ GHULYHG WKH %HUQRXOOL(XOHU HTXDWLRQ IRU EHQGLQJ RI EHDPV ZKLOH D VWXGHQW RI -RKQ %HUQRXOOL 7KH ZRUN ZDV QRW SXEOLVKHG XQWLO fN 1R VSHFLILF GDWHV DUH DYDLODEOH IURP RUGLQDU\ VRXUFHV LWHPV DUH ORFDWHG ZLWK UHVSHFW WR UHODWHG HQWULHV RI WKH WDEOH

PAGE 212

7$%/( $ &RQWLQXHGf 0XVVFKHQEURHN SHUIRUPHG WKH ILUVW FRPSUHKHQVLYH H[SHUn LPHQWV RQ WKH VWUHQJWK RI PDWHULDOV +H REVHUYHG WKDW EHDPV DQG FROXPQV EHQG EHIRUH EUHDNLQJ DQG WKDW EUHDNLQJ VWUHQJWK YDULHV LQYHUVHO\ ZLWK WKH VTXDUH RI WKH OHQJWK DQG DOVR GHSHQGV XSRQ GHSWK DQG EUHDGWK (XOHU LQWURGXFHG WKH VWXG\ RI PHFKDQLFV VSHFLIn LFDOO\ WKH FRQFHSWV SRLQW PDVV DFFHOHUn DWLRQ DQG YHFWRUV ZKLFK KH FDOOHG JHRPHWULFDO TXDQWLW\f 'DQLHO %HUQRXOOL GHYHORSHG WKH SULQFLSOH RI PLQLPXP SRWHQn WLDO HQHUJ\ DOWKRXJK KH LV EHVW NQRZQ IRU KLV ZRUN LQ IOXLG PHFKDQLFV B (XOHU OHDUQHG KRZ WR VROYH DQDO\WLFDOO\ OLQHDU RUGLQDU\ GLIIHUHQWLDO HTXDWLRQV ZLWK FRQn VWDQW FRHIILFLHQWV 3ULRU WR WKDW WLPH RQO\ VHULHV VROXWLRQV ZHUH DYDLODEOH r (XOHU DIWHU PDNLQJ VRPH FDOFXODWLRQV RQ DQ DFWXDO EHDP GHIOHFWLQJ XQGHU LWV ORDG GLVFRYHUHG WKDW PRVW UHDO DSSOLFDWLRQV UHVXOW LQ VPDOO VWUDLQ $OO SUHYLRXV ZRUN KDG SUHVXPHG WKDW GHIRUPDWLRQV ZHUH ILQLWH (XOHU GHWHUPLQHG WKDW WKH HTXDWLRQ RI WKH HODVWLF FXUYH FRXOG EH REWDLQHG IURP WKH PLQLPXP SRWHQWLDO HQHUJ\ (XOHU GHULYHG WKH FROXPQ EXFNOLQJ IRUPXOD WKH UHVXOWV ZHUH FRPSDUHG WR 0XVVFKHQEURHNnV H[SHULPHQWV LQ -RKQ %HUQRXOOL REWDLQHG WKH ILUVW GLIIHUHQWLDO HTXDWLRQ IRU WKH PRWLRQ RI D V\VWHP E\ VWXG\LQJ WKH G\QDPLFV RI D ZHLJKWHG FRUG XVLQJ OXPSHG PDVVHV 1R VSHFLILF GDWHV DUH DYDLODEOH IURP RUGLQDU\ VRXUFHV LWHPV DUH ORFDWHG ZLWK UHVSHFW WR UHODWHG HQWULHV RI WKH WDEOH

PAGE 213

7$%/( $ &RQWLQXHGf 'n$OHPEHUW LQWURGXFHG WKH ILUVW SDUWLDO GLIIHUHQWLDO HTXDWLRQV +H DOVR ZDV WKH ILUVW WR VHSDUDWH FRQVWUDLQW IRUFHV IURP H[WHUQDO IRUFHV EXW QHYHU VWDWHG 'n$OHPEHUWnV 3ULQFLSOH ZKLFK ZDV JLYHQ ODWHU E\ (XOHU DQG /DJUDQJH (XOHU SXEOLVKHG WKH HDUOLHVW H[DPSOH RI ZKDW FDPH WR EH FDOOHG WKH 1HZWRQLDQ 0HWKRG $OVR GHILQHG ZHUH OLQHDU PRPHQWXP DQG NLQHWLF HQHUJ\ GHULYHG E\ LQWHJUDWLRQ (XOHU ZDV WKH ILUVW WR GHULYH 1HZWRQnV (TXDn WLRQV IRU D GLVFUHWH V\VWHP %\ DSSO\n LQJ WKHP WR HDFK SDUW RI WKH V\VWHP DOO WKH UHTXLUHG HTXDWLRQV ZHUH REWDLQHG 7KH H[DPSOH XVHG ZDV WKH WKUHHERG\ SUREOHP RI FHOHVWLDO PHFKDQLFV (XOHU H[WHQGHG WKH SUHFHGLQJ VWXG\ WR PHFKDQn LFDO V\VWHPV RI DOO NLQGV LH WR FRQn WLQXRXV V\VWHPV ([DPSOHV XVHG LQFOXGHG ULJLG ERG\ PRWLRQ DQG WUDQVYHUVH YLEUDn WLRQV RI D URG 5LFFDWL VXJJHVWHG WKDW WKH ZRUN GRQH WR GHIRUP DQ HODVWLF ERG\ ZDV UHFRYHUDEOH LQ SDUW (XOHU GHYHORSHG DQ HQWLUH WKHRU\ RI ULJLG ERG\ PRWLRQ +H GHILQHG FHQWHU RI PDVV DQG GLVWLQJXLVKHG LW IURP FHQWHU RI JUDYLW\ $GGLWLRQDOO\ KH REWDLQHG WKH LQHUWLD WHQVRU DQG GHVFULEHG LWV HOHPHQWV ZLWK UHVSHFW WR WKH PDVV r /DJUDQJH GHWHUPLQHG WKH LQILQLWH VHTXHQFH RI WKHRn UHWLFDO EXFNOLQJ ORDGV /DJUDQJH DWWHPSWHG WR ILQG WKH FROXPQ VKDSH ZKLFK ZRXOG KDYH WKH ODUJHVW EXFNOLQJ ORDG IRU JLYHQ KHLJKW DQG YROXPH +LV DQDO\VLV ZDV LQFRUUHFW WKLV UHVXOWV D FLUFXODU F\OLQGHU 1R VSHFLILF GDWHV DUH DYDLODEOH IURP RUGLQDU\ VRXUFHV LWHPV DUH ORFDWHG ZLWK UHVSHFW WR UHODWHG HQWULHV RI WKH WDEOH

PAGE 214

7$%/( $ &RQFOXGHGf (XOHU f f f GHWHUPLQHG WKH KHLJKW DW ZKLFK D XQLIRUPO\ KHDY\ FROXPQ EXFNOHV /DJUDQJH f§ SXEOLVKHG RQH RI WKH PRVW LPSRUWDQW ZRUNV LQ PHFKDQLFV 0FKDQLTXH $QDOLWLTXH ,Q WKLV ERRN KH V\VWHPDWL]HG PHFKDQLFV XVLQJ WKH FDOFXOXV RI YDULDWLRQV WR GHULYH JHQHUDO HTXDWLRQV ZKLFK DSSO\ WR DOO SUREOHPV RI PHFKDQLFV *HQHUDOO\ VSHDNn LQJ KH REYLDWHG WKH XVH RI FRPSOLFDWHG JHRPHWU\ UHSODFLQJ LW ZLWK SXUH DOJHEUD 0RUH VSHFLILFDOO\ KD LQWURGXFHG WKH PLVQRPHU 'n$OHPEHUWnV 3ULQFLSOH JHQn HUDOL]HG FRRUGLQDWHV DQG /DJUDQJHnV HTXDWLRQV *DXVV SURYHG WKH IXQGDPHQWDO WKHRUHP RI DOJHEUD LQ KLV GRFWRUDO GLVVHUWDWLRQ UHTXLULQJ WKH LQWURGXFWLRQ RI FRPSOH[ QXPEHUV +H ODWHU UHFRJQL]HG WKHLU YHFWRU FKDUDFWHU DQG DSSOLHG WKHP WR WKH VROXWLRQ RI FRPSOLFDWHG PHFKDQLFV SUREOHPV

PAGE 215

7$%/( $ 0$-25 &2175,%87256 ,1 7+( '(9(/230(17 2) 7+( &$/&8/86 2) 9$5,$7,216 $1' $1$/<7,&$/ 0(&+$1,&6 1(:721 6LU ,VDDF f (QJOLVK PDWKHPDWLFLDQ /(,%1(7= %DURQ YRQ *RWWIULHG :LOKHOP f *HUPDQ PDWKHPDWLFLDQ %(5128//, -RKQ -HDQ -RKDQf f 6ZLVV PDWKHPDWLFLDQ %(5128//, -DPHV -DFTXHV -DNREf f 6ZLVV PDWKHPDWLFLDQ 0866&+(1%52(. 3LHWHU YDQ f 'XWFK SK\VLFLVW DQG PDWKHPDWLFLDQ %(5128//, 'DQLHO f 6ZLVV PDWKHPDWLFLDQ (8/(5 /HRQKDUG f 6ZLVV PDWKHPDWLFLDQ $/(0%(57 -HDQ /H 5RQG 'n f )UHQFK DVWURQRPHU DQG PDWKHPDWLFLDQ /$*5$1*( &RPWH -RVHSK /RXLV f )UHQFK PDWKHPDWLFLDQ /(*(1'5( $GULHQ 0DULH f )UHQFK PDWKHPDWLFLDQ *$866 .DUO )ULHGULFK f *HUPDQ DVWURQRPHU DQG PDWKHPDWLFLDQ +$0,/721 6LU :LOOLDP 5RZDQ f ,ULVK PDWKHPDWLFLDQ -$&2%, .DUO *XVWDY -DFRE f *HUPDQ PDWKHPDWLFLDQ %5,26*+, )UDQFHVFR f ,WDOLDQ PDWKHPDWLFLDQ &/(%6&+ $OIUHG 5XGROI )ULHGULFK f *HUPDQ PDWKHPDWLFLDQ 72'+817(5 ,VDDF f (QJOLVK PDWKHPDWLFLDQ KLVWRULDQ :(,(5675$66 .DUO 7KHRGRU f *HUPDQ PDWKHPDWLFLDQ %/,66 *LOEHUW $PHV f $PHULFDQ PDWKHPDWLFLDQ 32175<$*,1 /HY 6 f 5XVVLDQ PDWKHPDWLFLDQ

PAGE 216

$33(1',; % $ 6,03/( 3522) 2) 7+( .8+178&.(5 7+(25(0 &RQVLGHU WKH PLQLPL]DWLRQ RI D JHQHUDO XQVSHFLILHG IXQFWLRQDO ZLWK UHVSHFW WR DQ QGLPHQVLRQDO YHFWRU [ VXEMHFW WR P Q LQHTXDOn LW\ FRQVWUDLQWV IXQFWLRQDO 0LQ >)[f@ [ LQHTXDOLW\ FRQVWUDLQWV JA[f M OP ZKHUH [ >[ [f [ @ f§ Q ,QHTXDOLW\ FRQVWUDLQWV DUH FRQYHUWHG WR HTXDOLW\ FRQVWUDLQWV E\ LQWURn GXFLQJ VODFN YDULDEOHV VA GHILQHG VXFK WKDW JM[f V[f M OP %Of RU A VM [f >f§JM [f @ R % f 6LQFH IRU DOO M WKH OHIWKDQG VLGH RI HTXDWLRQ %Of VXPV WR ]HUR IRU DUELWUDU\ YDOXHV RI $ P e $>J[f V"[f@ M O %f LW IROORZV WKDW P [e$Bf )[f e $ >J [f V [f @ ) [f

PAGE 217

1HFHVVDU\ FRQGLWLRQV IRU WKH DGMRLQHG FRVW IXQFWLRQ WR EH D PLQLPXP DUH GHULYHG IURP WKH UHTXLUHPHQW WKDW WKH ILUVW YDULDWLRQ YDQLVK V 7 : 7 7 ,! [ f§ V f§ $ f§ f§ [ f§ V f§ $ ,I DOO WKH YDULDWLRQV DUH WR EH LQGHSHQGHQW WKHQ 9 )[f $ 9 J[f M VXPPHG [ [ M $ V[f M P a JM[f 6M[f M OP ZKHUH UHSHDWHG LQGLFHV LQ D SURGXFW LPSO\ VXPPDWLRQ 7KH LQGHSHQGHQFH RI M DQG B$ FDQ EH DUJXHG KHXULVWLFDOO\ 6ODFN YDULDEOHV DUH DQ DUWLn ILFLDO FRQWULYDQFH XVHG WR IRUFH D GHVLUHG FRQGLWLRQ ZLWKRXW DOWHULQJ WKH RULJLQDO SUREOHP $Q\ FKDQJH LQ VA M[f ZKLFK VDWLVILHV WKH JLYHQ FRQVWUDLQW PXVW EH DFFRPSDQLHG E\ D FKDQJH LQ [f VXFK WKDW WKH QHW HIIHFW RQ LV ]HUR WKXV [BVB$f LV VWDWLRQDU\ ZLWK UHVSHFW WR M )XUWKHUPRUH VLQFH WKH VXP LQ %f LV HTXDO WR ]HUR IRU DQ\ DUELWUDU\ YDOXHV RI $A DQ\ DUELWUDU\ YDULDWLRQ RI ZLWK UHVSHFW WR $A PXVW DOVR EH ]HUR *LYHQ WKH GHILQLWLRQ RI V[f LQ %f WKH WKUHH QHFHVVDU\ f FRQGLWLRQV PD\ EH H[SUHVVHG DV 9 >)[f $J[f@ [ f§ $MJML[f 6M[! A % f % f

PAGE 218

7KH ILUVW FRQGLWLRQ ZLOO EH XVHG ZLWK WKH FRQFHSW RI D IHDVLEOH XVDEOH GLUHFWLRQ WR GHULYH WKH ILQDO FRQGLWLRQ UHPDLQLQJ EHIRUH WKH .XKQ7XFNHU FRQGLWLRQ PD\ EH VWDWHG $Q\ GLUHFWLRQ LQ [VSDFH LV IHDVLEOH LI DQ LQFUHPHQW $[ LQ WKDW GLUHFWLRQ UHGXFHV WKH FRVW IXQFWLRQ )B[f 'LUHFn WLRQ G LV VDLG WR DOVR EH XVDEOH LI DQ LQFUHPHQW $[ LQ WKDW GLUHFWLRQ FDXVHV QR FRQVWUDLQW YLRODWLRQ &RQVLGHU ILUVW WKH FRQFHSW RI D IHDVLEOH GLUHFWLRQ IRU DQ\ GLUHFWLRQ G WKH GLUHFWLRQDO GHULYDWLYH LV GHILQHG WR EH G) I G f 9 )[f GV [ f§ ZKHUH OLG __ ,Q DGGLWLRQ WR WKLV UHVWULFWLRQ RQ WKH QRUP RI GLUHFWLRQ £ OHW WKH LQFUHPHQW G[ EH SUHVFULEHG E\ G[ GVfG %f DQG WKH YHFWRU VFDODU SURGXFW EH ZULWWHQ DV WKH HTXLYDOHQW LQQHU SURGXFW VXFK WKDW G f 9 ) ( G7 9 ) + 9 ) Gf ; ; ; ,Q WHUPV RI DQ LQQHU SURGXFW QRWDWLRQ FO )7 D GO ‘ 9[) G! )RU DQ LQFUHPHQW LQ GLUHFWLRQ G RI OHQJWK GV WKH FKDQJH LQ WKH FRVW G) 9 ) Gf GV [ IXQFWLRQ LV

PAGE 219

%\ WKH OLQHDULW\ SURSHUW\ RI WKH LQQHU SURGXFW G) 9 ) Gf GV +HQFH WKH FRVW )[f LV GHFUHDVHG E\ DQ LQFUHPHQW LQ DQ\ GLUHFWLRQ G ZKLFK VDWLVILHV 9 ) Gf %f &RQYHUVHO\ D GLUHFWLRQ G LV IHDVLEOH LI 9A) Gf LH WKH DQJOH EHWZHHQ G DQG WKH GLUHFWLRQ RI VWHHSHVW GHVFHQW LV DFXWH )RU DQ\ GLUHFWLRQ WR EH XVDEOH DQ LQFUHPHQW LQ WKDW GLUHFn WLRQ PXVW FDXVH QR FRQVWUDLQW YLRODWLRQ 2Q WKH FRQVWUDLQW ERXQGDULHV J[f IRU M e DV GHILQHG LQ &KDSWHU ,,, :LWK DQ\ GLUHFWLRQ DJDLQ $ GHQRWHG E\ G WKH FKDQJH LQ JA [f ZLWK UHVSHFW WR DQ LQFUHPHQW G[ LV GHILQHG WKURXJK WKH GLUHFWLRQDO GHULYDWLYH ZKHUH VXFK WKDW 9Y J [f f L Gf GV 6LQFH GV GHQRWHV D YHFWRU QRUP E\ ZD\ RI WKH GHILQLWLRQ RI GMF JLYHQ LQ HTXDWLRQ %f LW LV D SRVLWLYH TXDQWLW\ +HQFH WKH VHQVH RI FKDQJH GJ LV SUHVFULEHG E\ WKH LQQHU SURGXFW 9 J Gf IRU ZKLFK WKHUH DUH e WKUHH SRVVLELOLWLHV

PAGE 220

Lf 9 J Gf GJ e ,W ZDV DVVXPHG WKDW DFWLYH FRQVWUDLQWV DUH EHLQJ GLVFXVVHG IRU ZKLFK JB [f 7KH FRQVWUDLQWV UHTXLUH DOO J !Ff WR EH QRQn SRVLWLYH $Q LQFUHPHQW G[ LQ WKH GLUHFWLRQ G XVHG LQ WKH LQQHU SURGXFW VDWLVILHV WKH FRQVWUDLQW ZLWK WKH LQHTXDOLW\ VLJQ %HFDXVH WKH GLUHFWLRQ GRHV QRW OHDG WR D FRQVWUDLQW YLRODWLRQ LW LV E\ GHILQLWLRQ XVDEOH $FWXDOO\ IRU VXFK D GLUHFWLRQ G HYHQ WKRXJK SRLQW [ OLHV RQ WKH ERXQGDU\ JA[f WKH SRLQW [ G[f GRHV QRW LLf 9 J Gf r GJ 8QGHU WKH DVVXPSWLRQ WKDW JB [f LV DFWLYH ZKHUH JA [f ZKHQ WKH LQFUHPHQW GLH LQ GLUHFWLRQ G FDXVHV QR FKDQJH LQ JA G LV WDQJHQW WR WKH FRQVWUDLQW VXUIDFH :LWK [ G[ XVHG LQ WKH LQQHU SURGXFW WKH SRLQW [ G[f PD\ VWLOO OLH RQ WKH FRQVWUDLQW ERXQGDU\ :LWK VXFK DQ LQFUHPHQW WKH FRQVWUDLQW PD\ SHUKDSV QRW EH YLRODWHG DQG LI QRW WKH GLUHFWLRQ LV WKHUHIRUH XVDEOH 1RWH WKDW IRU OLQHDU FRQVWUDLQWV WKH ZRUG PD\ LV GHOHWHG IURP WKH SUHYLRXV VHQWHQFH ,W LV LQFOXGHG EHFDXVH QRQOLQHDU FRQn VWUDLQWV UHTXLUH DQ DUJXPHQW FDOOLQJ IRU LQILQLWHVLPDO LQFUHn PHQWV D ILQLWH LQFUHPHQW OHDGV WR D SRLQW WKDW LV QRW RQ WKH FRQVWUDLQW ERXQGDU\ $ ILQLWH LQFUHPHQW IURP D SRLQW RQ D QRQn OLQHDU FRQVWUDLQW ERXQGDU\ UHVXOWV LQ WKDW FRQVWUDLQW HLWKHU EHFRPLQJ LQDFWLYH RQ EHLQJ YLRODWHG GHSHQGLQJ XSRQ ZKHWKHU WKH FRQVWUDLQW VXUIDFH LV FRQFDYH RU FRQYH[ UHVSHFWLYHO\

PAGE 221

LLLf 9 J Gf 2 GJ $ 7KDW WKH FRQVWUDLQW [f LV DFWLYH UHTXLUHV &6f ,I DQ LQFUHPHQW GB[ LQ WKH GLUHFWLRQ G XVHG LQ WKH LQQHU SURGXFW UHVXOWV LQ D SRVLWLYH GJA WKH FRQVWUDLQW KDV EHHQ YLRODWHG 7KHUHIRUH G LV QRW D XVDEOH GLUHFWLRQ 7KH UHVXOW RI WKHVH SRVVLELOLWLHV LV WKDW IRU DQ\ SRLQW [ RQ D FRQVWUDLQW ERXQGDU\ IRU D GLUHFWLRQ G WR EH XVDEOH LW PXVW VDWLVI\ 9 J Gf M H %f $ $ ,I WKH VHW RI DOO GLUHFWLRQV G ZKLFK DUH IHDVLEOH FRQWDLQV QR XVDEOH GLUHFWLRQV DW WKH SRLQW [ LQ TXHVWLRQ WKHQ [ [ASS DQG [ LV WKH FRQVWUDLQHG PLQLPXP SRLQW &RQGLWLRQV IRU D XVDEOH IHDVLEOH GLUHFWLRQ JLYHQ E\ %f DQG %f DUH XVHG ZLWK WKH JUDGLHQW FRQGLWLRQ %f WR SURYH WKDW WKH /DJUDQJLDQ PXOWLSOLHUV RI %f PXVW EH QRQQHJDWLYH DW WKH SRLQW ZKHUH [ LV D FRQVWUDLQHG RSWLPXP &RQGLWLRQ %f VWDWHV WKDW DW WKH FRQn VWUDLQHG PLQLPXP SRLQW WKH QHJDWLYH JUDGLHQW RI WKH FRVW IXQFWLRQ OLHV LQ D VXEVSDFH GHILQHG E\ WKH JUDGLHQWV RI WKH DFWLYH FRQVWUDLQWV LH \ Z E E E V! ,Q WKH IROORZLQJ LPSOLHG VXPPDWLRQ LV UHSODFHG E\ WKH V\PERO ( DQG WKH GHSHQGHQFH RI ) DQG J XSRQ LV LPSOLHG 5HVROYLQJ 9 ) LQWR FRP $ SRQHQWV DORQJ WKH 9 J GLUHFWLRQV YLD WKH LQQHU SURGXFW JLYHV ;

PAGE 222

7KXV P 9;) 9[J f 9 ) e M/ 9 J __9 M __ r $ \ \Mf ,, 9M M H 7 7KLV FRQGLWLRQ UHODWLQJ WKH /DJUDQJLDQ PXOWLSOLHUV WR DQ LQQHU SURGXFW LV XVHG ZLWK WKH FRQGLWLRQV QHFHVVDU\ IRU WKH H[LVWHQFH RI D FRQVWUDLQHG PLQLPXP WR VKRZ WKDW $ $W D FRQVWUDLQHG RSWLPDO SRLQW IRU HYHU\ IHDVLEOH GLUHFWLRQ G WKH FRVW GHFUHDVHV DQG 9 ) Gf G LV IHDVLEOH [ WKH FRQVWUDLQWV DUH YLRODWHG RWKHUZLVH QRW RSWLPDOf 9 J G! G LV QRW XVDEOH e $VVXPH WKDW VRPH M $A LV QHJDWLYH DQG D IHDVLEOH EXW QRW XVDEOH GLUHFn WLRQ £ H[LVWV ,I G LV WDNHQ WR EH FRLQFLGHQW ZLWK 9 ) WKLV GLUHF [ WLRQ LV IHDVLEOH DQG XVDEOH FRQWUDU\ WR WKH RULJLQDO DVVXPSWLRQ 7KDW $ LV SURYHG E\ FRQWUDGLFWLRQ 0DWKHPDWLFDOO\ VWDWHG *LYHQ 9 ) Gf IRU DOO G Y[JM Gf IRU DOO M H $VVXPH $ IRU VRPH M H $ Lf $ 9 ) 9 Jf [ [ M

PAGE 223

LLf &RQVLGHU G 9 ) ; WKHQ B9[) Gf 9[) 9[)f __9[) __ ! G LV IHDVLEOH LLLf %XW 9 J Gf 9 J 9 )f ; ; ; G LV XVDEOH E\ YLUWXH RI Lf 7KLV VD\V WKH GLUHFWLRQ RI VWHHSHVW GHVFHQW LV D IHDVLEOH GLUHFWLRQ ZKLFK LV DOVR XVDEOH FRQWUDU\ WR WKH JLYHQ FRQGLWLRQ LYf 7KHUHIRUH ; M H 7 4(' $ 7KLV FRPSOHWHV WKH SURRI RI WKH .XKQ7XFNHU FRQGLWLRQV IRU WKH H[LVWHQFH RI FRQVWUDLQHG PLQLPXP RI )[f VXEMHFW WR J [f M } f P 7KH\ DUH Lf J [ f 6M9s37n f LLf ; LLLf $MJMA37f r LQ ; [M 9M : ‘

PAGE 224

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n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

PAGE 225

/LVWLQJV RI QLQH VXEURXWLQHV DUH FRQWDLQHG LQ WKLV DSSHQGL[ 7KH ILUVW VL[ UHSUHVHQW D SURJUDP IRU VROYLQJ WKH PLQLPXP GHIOHFWLRQ EHDP SUREOHP VXEMHFW WR VWUHVV FRQVWUDLQWV 7KH RUGHU RI SUHVHQWDn WLRQ LV 0$,1 ',)(4 &255& 48$6, 5(&,3 287387 )ROORZLQJ WKHVH DUH WKH WKUHH XVHVXSSOLHG VXEURXWLQHV UHTXLUHG WR VROYH WKH PD[LPXP EXFNOLQJ ORDG SUREOHP VXEMHFW WR RQO\ JHRPHWULF DQG PDWHULDO ERXQGV 7KH\ DUH SUHVHQWHG LQ WKH IROORZLQJ RUGHU 0$,1 ',)(4 &237,&

PAGE 226

,2& ,03/,&,7 5($/r$+= f &200&148&200;(555))'*}+(0$;(0$;/ &*00&12$7$'}$&6,*7$8 &O 0(16 ,21 *fWRf7O&Of}3^Of}$If%IFf%/f LH5f&f&*Of5-f
PAGE 227

7LOf ,-0 = 7LOf < } f r= r r f
PAGE 228

f8 r }7}8 }}7} }8 r } 7}8 n7f8 Ff && = ,, r m, f§f ); O=fr< , f < ^ f &* } 8 m-f &2 $*2 8 r & f f 3+,,.f 8 r8 ) ; 8 r r fr 8f :5,7( f = } 8 r 3+,/f/ }f )250$7 n2n ;); )'f &217,18( *2 7& &217,18( :5,7(f )250$7n}&;n(5525 (;,7 & 675(66 &21675$,17 (;&/8'(6 $// *(20(7 5,&$//< $'0,66,%/( &21752/}f &217,18( 6723 (1&

PAGE 229

VXEURXWLQH GLIHJ 11&13$5,0,7(5*}:3'*rf ,03/,&,7 5($/06$+}4f§= f &2002148&200;(5525 2 ) )'* + ( 0$; (0,$; %/2 &20021'$7$ %' $&6 ,* 7$8 ',0(16,21 *1&f :1& f 313$5 f '*1&1&f ',0(16,21 6 f } 7$8% W f )81& W 8! 8) ;f 8r8 );^ 8r r fr8f ); O;fr: f:f )// )81&W%'$&); f 8 %' 8 $& 3+, 3// 03, 86 );%'mrf`rr ,)W86/7$&2586&7 ` *2 72 3+ 6 )81&W%&86);f )^3+,6*73+, f *2 72 & 8 86 3+, 3+,6 03, )/8 )81&W%&O});f ,)W3/8*7 3+, f *2 72 r 3+, 3/8 03, &217,18( 86 r);frr ,)W86O/7 %'&586*7 f &2 72 3+ ,6 )81&86}});f )3+,6*73+,f *2 72 8 86 8 3+, 3+,6 03,

PAGE 230

QR QR RRRRRR QRQ &217,18( 1 ,1 8%16/ ,6,*r:,f$&frr 8%66/ 7$8r:f$& 8168 6,*r:f f rr 8%668 7 $8 r: f ,)8%168/($1'8%668/(f *2 72 675(66 &21675$,176 $&0,7 0 *(20(75,&$//< $'0,66,%/( 1% :5,7(f ;81688%668 )250$7n; n; f)& n8%168 n} }n8%668 *2 7& &217,18( 675(66 &21675$,176 $&0,7 *(20(75,&$//< $'0,66,%/( ,)8,%16/*(%'*58,%66/*(%'f *2 72 675(66 &21675$,176 $&0,7 $// *(20(75,&$//< $'0,66,%/( 1% *2 72 &217,18( 8%168 6,*r: f 8668 7 $8r:f ,)8%668*7 f 8 1 6 8f *2 72 ,)8%168R/7$& f *2 72 &21752/ }' f f &21752/6 &21752/

PAGE 231

RQ RRRR RU! & /, 833(5 %281'$5< ,6 8/% /( 8 /( & %281'$5< ,6 '(),1(' %< 0$;,080 1250$/ 675(66 8/( 8%168 ,1% 1% *2 7& & F &217,18( )8%668/7 $& f *2 72 8 833(5 %281'$5< ,6 8/% /( 8 /( %281'$5< ,6 '(),1(' %< 0$;,080 6+($5 675(66 8/% 8%668 ,1% 1% ** 7& &217,18( ,)8%66/*7 R8O(16/ f *2 72 8 /2:(5 %281'$5< ,6 & 8O/% /( 8 /( %281'$5< ,6 '(),1(' %< 0$;,080 1250$/ 675(66 8O/% 8%16/ ,1% 1% *2 7& &217,18( &

PAGE 232

R R R / /&:(5 %281'$5< ,6 2 8O/% /( 8 /( %281'$5< ,6 '(),1(' %< 0$;,080 6+($5 675(66 2,/6 8%66/ ,1% 12 & F F F F ,1% 86,*$&f*78.7$8W$&f F ,1% 86,*}$&f/78O7$8I$&f F 1% O 1& $'0,66,%/( &21752/ F 1% 1& $&7,9( 675(66 &21675$,17 F 1% 8O/% /( 8 /( $1' 8 (4 $& F 1% 8/% /( 8 /( $1' 8 (4 F F F F 388 )81&,OWOM);f ,)W3/8*7 3+, f & 72 & 8 8 3+, 388 03, &217,18( /6 ) ;frm ,)1*(4f &2 72 ,)86/7 $&586*7 f *2 72 & 3+ 6 )81&O86});f ,)3+,6*73+, f *2 72 & 8 / 86

PAGE 233

* 72 )+, 3+,6 13, ,)W1%(4f *& 7& & L ,),8/%/786$1'86/7 } f *2 72 )8 / % )81& M 8 / %});f ,)38/%*73+, f *2 72 8 8 8/% )+, 38/% 33, *2 7& 38/ )81& } $ & r ) ; f )38/*73+, f *2 72 8 8 $& )+, 38/ +3 &217,18( 86 r);$&rrffrr ,)1((4Of *2 72 ,)W86O/7 %'&586*7 f *2 72 3+ ,6 )81&86$&);f ,)^3+,6*73+, f *2 72 86 $& 3+, 3+,6 13, ** 7& ,),8,/%/786$1'86/7 f *2 72 38/ )81&^8/$&});f ,)38/%*73+, f *2 72 8 8 / % $&

PAGE 234

3+, 38/2 On3, &217,18( *f :f *f :f8Orr f r8 f f : f f 8 r8 f O; f 8rrfr8f  f :cf &2 &2 O &* f /,8 &* f & f &* f & & & & 33, f m m D 8 $1& 8 $5( &2167$176 F 1 3 f m m 8 ,6 127 &2167$17 F 3 3 a f f f 8 ,6 127 &2167$17 F 33, mm f 8 ,6 21 $ 675(66 &21675$,17 %281'$5< F 33, 8 ,6 21 $ 675(66 &21675$,17 %281'$5< F F F 33,* 03, *2 7& }f 03,* &* f O 8rrfr8*2 7& &217,18( /,8 O 8Orr fr8f &8: r8:f

PAGE 235

&8: &*, f &* f &* V f &* f &* M f &*, f &2 7& &217,18( 8,8 &8: &8: &* r f & } f &* f &*f &* W f &* f *& 7& &217,18( ,),1%(& & f &* f &*  f && 7& &217,18( &* &*, f &* f &*, f *& 7& &217,18( ,) ,1(& &* f r8:f 8 r'8: 8 r / : r F; fr8,8r'8 ,: '*,fm&8:'8: }:O fr8,8r'8: OO8rrfr8f }r : fr8,8r'8: Y^W8OmrfrO8rmff r8: f }8 :, f 8 r&8: 8r'8: O;fr8,8r'8: '*,fr&8:'8: :, f}8,8r&8: 8,8r8 a : f m 2 ,8 r 8 1 f *& 72 6 ,*m&,f8 8 6 ,* 6 ,*r* `8 8 7 $ 8 r & f 8 7$8 7$8r*f8 f *& 72 r6,*r*f8r8r$&f ,, 8rrfr$&f

PAGE 236

&* 2 f r6 ,*8 &*} f r6,*r*f8r8r$&f *& 7* &&0,18( &*f f^ 8Orr fr$& &* f r7$8WW* f 8r$& f &*, f 7$8 &&f r7$8r*f8r$&f &217,18( ;&6 6,* r :f ;&7 7$8r:f 8% ;&6;&7 8% ; &7} r ; & 6 *& ; ;8 r,; 6,*(,;f &6457;&;8 f 7$8%,;f ;&7;8 &&0,18( = ; f r+ :5,7(}& f =}88); & )250$7 ;} n; nr&m; n8,n}'mW;r n8 n}'r;}f) ;rA f 9Y5, 7(} f 8%+8%, ,1%1%03, )250$7 ; '}; ;f :5,7(& f 6 ,*% ,; ,; f 85,7(f 7$8%,;f ,; f )250$7;& f )250$7;* f 5(7851 &&17,18( 5(7851 (1*

PAGE 237

68%5287,1( &&55& 1 0 1& 13 13$5 ,7(5 7 3 %/ %5
PAGE 238

68nG 5&87 1 ( 48$6 1 0 }1& 1,},03 1 3$5 1, 7 3517 *:73$%%/%5& & } 5 } < } r f ,03/,&,7 5($/r,$+=f &&00&148&200;(55252))'*+(,0$;(0$;%/2 &2.0&1'$7$%'$&6,*7$8 &O 0(16 ,21 *1&f:1&f}713f313$5f$1&1&f%1&}1&f %/1&}1&f %51&M1&f &1& 1& f'*1&1&f5131&f <1 3 }1&f'1&}1&1, f 0 0 &$// 29(5 )/^--f ,7e5 ,0 &$// &455& 101&1313$5,7(573%/%5
PAGE 239

*,f Kr* f & '$%6,:8 f f§ f f ,)(0$;4f (0$; ,0 -, &&0,18( ) I 3517f &$// &87 387 1 0 1& 13 ,0 3517 7(5 : 7} '*
PAGE 240

(0$; (0$; *& 7& ,) (0$; %/&(0$;f :5,7( f ,7(5 )*50$7+,167$%,/,7< ,1 ,7(5$7,21 f :5,7(f :5,7(f 7 f O < ,-f1f, }13f )250$7 ;+;;+<;+<;+<;+<;+<;+<;+< ;+<;+<;+
PAGE 241

&* 1 %W,-f % W, r -f $ ,.fr&., 0 f && 1 && 1 $W L-f % -f ,) ^I2'.ff ; *2 72 ; && 1 5WInO f &2 1 50,f 50,f%/8-fr*-f %,}-f ;}$,-f &2 1 &2 1 $ -f f &2 1 $ -f $ ,f%/ ,. fr%. $W ,-f $ ,--%5, ,-f ... &$// &9)5)/---f )-f :5,7( f ...,7(5}0 &$// 5(&,3 100&0,3517, 7(503$6:$%&'*f . &$// 29(5 )/--f ,)---f§f :5,7( f ..., 7(5 &2 1 f f &2 1 f f$ fr5W 0f &2 1

PAGE 242

<0O ,f <0O ,f *, f 50, f f &2 0 0 ,' 0a ,0 ', ,' &2 1 *,f &2 1 *,f *,f ,,'f r5^ ,&f &2 1 5,',! 5,', f* f
PAGE 243

68%5287,1( 5(&,) ^ 1 0 1& } ,0 3517 7(5 03$6 :$ %& '* f ,03/,&,7 5($/r^$+=f &201&148&200;(55252))2*+(0$; } (0$; / &200&1'$7$%'$&6,*7$8 &,.(16,21 :1&f }$1&1&f%1&W1&f}&O1&}1&f'*1&1&f &* 1 &2 1 & W fn $ f && 1 )$&7 $ } f $,},f ,))$&7f :5,7( f ,0 )250$7+2=(52 21 &, $*21$/ 52: +0$75,; f 6723 &2 }1 $^, -f $ ,--)$&7 &2 1 ,).,f ,)$.,ff )$&7 $.,f $ .,f &2 1 $^.-f $.-f)$&7r$ -f &217,18( 1 3 $ 6 680 &2 1 f &2 1 &* f m &2 1 '*, -f &*,-f$,.fr&.0 1 && 1

PAGE 244

&2 }1 680 680 '$%6&* ,} f f '$%6'*-,ff ,)^ ,3517f :5,7( f 680 )250$7+ 680 2) $%6 2) 2))',$* 7(506 (f ,) 680 2) )2* f ,) 13$6 f 35(55 r35(55 ,) 680 35(55f &2 1 &2 } 1 $ f % ,f ,) ,3517 f :5,7( f )250 $7 ), 86(' 1(;7 72 /$67 5(&,352&$/f *2 72 ,) 13$603$6 f 35(55 680 13$6 13$6 &2 1 &2 & 1 %, -f $ f &2 1 &* , f &* , f &2 & 1 && 1 :-f & &2 1 :-f :-f$., fm '*}. f &2 }1 $-,f $, f:f *2 7& 5(7851

PAGE 245

68%5287,1( 287387 1 0 1& 13 ,0 ,3517, 7(5 } : 7 } '* 9f ,03/,&,7 5e$Or$+ 2= f &2In0&148&200; (5525 2) )2* +(0$;(0$; %/2 &200&1'$7$%&$&6,*7$8 ',0(16,21 *1&f:1&f}713f}'*1&1&f}<131&f ,) ,0f :5,7( f ,7(5 :5,7( f 0O 0O &* 0O :5,7( f 7 ,f<,-f1 f ,) ,3517f :5,7( f ,7(5 :5,7( f :5,7( f *2 7& :5,7( f ;& f 1 f :5,7( f ; :f 1 f ,) ,3517f &* 1 :5,7( f &*,f 1 f 5(7851 )250$7 +62/87,21 $)7(5 ,7(5$7,21 f )250$7 ;+;;+<;.<;+<;+<;+<;+<;+< ;+<;+<;+
PAGE 246

,03/,&,7 5($/r$+= f 5 ( $ / } 08 W.0 W0/ / $0 &200&148&200;(5545}2))2*}+W(,0$;(0$;}%/2 &20021'$7$ $8$/(8}(/585/08r.0'868W/$0 &20 0&17&& 886 $6) ;+7 ',0(16,21 *f}:f7f3f$f 'f } %/ f %5 f & f '* f 5f
PAGE 247

&2 ,&2 } (/,r-f 25 , I -f &217,18( % / f %/f +5f (5f % 5 } f (/f %/^ M f &217,18( ), ), 3 3 } 3, /$1 &2 & 13 7,f ,O-0 = 7, f &7 '&&63*=f 67 '6,13r= f 9 f &7 < f 67 <, f &7 < f /$1 = fr.008 <,f &}&7
PAGE 248

)250$7 } rn2n &; n ',& 127 &219(5*( (0$; n'f*7 (5525 r&f *2 7& :5,7( f (0$;(5525 )250$7 n }n2n ; n &219(5*(' (0$; f f /( (5525 r f &217,18( &,) /$0
PAGE 250

68%5287,1( ',)(* 1}1&13$5,0, 7(5*:3'*f ,03/,&,7 5($/r$+=! 5($/r 08.00//$0 &201&148&200;(55252))'*+(0$;(0$;%/2 &21 0 &18$7 $ $8$/(8(/}585/08.0'868/$0 &21 1&17 &2 8OI6OW $6)};,7 ',0(16,21 *1&f:1&f313$5f'*1&}1&f & 0867 63(&,)< 0/ )&7;f 0/ (;8 86 86 ;1 r:frIFf ;' .0r:f 08f}86r86 $5* ;1;' = '6,*186 $5* f 86 &$%6$5*f frr(;8 86 =r86 ,)W86O*7$/ f *& 72 8 $/ &8: &8: &8: &2 *$0 *& 7& ,)86*($8 f *& 72 & 8 86 &8: 86 r:> f f &8: 86Or: W f f &8: (;8r86m.0.0r: f 08f *$0 &2 *& 7& &217,18( /, $8

PAGE 251

&8: &8: &8: *$0 &217,18( 8 86 8 86 *$0 *$0 f : ^ f f :f8r8m8f f :fr: f f 0 0 0/ 8 r8 f ^ f f§ : f r : ^ f f : f ^8m8m8f & f f§ WR ^ f m: ^ f &217,18( &2 && O &* f &*f &*8 r*f8 & } f &*8} &8: &* f '*8r&8: &* f & 8 r & 8 : &* } f : f &*^ f : f &*8 0 r 8 &* f '*8r &8: &* f *8A &8: & f '*8r&8: &*^ f : f &*^ f : f &*8 Rr*f8 8r8r8f

PAGE 252

&* W f &* W f &* f &* f &* Uf &217,18( 5(7851 (1& 2 *8r &8: '*8}&8: '*8r&8: : f :  f 8r8r8f

PAGE 253

68%5287,1( &&55& 101&1313$5,7(573%/%5
PAGE 254

%,%/,2*5$3+< $VKOH\ + DQG 0F,QWRVK 6 & -U $SSOLFDWLRQ RI $HURHODVWLF &RQVWUDLQWV LQ 6WUXFWXUDO 2SWLPL]DWLRQ 3URF WK ,QWHU &RQJUHVV $SSO 0HFK $VKOH\ + 0F,QWRVK 6 & -U DQG :HDWKHULOO : + 2SWLPLn ]DWLRQ XQGHU $HURHODVWLF &RQVWUDLQWV >LQ 3URF $*$5-' 6\PSRn VLXP RQ 6WUXFWXUDO 2SWLPL]DWLRQ $' f@ $VLPRY $VLPRYnV %LRJUDSKLFDO (QF\FORSHGLD RI 6FLHQFH DQG 7HFKQRORJ\ 'RXEOHGD\ %DUQHWW 5 / 0LQLPXP :HLJKW 'HVLJQ RI %HDPV IRU 'HIOHFWLRQ 3URF $6&( (QJ 0HFK %DUQHWW 5 / D 0LQLPXP :HLJKW 'HVLJQ RI %HDPV IRU 'HIOHFWLRQ 3URF $6&( (QJ 0HFK %DUQHWW 5 / E 0LQLPXP 'HIOHFWLRQ 'HVLJQ RI D 8QLIRUPO\ $FFHOn HUDWLQJ &DQWLOHYHU %HDP 7UDQV $60( $SSO 0HFK ( %DUQHWW 5 / 6XUYH\ RI 2SWLPDO 6WUXFWXUDO 'HVLJQ ([SHU 0HFK A $$ %DUQHV ( 5 1HFHVVDU\ DQG 6XIILFLHQW &RQGLWLRQV IRU D &ODVV RI 'LVWULEXWHG 3DUDPHWHU &RQWURO 6\VWHPV 6,$0 &RQWURO %DUVWRQ ( 0 $ 0LQLPD[ 3ULQFLSOH IRU 1RQRYHUGDPSHG 6\VWHPV ,QWHU (QJ 6FL %HOODP\ 1 : DQG :HVW 0 0HWKRGV RI 3URILOH 2SWLPL]DWLRQ E\ ,WHUDWLYH $QDORJ &RPSXWDWLRQ 7KH &RPSXWHU %HUJ 3 : &DOFXOXV RI 9DULDWLRQV >LQ +DQGERRN RI (QJLQHHUn LQJ 0HFKDQLFV HG : )OXJJH 0F*UDZ+LOO@ %HUJDPLQL 0DWKHPDWLFV /LIH 6FLHQFH /LEUDU\ 7LPH/LIH %RRNV %KDUJDYD 6 DQG 'XIILQ 5 'XDO ([WUHPXP 3ULQFLSOHV 5HODWLQJ WR 2SWLPXP %HDP 'HVLJQ $UFK 5DWL 0HFK $QDO %ODVLXV + 7UDJHU NOHLQVWHU 'XUFKEHLJXQJ XQG 6WDEH JURVVWHU .QLFNIHVWLNHLW EHL JHJHEHQHP 0DWHULDOYHUEUDXFK =HLW IXU 0DWK XQG 3K\V

PAGE 255

%OLVV $ &DOFXOXV RI 9DULDWLRQV 2SHQ &RXUW 3XEOLVKLQJ &RPSDQ\ %OLVV $ /HFWXUHV RQ WKH &DOFXOXV RI 9DULDWLRQV 8QLYHUVLW\ RI &KLFDJR 3UHVV %RORWLQ 9 9 1RQFRQVHUYDWLYH 3UREOHPV RI WKH 7KHRU\ RI (ODVWLF 6WDELOLW\ 3HUJDPRQ 3UHVV %RO]D /HFWXUHV RQ WKH &DOFXOXV RI 9DULDWLRQV &KHOVHD 3XEn OLVKLQJ &RPSDQ\ %R\NLQ : + -U DQG 6LHUDNRZVNL 5 / 5HPDUNV RQ 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH $SSOLHG WR D 6WUXFWXUDO 2SWLPL]DWLRQ 3UREOHP 5R\DO $HUR 6RF %UDFK 5 0 2Q WKH ([WUHPDO )XQGDPHQWDO )UHTXHQFLHV RI 9LEUDWn LQJ %HDPV ,QWHU 6ROLGV 6WUXFW %UDFK 5 0 DQG :DOWHUV 0D[LPXP )UHTXHQF\ RI %HDPV ,QFOXGn LQJ 6KHDU (IIHFWV 1RWUH 'DPH 5HSRUW $' f %UHDNZHOO 9 7KH 2SWLPL]DWLRQ RI 7UDMHFWRULHV 6,$0 %U\VRQ $ ( -U DQG 'HQKDP : ) $ 6WHHSHVW $VFHQW 0HWKRG IRU 6ROYLQJ 2SWLPXP 3URJUDPPLQJ 3UREOHPV 7UDQV $60( $SSO 0HFK ( %U\VRQ $ ( -U DQG 'HQKDP : ) 2SWLPDO 3URJUDPPLQJ 3UREOHPV ZLWK ,QHTXDOLW\ &RQVWUDLQWV ,, 6ROXWLRQ E\ 6WHHSHVW$VFHQW $,$$ %U\VRQ $ ( -U 'HQKDP : ) DQG 'UH\IXV 6 ( 2SWLPDO 3URJUDPPLQJ 3UREOHPV ZLWK ,QHTXDOLW\ &RQVWUDLQWV 1HFHVVDU\ &RQGLWLRQV IRU ([WUHPDO 6ROXWLRQV $,$$ %U\VRQ $ ( -U DQG +R
PAGE 256

&KHUQ 0 DQG 3UDJHU : 2SWLPDO 'HVLJQ RI %HDPV IRU 3UHn VFULEHG &RPSOLDQFH XQGHU $OWHUQDWLYH /RDGV 2SW 7KHRU\ $SSOQV &LWURQ 6 (OHPHQWV RI 2SWLPDO &RQWURO +ROW 5LQHKDUW DQG :LQVWRQ &ODXVHQ 7 OEHU GLH )RUP DUFKLWHNWRQLVKHU 60XOHQ %XOO SK\VPDWK GH On$FDGPLH B &XQQLQJKDP : ,QWURGXFWLRQ WR 1RQOLQHDU $QDO\VLV 0F*UDZ +LOO 'HQQ 0 0 2SWLPL]DWLRQ E\ 9DULDWLRQDO 0HWKRGV 0F*UDZ+LOO 'H 6LOYD % 0 ( 0LQLPXP :HLJKW 'HVLJQ RI 'LVNV 8VLQJ D )UHn TXHQF\ &RQVWUDLQW 7UDQV $60( (QJ ,QGXVWU\ % 'H 6LOYD % 0 ( 2SWLPDO 9LEUDWLRQDO 0RGHV RI D 'LVN 6RXQG 9LE 'L[RQ / 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH $SSOLHG WR WKH 3URILOH RI D %HDP 5R\DO $HUR 6RF 'L[RQ / )XUWKHU &RPPHQWV RQ 3RQWU\DJLQn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n WLRQ >LQ 2SWLPDOLW\ &ULWHULD LQ 6WUXFWXUDO 'HVLJQ HG : 3UDJHU DQG 3 9 0DUFDO $))'/75@ (OVJROF / ( &DOFXOXV RI 9DULDWLRQV $GGLVRQ:HVOH\

PAGE 257

)HLJHQ 0 0LQLPXP :HLJKW RI 7DSHUHG 5RXQG 7KLQZDOOHG &ROXPQV 7UDQV $60( $SSO 0HFK ( )R[ 5 / 2SWLPL]DWLRQ 0HWKRGV IRU (QJLQHHULQJ 'HVLJQ $GGLVRQ :HVOH\ )R[ 5 / DQG .DSRRU 0 3 6WUXFWXUDO 2SWLPL]DWLRQ LQ WKH '\QDPLF 5HVSRQVH 5HJLPH $ &RPSXWDWLRQDO $SSURDFK $,$$ )XQJ < & )RXQGDWLRQV RI 6ROLG 0HFKDQLFV 3UHQWLFH+DOO *DMHZVNL $ DQG =\F]NRZVNL 0 2SWLPDO 'HVLJQ RI (ODVWLF &ROn XPQV 6XEMHFW WR WKH *HQHUDO &RQVHUYDWLYH %HKDYLRU RI /RDGLQJ =$03 *DOODJKHU 5 + DQG =LHQNLHZLF] & HGV 2SWLPDO 6WUXFWXUDO 'HVLJQ -RKQ :LOH\ DQG 6RQV *HOIDQG 0 DQG )RPLQ 6 9 &DOFXOXV RI 9DULDWLRQV 3UHQWLFH +DOO *HOODWO\ 5 $ (G 3URF $*$5' 6\PSRVLXP RQ 6WUXFWXUDO 2SWLPLn ]DWLRQ $' f *HOODWO\ 5 $ DQG %HUNH / 2SWLPDO 6WUXFWXUDO 'HVLJQ $))'/75f *HUDUG 2SWLPXP 6WUXFWXUDO 'HVLJQ &RQFHSWV IRU $HURVSDFH 9HKLFOHV $,$$ 6SDFHFUDIW *KRVK 7 '\QDPLFV DQG 2SWLPL]DWLRQ RI D +XPDQ 0RWLRQ 3UREOHP 3K' 'LVVHUWDWLRQ 8QLYHUVLW\ RI )ORULGD *MHOVYLN $ 0LQLPXP :HLJKW 'HVLJQ RI &RQWLQXRXV %HDPV ,QWHU 6ROLGV 6WUXFW B *RXOG 6 + 9DULDWLRQDO 0HWKRGV IRU (LJHQYDOXH 3UREOHPV 8QLYHUn VLW\ RI 7RURQWR 3UHVV *UHHQKLOO $ 7LWOH 1RW .QRZQ 3URF &DPEULGJH 3KLO 6RF B +DONLQ + 7KH 3ULQFLSOH RI 2SWLPDO (YROXWLRQf >LQ 1RQOLQHDU 'LIIHUHQWLDO (TXDWLRQV DQG 1RQOLQHDU 0HFKDQLFV HGV 3 /DVDOOH DQG 6 /HIVFKHW] $FDGHPLF 3UHVV@ +DQVRQ 1 &RPSDULVRQ RI 7ZR 2SWLPL]DWLRQV RI D 1RQOLQHDU %RXQGDU\YDOXH 3UREOHP $,$$ \

PAGE 258

+DXJ ( DQG .LUPVHU 3 0LQLPXP :HLJKW 'HVLJQ RI %HDPV ZLWK ,QHTXDOLW\ &RQVWUDLQWV RQ 6WUHVV DQG 'HIOHFWLRQ 7UDQV $60($SSO 0HFK ( +DXJ ( 6WUHHWHU 7 DQG 1HZHOO 5 6 2SWLPDO 'HVLJQ RI (ODVWLF 7UXFWXUDO (OHPHQWV $' f +HQQLJ 5 DQG 0LHOH $ 6HTXHQWLDO *UDGLHQW5HVWRUDWLRQ $OJRn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n QHHULQJ $FDGHPLF 3UHVV .HOOHU % 7KH 6ORSH RI WKH 6WURQJHVW &ROXPQ $UFK 5DWL 0HFK $QDO B .HOOHU % DQG 1LRUGVRQ ) 7KH 7DOOHVW &ROXPQ 0DWK 0HFK .RPNRY 9 2SWLPDO &RQWURO 7KHRU\ IRU WKH 'DPSLQJ RI 9DULDWLRQV RI 6LPSOH (ODVWLF 6\VWHPV /HFWXUH 1RWHV LQ 0DWKHPDWLFV 6SULQJHU9HUODJ

PAGE 259

.RSS 5 ( 3RQWU\DJLQ 0D[LPXP 3ULQFLSOH >LQ 2SWLPL]DWLRQ 7HFKQLTXHV HG /HLWPDQQ $FDGHPLF 3UHVV@ .RSS 5 ( 2Q WKH 3RQWU\DJLQ 0D[LPXP 3ULQFLSOH >LQ 1RQ OLQHDU 'LIIHUHQWLDO (TXDWLRQV DQG 1RQOLQHDU 0HFKDQLFV HGV 3 /DVDOOH DQG 6 /HIVFKHW] $FDGHPLF 3UHVV@ .RZDOLN 6 )HDVLEOH 'LUHFWLRQ 0HWKRGV >LQ 3URF $*$5' 6\PSRVLXP RQ 6WUXFWXUDO 2SWLPL]DWLRQ $' f@ /DQF]RV & /LQHDU 'LIIHUHQWLDO 2SHUDWRUV 9DQ 1RVWUDQG &RPSDQ\ /HLSKRO] + + ( 2Q WKH 6XIILFLHQF\ RI WKH (QHUJ\ &ULWHULRQ IRU WKH 6WDELOLW\ RI &HUWDLQ 1RQFRQVHUYDWLYH 6\VWHPV RI WKH )ROORZHUORDG 7\SH $60( 3DSHU $30( /RYLWW : 9 /LQHDU ,QWHJUDO (TXDWLRQV 'RYHU 3XEOLFDWLRQV /XHQEHUJHU 2SWLPL]DWLRQ E\ 9HFWRU 6SDFH 0HWKRGV -RKQ :LOH\ DQG 6RQV 0DGD\ & $ &ODVV RI 0LQLPXP :HLJKW 6KDIWV $60( 3DSHU '(7 0DUWLQ % 2SWLPDO 'HVLJQ RI (ODVWLF 6WUXFWXUHV IRU 0XOWLn SXUSRVH /RDGLQJ 2SW 7KHRU\ $SSOQV A 0DUWLQ % 7KH 2SWLPDO 'HVLJQ RI %HDPV DQG )UDPHV ZLWK &RPSOLDQFH &RQVWUDLQWV ,QWHU 6ROLGV 6WUXFW B 0DVXU ( ) 2SWLPXP 6WLIIQHVV DQG 6WUHQJWK RI (ODVWLF 6WUXFn WXUHV 3URF $6&( (QJ 0HFK 0D\HGD 5 DQG 3UDJHU : 0LQLPXP :HLJKW 'HVLJQ RI %HDPV IRU 0XOWLSOH /RDGLQJ ,QWHU 6ROLGV 6WUXFW B 0F&DUW % 5 +DXJ ( DQG 6WUHHWHU 7 2SWLPDO 'HVLJQ RI 6WUXFWXUHV ZLWK &RQVWUDLQWV RQ 1DWXUDO )UHTXHQF\ $,$$ 0F,QWRVK 6 & DQG (DVWHS ) ( 'HVLJQ RI 0LQLPXP0DVV 6WUXFn WXUHV ZLWK 6SHFLILHG 6WLIIQHVV $,$$ A 0F,QWRVK 6 & :HLVKDDU 7 $ DQG $VKOH\ + 3URJUHVV LQ $HURHODVWLFLW\ 2SWLPL]DWLRQf§$QDO\WLFDO YHUVXV 1XPHULFDO $SSURDFKHV 6WDQIRUG 8QLYHUVLW\ 5HSRUW 68'$$5 1R $' f

PAGE 260

0F1HLOO : $ 6WUXFWXUDO :HLJKW 0LQLPL]DWLRQ 8VLQJ 1HFHVVDU\ DQG 6XIILFLHQW &RQGLWLRQV 2SW 7KHRU\ $SSOQV 0HORVK 5 DQG /XLN 5 $SSUR[LPDWH $QDO\VLV DQG $OORFDWLRQ IRU /HDVW :HLJKW 6WUXFWXUDO 'HVLJQ $))'/75f 0LHOH $ :HOO + DQG 7LHW]H / 0RGLILHG 4XDVLOLQHDUn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n OHP $,$$ Mf 3ODXW 5 + E 2Q WKH 2SWLPDO 6WUXFWXUDO 'HVLJQ IRU D 1RQFRQn VHUYDWLYH (ODVWLF 6WDELOLW\ 3UREOHP 2SW 7KHRU\ $SSOQV @B 3ODXW 5 + F 2SWLPDO 6WUXFWXUDO 'HVLJQ IRU *LYHQ 'HIOHFn WLRQ 8QGHU 3HULRGLF /RDGLQJ 4XDUW $SSO 0DWK 3RQWU\DJLQ / 6 %ROW\DQVNLL 9 *DPNUHOLG]H 5 9 DQG 0LVKFKHQNR ( ) 7KH 0DWKHPDWLFDO 7KHRU\ RI 2SWLPDO 3URFHVVHV -RKQ :LOH\ DQG 6RQV 3RSH & DQG 6FKPLGW / $ HGV 6WUXFWXUDO 'HVLJQ $SSOLFDn WLRQV RI 0DWKHPDWLFDO 3URJUDPPLQJ 7HFKQLTXHV $*$5'RJUDSK $' f 3UDJHU : 2SWLPDOLW\ &ULWHULD 'HULYHG IURP &ODVVLFDO ([WUHPXP 3ULQFLSOHV >LQ $Q ,QWURGXFWLRQ WR 6WUXFWXUDO 2SWLPL]DWLRQ 8QLYHUVLW\ RI :DWHUORR 3UHVV@

PAGE 261

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n JUDPPLQJ 3DUW 6,$0 $SSO 0DWK 4 3DUW ,, 5R[LQ ( $ *HRPHWULF ,QWHUSUHWDWLRQ RI 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH >LQ 1RQOLQHDU 'LIIHUHQWLDO (TXDWLRQV DQG 1RQn OLQHDU 0HFKDQLFV HGV 3 /DVDOOH DQG 6 /HIVFKHW] $FDGHPLF 3UHVV@ 5R]RQRU / 3RQWU\DJLQnV 0D[LPXP 3ULQFLSOH LQ WKH 7KHRU\ RI 2SWLPXP 6\VWHPV $YWRPDW L 7HOHPHNK >(QJOLVK WUDQVODWLRQ $XWRPDWLRQ DQG 5HPRWH &RQWURO @ 5R]YDQ\ 1 $QDO\VLV YHUVXV 6\QWKHVLV LQ 6WUXFWXUDO (QJLQHHULQJ &LYLO (QJ 7UDQV ,QVWLWXWLRQ RI (QJLQHHULQJ $XVWUDOLD &( 5XELQ & 3 0LQLPXP :HLJKW 'HVLJQ RI &RPSOH[ 6WUXFWXUHV 6XEMHFW WR D )UHTXHQF\ &RQVWUDLQW $,$$ A 5XGLVLOO & 6 DQG %KDWLD 2SWLPL]DWLRQ RI &RPSOH[ 6WUXFWXUHV WR 6DWLVI\ )OXWWHU 5HTXLUHPHQWV $,$$ B (UUDWD Mf 6DOLQDV 2Q 9DULDWLRQDO )RUPXODWLRQV IRU 2SWLPDO 6WUXFWXUDO 'HVLJQ 3K' 'LVVHUWDWLRQ 8QLYHUVLW\ RI &DOLIRUQLD /RV $QJHOHV 6DOXNYDG]H 0 ( 2SWLPL]DWLRQ RI 9HFWRU )XQFWLRQDOV 3URJUDPn PLQJ RI 2SWLPDO 7UDMHFWRULHV (QJOLVK WUDQVODWLRQ $XWRPDn WLRQ DQG 5HPRWH &RQWURO

PAGE 262

6FKPLGW / $ -U $XWRPDWHG 'HVLJQ ,QWHU 6FL 7HFK DQG 6FKPLGW / $ -U 6WUXFWXUDO 6\QWKHVLV >LQ &RPSRVLWH 0DWHn ULDOV :RUNVKRS HGV 6 : 7VDL & +DOSLQ DQG 1 3DJDQR 7HFKQRPLF 3XEOLVKLQJ &RPSDQ\@ 6KHX & < (ODVWLF 0LQLPXP:HLJKW 'HVLJQ IRU 6SHFLILHG )XQGDn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n OHPV ZLWK D %RXQGHG 6WDWH 6SDFH $,$$ B 6WURXG : 'H[WHU & % DQG 6WHLQ 0 $XWRPDWHG 3UHOLPn LQDU\ 'HVLJQ RI 6LPSOLILHG :LQJ 6WUXFWXUHV WR 6DWLVI\ 6WUHQJWK DQG )OXWWHU 5HTXLUHPHQWV 1$6$ 71' 7DGMEDNKVK DQG .HOOHU % 7KH 6WURQJHVW &ROXPQ DQG ,VR SHULPHWULF ,QHTXDOLWLHV IRU (LJHQYDOXHV 7UDQV $60( $SSO 0HFK ( 7D\ORU ( 7KH 6WURQJHVW &ROXPQ $Q (QHUJ\ $SSURDFK 7UDQV $60( $SSO 0HFK ( 7D\ORU ( 0D[LPXP 6WUHQJWK RI (ODVWLF 6WUXFWXUDO 'HVLJQ 3URF $6&( (QJ 0HFK 7D\ORU ( DQG /LX & < 2SWLPDO 'HVLJQ RI &ROXPQV $,$$

PAGE 263

7LPRVKHQNR 6 3 DQG *HUH 0 7KHRU\ RI (ODVWLF 6WDELOLW\ 0F*UDZ+LOO 7RGKXQWHU DQG 3HDUVRQ $ +LVWRU\ RI WKH 7KHRU\ RI (ODVWLFn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f ,QWHU &RQWURO :DVLXW\QVNL = DQG %UDQGW $ 7KH 3UHVHQW 6WDWH RI .QRZOHGJH LQ WKH )LHOG RI 2SWLPXP 'HVLJQ RI 6WUXFWXUHV $05 :HLVVKDU 7 $ $Q $SSOLFDWLRQ RI &RQWURO 7KHRU\ 0HWKRGV WR WKH 2SWLPL]DWLRQ RI 6WUXFWXUHV +DYLQJ '\QDPLF RU $HURHODVWLF &RQn VWUDLQWV 6WDQIRUG 8QLYHUVLW\ 5HSRUW 68'$$5 1R $' f :X &ROXPQ ,QVWDELOLW\ DQG 1RQFRQVHUYDWLYH )RUFHV ZLWK ,QWHUQDO DQG ([WHUQDO 'DPSLQJf§)LQLWH (OHPHQW 8VLQJ $GMRLQW 9DULDWLRQDO 3ULQFLSOHV 'HYHORSPHQWV LQ 0HFKDQLFV 9RO 3URFHHGLQJV RI WKH WK 0LGZHVWHUQ 0HFKDQLFV &RQIHUHQFH 8QLn YHUVLW\ RI 3LWWVEXUJK =DUJKDPHH 0 6 2SWLPXP )UHTXHQF\ RI 6WUXFWXUHV $,$$ -M =LHQNLHZLF] & 7KH )LQLWH (OHPHQW 0HWKRG LQ (QJLQHHULQJ 6FLHQFH 0F*UDZ+LOO

PAGE 264

%,2*5$3+,&$/ 6.(7&+ -HUU\ & +RUQEXFNOH ZDV ERUQ LQ &OHDUZDWHU )ORULGD LQ 2Q JUDGXDWLQJ IURP 6WUDQDKDQ +LJK 6FKRRO LQ )RUW /DXGHUGDOH KH DWWHQGHG WKH $HURVSDFH (QJLQHHULQJ SURJUDP RI WKH 8QLYHUVLW\ RI )ORULGD 6WXGLHV WKHUH ZHUH LQWHUUXSWHG E\ D \HDUnV HPSOR\PHQW DW +RQH\ZHOOnV LQHUWLDO JXLGDQFH GLYLVLRQ LQ 6W 3HWHUVEXUJ )ORULGD +H UHFHLYHG KLV %6$( IURP WKH 8QLYHUVLW\ RI )ORULGD LQ )ROORZLQJ JUDGXDWLRQ KH ZDV HPSOR\HG E\ WKH %RHLQJ &RPSDQ\ LQ +XQWVYLOOH $ODEDPD :KLOH WKHUH KH ZRUNHG RQ HUURU DQDO\VHV WKH $SROOR6DWXUQ 9 RQERDUG FRPSXWHU IOLJKW SURJUDP DQG WUDMHFWRU\ VLPXn ODWLRQ 'XULQJ WKH HYHQLQJV KH ZDV D SDUWWLPH VWXGHQW DW WKH 8QLYHUn VLW\ RI $ODEDPD LQ +XQWVYLOOH UHFHLYLQJ DQ 06( LQ (QJLQHHULQJ 0HFKDQn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

PAGE 265

, FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 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c *HQH : +HPS $VVRFLDWH 3URIHVVRU RI (QJLQHHULQJ 6FLHQFHV

PAGE 266

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


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E4JRN74L9_PIX9XJ INGEST_TIME 2012-03-09T05:08:28Z PACKAGE AA00003942_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES