Washington Academy of Sciences

Founded in 1898

EXECUTIVE COMMITTEE

President

James E. Spates

President-Elect

Robert H. McCracken

Secretary

Donald O. Buttermore

Treasurer

R. Clifton Bailey

Past President

Ronald W. Manderscheid

Vice President (Membership Affairs)

M. Sue Bogner

Vice President (Administrative Affairs)

Jo-Anne A. Jackson

Vice President (Junior Academy Affairs)

Marylin F. Krupsaw

Vice President (Affiliate Affairs)

John G. Honig

Academy Members of the Board of

Managers

William M. Benesch

Carl E. Pierchala

Lawson M. McKenzie

Marcia S. Smith

Jean K. Boek

Thomas N. Pyke

BOARD OF AFFILIATED

SOCIETY REPRESENTATIVES

All delegates of affiliated

Societies (see inside rear cover)

EDITORS

Irving Gray

Joseph Neale

Lisa J. Gray, Managing Editor

ACADEMY OFFICE

1101 N. Highland St.

Arlington, Va. 22201

Telephone: (703) 527-4800

The Journal

This journal, the official organ of the Washing-

ton Academy of Sciences, publishes historical

articles, critical reviews, and scholarly scientific

articles; proceedings of meetings of the Acad-

emy and its Executive Committee; and other

items of interest to Academy members. The

Journal appears four times a year (March, June,

September, and December)—the December is-

sue contains a directory of the Academy mem-

bership.

Subscription Rates

Members, fellows, and life members in good

standing receive the Journal without charge.

Subscriptions are available on a calendar year

basis only, payable in advance. Payment must

be made in U.S. currency at the following rates:

U‘S. and Canada... .:.; $19.00

POSIT. eae 22e

Single Copy Price......: 7.50

Back Issues

Obtainable from the Academy office (address

at bottom of opposite column): Proceedings:

Vols. 1-13 (1898-1910) Index: To Vols. 1-13

of the Proceedings and Vols. 1—40 of the Journal

Journal: Back issues, volumes, and sets (Vols.

1-75 1911-1985) and all current issues.

Claims for Missing Numbers

Claims will not be allowed if received more than

60 days after date of mailing plus time normally

required for postal delivery and claim. No

claims will be allowed because of failure to no-

tify the Academy of a change in address.

Change of Address

Address changes should be sent promptly to the

Academy office. Such notification should show

both old and new addresses and zip number.

Published quarterly in March, June, September, and December of each year by the

Washington Academy of Sciences, 1101 N. Highland St., Arlington, Va. 22201. Second

class postage paid at Arlington, Va. and additional mailing offices.

INTRODUCTION

Operations Research (O.R.) is intimately involved with both the history of

the United States of America and with its future. Had it not been for the

contributions made by O.R. technology to the victory of the Allied Forces,

our nation as we know it might have ceased to exist. As we face a future in

which our national well-being depends upon our scientific and commercial

competitiveness on an increasingly-level playing field (we have no monopoly

on intelligence, energy or knowledge), timely and effective application of O.R.

technology may well be a deciding factor.

As explained in the paper “Historical Perspective’, by Donald Gross of

The George Washington University, the name ‘operations research’ was first

used by the British air ministry to describe work on the use of radar as an

early warning system during the years 1937-39. The roots of O.R., however,

go back even further—to the work of F. W. Lanchester and others in support

of the World War I war effort.

As Saul Gass of the University of Maryland at College Park points out in

his paper “The Current Status of Operations Research and a Way to the

Future”, O.R. work was being done in Washington D.C. within a year after

Pearl Harbor and continues to flourish in our area.

As O.R. spread around the world its rich mixture of applications efficiency

and effective solution of a wide class of operational problems, the Washington

area continued host to a critical mass of O.R. expertise. Some of the recent

O.R. accomplishments given in the report to the National Science Foundation

by the Committee on the Next Decade in Operations Research (CONDOR)

and published as ‘Operations Research: The Next Decade”’ are cited in this

journal by authors Gass, Gross and Harris, members of CONDOR.

Carl M. Harris of George Mason University and Richard H. F. Jackson of

the Center for Manufacturing Engineering, National Institute of Standards

and Technology (formerly National Bureau of Standards) in ‘“‘Operations Re-

search: Some Trends and Issues for the Future” continue development of the

theme introduced by Gross and Gass that high-speed digital computers have

provided a quantum leap in the ability to exploit O.R. theory and applications,

and carry them forward into the search for computer-aided improvements in

_ manufacturing productivity and in the modeling of processes under uncer-

tainty.

Gass, Gross, Harris and Jackson are professionally positioned in the aca-

demic area of our society, where the mission is to encourage research and

transmission of knowledge. No portrayal of a discipline, however, can be

complete without mention of the viewpoint of practitioners in the everyday

work world. Peter Malpass provides such a viewpoint in his paper ‘“‘O.R. as

Fun’. He is a former president of the Washington Operations Research/

Management Science Council (WORMSC) and head of the ‘Fun Bunch’, a

group of local O.R. practitioners who get together on a regular basis to

organize O.R. projects (apart from the ones they carry out in their day-to-

day employment) for the sheer fun of it. |

It is often said that Operations Research training provides an excellent basis —

for solving problems in management, science, engineering, or any other field

in which optimization of resource utilization is desired. We hope that the

papers in this issue will show the reader why this is so.

Marilyn T. Welles, Guest Editor

The MITRE Corporation

Trustee, Washington Operations Research/Management Science Council

Journal of the Washington Academy of Sciences,

Volume 79, Number 2, Pages 47-60, June 1989

Historical Perspective

Donald Gross

Department of Operations Research

The George Washington University

Washington, DC 20052*

ABSTRACT

Operations Research is defined and its history and prehistory given. Events setting the

stage for operations research and its early beginnings during World War II are presented.

Also highlighted are the post World War II growth of the profession, the development

of educational programs in operations research, and the emergence of professional so-

cieties in the field.

What Is Operations Research?

There are approximately 30,000—40,000 operations research professionals

in the world and if each were asked for a definition of operations research

(OR), one would probably end up with 30,000-—40,000 different definitions.

I believe the following conveys the essentials of what the field is all about:

Operations Research is the application of the principles of mathematics,

science and engineering for describing, understanding and improving the

operation of complex systems.

Emphasized are the key words, operation and complex systems. The object

of the OR analyst’s attention is a complex system, a system made up of people,

machines (in the general sense of hardware devices) and information. The

analyst’s goal is to study such systems so that they can be operated in the best

possible manner.

The following article that appeared in the Business Section of the Wash-

ington Post in 1985 entitled, ‘City Turns to High Tech to Unsnarl Its Traffic”’

*On leave 1988-89 to the National Science Foundation, Operations Research and Production Systems

Program, Division of Design and Manufacturing Systems Engineering.

48 DONALD GROSS

serves as an excellent example of the type of system that, for maximum

performance, requires an operations-research-type of analysis.!

‘“. . . the city is going to spend $25 million over the next five years to trans-

form its aging signals into a computer-controlled network of lights that can

be programmed to ease rush-hour congestion and improve traffic flow.”

The goal was to expedite the city’s traffic flow through improved operation

of its network of traffic signals, an objective that required a sizable investment.

“The new system would link the city’s traffic signals in a giant computer

network that could monitor each signal and program it to display red or

green depending on the intensity of traffic . . . Bringing traffic lights into

the information age is not a trivial exercise. Urban traffic, with its elaborate

ecology of main roads, side streets, cars, buses and pedestrians, is an enor-

mously complex systems problem. Everything interacts: everything is inter-

dependent. A traffic light doesn’t just control an intersection; it’s also part

of a system—one pedestrian’s green is a motorist’s red.”

An integrated traffic signal network (such as the 1,200 intersections in the

downtown Washington area slated for control) is a very complex system of

people (pedestrians and motorists), machines (motorized vehicles, traffic lights,

computers) and information (amount of traffic congestion at intersections,

time of day, weather conditions). Furthermore, intersections are not inde-

pendent because traffic at a given intersection can be influenced by and can

influence several nearby intersections.

There must be criteria or performance measures for the system so that

proposed operating strategies can be evaluated.

“Traffic control experts estimate the system could cut travel times by as

much as 25 percent along some routes, as well as account for $4 million in

fuel savings by minimizing the number of automobile starts and stops.”

There are often various measures of system performance possible for complex

systems. Travel times and fuel savings were two key measures highlighted for

the Washington study.

“In other words, traffic control is the mathematical art of compromise. Reds

and greens have to be balanced against pedestrians and machines. But it

takes high powered computers and sophisticated software to assure that those

compromises can be modeled, calculated and then successfully implemented

in the real world of traffic.”

To accomplish the goal of improving the operation of a complex system, a

model (usually mathematical) of the real world must be developed, relating

the operating decision variables to the measure or measures of system per-

formance. In the DC traffic example, the decision variables were the timing

of red-green at each intersection and the measures of system performance

were such things as travel times and fuel savings.

HISTORICAL PERSPECTIVE 49

Building the model is only half the job. The remaining half is to utilize the

model to find the best values of the decision variables (for example, the optimal

red-green times for each intersection). To do this, as pointed out in the above

quote, computers are almost always an indispensable ingredient of any OR

study because of the complexity of the systems to be analyzed.

Washington readers may well be wondering about the status of the three-

year-old downtown DC traffic control project. As of January 1989, 700 in-

tersections were being tested with the new hardware and software, and it is

anticipated that all 1,200 intersections governed by the downtown traffic con-

trol system will be in operation by the end of calendar year 1989.* Thus, after

the analysis and recommendations, it is almost always necessary to have a test

or evaluation phase of the study prior to implementation. The steps of a

complete OR study can be summarized as follows.

1. Formulate the problem (isolate the decision variables and determine system goals

and performance measures)

. Build the model (relate the system performance measures to the decision variables)

. Utilize the model to find the best settings for the decision variables

. Test and evaluate the model

. Implement the recommendations

Nn & W bdo

Prehistory

The industrial revolution (a more accurate name would be industrial evo-

lution because of its gradual nature over a relatively long period of time) sets

the stage for the eventual arrival of operations research. The industrial rev-

olution is generally acknowledged to have begun in England in the latter half

of the eighteenth century with the invention of Watt’s steam engine (1760),

although there were harbingers such as John Kay’s invention, in 1733, of a

mechanical weaving device called the flying shuttle.’ Shortly after Watt’s steam

engine, a number of inventions appeared in rapid succession in the textile

industry including Hargreaves’ spinning jenny, Arkwright’s water frame,

Crompton’s ‘‘mule,’”’ Cartwright’s power loom and, in the United States, Eli

Whitney’s cotton gin.

The end of the eighteenth century saw the factory system in place in the

textile industry. The first quarter of the nineteenth century yielded steam

locomotion, both rail and sea. In the late 1820’s, a civilian engineering profes-

sion emerged (prior to this, engineering was strictly military). Civil, mechanical

and mining engineers made up this fledgling group.

In the 1880's, a particularly significant event occurred. Frederick W. Taylor,

working for a steel company in Philadelphia, studied how men shoveled raw

materials and machined metals, recording the time it took to perform the

various tasks involved. This first time-and-motion study is generally regarded

as the beginning of industrial engineering (IE), and Taylor is often referred

to as the “‘father’’ of industrial engineering, a field out of which much of OR

eventually developed.

50 DONALD GROSS

In 1911, Frank and Lillian Gilbreth employed time-and-motion study to

effect methods improvement and work simplification in the brickyard industry,

thus significantly increasing productivity. They were also productive in their

non-working hours, having had twelve children (the book and movie ““Cheaper

by the Dozen”’ was based on their lives).

Earlier attempts to accurately predict man hours of labor needed to accom-

plish particular tasks were reportedly made by Jean Perronet of France in

1760 and Charles Babbage of England in 1820,* although their efforts were

either not as elaborate as the Gilbreths’ and Taylor’s or not as publicized (or

possibly both), since it is Taylor and the Gilbreths who are associated with

the beginnings of industrial engineering. These IE studies pointed toward OR

development because they dealt with how people interacted with machinery,

and thus considered the machines and people together as a system.

Another significant milestone in the prehistory of OR occurred in Denmark

in 1917 when a Danish mathematician A. K. Erlang performed studies for

the Danish National Telephone Company using mathematics and probability

to study telephone traffic congestion problems. This marked the beginning of

the mathematical theory of waiting-line phenomena, now a major subarea of

operations research called queueing theory (queueing theory methodology is

used in vehicle traffic studies such as the Washington, DC, example cited

above).

The Early Years

Operations Research, generally acknowledged to have begun in the United

Kingdom during the early stages of World War II, focused on man-machine

systems problems associated with the war effort. Scientists utilized mathe-

matics and engineering techniques to attack these problems. However, J. F.

McCloskey, in the first of three excellent articles on the early years of OR,

points out that there were actually precursors of OR in military affairs in the

U.K. as early as 1904, when studies led to the redesign of naval vessels in

terms of strategic and tactical requirements.’ In 1915, Lord Tiverton performed

a comprehensive study of bombing strategy involving target selection, navi-

gation methods, weather and logistical problems.

In 1916, F. W. Lanchester developed sets of differential equations describing

outcomes of military battles as functions of variables such as size of forces,

concentration of forces, and firepower. Today, the Operations Research So-

ciety of America (ORSA) awards a yearly prize, named for Lanchester, for

the best piece of theoretical work in the field.

About that same time period (World War I), A. V. Hill, head of the ex-

perimental section of the British army’s Munitions Invention Department,

studied antiaircraft gunnery, developing procedures that increased the effec-

tiveness of antiaircraft fire.

Finally, in the United States, Thomas A. Edison, at that time head of the

U.S. Naval Consulting Board, developed strategies, using statistics, for evad-

ing and destroying submarines, a field of endeavor not usually associated with

Edison.

HISTORICAL PERSPECTIVE 51

The name “operational research”’ was first used by A. P. Row of the British

air ministry to describe the work on the employment of radar as an early

warning system during the years 1937-39. Hugh Larnder headed up the first

official OR group at the Royal Air Force Fighter Command headquarters.

(OR is still an acronym for operational research in the U.K., although the

name was modified to operations research in the U.S.)

In 1940, a British physicist, P. M. S. Blackett, scientific advisor to the Royal

Army Antiaircrft Command, formed an OR group consisting of two physiol-

ogists, two mathematical physicists, an astrophysicist, a surveyor, and later

added another physiologist, and two mathematicians. This ‘“‘mixed team”’ ap-

proach was necessary for analyzing the complicated man/machine/data sys-

tems characterized by radar networks for early warning and antiaircraft net-

works for defense. Because of the many disciplines represented in this group,

it became known as Blackett’s Circus. This mixed team approach characterized

OR groups well into the 1950’s, at which time scientists formally trained in

OR began to emerge from the universities.

Early OR groups also contained chemists, psychologists, economists and

engineers and worked exclusively on problems dealing with wartime opera-

tions, expanding from the initial air defense problems to a variety of other

operational-type problems facing the army, navy, and air force such as the

best placement of explosive mines, countermeasures against enemy subma-

rines, and the effectiveness of night versus day bombing raids.

One civilian OR group was established in the U.K. to study effects on the

civilian population of enemy air raids and strategies for air raid protection.

Even though this was not a military group, the problems studied, nevertheless,

were still associated with the war.

Blackett is often credited as being the father of OR, not only for his initial

work in air defense, but also for his efforts in 1941 of setting down methodology

developed over the previous years and making a case for OR as a discipline

requiring competently trained scientists. However, it wasn’t until the mid

1950’s that formal programs in operations research were offered at universities.

McCloskey, in the second of his three articles on the early years of OR,

points out that by comparison to the sophisticated analyses of today, the early

OR World War II studies were not much more than intuition and common

sense. Nevertheless, these studies were done by scientists, using available

scientific techniques, to treat data and to work on military operational prob-

lems from a scientific point of view.° The analysts worked closely with military

commanders and gave invaluable advice to designers of equipment about

technical difficulties of operation, to manufacturers on reliability problems

and to the military commanders themselves regarding the best way to utilize

and operate their very scarce resources of equipment and manpower.

Thus opertional research in the U.K. contributed to the effectiveness of the

Royal Air Force Fighter Command by analyses of air engagements, integration

of radar into early warning systems, strategies for interception during night

attacks, and weapon effectiveness. The increase in the probability of inter-

cepting enemy aircraft due to radar was estimated to be a factor of ten, and

a further factor of two was attributed to OR analyses.’

52 DONALD GROSS

In the Coastal Command, OR teams analyzed such problems as the best

procedures for finding enemy submarines, convoy protection, and the optimal

depth for exploding depth charges. For the Bomber Command bomber losses

and bombing effectiveness-night versus day were worked on as well as radar

evasion techniques. In civil defense, such problems as blast pressures for

human survival and the effect on the civilian population of enemy air raids

were investigated. Indeed, OR played a very significant role in the war

effort.

Meanwhile, across the Atlantic, OR was also emerging in the U.S., in no

small part due to visits by British scientists. By the time the U.S. entered

World War II, there were two operations research groups in the navy and one

(called operations analysis) in the army air corps.°

The Naval Ordinance Laboratory had a group studying the use of mine

warfare as an offensive weapon. The group was headed by Ellis A. Johnson,

a geophysicist, who later took a commission and served on active duty. In

1942, the Naval Ordinance Laboratory Operational Research Group was of-

ficially established.

The same year, upon request of the navy, another organization, the An-

tisubmarine Warfare Operations Research Group (ASWORG) was formed

by an M.I.T. physicist, Philip M. Morse, and was soon joined by other prom-

inent scientists who became the ‘‘founding fathers” of OR in the U.S.” Among

them were Robert F. Rinehart of the Case School of Applied Science (later

Case Institute of Technology and currently Case Western Reserve University),

John B. Lathrop, an insurance actuary, Bernard O. Koopman, of the Applied

Math Panel at Columbia University. Jacinto Steinhardt, a physical chemist

with the National Bureau of Standards, and George E. Kimball, a professor

of chemistry at Columbia (ORSA also has awards named for Kimball and

Morse). William Shockley of Bell Labs (later to win a Nobel prize for his

work on transistors) was also a member of ASWORG.

ASWORG became ORG (Operations Research Group) in late 1943 and

branched out to study all aspects of naval warfare.

In the fall of 1942, the Eighth Air Force created an operations analysis

(OA) section, consisting of mathematicians, physical scientists, social scientists

and lawyers, among them the well-known chemist/statistician William J. You-

din and the well-regarded economist Robert Dorfman. The section worked

on problems of bomb and fuse selection, bombing accuracy, battle damage

and loss, radar, radar countermeasures, and gunnery.

Many independent OA sections evolved in various Air Force units through-

out the U.S. and abroad. LeRoy Brothers, an engineer and later dean of

engineering at Drexel Institute of Technology in Philadelphia headed up a

group in New Delhi to serve the army air forces in the India-Burma theater.

Two OR units were established in the army (as distinct from the army air

corps), one attached to the signal corps working on radio direction finding

and radar and one in Hawaii, working on army, navy and air force problems

there.

As in the U.K., OR made a significant impact on the effectiveness of our

fighting forces during World War II.

HISTORICAL PERSPECTIVE 53

Post World War II

By the end of the war, a great deal of methodology had been developed

for studying operational problems involving personnel and equipment. Many

of the above mentioned “fathers” of OR stayed on with the government,

working in the peacetime military. Others left the government, some returning

to universities and others entering industry, which had produced a number of

rather large (for those days) organizations with fairly complex problems in-

volving operating strategies and tactics.

A landmark event also occurred at this time—the invention of the electronic

digital computer, without which, OR would be far different than it is today.

The original concept of an automatic digital computer can be traced back to

Charles Babbage (1791-1871). Babbage’s “‘analytical engine,’ however, was

never completed.'’ It was to be punched card controlled (punched cards having

been invented by J. M. Jacquard (1752-1834) for automatic weaving). Herman

Hollerith (1860-1929) further refined punched cards and utilized them to

mechanize the U. S. census counts.

In 1941, the Mark I computer was built at Harvard and used mechanical

counters from I.B.M. business data processing machines. The ENIAC (Elec-

tronic Numerial Integrator and Calculator) was developed at Aberdeen Prov-

ing Ground and demonstrated at the University of Pennsylvania in 1946. This

computer was a collection of 20 electronic adding machines. The machine was

designed to integrate the ballistic equations for producing firing tables, but

could also be used as a general purpose computer.

The next group of computers developed, the stored program computers,

were most significant as they were the first in which the actual program was

stored in memory along with the data. These were really the first generation

of “modern” digital computers, and the first of these was Cambridge Uni-

versitys EDSAC, built in 1949. The EDSAC was quickly followed by the

SEAC and SWAC of the National Bureau of Standards, Whirlwind I at M.I.T.,

and the Manchester computers, developed at Manchester University. The

early 1950’s saw the beginning of commercial use of stored program computers.

Even though the early stored program computers had far less computing

power than the first I1.B.M. or Apple personal computers, nevertheless, these

ushered in an era which has had as much, if not more, of an impact on society

than the industrial revolution. Their significance to the growth of OR was

paramount.

At the close of the war, the navy, realizing the value of the independent

scientists working in its Operations Research Group (ORG), transformed the

group into the peacetime Operations Evaluation Group (OEG), which was

administered under a navy contract to M.I.T."’ Jacinto Steinhardt stayed on

to become its director. In 1962, OEG was incorporated into the newly formed

Center for Naval Analyses (CNA), which continues to flourish today in Al-

exandria, Virginia.”

The army also realized the value of its war-time operations research effort

and, in 1948, established the Operations Research Office (ORO) via a contract

to the Johns Hopkins University.'? Ellis Johnson, formerly with the navy’s

54 DONALD GROSS

ORG, became ORO’s director. In 1961, ORO became a non-profit federal

contract research center under the name Research Analysis Corporation (RAC).

RAC, in 1971, was taken over by the General Research Corporation (GRC)

of McLean, Virginia. When ORO became RAC, Ellis Johnson went to Case

Institute of Technology as a professor and director of the Systems Research

Center, where he taught operations research courses."

Other wartime members of the navy’s ORG left for various destinations.

Philip Morse returned to M.I.T. and George Kimball to Columbia. They

collaborated to document ORG’s activities which led to one of the first texts

in OR, Methods of Operations Research, published in 1951. Kimball later

joined the management consulting firm of Arthur D. Little, as did John Lath-

rop. Robert Rinehart left ORG to serve as executive secretary of the Research

and Development Board for the Department of Defense. From there he spent

a short time at Duke University and after an additional short stay at the

Institute for Defense Analysis, returned to Case Institute of Technology.'®

Thus the OR methodology developed at ORG was spread and further de-

veloped at a variety of places after the war.

In 1947, the Weapons Systems Evaluation Group (WSEG) was established

to advise the secretary of defense and the joint chiefs of staff. The group had

a military director of three star rank and a staff consisting of three general

grade officers, one from each of the services. There was also a component of

civilian scientists under a civilian director of research. In 1956, a non-profit

corporation called the Institute for Defense Analysis (IDA) was formed and

administered by a consortium of universities.'’ IDA exists today, as a non

profit corporation, in Alexandria, Virginia, doing studies in both the military

and civilian sectors.

The very active OR endeavors in the army air corps during the war continued

to abound in the post-war peacetime air force. OR in the air corps was de-

centralized as opposed to the navy’s centralized ORG. One war-time group,

the Combat Analysis Office of the Statistical Control Branch of the air corps

in the Pentagon was concerned with sortie rates needed for air force planning.”®

A mathematical statistician, George Dantzig, was there devising techniques

to apply to such problems. Since these problems were associated with air force

planning or ‘‘programming,”’ the name programming was used to describe this

class of problems. The problems were characterized by linear functions, and

the name linear programming was coined.

Linear, or in its more general form, mathematical programming, is not to

be confused with the term computer programming (namely, a sequence of

instructions to “‘tell” the computer what to do). The former is a mathematical

theory of solving optimization problems, that is, problems where one desires

to minimize or maximize a function of several (usually many) decision variables

subject to constraints (inequalities involving the decision variables). When the

function to be minimized or maximized (called the objective function) is linear

and the constraints are linear, we have a linear programming problem. If the

objective function and/or the constraints are not all linear, we have a nonlinear

programming problem. If the variables must be integers (airplanes instead of

gallons of fuel for example), the problems are referred to as linear or nonlinear

integer programming problems.

HISTORICAL PERSPECTIVE 55

Of course, linear programming and computer programming, while not linked

semantically, are closely linked practically, since the computations required

for most linear programming problems can not be performed without the aid

of a computer. It was indeed fortunate that, at the same time Dantzig was

developing a computational method for solving linear programs, stored pro-

gram electronic computers were to become available. The work of Dantzig

and his colleagues became known as air force project SCOOP (Scientific

Computation of Optimal Programs). The SCOOP scientists were most keenly

interested in developments in electronic computation and a great deal of air

force money sponsored much of the early work in developing electronic com-

puters.

Although Dantzig’s work on linear programming began during the war, it

continued into the post-war period (it was about 1946 when the name SCOOP

was used and 1948 when the name linear programming was coined). In 1947,

Dantzig developed a very efficient procedure for solving linear programs known

as the simplex method, which is still the preferred method of solution in most

linear programming problems. The method is very compatible to programming

on computers and there are excellent computer packages available that can

handle problems with tens of thousands of variables and constraints.

In 1946, the air force let a contract to the Douglas Aircraft Company to

administer project RAND, “. . . a continuing program of scientific study and

research on the broad subject of air warfare with the object of recommending

to the Air Force preferred methods, techniques, and instrumentalities for this

purpose”? RAND stands for research and development. The SCOOP people

kept in close contact with RAND scientists, and Dantzig eventually went to

RAND in 1952. From RAND, Dantzig went to Berkeley and then to Stanford,

where he is today, still very active in teaching and research. He has a joint

appointment in the operations research and computer science departments.

Another prominent mathematician-operations researcher, Richard Bell-

man, was also at RAND from the mid-1950’s to the mid-1960’s. Just prior to

his coming to RAND, he developed dynamic programming, another powerful

mathematical optimization technique for certain classes of programming prob-

lems.*” He left RAND for the University of Southern California in 1965.

Also at RAND from 1948 to 1973 was Edward S. Quade. Originally trained

as a mathematician, Quade worked on analyzing military systems, often fo-

cusing on human interaction with hardware. This work came to be known as

system analysis, and Quade published many books in this area.*! Systems

analysis and management science are names used interchangeably with op-

erations research. Systems engineering and industrial engineering are closely

related fields.

OR Comes of Age

While it was natural for the military and the defense establishment to expand

their OR activities and organizations after the war because of the wartime

successes of the OR activities, OR also grew in private industry and univers-

56 DONALD GROSS

ities. Those wartime OR analysts who returned to their universities developed

OR courses and research programs.

M.1I.T. under Morse’s influence, Columbia with the return of Kimball and

Koopman, Case Institute of Technology with the eventual return of Rinehart

and acquisition from Wayne State University of Russell L. Ackoff, an architect

with a doctorate in the philosophy of science, and Johns Hopkins, through

its ORO influence, all offered pioneering study in OR.

Locally, George Washington University also entered the OR field early,

having established a navy-sponsored Logistic Research Project in 1949 and

three years later, an operations research group in the engineering school under

the direction of Glen D. Camp, a wartime analyst with OEG.

In the late 1940’s a rise of OR activity also took place in industry. It started

with management consulting firms such as Arthur D. Little, and Booz, Allen

and Hamilton, because of their natural affinity for analytically studying man-

agement decision problems. Arthur D. Little attracted Lathrop and Kimball.

Booz Allen Applied Research, a subsidiary of Booz, Allen and Hamilton,

drew George Shortley from ORO. Lathrop left Arthur D. Little for Lockheed

Aircraft, a company in an industry which embraced OR with a passion. By

the mid-1950’s, almost every aircraft company had an OR group—Bell, Boeing,

Douglas, Chance Vought, Northrop, Republic, to mention just a few.

Other industries also developed OR activities, and by the mid-1960’s OR

was firmly entrenched in automobile and related companies, banking and

investment, brewing, building products, chemical, communications, electron-

ics and computers, insurance, food, retailing, paper, petroleum, photography,

printing and publishing, rubber, textiles, tobacco, transport, and utilities.”

In addition to the aircraft industry, operations research groups were par-

ticularly prevalent in the oil companies, where linear programming techniques

were used for gasoline blending and production planning, and statistical meth-

ods were employed for service station site selection. The communications

companies, particularly Bell Labs (now AT&T Labs), carried on the pioneer-

ing work of A. K. Erlang and employed analysts who further developed

queueing theory.

OR also became established in state and local governments. Traffic analyses,

of course, were prime candidates for OR attention. One famous study was

done in the early 1950’s by the Port of New York Authority on tunnel traffic

between New Jersey and New York City.” But traffic management was only

one of a myriad of problems facing local jurisdictions that was amenable to

OR methodology. Schedules for garbage and leaf pick up, fire station location,

and police patrol schedules are a few such examples. RAND, branching out

from strictly military work, opened a New York City office to assist the city

in just such problems. Most large metropolitan area governments either have

in-house OR capabilities or utilize outside consultants.

Thus by the end of the ‘‘fabulous fifties,” the post-war spread of OR was

complete—in peacetime defense, in industry, and in federal, state and local

government.

How to best operate complex systems, the essence of operations research,

must be addressed by organizations in all sectors of society. This is what makes

HISTORICAL PERSPECTIVE 57

the field so exciting and attractive as a career choice for scientists. A manu-

facturer must decide how much to ship from each factory to each warehouse

so as to satisfy warehouse needs without exceeding factory capacities, and at

minimum costs. An automobile rental agency has exactly the same problem

in determining how to get cars from locations where there are too many to

those locations that are short. Scheduling policemen to beats takes a similar

type of methodology as scheduling crews to airline flights. Determining how

many clerks to have during peak hours in a post office applies to tellers in a

bank as well. Optimal search patterns for finding submarines apply equally

to searching for lost weekend sailors. The methodology for determining the

best sites for locating gasoline stations can also be used by fast food chains.

With the arrival of the information revolution, complex systems abound.

Developing methods to insure their best operation is a necessity for the country

to be productive and compete in the world marketplace.

Education

In addition to the six universities mentioned earlier (M.I.T., Columbia,

Case, Wayne State, Johns Hopkins and George Washington), a number of

others began offering OR programs by the mid-1950’s. Generally these were

options in industrial engineering, engineering administration or industrial

administration programs. Cornell, Georgia Tech, Carnegie Tech, and Stanford

were among schools with such OR programs.

In a 1959 report of the educational committee of ORSA, programs of thirty

colleges and universities are described, and most of these still have excellent

programs today.* In a 1986 update of educational programs, 120 programs

are listed, with 52 in engineering departments, 45 in business/management

departments and 23 in various other university departments such as mathe-

matics, statistics, or computer science.~

Most formal OR education is at the graduate level. There are a few un-

dergraduate OR programs; however, most are options in industrial engineer-

ing. Almost any physical science, mathematics or engineering major is good

preparation for grduate OR study. For example, at George Washington Uni-

versity, about 40% of the OR master’s students have engineering or physical

science undergraduate degrees, about 40% have mathematics or statistics

undergraduate backgrounds, and the rest represent a variety of undergraduate

majors from economics to business administration.

Professional Societies

There are two major professional societies in the OR field. The Operations

Research Society of America (ORSA) was established in 1952. Most of the

founding members were among those previously associated with wartime OR

and who went into peacetime military OR careers with organizations such as

OEG, ORO and the air force’s Operations Analysis Office (QAUSAF). Co-

58 DONALD GROSS

lumbia, M.I.T., and Case Institute of Technology were well represented as

was Arthur D. Little. The organizer of the movement to establish a profes-

sional society in operations research was Philip Morse, and he became ORSA’s

first president. There were 73 charter members. The society now has ap-

proximately 7,000 members.

In 1953, The Institute of Management Sciences (TIMS) was organized as

an international society dedicated to the scientific analysis of complex decision

problems. There were 69 charter members, and William W. Cooper, a pro-

fessor of industrial administration at Carnegie Institute of Technology (now

Carnegie Mellon University) was elected its first president. Cooper is well

known for his work with colleague Abraham Charnes (the seventh TIMS

president) in linear programming. TIMS now also has approximately 7,000

members. Two thousand TIMS members are also members of ORSA so that

the combined membership of the two societies totals approximately 12,000.

While most of the non-university ORSA charter members were from de-

fense-related OR groups, the non-university charter members of TIMS were,

for the most part, from private industry. Such companies as Celanese Cor-

poration, General Electric, IBM, Esso Standard Oil, Computer Research

Corporation, Bell Labs, Peat, Marwick and Mitchell, Remmington Rand and

Burroughs were among the organizations where many of the charter members

were employed. However, the early distinction between the military orien-

tation of ORSA and the industry orientation of TIMS has disappeared. Today,

both societies have members from academia, defense, private industry, and

civilian government.

ORSA and TIMS have many joint activities including two national meetings

each year and the publication of several jointly sponsored journals. TIMS has

an international meeting every two years and ORSA is a member of IFORS,

the International Federation of Operational Research Societies, made up of

national OR societies similar to ORSA, representing 32 countries throughout

the world with a total membership of approximately 30,000.

OR in Washington

The Washington, DC, metropolitan area is a particularly active area for

OR professionals. Because of OR’s origin in the military during World War

II, much of it taking place in Washington, and the post-war development of

peacetime defense-related OR organizations such as ORO, OEG, WSEG and

OAUSAF, which were located in Washington or its suburbs, Washington

probably has more OR analysts than any other single location in the world.

Although lacking the industrial OR scientists that can be found in other large

metropolitan areas, the federal government (both in the Department of De-

fense [DOD] and civilian agencies such as the Department of Energy [DOE]

and the Department of Transportation [DOT]) more than make up the dif-

ference.

A joint local TIMS-ORSA chapter, called the Washington Operations Re-

search/Management Science Council (WORMSC) puts on an active program

HISTORICAL PERSPECTIVE 59

of monthly meetings, an annual symposium and issues a monthly newsletter.

Most of the local universities offer courses in OR and related topics. George

Mason University offers a masters’s degree program in OR and George Wash-

ington University and the University of Maryland offer both master’s and

doctor’s degrees.

Washington area OR professionals are well represented among the officers

of ORSA and TIMS, as well as allied professional societies in mathematics,

statistics, business administration, economics, engineering, and science.

Conclusions

The past two decades saw the transformation of the Industrial Revolution

into the Information Revolution, bringing us much more complex systems

made up of people, information and high technology hardware. The nation’s

productivity depends on the efficient operation and management of such sys-

tems. Operations Research is not only well poised to play a major role in this

effort, but it is a necessary ingredient for its success.

References Cited

1. Schrage, M. 1985. City turns to high tech to unsnarl its traffic. Washington Post. March 18.

2. Brown, J. 1989. Dynatron Corp. Private telephone conversation. January, 17.

3. Encylopaedia Britannica. 1971. 12: 210-215.

4. Encyclopaedia Britannica, op. cit.

5. McCloskey, J. F. 1987, The beginnings of operations research: 1934-1941. Operations Research. 35:

143-152.

6. McCloskey, J. F. 1987. British operational research in World War II. Operations Research. 35: 453-

470.

7. McCloskey, op cit.

8. McCloskey, J. F. 1987. U.S. operations research in World War II. Operations Research. 35: 910-925.

9. McCloskey, J. F. and F. N. Trefethen. 1954. Operations Research for Management. The Johns Hopkins

Press, Baltimore.

10. 1971. Encyclopaedia Britannica. 1971. 6: 245-245A.

11. Engel, J. H. 1960. Operations research for the U.S. Navy since World War II. Operations Research,

8: 798-809.

12. Engel, J. H. 1985. A tribute to Dr. Jacinto Steinhardt. OR/MS Today. 12, No. 1, p. S.

13. Whitson, W. L. 1960. The growth of the Operations Research Office in the Army. Operations Research.

8: 809-824.

14. Page, T., Pettee, G. S. and W. A. Wallace. 1974. Ellis A. Johnson, 1906-1973. Operations Research.

22: 1141-1155.

15. Little, J. D. C. 1985. Obituary of Philip M. Morse. OR/MS Today. 12, No. 5, pp. 6-7.

16. Weiss, A. B. 1985. Obituary of Robert F. Rinehart. OR/MS Today. 12, No. 3, p. 5.

17. Pugh, G. E. 1960. Operations Research for the Secretary of Defense and the Joint Chiefs of Staff.

Operations Resarch. 6: 839-846.

18. Dantzig, G. B. 1988. Impact of linear programming on computer development. OR/MS Today. 12,

No. 4, pp. 12-17.

19. Specht, R. D. 1960. RAND—A personal view of its history. Operations Research. 8: 825-839.

20. Adomian, G.and A. O. Esogbue. 1984. The contributions of Richard Ernest Bellman (1920-1984) to

operations research and management science. OR/MS Today. 11, No. 4, pp. 6-7.

21. Miser, H. J. 1988. In memorium: Edward S. Quade, 1908-1988. OR/MS Today. 16, No. 5, pp 16-

Wks

22. Ackoff. R. L. and P. Rivett. 1964. A Manager’s Guide to Operations Research. John Wiley and Sons,

New York.

23. McCloskey, J. F. and J. M. Coppinger. 1956. Operations Research for Management, Volume 2. The

Johns Hopkins Press, Baltimore.

24. ORSA, 1959, Formal Educational Offerings in Operations Research. Education Committee of the

Operations Research Society of America. Baltimore.

25. ORSA. 1986. Educational Programs in Operations Research. Operations Research Society of America.

Baltimore.

Journal of the Washington Academy of Sciences,

Volume 79, Number 2, Pages 60-69, June 1989

The Current Status

of Operations Research and

a Way to the Future

Saul I. Gass

College of Business and Management, University of Maryland,

College Park, MD 20742

ABSTRACT

In this paper, we review the current status of Operations Research in terms of its

methodology, selected applications, and its world-wide activity, and conclude with a

discussion of the modeling process and its role in the future of Operations Research. As

noted by Gross in the first paper of this volume, Operations Research traces its origins

to just before World War II with the work done in Great Britain in developing operational

tactics for the then new radar devices. (Also, see the paper by Lovell.)' Based on the

successful British experience, Operations Research moved to the Washington, DC area

within a year after Pearl Harbor. With the war’s end, governmental Operations Research

groups continued their research in military operations: the Navy’s Operations Evaluation

Group, the Army’s Operations Research Office, and the new Air Force’s Directorate of

Management Analysis and Operations Analysis Office. As one who was fortunate to

become involved in Operations Research by working for the Air Force towards the end

of the “‘early days,” I have a fond view of the support given Operations Research by the

Washington area, both Federal and private. Thus, from my perspective, it is very appro-

priate for the Washington Academy of Sciences to publish this volume on Operations

Research for the region’s scientific community. The papers in this volume, all by local

Operations Research researchers and practitioners, attest to the value of the Washington

influence and the importance of Operations Research. In the following sections, we review

the current status of Operations Research in terms of its methodology, selected appli-

cations, and its world-wide activity, and conclude with a discussion of the modeling process

and its role in the future of Operations Research.

THE CURRENT STATUS OF OPERATIONS RESEARCH 61

Introduction

Although Operations Research is a comparatively young science and profes-

sion, we can be very proud of its accomplishments. Operations Research has

made a great and profound impact on how we manage all facets of business,

industry, and government. And, just as important, the philosophy of the

Operations Research decision model has changed forever the process by which

we define and analyze decision problems. Operations Research has arrived

and is a respected profession.

When I state that Operations Research has arrived, I mean that a wide class

of operational decision problems (problems which I think the pioneers of

Operations Research took as their challenge) have, for all intents and pur-

poses, been solved. The classic and basic set of operational decision problems

found in business, industry, and government can be and are being solved by

Operations Research procedures. Operations Research theory has been es-

tablished and is constantly being advanced; Operations Research techniques

have been codified and are accessible; Operations Research applications, which

are the life-blood of the profession, are on the increase. We find that special

issues Of Operations Research, the Journal of the Operations Research Society

of America, have been published in health and health care, military and space

applications, urban problems, scheduling and operations management, and

manufacturing. Also, in the Handbook of Operations Research, Vol. 2, by

Moder and Elmaghraby,’ we find chapters in urban service systems, the health

services, educational processes, transportation systems, military systems, elec-

tric utilities, the process industries, and the leisure industries. Operations

Research impacts just about all application areas.

Much Operations Research is going on throughout the world with great

success. For example, the July-August 1986 edition of INTERFACES was

dedicated to the theme ‘““Management Science/Operations Research Around

the World.” In that volume alone, we find Operations Research applications

from Greece, Ireland, Federal Republic of Germany, Thailand, Finland, and

Colombia. The two previous issues of INTERFACES (January-February and

March-April 1986) contained descriptions of Operations Research activities

in Brazil, Canada, The Peoples Republic of China, as well as in the United

States. Surveys and publications on how Operations Research techniques are

being used throughout the world show that we know how to solve—and we

do solve—the problems of production, inventory, scheduling, blending, fore-

casting, logistics, and so on.**°-°’8" Applications of Operations Research are

being reported in all the major journals. A recent study showed that, on the

average, 36% of all Operations Research/Management Science articles pub-

lished in the major Operations Research/Management Science journals dis-

cuss real-world applications, with each issue having at least two such articles."

The program of the 1987 International Federation of Operational Research

Triennial Conference, held in Buenos Aires, reinforces the view that Oper-

ations Research is being used and is well-accepted. The program’s national

contributions ran from Argentina to Switzerland—from theory to practice,

with sessions and papers on the full range of Operations Research topics by

62 SAUL I. GASS

speakers from most of the 36 national societies that make up the Federation.

The profession can be proud of its individual and collective accomplishments

in the name of Operations Research. I am very positive about Operations

Research past and Operations Research present. And, from the paper by

Harris and Jackson in this issue, one can only be confident that Operations

Research will be an integral part of our industrial and technological future.

Operations Research Methodologies and Applications

It is difficult to circumscribe the methodological limits of Operations Re-

search techniques. We may say that any mathematical, logical or scientific

procedure that aids one in making decisions falls within the domain of Op-

erations Research. For example, statistical and probabilistic techniques are

widely used in Operations Research studies. However, there are specific meth-

odologies developed over the past fifty years that combine to form the trade-

mark of an Operations Research analyst. Here we have linear, nonlinear,

integer and mathematical programming; optimization theory; network flow

analysis; PERT/CPM; stochastic processes; theory of queues; decision anal-

ysis; game theory; search theory; dynamic programming; simulation and gam-

ing; inventory theory; forecasting; reliability theory; Markovian decision pro-

cesses; and heuristic problem solving. The use of many of these techniques

by Operations Research analysts is now standard practice. In addition, we

find them being used by other professions and being incorporated into other

disciplines, for example, by computer science and econometrics. It is not the

place here to describe the essence of these techniques (the interested reader

should consult Moder and Elmaghraby’ or any textbook in Operations Re-

search). But, it is appropriate to review some of the rich application base of

these methodologies. We do that by citing, in an edited form, some of the

recent Operations Research accomplishments, given in the report to the Na-

tional Science Foundation by the Committee On the Next Decade in Oper-

ations Research (CONDOR), and published as‘‘Operations Research: The

Next Decade’’'!' (Gass, Harris and Gross were members of CONDOR), and

by citing recent application reports published in Jnterfaces.

Some Operations Research Accomplishments

A Corporate Operations Research System

Citgo Petroleum Corporation developed a comprehensive analysis system

that combines such Operations Research disciplines as mathematical pro-

gramming, forecasting, and expert systems with statistics and organizational

theory, and applied it to crude and product acquisition, refining, supply and

distribution, strategic and operational market planning, inventory control, and

accounts receivable and payable.” Citgo credits this Operations Research

system with turning a 1984 operating loss of over $50 million into a 1985

operating profit of over $70 million.

THE CURRENT STATUS OF OPERATIONS RESEARCH 63

Logistics Management Using Network Flow Models

Network flow problems are concerned with the movement of goods such

as oil, telephone messages or products from sources to destinations over a set

of transportation links (arcs) and through intermediate transshipment points

(nodes). It is now possible to solve huge network flow problems, and as a

result, important new applications have emerged. Companies such as Agrico,

Ciba-Geigy, W. R. Grace, International Paper, Kelly-Springfield, Owens-

Corning Fiberglass, Quaker Oats, and R. D. Sloan have successfully coupled

their data gathering systems with network flow models to improve the cost

and service effectiveness of logistics decisions. For instance, Agrico’’ reported

a decrease in the net working capital requirements of 13% and a 5-year savings

of $43 million, while Kelly-Springfield'* reported savings of over $8 million

annually.

Management of Emergency Services with the Hypercube Queueing Model

A computer-based multiserver queueing model is now routinely used to

deploy emergency services personnel in New York City, San Diego, Sacra-

mento, Dallas, Portland, Caracas and Rotterdam.” Productivity improve-

ments are on the order of 10-15%.

Scheduling and Distribution using Lagrange Multipliers

Lagrange multipliers are used to relax complicating constraints in difficult

combinatorial optimization problems. This approach has grown from theory

to a proven tool in a number of large-scale applications. A model for scheduling

and distributing industrial gases at Air Products Chemicals, Inc.'° used this

technique to save 6-10% of operating costs, amounting to annual benefits of

about $2 million.

Trucking Operations by Mathematical Programming

Enormous progress has been made using large-scale mathematical pro-

gramming models to route raw materials, components, and finished goods

optimally among production plants and warehouses. One such technical

achievement is the use of approximation methods to analyze models with

nonconvex cost curves representing the economies of scale that typically arise

in trucking operations. General Motors used this method in more than 40

plants to achieve a 26% logistics cost savings, for an annual saving of $2.9

million.!

Improving the Water Supply by Simulation Modeling

Simulation models were used to describe the water distribution system of

the Netherlands.'* These models were part of a broad analysis focused on

building new facilities and changing operating rules to improve water supply,

64 SAUL I. GASS

as well as on the adjustment of prices and regulations to reduce demand. The

analysis is credited with saving hundreds of millions of dollars and reducing

agricultural damage by about $15 million per year.

Inventory Control at Blood Banks

An inventory control model for the inventories of blood at blood banks in

Long Island also schedules blood deliveries according to statistical estimates

of the needs of each bank and uses actual requirements to adjust deliveries.!°

It forecasts short-term shortages and surpluses to control movement of blood

from adjoining regions. As a result, reductions of blood wastage of 80% and

of delivery costs by 64% were realized.

Highway Maintenance by Markov Decision Processes

A large-scale Markov decision process model was used to develop optimal

maintenance policies for each mile of the 7,400 mile network of highways in

Arizona.” The model integrates management policy decision, budgetary pol-

icles, environmental factors, and engineering decisions. During the first year

of implementation the model saved $14 million, almost one-third of Arizona’s

highway preservation budget. The forecast for future savings is about $25

million.

Improving Utilization of Air Force Cargo Aircraft

The Deployable Mobility Execution System was developed for the U.S. Air

Force to improve aircraft utilization and responsiveness in airlift operations.7!

The model uses a modified cutting stock heuristic to generate feasible cargo

loads, which the planner can modify using detailed interactive graphics. In

three tests during military exercises, the system reduced load-planning man-

hours by 90% and increased aircraft utilization by 10%. During the Grenada

rescue operation, the system saved over $2.5 million in flying-hour costs and

provided timely planning. Projected annual savings for peacetime exercises

are estimated to exceed $20 million.

The Merging of Public Safety Organizations

The city of Grosse Pointe Park, Michigan, planned to totally merge its

police and fire departments, which also provided emergency medical services.”

Almost all personnel were trained as both policemen and firefighters, and

many as emergency medical technicians. The firefighters opposed the merger

and forced a referendum. In just 17 days, an analysis was done that found

that a merger would reduce operating costs by as much as $100,000 per year,

further reduce response times, and increase patrol coverage, and add an ad-

ditional detective.

THE CURRENT STATUS OF OPERATIONS RESEARCH 65

The Process of Operations Research and Model Evaluation”

In my earlier remarks, I stated that Operations Research is a respected

profession and that we are responsible for the effective and efficient solution

of a wide class of operational problems. The emphasis here is on the phrase

‘operational problems.’”’ This is what Operations Research was originally

designed to do—solve operational problems. During the late 1960’s, under

the assumption that technology can solve many of our social problems, there

was an attempt to apply systems engineering and Operations Research pro-

cedures to the full range of governmental policy-analysis problems: to prob-

lems in health, education, welfare, energy, criminal justice and public safety,

environment, transportation, and urban planning.**>”°’8 The promises of

such technology fixes caused a great deal of public funds to be allocated to a

wide range of projects. Although there were some successes, the objectives

of many of the projects were not achieved. The combined wisdom of those

who have studied Operations Research, as applied to policy-analysis problems,

yields the following list of difficulties in applying Operations Research in this

important area:

MODELING PROJECT CONCERNS

Insufficient problem definition

Inadequate user participation in problem definition

Inadequate agreement on model specifications

Insufficient user participation in model planning

The model was not responsive to the user’s needs

Poor user/developer coordination; resistance to change

Inadequate monitoring by the user agency

Insufficient data for current and future applications

Inadequate evaluation (testing and validation) procedures

Inadequate documentation

The vanishing advocate

It is not my purpose here to discuss the whys of these concerns, but for

those of you who are not familiar with the story, I strongly suggest that you

review the literature on these activities.*°7’8”? What I do want to discuss is

a major concept that evolved due to the concerns of such work—a concept

that I think is essential to the future success of Operations Research. It is the

idea of model evaluation, and the imbedding of the process of evaluation into

the mainstream of model formulation, development and implementation.

The modeling projects that I have been referring to were designed to support

policy decisions of governmental agencies. The theoretical bases of such models

were cloudy, at best. Concerns were raised as to the validity of the model

results, and to the whole question about the role of decision-aiding models in

government. Out of this grew the concept and process of model evaluation,

as applied to computer-based modeling systems. I would like next to discuss

what evaluation is all about, its role in the practice of Operations Research,

and why all Operations Research analysts and others should be interested

In it.

66 SAUL I. GASS

Over the past few years, model evaluation has been a well-studied activ-

ity 24903192334 but, Iam afraid, not a well-known idea. Briefly, model eval-

uation is a process by which interested parties, who were not involved in a

model’s origins, development, and implementation, can assess the model’s

results in terms of its structure and data to determine, with some level of

confidence, whether or not the results can be used to aid the decision maker.

In other words, model evaluation equates to an independent assessment of

the model and its use. Not all models need to be or should be evaluated. My

concern is with modeling projects that you and I would agree are complex

and involve critical decision components, for example, the Strategic Defense

Initiative, air traffic control, integrated manufacturing and logistics systems,

and governmental policy analysis. There are three basic reasons why I advocate

the evaluation of such complex models:

(1) For many models, the ultimate decision maker is far removed from the modeling

process and a basis for accepting the model’s results by such a decision maker

needs to be established,

(2) For complex models, it is difficult to assess and to comprehend fully the inter-

actions and impact of a model’s assumptions, data availability, and other elements

on the model structure and results without some formal, independent evaluation,

and

(3) Users of a complex model that was developed for others must be able to obtain

a clear statement as to the applicability of the model to the new user problem

area.

Although a discussion of the full details of an evaluation process is not

appropriate here, I would like to illustrate the essential features of one ap-

proach that involves the following five major evaluation criteria:**”?

EVALUATION CRITERIA

1. DOCUMENTATION

2. VALIDITY

Theoretical Validity

Data Validity

Operational Validity

3. COMPUTER MODEL VERIFICATION

4. MAINTAINABILITY

Updating

Review

5. USABILITY

Briefly, these criteria are concerned with the following aspects of the model-

ing process: (1) Documentation must describe what the developers have done,

and why, and how; documentation should be sufficient enough to permit

replication by independent evaluators of the model’s results and claims. (2)

There is no standard way to validate a decision model, but most investigators

will agree that any such evaluation encompasses the concerns of theory (does

the theory fit the problem environment?), data (are the data accurate, com-

THE CURRENT STATUS OF OPERATIONS RESEARCH 67

plete, impartial and appropriate?), and operations (how does the model di-

verge from the users’ perception of the real-world?). (3) Verification attempts

to ensure that the computer-based model runs as intended, that is, the com-

puter program accurately describes the model as designed. (I want to note

that my view of the modeling process emphasizes model verification and

validation as explicit steps of model development.) (4) The long term use of

many government models calls for a preplanned and regularly scheduled pro-

gram for reviewing the accuracy of the model over its life cycle, which includes

a process for updating and changing model parameters and model structure.

(5) The usability of a model rests on the availability of data, the ability of the

user to understand the model’s output, model portability, and costs and re-

sources required to run the model.

Evaluation is usually applied to a modeling project after the model has been

developed and, more often than not, after implementation. The view that

evaluation is done at the end of a project is how I first addressed the issue

and it is how it is usually practiced. The three reasons for evaluation cited

above are based on the premise of post evaluation. But, only reason (3)

constrains us to a post evaluation. My current view is that the emphasis on

performing evaluation activities at the end of a project is wrong. Post eval-

uation of a model is costly; difficult to accomplish, as many of the principal

analysts are no longer available; and, as the implemented model is ever

changing, the evaluation team usually ends up evaluating an outdated model.

Post evaluation is like closing the barn door after the horse has been stolen;

corrective actions should be ongoing and continual. The benefits from a post

evaluation are not proportional to the cost. The funds and resources are better

applied to what might be called “ongoing evaluation.”’ This means that Op-

erations Research analysts must carry out the project tasks under the as-

sumption that they are operating under a continuing and ongoing evaluation.

Working under that assumption would force them to perform verification and

validation activities consistent with the aims and resources of the project, and

to document the results. It would force them to insist on the correct level of

funds and personnel to do the work properly the first time. In essence, I am

saying that if the Operations Research analysis team conducted its work as if

it expected an independent evaluation to take place, the team would neces-

sarily do a more thorough and professional job. The need for a post evaluation

would disappear, as the evidence that the analysts furnished would state ex-

plicitly the assumptions, limitations and the proper uses of the model. An

evaluator would then only need be concerned with independent replication

and extensions of the model to other problem areas. The concept of developing

and implementing a model-based project under the framework of an ongoing

evaluation protocol is an important message that I feel must be adopted by

the current and next generation of Operations Research analysts, and by the

managers who are responsible for the funding of these projects.

From my experiences and from what I perceive to be the need for profes-

sional guidelines for all aspects of a modeling project, I believe that, along

with ongoing model evaluation, we have to impose a life-cycle view of model

SAUL I. GASS

development on a modeling project. My structure of a model’s life-cycle in-

cludes thirteen interrelated phases, as follows:

—Embryonic (Initiation)

—Feasibility

—Formulation

—Data

—Design

—Software development

—Validation

—Training and Education

—Installation

—Implementation

—Maintenance and Update

—Evaluation and Review

—Documentation and dissemination

The importance of the model life-cycle approach is that it forces the model

developers and users to face up to the needs (personnel, funds and time) of

each modeling activity. (See Gass* for details of the phases.)

Operations Research projects need to apply life-cycle management proce-

dures and ongoing evaluation techniques. The joining together of project

management and model evaluation protocols is what I believe is required to

nurture and sustain Operations Research. This systemic view of how we should

manage the Operations Research modeling process is what will bring Oper-

ations Research from its successful past and present into the future.

References Cited

. Lovell, B. 1988. Blackett in War and Peace. J. Op. Res. Soc., 39, 221-233.

. Moder, J. J. and S. E. Elmaghraby (eds.). 1978. Handbook of Operations Research, Vols. 1 & 2,

Van Nostrand Reinhold, NY.

. Thomas, G. and J.-A. DaCosta. 1979. A Sample Survey of Corporate Operations Research, Interfaces,

9, 102-111.

. Muller, W. and C. B. Tilanus. 1976. Linear Programming from a Management Point of View, a Survey,

Netherlands, Euro. Jrnl. of OR, 2, 223-231.

. Kivijarvi, H., P. Korhonen and J. Wallenius. 1986. Operations Research and its Practice in Finland,

Interfaces, 16, 53-59.

. Dam van, W. B. and C. B. Tilanus. 1983. Mathematical Programming in the Netherlands, Report

BDK/ORS/83/10, Eindhoven Univeristy of Technology, Eindhoven, Netherlands.

. Peterson, R., D. W. Conrath, and C. T. L. Janssen. (eds.). 1973. Management Science in Canada,

Management Science, 20, Part II.

. Bartholdi, 111, John J. 1986. Operations Research in China, Interfaces, 16, 24-30.

. Pollock, S. M. and K, Chen. 1986. Strive to Conquer the Black Stink: Decision Analysis in the People’s

Republic of China, Interfaces, 16, 31-37.

. Lingaraj, B. P., F. M. E. Raiszadeh and Y. Kathawala. 1986. Publications on Real-World Applications,

paper presented at the TIMS XXVII International Conference, Gold Coast, Australia.

. CONDOR Committee. 1988, Operations Research: The Next Decade, Ops. Res., 36, 619-637.

. Klingman, D., N. Phillips, D. Steiger and W. Young. 1987. The Successful Deployment of Management

Science Throughout Citgo Petroleum Corporation, Interfaces, 17, 4—25.

. Glover, F., G. Jones, D. Karney, D. Klingman and J. Mote. 1979. An Integrated Production, Dis-

tribution, and Inventory Planning System, Interfaces, 9, 21-35.

. King, R. H. and R. R. Love, Jr. 1980. Coordinating Decisions for Increased Profits, Interfaces, 10,

4-19.

15:

16.

THE CURRENT STATUS OF OPERATIONS RESEARCH 69

Larson, R. C. 1974. A Hypercube Queueing Model for Facility Location and Redistricting in Urban

Emergency Services, Comput. & Opns. Res. 1, 67-95.

Bell, W. J., L. M. Dalberto, M. L. Fisher, A. J. Greenfield, R. Jaikumar, P. Keida, R. G. Mack and

P. J. Prutzman. 1983. Improving the Distribution of Industrial Gases with an On-Line Computerized

Routing and Scheduling Optimizer, Interfaces, 13, 4—23.

. Blumenfeld, D. E., L. D. Burns, C. F. Daganzo, M. C. Frick and R. W. Hall. 1987. Reducing Logistics

Costs at General Motors, Interfaces, 17, 26-47.

. Goeller, B. F. and the PAWN Team, 1985, Planning Netherlands’ Water Resources, Interfaces, 15,

3-33.

. Brodheim, E. and G. P. Prastacos. 1979. The Long Island Blood Distribution System is a Prototype

for Regional Blood Management, Interfaces, 9, 3-20.

. Golabi, K., R. B. Kulkarni and G. B. Way. 1982. A Statewide Pavement Management System,

Interfaces, 15, 3-33.

. Cochard, D. D. and K. A. Yost. 1985. Improving Utilization of Air Force Cargo Airlift, Interfaces,

15, 53-68.

. Chelst, K. 1988, A Public Safety Merger in Grosse Pointe Park, Michigan—A Short and Sweet Study,

Interfaces, 18, 1-11.

. Gass, S. I. 1987. Managing the Modeling Process, European Jrnl of OR, Vol. 31, No. 1.

. Gass, S. I. 1983. Decision-Aiding Models: Validation, Assessment, and Related Issues, Ops. Res. 31,

603-631.

. Hoos, I. R. 1972. Systems Analysis in Public Policy, University of California Press, Berkeley, Calif.

. Greenberger, M., M. A. Crenson and B. L. Crissey. 1976. Models in the Policy Process, Russell Sage

Foundation, NY, NY.

. Kraemer, K. L. and J. L. King. 1986. Computer-Based Models for Policy Making: Uses and Impacts

in the U. S. Federal Government, Operations Research, 34, 501-512.

. Brewer, G. 1973. Politicians, Bureaucrats, and the Consultant, Basic Books, NY, NY.

. Brewer, G. and M. Shubik. 1979. The War Game, Harvard University Press, Cambridge, Mass.

. Gass, S. I. 1977. Evaluation of Complex Models, Comput. & Opns. Res. 4, 27-35.

. Gass, S.I. (ed.). 1979. Validation and Assessment Issues of Energy Models, Proceedings of a Workshop,

National Bureau of Standards Special Publication 564, Washington, DC.

. U.S. GAO. 1979. Guidelines for Model Evaluation, PAD-79-17, Washington, DC, January.

. Gass, S. I. and B. W. Thompson. 1980. Guidelines for Model Evaluation: An Abridged Version of

the U. S. General Accounting Office Exposure Draft, Operations Research, 28, 431-439.

. Wood, D. O. 1979. Energy Model Evaluation and Analysis: Current Practice, in Gass [31].

. Gass, S. I. 1979. Computer Model Documentation: A Review and an Approach, NBS Special Pub-

lication 500-39, National Bureau of Standards, Washington, DC.

Journal of the Washington Academy of Sciences,

Volume 79, Number 2, Pages 70-85, June 1989

Operations Research: Some

Trends And Issues For The Future

Carl M. Harris

Department of Operations Research & Applied Statistics

George Mason University

Fairfax, VA 22030

and

Richard H. F. Jackson

Center for Manufacturing Engineering

National Institute of Standards and Technology’

Gaithersburg, MD 20899

ABSTRACT

As computers and computing have come of age in the decade of the 1980s, operations

research has begun to flourish as never before. In a certain sense, it has become the

computational science, and its future is therefore intimately tied to the computer and what

computing will look like in the years to come. In this paper, we highlight the importance

of this relationship, and particularly some of the major issues that the operations research/

computer science interface brings forth. These include the emerging importance of com-

putational experimentation, especially in the use of OR software; the role of operations

research in the search for computer-aided improvements in manufacturing productivity;

and the change that computers have and are bringing about in the modeling processes

under uncertainty.

1 Introduction

Previous papers in this special issue of The Journal of the Washington Acad-

emy of Science discussed the past and the present status of operations research.

‘Formerly, National Bureau of Standards

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 71

In this paper, we present a view of the future of this field. It seems appropriate

to begin by summarizing the view presented by the members of the Committee

On the Next Decade in Operations Research (CONDOR) in their report

‘‘Operations Research: the Next Decade”’ (see Committee, 1988).

The authors of the CONDOR report, an impressive collection of 24 op-

erations research professionals from 15 U.S. universities, 1 foreign university,

and 3 U.S. companies, listed some significant operations research accomplish-

ments and sketched some important opportunities for the next decade. Their

perspective in producing this outlook on a research agenda was to “‘identify

opportunities for

@ building intellectual capital in the form of basic theoretical knowledge,

® stimulating the development of new subfields,

@ responding to the needs of the field’s many application contexts and to those of

society at large, and

@ improving practice.”

They highlighted five major areas of operations research: optimization,

stochastic processes, manufacturing and logistics, the operations research/

artificial intelligence interface, and operational and modeling science. Of

course, their list was not meant to be all inclusive, just as this report is not.

The CONDOR members felt however that over the next ten years the high-

lighted areas would be particularly fruitful. In discussing these topics, they

were less concerned with the degree to which theory and practice are mixed

in individual instances than with emphasizing that both are important and

neither can be overlooked. It is the feedback to the agendas of the theoreticians

from the never ending supply of application areas and problems requiring

solution that makes this field so attractive. They felt that in operations re-

search, both theory and practice commingle in such a way as to make for a

happy and successful mixture.

Although we certainly agree with the authors of the CONDOR report, our

perspective in writing this article is different from theirs. Ours is that of

practitioners and academicians from the Washington metropolitan area, and

is motivated by certain issues and well-developed, well-reported trends we

see in this country.

2 Operations Research and the Information Age

The first of these to be discussed is the trend alluded to in the title of

this special issue, the information explosion. This information explosion has

resulted in what has been called the information age, and is a direct result of

the increasing availability of cheaper and faster computing. The incredible

advances in processing times and storage mechanisms and sizes have encour-

aged the collection and dissemination of increasing amounts of data and in-

formation. In fact, the paradigm for scientific research has traditionally been

theory and experimentation. With the advent of cheaper, faster, and friendly

72 CARL M. HARRIS AND RICHARD H. F. JACKSON

computing, the new paradigm has become theory, computation, and experi-

mentation.

The techniques, ideas, methods, and models developed by operations re-

searchers over the years have so far been successful in keeping up with the

needs of this information age. This pace must be maintained, of course. In

addition, we must respond to new challenges, one of which is the increasing

emphasis on computation in our own field. For example, as computing be-

comes cheaper and cheaper, it is tempting to resort to a computer or a com-

puter solution to a difficult problem before attempting analytic approaches.

This must be resisted. Whenever an analytic approach is available, it should

be pursued first. Other challenges confronting operations researchers in this

increasingly computational world are in making computers work better with

each other and with humans.

2.1 The Interface with Computer Science

It was not until computers came of age that operations research began to

flourish, too. It has become the computational science, providing solutions to

almost all the information and data analysis problems arising with expansion

of the information age, e.g., data visualization, data animation, data reduction,

data analysis, data modeling, and even studying computers as a computational

device, and using them as a laboratory. Some of these issues were addressed

in Greenberg (1988). In his article, Greenberg identified the following 12

areas which he felt collectively cover (though neither exclusively nor exhaus-

tively) the interface:

design and analysis of algorithms,

heuristic search and learning,

parallel computation,

simulation,

computational probability and analysis,

information storage and retrieval,

decision support systems,

database theory, optimization, and integration,

cognitive modeling and analysis,

representability and computational logic,

fuzzy systems, and

telecommunications.

To this list we would add computational experimentation with operations

research software. Although some would argue it is included in the first cat-

egory above, analysis of algorithms, we feel it is important enough to be

singled out and discussed further.

2.2 Computational Experimentation

Controversy often surrounds the reporting of results of computational ex-

perimentation with operations research software. Subtle points with unrec-

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 73

ognized but profound effects are sometimes overlooked, and testing proce-

dures are sometimes inadequate. Papers often appear with overlapping and

conflicting claims of computational performance of new algorithms being pro-

posed. Sometimes papers appear with unsubstantiated claims of performance.

Worse, papers occasionally appear containing performance claims without

sufficient information about the experiment or the software being tested for

others to verify the claims. It is increasingly the case that not just pride of

authorship is at stake, but also commercial interests, as the state-of-the-art of

operations research algorithms and software has improved along with their

use in business, industry, and government.

Some progress has occurred. Organizations like the Committee on Algo-

rithms of the Mathematical Programming Society have been formed to serve

as a focal point for information about software and algorithms, to develop

suitable methods for comparing algorithms and software, and to encourage

the use of standards and guidelines for developing, documenting, and testing

new algorithms and software. This committee and its publications have been

instrumental in improving the quality of computational experimentation. It is

rare nowadays (though not unheard of) to see statements like the following

that appeared in a published paper in 1968:

“Since the methods were coded for different machines in different languages

by different programmers, there is little point in giving a detailed assessment

of the results, particularly as so many of the problems were degenerate.

However the results show that... .”

One reason such statements have become rare is that many algorithm de-

velopers understand that computational testing of new algorithmic ideas is an

integral part of the development process and that this testing is as much a

part of the process of experimentation as it is in other branches of science.

As such, these experiments must be conducted with rigor, following the dic-

tates of the scientific method. The development of this view of computational

experimentation was aided in part by the publication (Crowder, et al., 1979)

of guidelines for reporting results of computational experiments. Since their

publication, some journals of the operations research community have adopted

them as part of their editorial policy. In many cases, editors, associate editors,

and referees have privately adopted them to aid in the publication process.

Recently, the Mathematical Programming Society has undertaken an effort

to update and revise these guidelines (Jackson, et al., 1989). This new report

will expand on some of the earlier work by Crowder, et al., and will also

become part of the editorial policy of some of the journals of the operations

research community.

The important point to be made here is that research, development, and

testing with operations research software are all important, and all must pro-

ceed according to the scientific method. In other words, just as we require

mathematical analysis to be thorough, we must also require computational

experiments to be thorough. Further, published results of both theoretical

and computational analyses must provide sufficient information for others to

check or substantiate the reproducibility of the results. But what is a thorough

74 CARL M. HARRIS AND RICHARD H. F. JACKSON

computational experiment? When must a computational experiment be per-

formed, and how thorough must it be? Also, what about protecting com-

mercial, proprietary interests?

It is not the purpose of this paper to explain in detail how properly to

conduct a thorough computational experiment with operations research soft-

ware. Much has been written elsewhere on this topic. See, for example, Green-

berg (1979), Hoffman et al. (1987), Mulvey (1982), and Schittkowski (1985).

Suffice it to say here that one must develop a sound experimental design,

including identification of objectives, influential variables, performance mea-

sures, and test problems. Understanding how to do this in all cases is an

important area for further research. This is an exciting area which was also

identified by the authors of the CONDOR report.

When to perform a thorough computational experiment depends on the

performance claims one wishes to make. The guiding principle is that the

computational results presented must be sufficient to justify the performance

claims made. There are basically three types of computational studies:

@ preliminary testing to demonstrate feasibility and promise of a new idea,

@ more detailed experimentation to assess the strengths and weaknesses of an

implementation, and

@ detailed comparison of the performance of an implementation with prominent

methods already available.

The amount of computational testing required varies from minimal in the first

case to thorough in the last. However, as mentioned above, we have to learn

more about how to perform these thorough experiments effectively. For ex-

ample, there is yet to be developed a common understanding of the sources

of randomness inherent in testing a piece of operations research software.

Proprietary software presents special problems. Requiring public repro-

ducibility sometimes is perceived as running counter to corporate commercial

interests. Full reproducibility of experimental results requires a listing of the

program and a detailed description of the inputs and the test problems used.

This obviously presents a problem when either one or both are proprietary.

Crowder, et al. (1979) attempted to resolve this conflict by requiring that ‘“‘an

absolute, reasonable, and scientifically justifiable criterion should be that the

authors themselves be able to replicate their experiment.” This of course is

sound but, as it turned out, did not go far enough. While it addressed the

need for replication as part of the scientific method, it left unexplored one of

the main reasons for publishing: to inform others so that they may replicate,

verify, modify, and build. Thus the scientific community was dissatisfied. Full

disclosure is the process by which scientific progress is achieved. In commercial

enterprises, this objective is violated intentionally: the guiding principle is

rather to withhold as much information as possible.

It may be some time before this issue is resolved satisfactorily for all con-

cerned. In their forthcoming paper, Jackson, et al., (1989) propose a solution

based on a clear separation of commercial and scientific presentations. Even

this, if accepted, will likely suffice only until some new idea emerges in the

debate. For example, the recent success on the part of AT&T to patent some

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 75

of the mathematical ideas developed by their scientists adds a completely new

dimension to the debate, one whose ramifications will be felt for many years

to come.

2.3 Strengthening the Interface

It should be clear from this discussion and from Greenberg’s list that the

interface between operations research and computer science is an active and

fruitful area of research and that much can be gained by continuing this

collaboration. On the other hand, while this has been an active area, and

while operations research methods and analysis techniques have developed

pari passu our ability to compute and collect data, it has not always been the

case that the most appropriate amount or type of analysis has been brought

to bear during the development of data bases and data-driven models and/or

systems. Operations research is the field aimed primarily at providing decision

makers with aids in understanding, analyzing, simplifying, visualizing, and

forecasting data. Techniques, algorithms, and models developed by operations

researchers are techniques developed with this goal in mind.

If these techniques, algorithms, and modeling methods have not been em-

ployed in as many areas as they should, it is not, we believe, because of a

lack of methods or a desire to employ these methods, but rather because we

need more ‘‘missionaries” (Blumstein, 1988). In his article, Blumstein lists

three principal approaches for maintaining the strength of our field:

@ enhancing the power of our methods so that they are all the more powerful and

apply to broader classes of problems—the task of our ‘algorithmists’ and meth-

odologists,

@ reaching out to developing methodological fields . . . and bringing them within

our community and our aggregate toolkit—the task of [our profession’s] leaders,

and

@ identifying problem areas whose structure and organizational readiness make

them appropriate as the next beneficiaries of our approaches and methods—the

task of our ‘missionaries.’

Blumstein sees the public sector as a primary target for the next generation

of operations research/management science missionary work. We agree, but

also see important targets of opportunity in the private sector. An increasingly

important one is manufacturing, specifically automated flexible manufactur-

ing, the next issue to be discussed here.

3 Operations Research and Manufacturing

A growing national debate has focused on the decline of U.S. industry’s

competitiveness and resultant loss of marketshare in the global marketplace.

The debate has identified many possible culprits responsible for this decline,

ranging from budget deficits to short-term, bottom-line thinking on the part

of U.S. management. As America’s industry faces greater and greater eco-

76 CARL M. HARRIS AND RICHARD H. F. JACKSON

nomic challenges, increasing emphasis will be placed, not just on technological

solutions to these problems, but on our effectiveness in managing these new

technologies. While operations research and management science methods

may be helpful to those addressing these managerial and economic issues,

these methods surely can help improve productivity in the manufacturing

sector. These techniques can be helpful in the classical ways, e.g., scheduling

routing, layout, and process modeling. In addition, they are critical to attempts

to implement the new advanced manufacturing technologies and make them

operate efficiently. Since advanced manufacturing is such a topical area, and

since it is tied so closely to the economic health of the nation, we pursue it

in slightly more depth below.

3.1 Flexible Automated Manufacturing

Manufacturing plants typically consist of combinations of people and ma-

chines, working together to maximize corporate profits from the goods they

produce. Many of these plants are plagued by large work-in-process inven-

tories, low utilization of equipment, insufficient throughput, and excessive

delays, all of which decrease profits. Hopes for alleviating these problems

were raised when computer-controlled robots, machine tools, and transporters

became commercially available and sophisticated computer software tech-

nology was developed to support it. Many companies made large capital in-

vestments in these new Computer Integrated Manufacturing Systems (CIM),

expecting that their integration into existing plants would increase profits and

world-market shares.

Not only has this not happened, and the problems not been resolved, but

introducing CIM into existing factories has had a negative impact in some

cases. There are three major reasons for this surprising phenomenon. First,

integrating equipment from different vendors was far more difficult than orig-

inally anticipated. Second, the continued use of existing planning and sched-

uling strategies often exacerbated the problems mentioned above. Finally,

existing data management and communication strategies are inadequate to

handle the increased dependency on “data” in these CIM environments. Clearly,

then, we must: |

@ design better techniques for solving these old problems,

@ identify and solve the new problems, and

@ develop an integrated framework for solving all of them.

3.2 The Automated Manufacturing Research Facility

One approach to this was developed in the Center for Manufacturing En-

gineering at the National Institute of Standards and Technology (formerly the

National Bureau of Standards). NIST is fundamentally committed to pro-

moting the development of standards for automated manufacturing systems

and to transferring technology to American industry. To meet this responsi-

bility, the Center for Manufacturing Engineering established an experimental

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 77

test bed, the Automated Manufacturing Research Facility (Simpson, et al.,

1982).

Physically, the AMRF contains several robot-tended machining worksta-

tions, a cleaning and deburring station, an inspection station, a material han-

dling system, factory control software, database management systems, and

the communications support to integrate it. This equipment includes donations

and purchases from four different robot manufacturers, three machine tool

vendors, and every major computer company. Basic principles from physics,

computer science, behavioral sciences, control theory, operations research,

and engineering disciplines have been used to transform these individual com-

ponents into a fully integrated, flexible, small batch manufacturing system.

When originally conceived in the late 1970s, the AMRF was unique. Since

that time, and since having solved some of the hardware and communications

problems in the AMRF, more automated facilities have begun to appear.

During this time, most of the resources expended on automated manufacturing

have concentrated on demonstrating feasibility of the idea of fully flexible,

integrated, automated manufacturing facilities. It is now time to turn our

attention to the difficult tasks confronting us in the areas of software engi-

neering, data handling, data analysis, process modeling, and optimization.

That is, having demonstrated that the idea works, we must now make it work

smoothly and efficiently. We believe operations research 1s the discipline that

provides the tools to achieve this end. In order to understand this, it is im-

portant to understand the hierarchical structure of modern automated man-

ufacturing control systems.

To achieve maximum flexibility and modularity, the AMRF control system

was:

@ partitioned into a five-level functional hierarchy in which the control processes

are completely data driven and communicate via NIST-developed hardware and

software interfaces;

@ designed to respond in real-time to performance data obtained from machines

equipped with sophisticated sensors; and

® implemented in a distributed computer environment using state-of-the-art tech-

niques in software engineering and artificial intelligence.

The hierarchical command/feedback control structure ensures that the size,

functionality, and complexity of individual control modules is limited. Al-

though the flow of control in this hierarchy is strictly vertical and between

adjacent neighbors, it is necessary and even desirable to share certain classes

of data across one or more levels. In addition, each control level is completely

data-driven. That is, the data required to perform its functions are separated

from the actual control code. All data are managed by a distributed data

administration system (Barkmeyer, 1986) and transmitted to and from control

processes via a communication network.

Several hierarchical models for controlling shop floor activities have been

proposed. The models typically decompose manufacturing activities into five

hierarchical levels: facility, shop, cell, workstation, and equipment. Activity

at each of these levels is data-driven, and each can be expanded to fill out

78 CARL M. HARRIS AND RICHARD H. F. JACKSON

the tree. This structure provides a convenient mechanism for describing the

functions of an automated facility and the databases needed to meet manu-

facturing requirements.

3.3 Optimization and Flexible Automated Manufacturing

An important question to be addressed now in flexible automated manu-

facturing is whether it is possible to mimic this hierarchical approach to control

and decompose the planning process along the same lines for the facility as a

whole. Traditionally, planning and scheduling problems have been analyzed

without decomposing them into subproblems. See, for example, Baker (1974),

Conway, et al. (1967), and Elmaghraby (1973). The operations research lit-

erature contains many articles proposing mathematical programming, simu-

lation, heuristic, and other approaches for solving these problems under special

conditions. Due to their computational requirements and restrictive assump-

tions, these approaches tend to have limited applicability in a real manufac-

turing environment. They are typically unable to provide useful solutions in

a timely manner. It is possible that a hierarchical approach will provide the

mechanism for solving these problems in the timely and efficient manner

required. Some progress has been made toward this goal.

There are several advantages to a hierarchical approach to process planning.

As with any hierarchical representation of data, it is easy to use, understand

and exploit. Another advantage is the way in which the problem domain is

automatically partitioned into regions associated with classes of machinable

processes. This last, of course, provides much improvement in the speed of

search procedures since the search need only be performed within one of the

partitions. The challenge to operations researchers in this area, then, is to

develop the tools that can be used to schedule, plan, and control systems in

this hierarchical, real-time environment. Some work has been done in this

area. See, for example, Jackson and Jones (1986), Jackson and Jones (1987),

Davis and Jones (1988), and Davis, et al. (1989).

Artificial intelligence techniques have been used to address some of the

problems in automated flexible manufacturing in a way that mimics practice.

Planners and decision makers generally decompose difficult problems into

subproblems for experts to solve. The decision maker must then combine

these individual solutions to obtain the solution to the original problem. Ar-

tificial intelligence researchers have attacked problems this way but with little

success so far. See Lawrence and Morton (1986), Fox (1983), Steffen and

Green (1986), Wysk and Yang (1986), Chiodini (1986), Maley, et al. (1986).

Others have worked to combine principles of hierarchical control with artificial

intelligence techniques, but again results have been limited. For more on this,

see Gershwin (1986), O’Grady and Menon (1986), Akella, et al. (1984), Bitran

and Hax (1977), Bitran, et al. (1981), Bitran, et al. (1982), Davis (1984), Nau

(1986), and Shaw (1986). An exciting opportunity lies in combining the an-

alytical techniques and modeling approaches of operations research with the

rule-based approaches of artificial intelligence to produce a result that is more

than just feasible, but is optimal with respect to some objective. Too often,

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 79

rule-based systems are proposed as solutions to complicated problems before

other analytic approaches have been considered.

We believe that in order to be successful in solving the kinds of problems

confronting this nation’s manufacturing industry, an interdisciplinary approach

to problem-solving is needed. This interdisciplinary approach must concentrate

on obtaining the most efficient, cost-effective solutions to problems in ad-

vanced manufacturing as is possible. Too often, system designers are content

to settle for feasible solutions to a problem with little or no effort expended

to find optimal or even improved solutions. This cannot continue if the U.S.

manufacturing industry is to survive in the global marketplace.

Operations research holds one key to success in these areas. An exciting

possibility for operations researchers exists in capitalizing on the existence of

the Automated Manufacturing Research Facility at the National Institute of

Standards and Technology to establish one of the Operations Research lab-

oratories discussed in the CONDOR report. The members of CONDOR noted

that the quality of operations research practice depends ultimately on how

well the supporting theory and its implementing model match the key features

of the real-world operation in question. They recommended the establishment

of one or more laboratories dedicated to facilitating empirical work in oper-

ations research. Such a laboratory would emphasize controlled experimen-

tation, an activity typically too interruptive for a working industrial plant.

These are the same reasons the AMRF was established in 1981: to have a

controlled laboratory for experimentation in advanced metrology, interface

and communications protocols and standards, and real-time control systems

to support advanced manufacturing in this country. Now that it has reached

a level of completion as a fully flexible, integrated automated manufacturing

research facility, it can and should be used to advance our knowledge in areas

that can improve on the feasible solutions sought and obtained during its

development.

4 Computational Stochastic OR

Back on the methodological side, computers have truly changed the very

nature of stochastic operations research modeling. Heretofore, a lack of avail-

able solution procedures has often prevented the application of key theoretical

results from the probability literature. Particularly in the study of large-scale,

random systems, analyses aimed at obtaining exact answers have been typically

ineffective, since the evaluation of real systems depends upon easy access to

numbers.

Various numerical procedures and approximation techniques have, in fact,

been tried (with limited success) over the years for deriving information on

the major performance measures of stochastic systems. But now the ready

availability of inexpensive and friendly computing makes it easier to turn

theory into real problem solutions. Probably the most pronounced recent

trend in this field is the application of advances in numerical procedures and

80 CARL M. HARRIS AND RICHARD H. F. JACKSON

their computer implementation to support the effective application of prob-

ability methods in the modeling of dynamic, large-scale systems. And, we are

indeed going to see an increasing impact of computing on stochastic operations

research in the forthcoming years.

One of the biggest challenges today is the application of probabilistic and

statistical methods to the modeling of complex computer/communication net-

works. Not only has the technological revolution changed our ways of solving

problems, but it has also offered a totally new and critical area of application.

Of further special concern are two classes of applications more associated with

industry. These are stochastic network analyses of manufacturing problems

(commonly associated with flexible manufacturing) and the operational model-

ing of large maintained systems of complex engineering equipments. Taken

together, these three modeling environments require the wide-spread use of

stochastic processes of a fairly general nature.

4.1 Queueing and Probability Distributions

It turns out that queueing models are particularly common in these problem

areas. This is, of course, not at all surprising given the historical role of

queueing in telephony and its later application to production line and logistics

activities. For example, it has been long known that the modeling of preventive

maintenance and its interaction with inventory stocks and generalized logistics

flows leads to very complex numerical problems. This becomes even truer

when the systems are allowed to be large and possibly time varying. Further

complications include nonstandard provision of service, non-Markovian prob-

ability structures, and large state-space problems, as examples.

But it is in the area of queueing where we have probably seen the most

progress to date on the use of numerical methods. We make special note here

of the recent work of Neuts (major portions of which are documented in his

1981 book and the referenced 1975 paper). He has coined the phrase “‘com-

putational probability” to describe these directions of research, defined as the

study of stochastic models with special attention to algorithmic feasibility over

a realistic range of parameter values. In his own research, Professor Neuts is

able to avoid transform solutions and complex variable methods for prime

classes of problems by the creative use of iterative (computer-based) matrix

methods. But it should be clear that such methods are still very heavy on

computations. The obstacles to reasonable solutions using Neuts-type argu-

ments are often formidable, though the effective use of contemporary com-

puter technology has made them quite attractive. However, the very nature

of the underlying complexity of iterative matrix models will often diminish

their usefulness, particularly since it may be impossible to estimate population

parameters in any practical way. A major direction of recent research, there-

fore, has been to find related computationally feasible methods for which

statistical inference procedures are easier to develop. (See Albin and Harris,

1987, for a more complete picture of recent research directions. )

But, clearly, the numerical analysis of queueing systems carried out by Neuts

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 81

using phase-type distributions is a major computational advance. The work

of Harris and others on generalized hyperexponential (GH) distributions pro-

vides a numerically and statistically attractive relative of the phase types (see

Harris and Sykes, 1985 and Botta and Harris, 1986), and their use in single-

server queueing systems is well documented in Harris (1985).

The GH distributions are linear combinations of exponential distribution

functions with mixing parameters (positive and negative) that sum to unity.

The denseness of the class GH with respect to the class of all CDFs defined

on the positive real line has been established in Botta and Harris (1986) by

showing that a GH distribution can be found that is as close as desired, with

respect to a suitably defined “‘distance”’ measurement, to any given CDF. A

nonlinear optimization routine has been developed for the likelihood esti-

mation of the GH parameters and is described in Harris and Sykes (1985).

Earlier papers by Kaylan and Harris (1981) and Mandelbaum and Harris

(1982) provide a history of the evolution of the algorithm and its original

application to mixtures of exponential and Weibull densities. But some critical

problems have not been solved in the use of GH distributions.

Possibly the major unresolved problem in the use of phase-type and gen-

eralized hyperexponential distributions is determining the number of com-

ponent terms. In most mixture work, it is implicitly assumed that M, the

number of components in the mixture, is known before the estimation of

parameters is attempted. In many cases this will be so, since the decision to

fit a mixture distribution will be based upon theoretical knowledge of the

application at hand, for example, the known existence of two or more species,

the presence of two sexes, etc. However, there are, situations where the

decision to apply a mixture distribution must be based upon, or at least sup-

ported by, the sample data, and so questions arise as to the appropriate value

of M, and in particular whether M = 1, in which case a mixture would be

unnecessary. Several proposed methods may be helpful in such situations,

particularly where mixtures of normal distributions are being considered. These

techniques fall fairly naturally into two classes: the first contains informal

graphical techniques, such as examination of sample histograms, etc., while

the second includes the more formal hypothesis testing variety of technique.

(See Everitt and Hand, 1981.)

However, we stress that graphical techniques (for regular mixtures) are

informal, even more so when the mixtures are generalized. There was a slight

attempt at graphing in Harris and Sykes (1985), but statistical precision was

certainly lacking. More formal methods have indeed been proposed for testing

hypotheses on the number of components in the mixture.

To reinforce the point regarding the difficulty of determining an appropriate

number of terms (say M) for a GH distribution fit to data, we offer a heuristic

approach due to Kumaresan et al. (1984). The very simplicity of their method

is further evidence about the lack of satisfactory procedures in the literature.

They say that: if M is not known a priori, an estimate of M can be found as

follows. Choose M = 1, and find the subset of size unity that best fits the

data. Call the corresponding minimum error £;. Then, choose M = 2 and

find the best subset of size two and the corresponding minimum error £).

82 CARL M. HARRIS AND RICHARD H. F. JACKSON

Repeat the procedure until the rate of decrease of the error with increasing

values of M is small, consistent with the modeling of broadband noise. The

integer i at which E£; shows the significant drop in rate of decrease is taken

as M.

There is another important side to the use of numerical methods in prob-

ability modeling, namely, the incorporation of the properties of the stochastic

variables directly into the department of the computational technique. A good

example is in rootfinding, particularly as applied to Markov processes and

queues, where there has been frequent controversy over the years regarding

the use of numerical procedures. However, rootfinding in queueing is so well

structured that problems do not occur. Fundamental properties possessed by

queueing models eliminate classical rootfinding problems. Most importantly,

we now know that uniqueness of roots is common within simply determined

regions in the complex plane, and it is thus possible to design extremely

efficient algorithms for locating the roots. (See Albin and Harris, 1987.)

4.2 Simulation and Stochastic Networks

When exact mathematical solutions to problems of realistic size prove in-

tractable, use is often made of Monte Carlo simulation, leading to increased

emphasis on empirical studies, data collection, and computational schemes.

Simulation is one of the most widely used methods of operations research.

The past decade has seen significant improvement in the quality of simulation

software, as well as in the analysis of the computations required to obtain a

given Statistical precision to simulation output. During the coming decade the

importance of simulation will grow still further as the need to analyze complex

stochastic systems increases.

Although simulation is widely used, a continuing limitation has been an

inability to approach system design without requiring an inordinate amount

of simulation computation. Several recent developments suggest that the next

decade will see major advances in simulation optimization. In particular, ef-

ficient algorithms for derivative estimation, based on perturbation analysis

and likelihood ratio methods, as well as recent advances in stochastic ap-

proximation algorithms (which can be used for simulation optimization) hold

great promise of offering significant improvement in the optimization of dis-

crete-event dynamic systems. Improvements in efficiency may also be realized

as a consequence of research in variance reduction, especially in the application

of large deviation results (a probability theory for rare events) to importance

sampling. Improvements in output analysis (particularly involving correlated

data), development of robust termination criteria for simulation experiments,

and enhanced graphic display of data will also be important.

Probability researchers have been widely criticized for studying systems that

are not relevant to the real world. However, it seems that this criticism 1s

largely due to the complexity of available results in the literature rather than

due to the system themselves. If one looks at some of the applications of

queueing, for example, one finds more complex systems than those in the

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 83

general operations research and applied probability literature. The distinction

is in the nature of analysis. The results found in the applied area literature

are applicable, though approximate. A large percentage of results found in

the more mathematical literature are less useful, though rigorous. Therefore

if we want the major methods of applied probability to be important problem-

solving tools, then numerical techniques should be put to increasing use when-

ever necessary and possible.

Both simulation and mathematical modeling of stochastic networks are of

particular concern to researchers today. As an example, we note recent re-

search on the development of feasible computational procedures for modeling

systems of single-server queues which obtain service in a rotating or cyclical

fashion, as might be true in some computer/communication systems with

polling or token-passing. Successful work to date has produced an algorithm

for modeling Poisson input, single-server queues where original service times

are extended to include a system-state-dependent, totally general vacation (or

inavailability time) after each service. (See Harris and Marchal, 1988.) The

next step in the sequence would be the modeling of cyclic-service queues and

their application to realistic computer networks.

To explain the relationship between the vacation model and the cyclic-

service problem and to give our thoughts on further work, we offer the fol-

lowing. Consider a group of n queues all served by a single central server,

modeled as an n-dimensional Markov chain. We need to derive a “service

policy” to specify which queue the server next attends. If that policy depends

only on the current state of the system (even if non-zero time is required to

move from queue i to queue }), the Markovian nature of the chain is preserved.

When the server is presented with a choice among the stations because there

are customers present at more than one, it next attends one determined by a

decision rule which depends only on the current state of the system. For

example, it could remain at the current station until no one remains (exhaustive

service) or it could go to the next occupied station (round robin) or it could

randomly select from among the occupied stations. In any event, once a

decision rule or service policy is selected, a well-defined set of transition

probabilities can be found for the Markov chain.

5 Concluding Remarks

We end this paper with an anecdote. When one of the authors of this paper

was an undergraduate, he worked in the university computing center as a

programmer and computer operator to pay his bills. During this time he was,

like most undergraduates, agonizing over career choices. At the time, a wise

and experienced woman was head of the computing center. She counseled

that, while computers and computing can be fun and fascinating in their own

right, the real excitement was to be found in disciplines which could exploit

these new tools fully. Thus he was guided toward operations research and has

never regretted the decision.

84 CARL M. HARRIS AND RICHARD H. F. JACKSON

References Cited

Akella, R., Choomg, Y. and Gershwin, S., “Performance of Hierarchical Scheduling Policy”, IEEE

Transactions on Components, Hybrid and Manufacturing Technology, CHMT-7, pp. 225-240, 1984.

Albin, S. L. and Harris, C. M., Statistical and Computational Issues in Probability Modeling, special issue

of Annals of Operations Research, vols. 8 and 9, 1987.

Baker, K. R., /ntroduction to Sequencing and Scheduling, John Wiley, New York, 1974.

Barkmeyer, E., Mitchell, M., Mikkilineni, K. P., Su, S. Y. and Lam, H., ““An Architecture for Distributed

Data Management in Computer Integrated Manufacturing”, NBS Technical Report NBSIR 86-3312,

National Bureau of Standards, Gaithersburg, MD, January 1986.

Bitran, G. R., Haas, E. A. and Hax, A. E., “‘Hierarchical Production Planning: A Single Stage System”’,

Operations Research, vol. 29, pp. 717-743, 1982.

Bitran, G. R., Haas, E. A. and Hax, A. E., “Hierarchical Production Planning: A Two Stage System’’,

Operations Research, vol. 29, pp. 232-251, 1982.

Bitran, G. R., Hax, A.C., ““On the Design of Hierarchical Production Planning Systems”, Decision Sciences,

vol. 8, pp. 28-54, 1977.

Blumstein, A., “Missionaries, Algorithms, and Leadership,” OR/MS Today, vol. 15, no. 5, p. 4, October

1988.

Botta, R. F. and Harris, C. M., “Approximation with Generalized Hyperexponential Distributions: Weak

Convergence Results,” Queueing Systems: Theory and Applications, vol. 2, pp. 169-190, 1986.

Chiodini, V., “A Knowledge Based System for Dynamic Manufacturing Replanning,” Proceedings of the

Symposium on Real-Time Optimization in Automated Manufacturing Facilities, NBS Special Publication

724, National Bureau of Standards, Gaithersburg, MD, January 21-22, 1986.

Committee On the Next Decade in Operations Research, ‘‘Operations Research: the Next Decade,”

Operations Research, vol. 36, No. 4, pp. 619-637, 1988.

Conway, R. W., Maxwell, E. A. and Miller, K.W., Theory of Scheduling, Addison-Wesley, Reading,

Massachusetts, 1967.

Crowder, H. P., Dembo, R. S. and Mulvey, J. M., “On Reporting Computational Experiments with

Mathematical Software,” ACM TOMS, vol. 5, No. 2, pp. 193-203, 1979.

Davis, W. J., “Decision Making and Control Hierarchies for Production Systems,’ Technical Report No.

145, PLAIC, Purdue University, West Lafayette, IN, 1984.

Davis, W. J. and Jones, A. T., ““A Real-time Production Sheduler for a Stochastic Manufacturing Envi-

ronment,” /nternational Journal of Computer Integrated Manufacturing, vol. 1, No. 2, pp. 101-112, 1988.

Davis, W. J., Jackson, R. H. F. and Jones, A. W., “Real-Time Optimization in the AUtomated Manu-

facturing Research Facility,” in Current Research in the Movement, Storage, and Control of Material—

Volume 1, Springer-Verlag, Berlin, 1989.

Elmaghraby, S. E., (ed.), Symposium on the Theory of Scheduling, Springer-Verlag, Berlin, 1973.

Everitt, B. S. and Hand, D. J., Finite Mixture Distributions, Chapman and Hall, London, 1981.

Fox, M., Constraint-Directed Search: a case study in job shop scheduling, Doctoral Dissertation, Carnegie-

Mellon University, 1983.

Gershwin, S., ““An Approach to Hierarchical Production Planning and Scheduling’’, Proceedings of the

Symposium on Real-Time Optimization in Automated Manufacturing Facilities, NBS Special Publication

724, National Bureau of Standards, Gaithersburg, MD, January 21-22, 1986.

Greenberg, H. J., ed., Design and Implementation of Optimization Software, NATO Advanced Study

Institute Series, Sijthoff & Noordhoff, the Netherlands, 1978.

Greenberg, H. J., ‘Interfaces Between Operations Research and Computer Science, OR/MS Today, vol.

15, No. 5, October, 18-21, 1988.

Harris, C. M., “Mixed Exponential Approximations for GI/G/1 Waiting Times,” Computers and Oper-

ations Research, vol. 12, pp. 285-289, 1985.

Harris, C. M. and Marchal, W. G., ‘State Dependence in M/G/1 Server-Vacation Models,’ Operations

Research, vol. 36, No. 4, pp. 560-565, 1988.

Harris, C. M. and Sykes, E. A., Likelihood Estimation for Generalized Mixed Exponential Distributions,”

Naval Research Logistics, vol. 34, no. 2, pp. 251-279, 1987.

Hoffman, K. L., Jackson, R. H. F. and Telgen, J., eds., Computational Mathematical Programming,

Mathematical Programming Study 31, North Holland, Amsterdam, 1987.

Jackson, R. H. F., Boggs, P. T., Nash, S. G. and Powell, S., ““Report of the Ad Hoc Committee to Revise

the Guidelines for Reporting Computational Experiments in Mathematical Programming,” to appear in

1989.

Jackson, R. H. F. and Jones, A., (eds.), Proceedings of the Symposium on Real-Time Optimization in

Automated Manufacturing Facilities, NBS Special Publication 724, National Bureau of Standards, Gaith-

ersburg, MD, January 21-22, 1986.

Jackson, R. H. F. and Jones, A. W., An architecture for decision-making for the factory of the future,”

Interfaces: Special Issue on Manufacturing, vol. 17, No. 6, pp. 15-28, 1987.

OPERATIONS RESEARCH: SOME TRENDS AND ISSUES 85

Kaylan, A. R. and Harris, C. M., “Efficient Algorithms to Derive Maximum Likelihood Estimates for

Finite Exponential and Weibull Mixtures,’ Computers and Operations Research, vol. 8, pp. 97-104,

1981.

Kumaresan, R., Tufts, D. W. and Schert, L. L., ““A Prony Method for Noisy Data: Choosing the Signal

Components and Selecting the Order in Exponential Signal Models,” Proc. IEEE, vol. 72, pp. 230-

233, 1984.

Lawrence, S. R. and Morton, T. R., “PATRIARCH; Hierarchical Production Scheduling,’ Proceedings

of the Symposium on Real-Time Optimization in Automated Manufacturing Facilities, NBS Special Pub-

lication 724, National Bureau of Standards, Gaithersburg, MD, January 21-22, 1986.

Maley, J., Ruiz-Mier, S. and Solberg, J., “Dynamic Optimization in Automated Manufacturing—A KnowIl-

edge Integrated Approach,” Proceedings of the Symposium on Real-Time Optimization in Automated

Manufacturing Facilities, NBS Special Publication 724, National Bureau of Standards, Gaithersburg,

MD, January 21-22, 1986.

Mandelbaum, J. and Harris, C. M., ‘““Parameter Estimation under Progressive Censoring Conditions for

a Finite Mixture of Weibull Distributions,” T/MS Studies in the Management Sciences, vol. 19, 239-260,

19872.

Marchal, W. G. and Harris, C. M., ‘““A Modified Erlang Approach to Approximating GI/G/1 Queues,”’

Journal of Applied Probability, vol. 13, pp. 118-126, 1976.

Mulvey, J. M., ed., “Evaluating Mathematical Programming Techniques,” Lecture Notes in Economics

and Mathematical Systems, No. 199, Springer-Verlag, New York, 1982.

Nau, D. S., “Hierarchical Abstraction of Problem-Solving Knowledge,” Proceedings of 1986 ASME Winter

National Meeting, Anaheim, CA, December, 1986.

Neuts, M. F., Matrix-Geometric Solutions in Stochastic Models, Johns Hopkins, Baltimore, 1981.

Neuts, M. F., “Probability Distributions of Phase Type,” in Liber Amicorum Professor Emeritus Dr. H.

Florin, University of Louvain, Belgium, pp. 173-206, 1975.

O’Grady, P. and Menon, U., ‘“‘A Hierarchy of Intelligent Scheduling and Control for Automated Manu-

facturing Systems”, Proceedings of the Symposium on Real-Time Optimization in Automated Manufac-

turing Facilities, NBS Special Publication 724, National Bureau of Standards, Gaithersburg, MD, January

21-22, 1986.

Schittkowski, K. S., ed., Computational Mathematical Programming, Proc. of Nato Advanced Study Institute

on Computational Mathematical Programming, Springer-Verlag, New York, 1985.

Shaw, M., “A Two Level Planning and Scheduling Approach for Computer Integrated Manufacturing”’,

Proceedings of the Symposium on Real-Time Optimization in Automated Manufacturing Facilities, NBS

Special Publication 724, National Bureau of Standards, Gaithersburg, MD, January 21-22, 1986.

Simpson, J., Hocken, R. and Albus, J., ““The Automated Manufacturing Research Facility”, Journal of

Manufacturing Systems, vol. 1, no. 1, pp. 17-31, 1982.

Steffen, M. S. and Greene, T. J., ‘‘Hierarchies of Sub-Periods in Constraint Directed Scheduling”, Pro-

ceedings of the Symposium on Real-Time Optimization in Automated Manufacturing Facilities, NBS

Special Publication 724, National Bureau of Standards, Gaithersburg, MD, January 21-22, 1986.

Wysk, R., Wu, S., Yang, N., ““A Multi-Pass Expert Control System for Flexible Manufacturing Systems”’,

Proceedings of the Symposium on Real-Time Optimization in Automated Manufacturing Facilities, NBS

Special Publication 724, National Bureau of Standards, Gaithersburg, MD, January 21-22, 1986.

Journal of the Washington Academy of Sciences,

Volume 79, Number 2, Pages 86-93. June 1989

Commentary

The Fun of OR

Peter Malpass

The Fun Bunch of OR*

1. Introduction: The What of OR

Few disciplines examine the non-technical reasons for their existence. For

over ten years, a small informal group in Washington has known but kept

carefully hidden the real reason for the continued existence of Operations

Research (OR) and the limited success its practitioners have achieved. That

reason is that OR is fun.

1.1 Definition.

In preparing this manuscript, the authors ran immediately into the most

fundamental un-answered question: ‘“‘What is OR?” in response, we replied

that part of the fun of OR for over 40 years has been trying to define it. At

our annual holiday party, we ascertained that everyone thought the definition

by Stafford Beers was the most thorough:

66 99

But in fact no one remembered it. Instead we had a beer chugging contest

to determine the best informal one at the table. As you can see, OR is a lot

of fun. Candidates included:

‘Research into Operations” (Somewhat uninformative)

“Quantitative Common Sense” (OK, but awfully biased towards mathematics)

*The Fun Bunch is an informal Special Interest Group of the Washington Operations Research/Man-

agement Science Council (WORMSC). The members meet regularly to explore technical issues, and

include, but are not limited to: Hebron Adams, Eloise Brooks, Ed Rattner, Mary Leitnaker, Pete Malpass,

Carla Bowers, Fred Sapp, Marilyn Welles, Doug Samuelson, Ann Harrison, Joe Jacobs, Kathleen Carley,

Bob LaCroix, Margaret Patrick and Myron Hatcher.

THE FUN OF OR 87

‘‘An aid for the executive in making his decisions by providing him with needed

quantitative information based on the scientific method of analysis’ (Morse &

Kimball—too sexist, and limited in its customer base)

‘“OR may be regarded as a branch of philosophy; as an attitude of mind towards

the relation between man and environment; as a body of methods for the

solution of problems which arise in the relationship.” (Kendall—kind of non-

descriptive and incapable of application. Is there a science that does not come

under this definition?)

“OR is the art of giving bad answers to problems to which otherwise worse

answers would be given” (Saaty—sort of negative, but better than most)

“OR is the application of logic and mathematics to a real world problem in such

a way that the method doesn’t get in the way of common sense”’ (Woolsey—

pretty good)

‘Ops Researchers are the CPAs of the non-bottom line world” (Fred Sapp—

difficult, but catchy)

“OR is the art, methods, and technology of solving problems that are (1) huge,

(2) here now, (3) multi-disciplinary, (4) capable of management, and (5) in

which the human component is key” (Hebron Adams and Ed Rattner—very

classical, but limited)

“OR is a license to use the Philosopher’s Stone (Use common sense, or more

accurately, turn anything into gold)” (Fred Sapp #2—classical, but very catchy)

“If someone will pay for it, we'll call it OR and do it” (Adams and Sapp—

pragmatic, but is it any different than anyone else?)

‘OR is applied common sense” (Unknown—the problem is differentiating com-

mon sense from conventional wisdom)

By now, it should be clear that OR is a very human application, although

most of what is taught in schools is mathematical techniques such as mathe-

matical programming, stochastic (probability) processes, statistics, decision

and game theory, inventory (yawn) control, and other computational tools.

Tools play a big part, but Woolsey* points out they can be a part of the

problem as easily as a part of the solution.

1.2 Purpose and Scope.

The purpose of this article is to identify and describe the fun of operations

research as distinguished by our unrepresentative sample of practitioners over

the years. We have begun by addressing the what of OR, and we shall proceed

with the why, how, when, where, and who. Why is easy: because some one

asked...

2. Methodology: The Who and How of OR

In discussing the fun of OR, the who and how are important environmental

aspects which make the field more or less comfortable for various personalities

and techniques.

*Professor Gene Woolsey, Colorado School of Mines.

88 PETER MALPASS

2.1 The Who of OR.

In the words of the marketeers among us: “‘the whole world is OR; every-

thing else is just around.” Since all of us in a sense solve problems to optimize

aspects of our lives all the time, it would certainly seem that there is a (perhaps

Zen) element of truth in the statement that we are all OR analysts. How

people get to be OR analysts, or at least call themselves operations researchers,

is another set of issues. Perhaps of all of the engineering and mathematical

disciplines, OR is the most difficult to rigorously define, and this is one of its

glories to many of its practitioners.

2.1.1 Academic Input.

The source of new OR analysts is heavily dependent upon the academic

component. Industrial Engineering (IE), Operations Research (OR), Man-

agement Science or Decision Science (MS), and Systems Engineering (SE)

departments all produce students qualified in some of the methods and most

of the concepts of operations research. None of them produce students qual-

ified in all of the methods, and few spend much time on the practice of the

methods taught. This is a bone of contention with practitioners who believe

that the practice is more important than the methods themselves.

As money has gotten tighter, however, we see a humanly deplorable but

pragmatically wonderful transformation: academics looking for real world

consulting and/or jobs to supplement their incomes. They also are learning

practice, which should result in a higher quality product as time goes on. Some

academics go a step further: Gene Woolsey is quoted as saying of his PhD

candidates’ topics, “If someone will pay for it, I'll sign it.”” Bryce Elkins of

CSC described it as, ‘““When money gets tight, integrity goes down the tubes.”

On the lighter side, Mary Leitnaker says, “‘the reason battles in academia are

so fierce is that the stakes are so small.’’ We see it as, ““When money gets

tight, performance (functionality and productivity) becomes the most impor-

tant factor (a buyer’s market knows quality).’”’ Having a thorough grasp of

the tools is good, but knowing how and when to apply them is better, and

that takes practice and usually, a little guidance. Since practice is usually more

fun than theory or rote learning of methods, we applaud these trends and

note that our discipline is becoming more fun.

2.1.2 The Rest.

What seems to be the most important factors for OR analysts are the same

things that make good psychologists: they listen well, and are perceived by

the client to want to help solve the problem. Imagination helps, but is not

required. Some of the best practitioners tend to come from non-OR training.

The fun bunch is characterized by quite a mixture: french majors, statisticians,

THE FUN OF OR 89

ex-grocery store operators, ex-elementary ed teachers, sociologists, etc. The

driving personality seems to be a person with a history devoid of great wealth,

although some achieve it. The tools have been procured via a master’s pro-

gram, selected coursework, or learning on the job (hard, but fair, as Jimmy

Carter used to say). You might be an OR analyst and not know it! We find

that people with an unconscious trend towards optimizing how they do things

(more than what they do), are the ones who seem to do well at OR.

2.2 The How of OR.

OR grew out of Industrial Engineering (how to do things faster, cheaper,

or better) and has itself spawned the disciplines of Management Science and

Decision Science (OR tools applied to business and organizations) and Systems

Engineering (OR tools applied to computer science and electrical engineer-

ing). OR tends to be characterized by the use of mathematical and procedural

(methods oriented) tools. The concepts, organization, and procedural tools

of OR have been adopted by many other disciplines in one form or another.

As an example, talking about Critical Path Methods or modeling for project

planning is familiar to most professionals. Since the other papers in this issue

are about the past, present and future of OR, it would be presumptious to

enumerate the tools of the trade here. Suffice it to say, the fun of OR is as

much in the application as in the development of tools.

Along with the rest of the sciences and engineering disciplines, OR has

been feverishly automating every possible tool. This includes the algorithmic,

or mathematically based, and the procedural, or human based. One concern

of the fun bunch and others is that OR, its children, and to a certain extent

its parent, have lately failed to keep a balance of emphasis on the human side

of problem solving to go with a marked emphasis on the automated or math-

ematical tools that assist humans in problem solving. This has been remedied

in an atypical fashion by the practitioner community, which has been successful

in poking fun at the structured mathematical side of the house in a variety of

ways. These include “‘The Fifth Column” series of articles in the Interfaces

journal, the parables of Doug Samuelson in the news rag OR/MS Today, and

the Social Sciences Applications Special Interest Group’s sessions on Anec-

dotal OR, Betting for the Professional, Models of Political Influence in the

Office, and so forth at the national meetings. On the darker side are the self-

flagellations like Russell Ackoff’s articles once a decade in the main journals

on topics like “OR is Dead, But They Haven’t Buried It Yet.”

One of the fun bunch synthesis ideas is that a lot of the fun is picking apart

other peoples’ work which, once presented, becomes work to pick apart too.

The fun of OR is that of fun never ends, so few projects ever are completed!

On reflection, or a glance at a textbook index, most of OR’s tools support

planning and controlling (two of five management functions) rather than eval-

uating per se. Therefore, it is apparent that OR is basically a constructive

discipline, and the critical evaluative aspects are merely quality control...

90 PETER MALPASS

2.2.1 Positive Hows.

To find out about people look in their garbage cans. The OR community

is littered with concepts and guidelines, and its preferred place in the scientific

method is as model developer and decision criteria applicator. Almost all OR

texts are begun with a chapter that says, ““OR = algorithms, techniques, and

procedures.”’ Most then go on to provide 30—40 chapters of algorithms and

techniques, and a summary chapter talking about the problems with quanti-

fying heuristics (procedures) that appear to be substantive expertise domain

knowledge that would make a great expert system, for example. This also

tells you where the big bucks are. The authors of this article are some of the

“still small voice” that worries about procedural and hybrid optimization as

well. There is always a concern that mathematicians will invade the field and

discover that OR can be used to create more OR, call it Applied OR, and

sometime later, the discipline truly will vanish into the realm of unused and

unusable mathematical trivia.

The OR journals have been characterized for years by the battle between

application and theory (mathematical proofs). Currently, Interfaces (joint OR/

MS) is very applied, and Operations Research journal is very theoretical. Since

you need both, it is a good solution. Because some things are good for you

even though they taste, feel, or otherwise seem bad, you get both. We also

note that though Pareto was an economist, his fame is from OR dissemination:

20% of the journal articles carry 80% of the value. Some folks say only 10%

are worth reading for since academics are somewhat less efficient . . . The

funny part is that the 10% of “‘value”’ is different for any two people even in

the same department.

Few disciplines overtly recognize a key factor in dissemination of knowledge:

institutional memory is transmitted at the coffee pot and over long lunches

trading “‘war stories.”’ The anecdote is a key feature in supplying the concep-

tual workings of the discipline. Perhaps the best anecdote is told by John

Seely Brown at Xerox, who was working on training aids for copier service

technicians and when interviewing the local guru was asked, how would you

attack the “intermittent copy quality” problem? Well he thought and tried

several very uneconomical suggestions, before he gave up. The guru asked,

‘Where do poor quality copies go?’’ And hence the source of some Garbage

Can modeling in OR . . . The fun of OR to Carla Bowers is the people. She

advises that an OR analyst was called in to handle a major skyscraper elevator

problem. It seemed like no matter how fast (up to discomfort) they got the

elevators, people complained about waiting for them. The OR analyst read

the studies, went up and down for a few days, and then, in the bathroom,

realized the answer. The building supervisor put up mirrors by the elevator

doors and in the elevators and posted notices about “‘new, improved” service.

The complaints stopped immediately. It turns out we don’t mind waiting when

we have such good-looking company.

Another formalized anecdotal dissemination mechanism is the Professional

Meeting. OR/MS is fortunate to allow such informal topics as Panel discus-

sions on theory and techniques in Problem Definition (the beginning of the

THE FUN OF OR 91

scientific method), and in the same meetings, a session on Anecdotal OR. As

an example of the fun of professional meetings, the anecdotal OR session

included a talk on ‘“‘What Homer Never Told You” which was selected readings

from a parody of the Iliad and Odyssey, where the Trojan Horse was con-

ceptualized as a Camel in year 1 of the Trojan War, but the Royal Operational

Safety and Health Administration (ROSHA) forced redesigns that resulted

in 9 years schedule overruns to complete the working prototype.

Paper two discussed institutional anecdotes as a measure of the environ-

ment: an organization espousing the KISS (Keep It Simple, Stupid) principle,

may not be a great place to do esoteric work! OR has a Parabilist (like a

probabilist, but less uncertain) who bi-monthly issues parables (analogies)

spoofing silly practices. He talked about couching dangerous straight forward

attacks in simple sounding parables to avoid undesirable backlash.

The final talk was overcome by a group of Young Revolutionaries who took

over the podium and demanded an OR anecdote contest. Fortunately they

brought bottles of wine for the first three places, and instituted a democratic

rating by applause evaluation system. It took only one example to crank up

the imaginations and memories of the audience who contributed 12 anecdotes

in 15 minutes. The winner recalled that he worked in an office in the Pentagon

as a lad where their job was to find out who had information needed by the

top brass and obtain it. At some point they started logging how many calls it

took to find the person or office where the information resided. The six month

average was about 4.3 calls. One day, a rascal suggested that they randomly

open the phone book or better still use random number table for the last 5

digits. They all participated for a few days and were amazed to find out that

the average number of calls for a random start to the phone tree was only 4.8

calls to get to the same office or expert. We consider this proof of the old

wisdom that for any two people in the United States, they probably have a

friend, relative or associate in common.

The second place anecdote was from the Fire Safety environment. It seems

that with both seat belts and smoke detectors there is a common phenomenon.

More people are getting injured in fires with smoke detectors because they

find out about the fire early enough that they think they can fight them. More

people are getting hurt in states with seat belts since they think they can drive

faster safely. The failing of the litigation approach to vice seems to be to

examine what a professor at George Mason University calls the ‘“Conservation

of Vice.” If you remove sugar from food, people eat more and the salt and

fat kill them. If you remove sexual outlets (crack downs on prostitution for

example), there is more violence at home. His concern was, when we remove

fat and salt and sugar from foods (on the near horizon), what will we do to

be “‘bad’’?

Networking is an established method of transmitting a professional infor-

mation base via the same anecdotes, and also formal state-of-the-art presen-

tations (evening meetings of the local chapter). Networking = people +

people + people. Unfortunately this tends to take a lot of work, and engineers

tend to be less social by conventional wisdom, than for example, secondary

teachers. Part of the fun of OR is to realize that your purely social luncheon

92 PETER MALPASS

with co-workers is probably going to support you in terms of people contacts

(and avoidances or methods to work around them) and might even carry some

great techniques.

Finally, the How of OR does seem to many people to be common sense.

If you stop to imagine at least one alternative to the immediate suggestion of

how to solve a problem, then suddenly you have optimization questions. The

tools and techniques for analyzing and evaluating optimum choices are the

main thrust of OR, but there are also less well known techniques for problem

definition, alternative solution approach generation, and the rest of the steps

in the scientific method. Learning ever more tools and techniques (maybe just

the learning) is a lot of fun too.

2.2.2 Negative Hows (Or How Not to).

OR follows the advice of Tom Robbins in Even Cowgirls Get The Blues,

who reminds us that we learn more from our mistakes than from our sucesses!

A routine column in Interfaces is ‘“The Fifth Column”’ or as Pogo said, ‘““‘We

have met the enemy and they are us!’’ This column specializes in horror stories

(war stories with a sanguine ending) to teach how not to do. An example is

‘On Modeling the Modelers . . .”’ in which Woolsey describes three ways to

fail and omits describing the way to succeed at problem solving. The basis of

the article is that there are technical, political, economic, and social environ-

ments interacting on every problem. Rarely is the problem single dimensional.

He notes the Beltway Bandit Buster, where the technical issues are clear and

feasible, but the political problem is untenable. If the OR analyst walks in

unaware of the political problem, (s)he becomes the solution by getting blamed

for failure (a 0 — 1 game). There are also untenable technical solutions with

easy political solutions, and similar problems. Our favorite is a system de-

veloped without user input which is then ignored by the users much to the

dismay of management.

Another heuristic we should all recall in these days of Total Quality Man-

agement is the rule of thumb that seven people hear about any good happening

(you do well or an ‘‘Attaboy’’), but 22 hear when you do bad (you blow it,

or an ‘“‘Awshucks’’).

3. The When and Where of or

OR is going on all the time, all around you. It’s everywhere! The driving

principle is ‘Don’t waste scarce resources!”’ and it does not seem to be limited

to those of us who remember the Depression or even the Recession of ’73.

If you are optimizing your approach to achieving goals, you are probably using

OR techniques. If not, maybe we could interest you in a course, or maybe a

few beers.

4. Summary

We close with a reminder that despite all the fun, this is a serious discipline,

and has great longevity and staying power. Three professionals made it to the

THE FUN OF OR 93

gates of Heaven, and were greeted by the gatekeeper, who discovered that

there was only room for one, and upon asking for a decision criterion, was

given that entry would be granted to the one practicing the oldest profession.

The MD immediately suggested that God made Eve from Adam by removing

a rib to complete creation, so surgery, hence, medicine was the oldest profes-

sion. The engineer remarked that engineering preceded surgery, since God

brought Order from Chaos. The OR analyst asked who they thought created

the chaos?

rt id

v Liabeanuniemirerinapiet apr ish oh spiehaatie

rls, whercbteAy sO) By ceaalaneeicee rberweee ev seit

srovniats arherot tahoe

shoiegitgablo orlt ga os ono od! oF baliterg

PRyAG dali pee ori duet aed ch (SGU ee AOTC

(a2 DRIES Lebhsloag stage dati bai bars eas

Havorhognds ddwsbudes vevlgng Sato scope Dao abet

as aswell Knows rechmnicjes “

MALTY Ree : a.nd the ped

‘4 : ‘¢} ‘ > ad rechma@ii

: Cespile all ine fun, f

nncwet + Staving mower, Three prod

ams

wo PPP Taree : : .. Ween

— . Varant

MA AMEE Rom Bes (hay cTT TT oo. Atatiin B. Witham

Vie ie Lee oe eee Je Aone A. Tackeod

; Washington, . 6.66.5... nies id ads PR - Vacant

; ladveteuded rue jobeitune rey Ae gave

¥ Ril

yawned ade ot oe shah vucarnt

Vaan

asi

Foreeat Peaster:

Caewtpe A at

anv Claetatedles Rusioces Wee! iat 0 Sov alee (om = Ab

Soaety of Mechanical Engincers, Wartington Section M4 Ow

piical Society ef W. Kunal Pune)

fe Society for iceland aabicasan t Branch Views

ne 7 American Military Fagineets, Washington Pon (hares A Suctingge

Tenaey 6 CHE Lagincon, National Capital Section Herat 4 ‘Peennds

onde Bicdogy and Medicine, DC Section Cyrm & Oheerting

enn S for Motate. Wishington Chapter... ., ifonc Vacanti

‘Amotiatiis of Dental Reweerch, Washington Section —. Tawi! Matick’

hatitute of Actusauiics and Astronautics, National Capiial Seccon Regionet C insiih

| 2s , A. lifer Wogher

TYETse . Afhert 1. Devtibe

yA => Lue ever Ricisard KR. (nak

sO Presiaee Ws - , a : Va wat

g4 Ve. - + Vecmet

; .

bebires (pe abavdscs ” zegwel

Tyr? abau artnet Albert Ghockraay

Ver adenciaerecdennse: Peggy A. Diwod

Balances vitae ..) Witlam K. Grew

rare a) ‘ Vacant

af Aretica, Waxtiogton Section. ehes Vacant

Bat Insitute of Mising, Metalurgieal aod Petioteuon Enqune=n. Pihngion Secthan V acai

al vlhieeiicbidl ibis AT Ce - Rotert 1 ihet ra'ben

obi , I rf Antica, MD-DC-VA Section. . ' caebvece Alltel LD Willows

- aye aT ey Pee —Fiean!

C. Py os = We ww « yah eee veies ot Wits & ) othe Plan

viet dailies ¢ MPN EN hd 08 Gh ens sua <3 bYadr, coer rr ty eer Reten |. Brady

~aycdnaer om Soxsety. Potomac Division, ............ Vacant

Pacapnadii ‘evearch, Mtropalinn Washington Chapter jlRomald W. Wandetacted

pene aeaeS he eee Serr veers th eek en ve yhuke ome es -» Stanley Deutch

20 hae halen : ee payer! FA hic Seay ge . Vacant

yet Wyss awe wie rrees Tha LA Reiphs 2 Cole

mlogica te hana a “4 see ty *s hea teres

: ‘ Lonnie Section... +o Ralph fh. Col

mae , iain ‘ TE a4 Vacate

‘) re 1 Bins adap W9A8s Fete whan Cert Pee hale

Te | Q ate tna ‘ hep opt 1

perp idee Pet 7 * je a<.6 iene

Bae jot Science, Tex “aay

DELEGATES TO THE WASHINGTON ACADEMY OF SCIENCES,

REPRESENTING THE LOCAL AFFILIATED SOCIETIES

SRE AE CPCECEY OE WN MMOMINDEOND, Go Spain y cet tev vec een see cnes nee vacsntcavacessearens Vacant

rented Cie ly WE WASMUIOTON 52. i ene eee sen ele eee censaneseeees Vacant

ee CREW OF WRN, inn cabin ch eso we oes cede ye wine eninncvevertab acs Austin B. Williams

MIE CNEL OL VV ABININRTON occ cs occ cs wt te pk ee teed ewe cee eeens Jo-Anne A. Jackson

eee e eC Y CO Wy OMIIMIOUON 5. csc cis pe ee nde date wpe nd wae eke eae cee ba ne eke ob as Vacant

SET ES ce A nm eg ga Vacant

ee SCIOTO OT WANMIIOUOE onc, oo ee os eee aR twa e cemae ess bacavenenseuedp ade nene Vacant

ee NCR O, TCC OSRICl OL COMMA cde espe s cnet eee ever swevescesvvvenwe Vacant

ee ROM tiie wk oe <0 cg RST webu eas EAN Ge AO ue ok RRO PRT Ode ws Vacant

ene aR OCU) We NR Sica ehy, wu se Dyin Fae Ws 0 US ween a bo Hil Kniide nlp oe te ee ees Vacant

meactety Of American Foresters, Washington Section ................2.0.eeecees Forrest Fenstermaker

Pen miGn SUtiely Ol PAPINEENS. .. 20... ene e ee ce cde eee nce rede cesaweseud es George Abraham

Institute of Electrical and Electronics Engineers, Washington Section................ George Abraham

American Society of Mechanical Engineers, Washington Section...................00e00 Michael Chi

Seameoeted! SOCITY OF WASHINGTON .... 6. 6c sc cc sc cccaneneneespeceeretyeweesed Kendall Powers

_emerican Society for Microbiology, Washington Branch .......... 0.000000 cctscerenewcevenee Vacant

Society of American Military Engineers, Washington Post...................... Charles A. Burroughs

American Society of Civil Engineers, National Capital Section................... Herbert A. Pennock

Society for Experimental Biology and Medicine, DC Section ..................... Cyrus R. Creveling

mcrcad socety for Metals; Washington Chapter ...5 02.6.0. c cee cee eee ee eens ea eb ee deuts Vacant

American Association of Dental Research, Washington Section..................... J. Terrell Hoffeld

American Institute of Aeronautics and Astronautics, National Capital Section....... Reginald C. Smith

emetean Meteorological Society, DC Chapter .. oi... 2.6.5 0660 cece cede ee anes A. James Wagner

eMC CIGIERY CE VV CUOI 5s. ke es eee ode cee are one eens Albert B. DeMilo

meecoustical Society of America, Washington Chapter..........6..055... 00s. ee neces Richard K. Cook

muenetican Nuclear Society, Washington Section... ......6.... 0266 cee c ce ece setae eeenee Paul Theiss

meriume oO: Food Technologisis, Washinetom SeCHON 666 .5..6 coven ce eco cet e nen ee eceeneanes Vacant

American Ceramic Society, Baltimore-Washington Section.......... Rani tate cen aetna ec a Vacant

ES Ee ir a Nee a nee al nee ee Vacant

Deen Raoty Of ociemee Club. 2-5-5 ocak sedi vee ad na hei wade cae baw ewes Albert Gluckman

American Association of Physics Teachers, Chesapeake Section ..................... Peggy A. Dixon

mpucal Society of America, National Capital Section... ..................5.000005 William R. Graver

American Society of Plant Physiologists, Washington Area Section..................00 eee eee Vacant

Washington Operations Research/Management Science Council ...................... John G. Honig

eutument society Of America, Washington Section. ... 2... 6.00606 6h0 cece sec e ee less eenes Vacant

American Institute of Mining, Metallurgical and Petroleum Engineers, Washington Section ..... Vacant

eee PML PU SIRQUMOBICTS eos g cee Sik os Ge Re ne se pe teen ese eee Hew nee Robert H. McCracken

Mathematics Association of America, MD-DC-VA Section....................0000- Alfred B. Willcox

a eR PORN Ne red Sac ve any So die Fate woh dea spied a ele es kew e Vacant

ret NTN ON ON yo ca Ap op Ghee cola Codec muna We angled dsb dae neaes Jane Flinn

meamierican Phytopathological Society, Potomac Division. ....... 2... 0.6.0.6. 2 ec cee ee eee eee Vacant

Society for General Systems Research, Metropolitan Washington Chapter .... Ronald W. Manderscheid

emian Pactote society. Potomac Ciapler oo sik ice ee cee cece sine eeeaae geen ess Stanley Deutsch

Peerica Pisteimcs soctciy, Potomac ChAPIEr 0... ok ec e snee cece eee de dees ete e eke canes Vacant

masocianon tor Science, Technology and Innovation... ..... 2.2... .022 sects eee ee eee Ralph I. Cole

eee uCNO MIC AN MSCIENW Jo oc cng ey we oe Geese ees pc css aeegeee crews Ronald W. Manderscheid

Institute of Electrical and Electronics Engineers, Northern Virginia Section............. Ralph I. Cole

Association for Computing Machinery, Washington Chapter................... cee eeeeeeeee Vacant

EMRE eemeENINC A OMNI Ge Se aca en dung dina be dadvcdeavascsseuaesecece Carl Pierchala

Delegates continue in office until new selections are made by the representative societies.

Washington Academy of Sciences

1101 N. Highland St.

Arlington, Va. 22201

Return Requested with Form 3579

i Sale ogg

<0 > oe ee es we ee —

pe TR Tr

«one

oad oe

ener

Pe atid

fe fa gee sean oe E

Meer isthe nga

a aed

ee len aleatiind

wpa aT I TTS”

weet

so%

qe ee

oer &

ee SN IN Pe AEE

inte Pig 51g GER AEG

pi NF OP eH

ST inthe Sab” Sea i

seca nearer net,

oe any a we

pees

waar ahem e ®

AE Te A NATO ET" OS

tera

LAs eae Se NER

wwe

wee

yr eee SS

vue

qe bene®

wey ——“

a epee MS.

nn rye.

ee ol

et eee

ree oes

peraw rss

peer es aT

Sted ta

Me iat TE

Se dod

‘ ee,

iy wa eT

“paar

5 Or re

“0

wetness

er eae Tol Tal Ck ea

reo sre

pathy ae

eras

wor

Ee as

fee

Fa oi bol a de

4 yrewores a ar

ate

vie as ee

sere

eT in eal

eee

Pe ap steno ae

NS AA

cee TRO 6 ee

4 yee

wees ee

Se es Diab dined

em Ne

Karen

I I ak died

pee Oe

a ashe

sp news ae oes

woeky drs

See OO OPTS ATO

494 eT a heel

ap wane se

yo fh panes

Pane

OREN TCS

Pe is

arena

mee RY Re

Ae ert

a abla

«tienen

eee SSS SHEEN

emi pwew ees?

gece ane RE

enews

Pea PAO

ag hw een

SG hg SIA aH GES tare FO

a PAO CAO A Sa ahng ipetettciet

TT Oe aint dtiahien

wT PEORIA

wer

pa PR Tt HPS

Ta paraicans ees en i Sd

ee ata daetaaeetetoe

pinay arent tae ete

ee Pere

Ps ok Be ih

yeu at

be PRT NN

Pe eeh ils

eqyeas eer END

ree See SN

ET lh eee

be a

purer

were *

gee RS

erate

oe es

pete

or ptr resem

Quit DEPOT AERO SINS Pe

TE anaes de ie PUG

selene TT ak dota

Seen Eee inet

ieee ae eae

ey ee a ae

+ beeerune

CG td” PEM

ce merece ee

Sqn eee ir

ste CURE”

ATE AEST FEE

regen ye eS

wer

ope reyes

ee a

petre To

ses eae

TS atte

eS ibd pont

Sens Sweets

ae as

Peak Mine ele att

ait way

i ater

paws ewe ae et -

Dal ae

: neeeeen tt ee

PRO SRe A

ae RN SE

peers el

pap eam ee

us mre we

eu? PERE OO em gee

“ Pearle Ee ESE

en eat Aree

ay eT PURO ; bioeeay

NOG hd pt ie Pehle

pda ab eae SAI R CD CO ite einpehetn ee

all POU DO BOD CTT AI TLE Pao mee

he Oe Fk POA Sa OE

>t rent

eS ae

pasate ery 5 ager

oeanpres wee

aa avenge Hest

CP pO DAH PLEO

own wig ees

a fos

Aa ase oe eee

tees

6) US

yee ere te ty, er

eebin ee

er ee oe papen ener aan

. . pala UST

a 7h Ae TY OER EP RIT

TS pepeeten sneer

den EE I ITE

eis OLE eae AEE SE TE

Cr a IP Lo da ditt al

Oe a ke nhl aad

PMT IAAT Tee

urease en

west

wiegage rte Oe

ee nt

eres

Suge See

wey eae Guyueracerwt eS

py eee paremeteeaes Sy cl a ee Ee UT ‘

wey epee MRR BN ree

renee ee tS Pa TT Se heh panei tt

ents tte a utenenwn. WLS wy ae

een Com ee Re

Sak PST RO ANT

PRE es ESS

alpen aa scot

ae tulid

”

fe FORE RNA Fa

eg ged Toned ae Bae eee? aepitaniesdenppecninnsene

Henan out

CD el ae

puree ys

wegen

fly OP PUPA rete capacities oe tel

pba TF

pow ieee

epee! ary

Ene eu ENE Te

mp Widnes yw ee sree ra oP ES

‘ RR a Lhintetin tnt

SOs, GRUP UT

ee hotels

we ae ee ~~

Spay pom a tesa Baha TAME CE TEES deena

ST ae genie tested

ebcgens e URED

pee es Oe

TE phcthl dh esate

Ch Pee “

RS Le ab eae ae

ee eal

2 re epapn pete Ona at eek qretenmaete atte

payee aD CSTE PETS

genes wee ne a

ever parent

Sy aqews et ~~“

yee NS CEE

ET ela a

Wipers

STE Se AE

Lee ae AIG

wut

he ol

eee

eos

ee Ce ie

erarey

Pete ESE

ee a a

mye see

ka a dal

weer

jes eae

Pg RA Ce are

eee

pyre

SD a sabes eee

a bate

St Ose eee

earn’ gs ae pee aereem ANE LOIN MENT ae

Segnalo pners™

parsers ree

Oe nt a al atl aliclad

PPC aod

Oe a te

aps ag rae

> on por, wets

Spee OF a