214 views

Uploaded by markhamsquare

Obituary: Evelyn Martin Lansdowne Beale, 8 September 1928-23 December 1985
By: M. J. D. Powell, Pub. in: Biographical Memoirs of Fellows of the Royal Society, Vol. 33 (Dec., 1987), pp. 23-45

Obituary: Evelyn Martin Lansdowne Beale, 8 September 1928-23 December 1985
By: M. J. D. Powell, Pub. in: Biographical Memoirs of Fellows of the Royal Society, Vol. 33 (Dec., 1987), pp. 23-45

Attribution Non-Commercial (BY-NC)

- lpSolveAPI
- Optimization Handout
- Optimization Scheduling LKAB
- Optimization
- Mar-31-2011-3-40-46-yehiag2002@yahoo.co.uk-DrYaheia (5)
- OT I Quiz 2
- MILP Unit Commitment Formulation
- Solving a Multi Objective Fuzzy Variable Linear Programming Problem Using Ranking Functions
- 6. 81-88
- HTC Paper
- MILP model for HEN Retrofit(Barbaro et al)-05.pdf
- QCh1
- MATH 415 (Operational Research II)
- Restrepo J. Masters
- A Disassembly Line Balancing Problem With Fixed Number of Workstations
- BI_09_Optimiz.pptx
- render_07
- Simplex Method (2-Stage)
- 1-s2.0-S0378779612000405-main_2
- Ch11

You are on page 1of 24

BY M. J. D. POWELL,F.R.S.

INTRODUCTION

school and university. He became a pioneer in the development of linear

programming methods at the Admiralty Research Laboratory (A.R.L.),

Teddington. He then joined the Corporation for Economic and Industrial

Research (C.E.I.R.) in 1961 in response to the challenge of applying

operational research and mathematical programming to industrial prob-

lems. C.E.I.R. became Scicon (Scientific Control Systems Ltd) but

Martin remained there, being its 'Scientific Adviser' finally, a title that

reflected his strong preference for advancing his subject in a benevolent

way despite the commercial pressures of industry. Regularly on Mondays

from 1967 he attended the Mathematics Department at Imperial College

as a visiting professor. There, at conferences and in his published work,

he communicated his extraordinary skill at extracting useful results

computationally from mathematical models of real problems. Most of his

papers on particular calculations and on particular techniques are sub-

stantial contributions to knowledge, but probably he will be remembered

best for his constant and active interest in the development of mathe-

matical programming systems for applying optimization algorithms

painlessly in practice. He wrote (1961 c)* that 'The most important

part of operational research is educated common sense, and computers

have absolutely no common sense', but he planned his systems so well that

this defect of computers was negligible. There are no secrets of his success

as he believed in open publication of useful discoveries. In all ways he

was generous and kind, subject to high standards of honesty and academic

integrity. He was devoted to his family and to the Christian faith.

* Numbers

given in this form refer to entries in the bibliography at the end of the text.

23

M R ca-

24 Biographical Memoirs

FAMILY LIFE

the elder child of Evelyn Stewart Lansdowne Beale and Muriel Rebecca,

nee Slade. Then his father, who had studied X-rays and crystal structure

with Sir William Bragg, F.R.S., was Senior Physicist and Research

Engineer at the research station of the Anglo-Persian Oil Company at

Sunbury. This work took the family to Persia for a year in 1932, about

one year after the birth of Martin's brother Julian. On their return they

moved to Markham Square, Chelsea, keeping a house in Middlesex, but

Martin was unwell. Eventually it was found that he had contracted

malaria in Persia, so he did not begin his formal education until 1934, at

the Montessori School in London, which he attended for two years. His

father left the oil business to set up a private consultancy firm for industry

with R. Denman, and worked brilliantly on a wide range of engineering

problems (The Times 1972). Throughout World War II his mother was

Chair of the London and Middlesex branch of the Women's Land

Army.

At school Martin showed an aptitude for figures, but his writing began

curiously as he had learnt the alphabet from wooden bricks when 2, so to

him the orientation of the letters was unimportant. He studied Latin with

his mother when 8, anticipating entry to St Aubyn's preparatory school

(Rottingdean, Sussex) in 1937, but he was at home throughout 1939-40

because of ill health. Home was now above Treyarnon Bay, for the Beales

had become captivated by this part of Cornwall during a holiday at

Constantine, and they had built Windhover House with the Denmans, an

addition to their properties in Markham Square and in Middlesex.

Martin studied at home with Julian and some other children during that

year, and this small school continued throughout the war, although

Martin returned to St Aubyn's in 1940, which had evacuated to Bettws-

y-Coed. There he aimed for a scholarship to Winchester in 1941, but

measles prevented him from travelling for the examination. After a

further year at St Aubyn's his mathematics master, the late E. Webber,

said to his wife: 'I cannot teach any more mathematics to Martin Beale, I

have taught him all I know.' Martin was at Winchester from 1942 to

1946, having gained the second scholarship in 1942. Although he was

the joint winner of the Richardson prize for mathematics there, he was

advised not to try to follow his father from Winchester to Trinity College,

Cambridge, because the competition might be too strong. Martin rejected

this advice and went up to Trinity with a scholarship in 1946.

Bridge was a very strong interest of Martin's at Cambridge. His intro-

duction to this game was a fascinating book on evaluating the strengths

of hands by counting losers, which he had found when visiting his

maternal grandparents at Stokenchurch in 1939. Earlier he and Julian

had spent much time on devising the structure and running of an

Evelyn Martin Lansdowne Beale 25

imaginary country but bridge replaced this activity. Hands were discussed

on journeys to Cornwall, and he devised a new bidding system for the

family, the usual four being Martin, Julian, Mrs Beale and her mother.

Both Martin and Julian played for Cambridge against Oxford, and Julian

went on to play for England in international matches. Having heard that

Martin would work in his group at the Admiralty Research Laboratory,

S. Vajda visited him at Cambridge, but apparently he showed more

interest in a book he had begun to write on bidding systems (which was

not published) than in his future employment. Martin did, however, join

the mathematics group of the A.R.L. from Cambridge, having gained

a first-class honours degree in mathematics in 1949 and a distinction in

the statistics diploma in 1950.

He spent the university vacations at home in Stanwell Moor or

Treyarnon Bay, and continued to live with his parents when he joined the

A.R.L., moving from Stanwell Moor to Whitehall, Wraysbury (also in

Middlesex) in 1952. On 1 July 1953 he married Violette Elizabeth Anne

Lewis (Betty) at St Mary's Church, Hampton. They had met in time to

celebrate his 22nd birthday together for she was a scientific assistant in

the A.R.L. Mathematics Group. This celebration was because she had

been sad to hear that Martin had spent his 21st birthday playing bridge,

but he soon encouraged her to play too. An early example of his concern

for teaching others occurred on a train from Waterloo when Betty sought

refuge in a 'ladies only' compartment as she did not wish to receive

instruction, but Martin, from the corridor, was determined to enlighten

her on the lead to defeat a contract that had occurred earlier in the

evening. She soon became part of the Beale family, visiting Windhover

several times before their marriage, and enjoying cycling, walking and

bridge with Martin. They became firm friends and dependent on each

other for the rest of his life.

Their first home was a small terrace house at Strawberry Hill,

Twickenham, within cycling distance of A.R.L., where Betty continued

working until a few weeks before the birth of their elder son, Nicholas,

in February 1955. Their daughter, Rachel, was born in May 1957. The

family spent 1958 in the U.S.A. (at Princeton except for a six-week

summer visit to Los Angeles), which not only helped Martin's career

but also provided initial contacts with several Americans who became

intimate friends of the family, including George Dantzig and Philip

Wolfe. On returning to England they were able to afford a car. Their

younger son, Marcus, was born in August 1960. In 1963 they moved to

Ember Lane, Esher, Surrey, from 3 bedrooms to 5, Betty having found

the house as Martin was so immersed in his work.

At their new home Martin would work on the sitting room floor, and

even there he enjoyed the company of his children provided they did not

disturb his papers. Many visitors came to Ember Lane, and E. Hellerman

(Dantzig & Tomlin 1987) writes: 'One evening, I think it was in 1967, my

26 Biographical Memoirs

wife and I were guests at the Beale home in Esher. After a fine dinner and

stimulating conversation, Martin, knowing that I had a background in

music, decided that he and his family would sing a Gilbert and Sullivan

operetta. Each member of the family was assigned one or more roles to

sing, and then to the music of a recording the performance ensued.

Needless to say, I had never heard a rendition of Gilbert and Sullivan

with more spirit and gusto. Martin's enthusiasm for the music spilled

over into the family.' This enthusiasm included composing 'Ping went

the bell' for the family and piano playing, especially the music of Haydn.

Later Rachel studied music at the Dartington College of Arts, and

Nicholas and Marcus too became stronger musicians than Martin.

Remarkably, Nicholas and Marcus also obtained scholarships from

St Aubyn's to Winchester, and then went up to Trinity College, Cambridge,

where they read mathematics and architecture respectively.

On Sundays the family regularly attended church at Weston Green,

near Esher, and then spent the rest of the day at Whitehall. There they

thoroughly enjoyed the creations of Martin's father, including a go-cart

and 'H.M.S. Octagon', which was a mock submarine conning-tower in

the garden that included an escape slide and telephone communication

with the house, but Martin did not inherit any engineering skills. They

swam in the swimming pool there, even on Christmas Day. Every

summer they visited Windhover, and there are photographs of massive

systems of sandcastles that the family constructed at Treyarnon Bay.

After Martin's appointment at Imperial College, Rachel, when writing

home from school, addressed her letters to 'Professor and Mrs Beale',

but the school office had been advised to write to 'Mr and Mrs Beale'.

When questioned on this anomaly, Rachel explained that her father was

a professor only on Mondays. From then on the family adopted this

distinction gleefully, and Betty recalls: 'When he became an F.R.S.

everyone was being announced as they greeted the President at a Con-

versazione:- " Lord and Lady X", "His Excellency the Ambassador for

Y", "Professor and Mrs Z",...; Martin gave our name as "Mr and Mrs

Martin Beale". I said later "You might have said Professor on this

occasion". "Why? " he replied simply, " It isn't a Monday".'

His father died in 1972 and for a few years his mother continued to run

both Whitehall and Windhover House, where she was visited regularly

by her descendents. She then decided to retain only Whitehall, but

Martin's attachment to Cornwall was so strong that he sold the house at

Ember Lane in 1977 to buy Windhover. Because he needed a base near

London too, Martin and his family also resided in a part of Whitehall.

In a typical week he would drive to Treyarnon Bay after work on the

Friday and return to Whitehall on the Sunday afternoon, much refreshed

by the weekend in Cornwall. Of course he was allowed leave from his

activities at Scicon, but his holidays were seldom more than two weeks

long. The 1978 holiday was tragic: his brother, Julian, during one of his

frequent visits to Treyarnon Bay, fell to his death in a cliff accident.

Evelyn Martin Lansdowne Beale 27

invariably attended all available lectures. Otherwise he participated

enthusiastically in excursions and other activities; for example, he was

among the group from the 1972 Figueira da Foz NATO Summer School

who on many evenings drove furiously in bumper cars at a local fair-

ground. Betty always accompanied him to the triennial symposia of the

Mathematical Programming Society (M.P.S.), and there they thoroughly

enjoyed meeting friends, always finding a way of offering hospitality,

even at the Massachusetts Institute of Technology in 1985 where they

shared a tiny room in a student dormitory. The 1973 symposium at

Stanford University was unusual as Martin arrived a week early in

California for a holiday. All the family travelled there, and also they all

attended the meeting in Budapest in 1976, where they were in the public

eye, as Martin was then Chairman of the M.P.S.

His high reputation brought him more and more commitments during

the last 10 years of his life, and his response was to work even harder as

he was determined not to let anyone down. After dinner he and Betty

would often enjoy about 3 games of backgammon in 20 minutes and then

he would spend the rest of the evening working. Immediately after a walk

or a swim in Cornwall he would usually return to his studies, even during

his 'holidays', but he never had a computer at home. He played intensely

too, not only at board and card games, but also in the garden and on the

beach, sometimes as a fierce shrieking cannibal in pursuit of his grand-

children. Betty remembers that a NATO workshop in Cambridge in

1980 allowed them to be together away from home for an unusually long

time: it lasted 12 days. Another visit that shows some of his personality

is recalled by R. V. Simons of Scicon:

In 1984 we went to India to give a course of lectures to the Oil and Natural

Gas Commission on optimization in the oil industry. From the moment that

we got on the plane it was evident that he knew how to enjoy himself and

to receive as well as to give. He was at home with the Indians, who were so

eager to learn, and entered fully into the spirit of a coach trip around

Ahmedabad and the surrounding countryside which was put on for us and

for the participants on the course. Back in Delhi he showed no dismay, if he

felt any, at the spartan accommodation which was provided for him at the

Indian Statistical Institute and proceeded to spend his spare time sight-

seeing, travelling around on the local buses rather than seeking out a more

salubrious form of transport.

Usually he would wear a jacket and tie on such excursions.

He was seriously ill at the beginning of 1985. Instead of a large cele-

bration of one gigasecond (109 seconds) of marriage to Betty on 9 March

1985 there was a small party. He was told at the end of August that his

condition was terminal, and his response was to go on being as energetic

as possible. Sir David Cox, F.R.S., of Imperial College writes: 'He was

a distinguished man in a deep sense of the term. While one remembers

him primarily as at the height of his activity it is impossible not to

2 RBM

28 Biographical Memoirs

beyond the limit of his strength under conditions of great discomfort, and

his dignity and courage during this period will not be forgotten by those

who witnessed them. In this period, as no doubt throughout his married

life, he was sustained by Betty's quite exceptionally devoted care.' He

was at work as usual during the last week of his life, but was taken by air

to his beloved home in Cornwall on 21 December where he died on

Monday, 23 December.

PROFESSIONAL CAREER

Research Laboratory in 1950 under S. Vajda who writes (Dantzig &

Tomlin 1987) 'I am sometimes praised for having introduced Martin to

Linear Programming. I did and I am glad of it. But there is no merit in

having done it. I have introduced LP to others as well, but they did not

nurture the seed the same way as Martin has done.' His early studies

yielded some incisive papers that are discussed later. Also much of his

time was spent on government projects. For example, K. Bowen (1986)

reports that in 1951 he was the statistical adviser of the High Frequency

Direction Finding Analysis Working Party, and that: 'In those days, we

had to battle against Martin's unworldly knowledge, as well as battling to

understand it, but we knew that we had a rare talent and we used it to the

full. To instance only one contribution, we owe to Martin the operational

and excellent Brooke bearing classification system.' (Details are given in

references 1961 a and b.) Unpublished work at A.R.L. included studies of

mine-sweeping performance and radioactive fallout. He attended the

nuclear tests on Christmas Island. The 1958 calendar year was spent

as a Research Associate in the Statistical Techniques Laboratory at

Princeton University where his colleagues included G. E. P. Box, H. W.

Kuhn, C. L. Mallows, A. Tucker and J. W. Tukey.

The mental concentration that he brought to the problems he addressed

shut out conventional behaviour. Indeed K. Bowen writes (1986):

'I remember, and they will never forget, his performance for a U.S.A.-

Canadian group of analysts in Room 39, in the Admiralty. With an array

of tabulations and diagrams on a long table, he started by kneeling on a

chair and finished on the table, wandering about the data on hands and

knees.' This single mindedness continued for at least 10 years, because

A. S. Douglas recalls (1986) from 1962 that: 'He never worried what

people might think of him. If it improved his thought processes he would

lie down and raise a leg or climb on the table or stuff his handkerchief in

his mouth. Peter Windley did once comment to me that he "wished

Martin would take his handkerchief out of his mouth when dropping one

of his pearls of wisdom" - but it was said in amusement rather than

annoyance. Indeed no one minded what he did, since he always came up

with useful and perceptive remarks, from whatever position and through

Evelyn Martin Lansdowne Beale 29

research is confirmed by S. Vajda: 'He was highly appreciated at A.R.L.

When members had to be assessed, it might have been asked " Is (s)he as

good as Beale?" Few were.'

C.E.I.R. (U.K.) was founded by A. S. Douglas and M. G. Kendall in

1960 as a subsidiary of C.E.I.R. in Washington. Martin joined it from

A.R.L. in 1961, being no. 29 on its pay-roll. His activities there until 1964

are well described by A. S. Douglas (1986). They included modelling the

diffusion equations that govern the output of oil-fields, and the simulation

of air-battles for the Shape Technical Centre. He also began his very

important work on the development of mathematical programming

systems for the solution of general calculations, because such systems

were a major part of the commercial success of the parent company in the

United States. He led the British C.E.I.R. group that worked in this field

including M. Fieldhouse, R. E. Small, P. Windley and M. Jeffreys. One

of the team on the American side was E. Hellerman who writes (Dantzig

& Tomlin 1987) about Martin: 'His insight into how algorithms could be

implemented on a computer was phenomenal. He was the dynamo

behind the extensions to the LP/90/94 System in the areas of Separable

Programming, Mixed Integer, Dantzig-Wolfe Decomposition, and a

host of other practical algorithms.'

Martin described these algorithms and his insight very clearly in his

book Mathematical programming in practice, which was published by

Pitman in 1968. His reputation was now well established, and he was

much in demand as a speaker at conferences. In this year C.E.I.R.

became Scicon, now a subsidiary of British Petroleum, and work had

begun on the mathematical programming system UMPIRE, which

superseded LP/90/94. J. A. Tomlin, who joined C.E.I.R. in 1968, writes

(Dantzig & Tomlin 1987):

UMPIRE was built in an informal atmosphere which would horrify

today's 'structured everything' advocates. In the first year or two there were

five or six of us working for Martin on the project, which is said to have

begun when Rita Walton was called into his office to write FORMC. The

three or four more junior team members shared an office, and I think Martin

found us a bit unruly. There was a high non-English quotient - at various

times there were members from America, Australia, Canada and India, all

conspiring with his German secretary - and we often found Martin's very

English mannerisms amusing. I'm sure he knew this, but never, of course,

referred to it. We encouraged Martin's eccentricities. When we hit a snag,

or a new idea looked promising, he had a way of climbing on the nearest desk

or mangling the end of his tie, while he mentally solved the problem to

his own satisfaction before enlightening us. This was often much more

entertaining than the work we were supposed to be doing, which Martin

sometimes poured on at a fierce pace, assuming that everyone could keep up

with him. One would sometimes dread those mornings when he would come

in with a sheaf of papers full of new ideas, and often code to be implemented.

2-2

30 Biographical Memoirs

are given later. R. S. Hattersley of the SCICONIC team recalls: 'Martin

was always refreshing to work with because he would approach any

problem with an open mind, and always preferred to try a new idea than

force an existing solution to fit. He produced ideas at a formidable rate

without pausing for weekends, holidays, and it sometimes seemed, sleep.

We were often fighting a losing battle to understand and test them all

but we didn't like to disappoint him by having no results to report.

Fortunately, he was always ready to stop and explain his thoughts as

often as necessary, and moreover he would be genuinely concerned if he

thought we were working too hard and promise to restrain his inventive-

ness.' Another view of the latter part of his career at Scicon is from his

secretary, B. Peberdy:

What always struck me was Martin's ability to be friendly with and make

welcome eminent people or the most lowly person in the company. Although

he was a splendid person to work for, he also expected people to work as

hard and as efficiently as he thought them capable of. He did not suffer

carelessness or fools gladly. At the same time he was very patient if people

could not easily understand something. His total unstuffy, non-pompous,

modest and rather humble manner belied a joy-filled and uninhibited

personality, which came as a great surprise to people who didn't think past

the mathematical genius to find this delightful person within. It was not

uncommon to see him skipping down the corridors, skipping gleefully

because he had solved an annoying mathematical problem.

In view of Martin's achievements and leadership at Scicon, it is

remarkable that he also found time to make a very substantial contribution

to teaching at both undergraduate and postgraduate level at Imperial

College. His work there is described by J. T. Stuart, F.R.S.:

Martin Beale's association with the Department of Mathematics goes

back to 1967 and he was a Visiting Professor from October of that year until

his death in December 1985. Visiting Professors at Imperial College often

play a very important role but I doubt if there are any who can match the

regularity with which Martin came to the College and prosecuted both

teaching and research work over such a long period. It is fair to say that he

came on Monday of almost every week of the year during those 18 years. He

taught third year undergraduate courses on operational research and related

areas, and participated regularly in the teaching programme of our M.Sc.

course in Statistics. In addition, he had many research students who

graduated with Ph.D. degrees from our Department after having studied

under his direction. Thus, in a real sense we came to regard Martin Beale

as a permanent member of our Department and many of us feel he achieved

as much in one day a week as might reasonably be expected of a full-time

member of the staff, he was that active in both research and teaching.

Further, Sir David Cox writes:

Our Visiting Professor scheme is designed to foster contacts with research

workers outside the academic sphere and particularly with industry. Martin

Evelyn Martin Lansdowne Beale 31

entirely characteristically took his duties extremely seriously and over the

years lectured regularly to final year undergraduate students of mathematics

on optimization and operational research and to M.Sc. students of statistics

on numerical optimization. In addition and very importantly he supervised

a succession of doctoral students, 10 over the period in question. In the initial

period I think students found his lectures difficult, particularly because

Martin's style was not suited to the habits of British undergraduates who

tend to work not so much from textbooks as from lecturers' notes which are

expected to be given in a form lending themselves to coherent note-taking.

An anecdote from those early years in many ways characteristic of the man

is told by an ex-colleague who attended one of Martin's courses. One of the

students was rather noticeably asleep. Martin at first made no comment, but

towards the end of the period said to my colleague "Please wake Mr X. He

will find this next bit interesting ". I feel sure this was said out of real concern

and was not to be taken sarcastically.

His approach to teaching at postgraduate level is recalled by E. M.

Aghedo, who was one of Martin's research students:

He was more of a father than a supervisor to me. He had the outlook of

the legendary Chinese Martial Arts Teacher who would punch hard to make

the pupil learn to punch and praise when he had learned. Martin Beale had

the magic of making the most difficult of subjects a child's play. I was taught

to be critical of any result by not being deceived by its beauty but being

careful to explore fully its usefulness. He usually set me on a job like so:

"Investigate the correctness of this concept. If correct, why, and if not, why

not." Such was the man Martin! He was a very thorough man who believed

in a high standard of work and was very uncompromising about this.

Of course he also gave much encouragement and guidance to young

researchers at Scicon, as shown in a tribute from R. M. Hattersley:

'Martin taught me, with patience and kindness, most of what I know

about mathematical programming. He promoted in me a lasting interest

in the field, and for that I am in his debt, as no doubt are many others

before me.'

The research of very many people was helped by his participation in

conferences. His invited lectures were always useful but not always easy

to follow, because in a single talk he would often give details of many

different subjects. They were thoroughly enjoyable, however, as Martin

excelled at conveying his enthusiasm, and he would not expose a topic

unless he believed that it was valuable. The contents of these lectures are

considered later in this memoir. When listening to papers he sat near

the front of the audience, he was always attentive, and usually he asked

questions. He tried much harder than most people to understand all that

was said, and frequently he sought further information from speakers

after their presentations. These discussions were often of more help to the

speakers than to Martin, but his thirst for knowledge of mathematical

programming algorithms that might be useful was unquenchable. At

conferences he hardly ever missed an opportunity to listen to a relevant

paper and he did not skip talks that were given by unknown researchers.

32 Biographical Memoirs

Martin contributed much to professional societies and to official com-

mittees, including the Subcommittee on Computing of the Defence

Scientific Advisory Council, the Operational Research Methodology

Committee and the Council of the Assessments Board, the Mathematical

Sciences Subcommittee of the University Grants Committee, and three

S.E.R.C. (Science and Engineering Research Council) bodies, namely

the Mathematics Committee, the Operational Research Panel and the

Science Board Computing Committee. It is stated by A. S. Douglas

(1986) that: 'As a committee member he was always admirably to the

point and never long-winded, but he did not miss much and was always

willing to speak up when he felt something needed saying-a very real

support for the Chairman and sometimes a useful discipline for him if he

got his summing up wrong!'

He joined the Royal Statistical Society in 1950, being Honorary Secre-

tary from 1970 to 1976 and a Vice-President from 1978 to 1980. He was

an active member of both the Institute of Statisticians and the Operational

Research Society. For many years he was Treasurer of the Statistical

Dinner Club, a rather venerable institution that he served with great

devotion. He contributed much to the Institute of Mathematics and its

Applications, giving several invited talks at conferences, being on the

editorial board of its Journal of Applied Mathematics, and serving on

Council from 1980 to 1985, first as an ordinary member and then as a

Vice-President. His part in the Mathematical Programming Society is

described by P. Wolfe (Dantzig & Tomlin 1987):

In 1964, he was the leading spirit in organizing an 'International Sym-

posium on Mathematical Programming' in London, the first such meeting

held outside the United States. Subsequently, finding a small surplus of

funds in their treasury, the organizers designated it the 'International Pro-

gramming Fund' and chose a small international committee to hold it. This

was the first step in the identification of Mathematical Programming as a

professional specialty, and eventually led to the formation of the Mathe-

matical Programming Society in 1972. The newly formed society recognized

Martin's achievements in many ways. The first election held by the Society

had to choose both its first Chairman and his successor. George Dantzig was

the first Chairman and Martin the second, serving from 1974 to 1976. Prior

to that, he was asked to join the board of senior editors of the Society's new

journal, Mathematical Programming,a small group which included two who

later became Nobel laureates. He served on the Council of the Society from

1982 to 1985 and otherwise on several of its committees. Whether he had an

official role or not, he was regularly consulted on all matters of policy, and

his advice was followed.

The academic excellence of his work was acknowledged by his election

to the Royal Society in 1979, and by the award of the Silver Medal of the

Operational Research Society in 1980. He was a member of Council of

the Royal Society in 1984-85.

Evelyn Martin Lansdowne Beale 33

PUBLISHED WORK

several techniques that have become standard procedures in mathematical

programming calculations, descriptions of mathematical models for a

wide range of practical applications, and explanations of the efficient use

of optimization methods. Initially his writings just show remarkable skill

and ingeniousness in response to real needs and academic questions, but

Martin became more and more concerned that computer users generally

should benefit from advances in suitable optimization algorithms.

Therefore his later work includes many excellent expository conference

papers, in addition to valuable new work that he continued to publish

throughout his career.

Most researchers in mathematical programming propose new algo-

rithms and study their theoretical properties, which is a far cry from the

successful solution of real optimization problems, because in practice it is

usual to prefer methods that are readily available through easy-to-use

computer programs. Therefore the development of new methods, the

provision of general software for these methods, the use of the software

for particular calculations, and the construction of the mathematical

models that give the calculations are each highly important tasks. Martin

was expert at all of them, but this is a story to unfold gradually. We begin

with his first three papers and the research that stemmed from them,

which provide a trail through his work on the algorithm development

side.

His first paper, published in 1954, presents an independent discovery

of the dual form of the simplex algorithm for linear programming in the

following original way. One requires the least value of a linear function of

n non-negative variables subject to m < n linear equality constraints. All

but one of these constraints are used to eliminate (m- 1) of the variables,

leaving a problem in (n-m+1) variables with one equality constraint

that can be solved explicitly. If the values of the eliminated variables at

this solution are non-negative then the calculation is complete. Otherwise

one member of the set of eliminated variables is altered and a new iteration

is begun, the change being such that each iteration increases the value of

the objective function except in degenerate cases. Thus convergence is

usual but reference 1955b gives an early example of cycling in linear

programming calculations.

This work led to one of the first algorithms for quadratic programming

(1955a, 1959c), where the calculation of the previous paragraph is

generalized by allowing the objective function (F say) to be quadratic, but

the constraints are as before. A linear approximation to F allows any

linear programming method to provide a fruitful direction of search in

the space of the variables, but now the best step along this direction may

be determined by the curvature of F instead of by a constraint boundary.

The key idea is to absorb this possibility into the linear programming

34 Biographical Memoirs

direction to be a new variable, which gives one more variable and one

more linear equality constraint. Making the new variable zero provides

the optimal step that is determined by the curvature of F, and the new

variable can be dropped when it is no longer useful. Holding the new

variable at zero is equivalent to making future search directions conjugate

to the search direction that defined the new variable, so in fact Martin

proposed one of the earliest conjugate direction algorithms, but he does

not mention conjugacy explicitly until his survey paper (1967 a).

However, he observed in 1955 (1955 a) that the key idea is also suitable

for minimizing the sum of the largest t terms from a set of s linear

functions (where s and t are any positive integers with t < s), but it seems

that this suggestion has not been taken further.

The use of conjugate search directions is the only technique for mini-

mizing smooth unstructured functions of several variables that receives

much attention in his publications, and it is the subject of some substantial

work. Much time was given to CGAP (conjugate gradient method of

approximation programming), which minimizes a general differentiable

function subject to general constraints. CGAP is particularly efficient

when, for any fixed values of a few of the variables, the remaining

calculation is an LP (linear programming) problem. Changes to nonlinear

variables are derived from linear approximations to nonlinear terms, and

the constraints usually allow some of these variables to be 'independent',

which means that their values are free so other variables are adjusted to

satisfy the constraints. Conjugate search directions are used when the set

of nonlinear independent variables is unchanged and when no derivative

discontinuities arise from the LP problem that determines the linear

variables; otherwise the steepest feasible descent direction is preferred.

Many other details are important to efficiency (1974e, 1980b (written in

1976 with A. S. J. Batchelor)) so a later review (1978d) is recommended

for a clear summary of the main ideas. Here a close relation to the reduced

gradient method is explained (see also 1982b), and it is stated that

'Serious use of the current version of the system has been confined to a

sequence of oil production problems for the Kuwait Oil Company'. Later

Martin seems to have decided that more successful procedures had

become available for nonlinearly constrained calculations, because

CGAP is not mentioned in his 1985 survey of mathematical programming

systems (1985 a).

In reference 1972b he proposed the first conjugate gradient procedure

for unconstrained optimization that achieves quadratic termination from

a general initial search direction. The method, which is now employed in

several optimization algorithms, just adds to each conjugate search

direction the multiple of the initial search direction that gives orthogon-

ality to the initial change in gradient. It is useful for 'restarts' that

Evelyn Martin Lansdowne Beale 35

function is not quadratic, and it can provide conjugacy in constrained

calculations when some earlier changes of variables have been restricted

by constraint boundaries (1978e, coauthor R. Benveniste). Further work

on conjugate gradients includes a device for achieving the least value of

a quadratic function without any exact line searches (1985d, coauthor

O. S. Brooker).

When solving nonlinear calculations Martin usually preferred to apply

the 'separable programming' technique. In its basic form it requires each

nonlinear function to depend on only one variable, and often this con-

dition can be achieved by a reformulation of the calculation or by the

introduction of some new variables. He found that the method is suitable

for a wide range of problems (see 1965b and 1968, for example), and he

increased its usefulness greatly by his work with J. A. Tomlin (1970g) on

'special ordered sets', which are as follows.

The separable programming technique makes a piecewise constant

or a piecewise linear approximation to each nonlinear function of one

variable. Thus each section of an approximation depends on one or two

values of the underlying nonlinear function. There is a non-negative

variable for every function value that is used. These variables are weights

that sum to one, and the piecewise constant or piecewise linear cases are

obtained by allowing only one or two adjacent variables to be non-zero.

The success of this approach in practice depends on how these com-

binatorial conditions on the variables are achieved. It is usual to employ a

'branch and bound' method: here one solves a sequence of calculations

where the constraints of each one are only linear equations and simple

bounds on the variables (that can force some of the variables to take

prescribed values). Usually a solution of a trial calculation does not satisfy

the combinatorial conditions, and then it may be replaced by two new

trial calculations whose constraints define disjoint regions that exclude

the solution that has just been found but that contain the required solu-

tion. Special ordered sets provide highly successful ways of making

these splits. Specifically, if only one of the m variables {v': i = 1,2, ..., m}

may be non-zero, but if vi and Vj are both non-zero in a trial solution

where i < j, then the '(S1) set' strategy requires v1 = v2 = ... = vk = 0 in

one of the new calculations and vk+1 = vk+2 = ... =vm = 0 in the other

one, where the integer k is chosen from the interval [i,j- 1]. Alternatively,

in the case of piecewise linear approximation where two adjacent variables

may be non-zero, if the trial solution is infeasible because i < j-1, then

the '(S2) set' strategy employs the alternatives v1 = v2 = ... = Vk_1 = 0,

or vk+l = vk+2 = ... = vm = 0, where now i+1 < k <j-1.

His view of the merit of this approach changed over the years. In

his book (1968), he writes: 'Separable programming is probably the

most useful nonlinear programming technique', and his first paper

36 Biographical Memoirs

separable terms, and I have often extolled the virtues of separable pro-

gramming. It is therefore sad for me to have to admit that we have made

little use of separable programming recently.' His first two papers with

J. J. H. Forrest (1976a, 1978c), however, provided ideas and techniques

that helped to restore separable programming to first place among his

methods for nonlinear functions. They give careful consideration to the

choice of the integer k of the previous paragraph and find good ways of

selecting it. They propose an interpolation technique for adding more

function values automatically where it is advantageous to improve the

accuracy of a piecewise approximation. They introduce 'linked ordered

sets' to process products of the form uf(x) directly, where u is a variable

and where f(x) is a function of one variable that is treated by separable

programming (see also 1980d). Their later paper (1978c), which is much

easier to read than reference 1976a, gives several numerical examples to

show that separable programming is suitable for calculating global

solutions of non-convex optimization problems, because the approxi-

mations to nonlinear functions apply throughout their ranges. CGAP,

however, employs local approximations that are usually valid only for

small changes to the variables. This is the point that seems to have been

decisive in Martin's eventual preference for separable programming over

CGAP.

Of course the goal of this work was to improve the efficiency of actual

computer programs for solving real problems, so Martin became highly

expert on the practical details of techniques for integer programming,

especially the branch and bound method. In an early survey article (1965 a)

he writes 'So far, cutting plane methods for integer programming have

been easily the most successful', but he reported enthusiastically in 1974

(1974b) that 'The most dramatic improvements during the last 3 years in

the capability of general mathematical programming systems have been in

the field of integer programming' and that 'Branch and bound strategies

have been developed to the point where they can solve many problems

completely in a moderate time.' The splitting of one continuous problem

into two (branching) by the branch and bound method has been men-

tioned already, in order to satisfy combinatorial constraints, and the other

main ingredient (bounding) is the maintenance of a bound on the final

value of the objective function that allows one to rule out most of

the continuous trial problems from further consideration before many

branches have been made. Ideally one would not explore branches if

some further work on another part of the tree would show that they

cannot yield the required solution. Therefore the development of suitable

practical strategies for selecting the continuous problem to solve next is

crucial to efficiency. His early strategies (1965c) were unexciting, but

highly successful techniques for branching and bounding are described in

papers written between 1974 and 1985 (1974b, 1977a, 1979b, 1985b).

Evelyn Martin Lansdowne Beale 37

The work on special ordered sets, for example, shows that much effort

is needed to include in a mathematical programming system the ability to

gain efficiency from the structure of an optimization calculation. There-

fore, because most problems are now solved by general systems, it is

usual to ignore structure unless there are already features in a system that

can take advantage of it. However, in the days when new programs were

written for new calculations and when computers were much slower,

more attention was given to special structures. In particular Martin

studied the 'purchase-storage problem' with G. Morton (1958): here one

can buy commodities at a seasonal price, there is a known market for them

in a sequence of time periods, and they can be stored for later selling but

there are capacity constraints on the warehouses. It is shown that the

optimal profit up to the end of a time period can be calculated easily from

the optimal strategy up to the end of the previous time period, and that

some parametric programming can be included too, parametric pro-

gramming being the study of the effect on the solution of changes to

the coefficients of the calculation. In reference 1959a he extends the

'transportation problem' to the case when each variable in the objective

function is replaced by a convex function of the variable, but the

proposed algorithm seems not to have been very successful. In reference

1963 a he addresses the important case when a few variables of a linear

programming calculation occur in many constraints, but otherwise the

problem can be partitioned into several small disjoint problems; it

describes a primal algorithm that takes advantage of this structure.

Another variation of the transportation problem is considered (1963 b):

here there are so many sources and destinations that it is impracticable

to store every source-destination distance, so instead one works with

distances between relatively few key towns and from sources and

destinations to their neighbouring key towns. He notes (1980e) that the

special ordered set technique can be applied to a zero-one variables

problem whose objective function is the ratio of two linear terms

(fractional programming). After 1963, however, most of his work on

structured calculations became more general, and it provided valuable

contributions to the following mathematical programming systems.

Reference 1965c (coauthor R. E. Small) describes the method for

integer programming that is incorporated in the C.E.I.R. LP/90/94

system. This system also solves 'decomposable' linear programming

problems efficiently (1965d, coauthors P. A. B. Hughes & R. E. Small),

where 'decomposable' means that, except for a few constraints that are

treated specially, the problem can be separated into several small

independent subproblems, so the situation is similar to the one that was

studied in 1963 (1963a). Several other features of LP/90/94 are

mentioned (1968). Then the UMPIRE system was developed at Scicon

(1970c); this work included an independent discovery of the use of

product forms and triangular decompositions of sparse basis matrices in

38 Biographical Memoirs

advantage of sparsity are special ordered sets for separable programming

(1970g, 1976 a), and GUB (generalized upper bound) sets, a GUB being

a bound on the sum of some of the variables subject to the condition that

a variable may not be present in more than one GUB (see 1970c and

1975b, for example). UMPIRE was followed by the SCICONIC system,

which is the main subject of reference 1978 a. It solves linear, separable,

integer and parametric programming problems. A key feature of

SCICONIC is the attention that is given to the interface between the

system and the user: about 15000 lines of computer code are used to

control input and output, to check some of the input, and to convert the

input to a form that is suitable for the main calculation. Martin

emphasized the importance of such interfaces (1970b, 1978a), because

they are crucial to the work that is needed to apply a computer system to

a new optimization calculation. Also he held strong views on the choice

of notation for elements of matrices, and he made some specific

suggestions (1974c, 1978a, 1980c) to help the formulation of complex

mathematical programming problems.

Many of the invited papers that he presented at conferences describe

the main techniques of these systems, and he believed that wide publicity

for useful developments in mathematical programming is far more

important than commercial exploitation. These talks have helped many

optimization calculations generally to be solved by suitable methods. The

most useful algorithms that were available 20 years ago are described

clearly in his book (1968), and a sequel to it will be published soon by

Wiley. He seldom included original work when he addressed a general

audience, to concentrate on features that he knew from experience to be

highly valuable, but at many meetings he also offered a submitted paper

on new research. Thus his survey articles are rather repetitive: in

particular there are many descriptions of special ordered sets for separable

programming (1974b, 1976b, 1977a, 1978d, 1979b, 1980d, 1985b), and

of the advantages of factored forms for the basis matrices of linear

programming (1969c, 1971, 1974b, 1975b, 1976b). He always spoke

enthusiastically about the techniques he believed to be best, and this

point of view caused him to make the following delightfully immodest

comment about his quadratic programming algorithm (1967 a): 'I am not

going to make any pretence of being impartial between these methods. I

will content myself with explaining how my own method works and why

I think it has great advantages over all other methods.'

His publications include a wide range of examples of applications of

optimization methods to investigate real problems. In many cases the

Scicon computer systems were used for the actual numerical calculations,

but mainly these papers consider the construction of suitable mathe-

matical models. In 1961 he considered the accuracy of direction finders

(1961a, b) the problem being to estimate the variance that occurs in

Evelyn Martin Lansdowne Beale 39

Coen & A. D. J. Flowerdew) studies the cheapest mix of raw materials

that is sufficient to produce prescribed quantities of iron from four blast

furnaces and a sinter plant. In reference 1966b (coauthors P. A. B.

Hughes & S. R. Broadbent) the prediction of television audiences and

newspaper readership from past ratings is considered to help advertisers.

Reference 1970 a describes the development of a computer package for a

Farm Advisory Service that is designed to give some assistance to farmers

on such things as crop rotation and the scheduling of labour. A

combinatorial problem with several zero-one variables is considered

briefly (1972 a, coauthor J. A. Tomlin), and in 1974 there appeared a long

discussion of the formulation of a military transportation problem

(1974c, coauthors G. C. Beare & P. Bryan-Tatham). A model for

investigating the consequences of different uses of limited resources in

the national health service is described in reference 1974a (coauthors

A. G. McDonald & G. C. Cuddeford), and it is mentioned again in

reference 1975a. This paper, however, gives more attention to the

cheapest way of expanding a pipeline network to provide greater flow

capacities between prescribed nodes, which is an integer programming

problem because the number of different available diameters of new pipes

is small. The success of the health care model (1974 a, 1975 a) is discussed

(1977b, coauthor J. Sullivan), and it is noted that a major deficiency is

that there is hardly any variety in the treatment of patients who are as-

signed to a single class. The optimal relocation of government depart-

ments is the subject of reference 1978 b (coauthors P. S. Ayles, R. C. Blues

& S. J. Wild): after the departments are grouped by using 'Communica-

tions Cluster Analysis', there is a linear (or integer) programming

problem to solve, except for product terms. Ways of treating these terms,

including special ordered sets, are compared. He addresses the plan-

ning of a network of pipes to bring natural gas ashore from wells, where

the system may include compressors (at a price) to boost flows (1979c,

1983). Oil field exploration is considered (1986a) from a stochastic point

of view: several statistical questions are discussed, for example the pre-

diction of the undiscovered resources in an area by extrapolating from the

number of successes and failures of drillings that have already been made.

The advantages of constructing and optimizing such models is a main

theme of the 1980 Blackett Memorial Lecture (1980f). Here Martin

distinguishes between routine calculations where the usefulness of

numerical results has been established, and initial speculative attempts at

modelling. He emphasizes that often in the latter case the main benefits

of an optimization calculation are that its formulation concentrates the

mind on the principal features of the field of study, and that unacceptable

numerical results may yield improvements to the model. Because this

point of view is tenable only if the optimization itself does not require

much effort, a major purpose of this paper (and of 1969a, 1984b and

40 Biographical Memoirs

mathematical programming systems has made the optimization of the

parameters of a wide range of models easy and reliable. Another

interesting and general paper on modelling is reference 1980c: it gives

particular attention to avoiding pitfalls that can cause severe loss of

efficiency in subsequent optimization calculations. On the theoretical side

of operational research he contributed a short proof that, in multi-

objective assignment problems, efficient solutions cannot be dominated

by linear combinations of efficient solutions (1984a).

His expert knowledge of both mathematical programming and statistics

has provided some papers that do more than applying optimization tech-

niques to statistical problems. In particular references 1962 (coauthor

G. P. M. Heselden) and 1966a address Blotto games, where 'The

Colonel Blotto game is a military deployment problem found in Caliban's

Weekend Problems Book'. An example that he gives is the siting of

missiles to defend targets against bomber attacks, when the total number

of missiles is fixed and is fewer than necessary to defend all the targets

adequately. Such games are combinatorial problems that can be so large

that one aims only for an approximate solution. He recommends allowing

the deployments to be probability distributions to achieve convexity, for

this makes the calculation much easier. The main idea is illustrated by the

remark that the average of 3 missiles and 5 missiles is not 4 missiles but

3 missiles half the time and 5 missiles half the time. The optimal solution

of the new problem usually suggests a near optimal solution of the

original calculation. Cluster analysis is the subject of reference 1969b. An

algorithm for assigning points to clusters is proposed that reduces the

number of clusters iteratively, and that exchanges points between clusters

on each iteration until a local minimum is found. There is a discussion of

statistical tests for choosing the number of clusters. Statistics is also

merged with optimization in a paper on stochastic programming (1980 a,

coauthors J. J. H. Forrest and C. J. Taylor). Here commodities are

manufactured, stored and sold in a sequence of time periods, the only

deviation from linear programming being a random element in the

demand for current and future purchases that depends on previous sales

(which are known). Some choices of suitable distributions for the

randomness are considered, and they yield nonlinear functions that are

treated by special ordered sets; consequences of these choices are

illustrated by small numerical calculations. This difficult work is extended

(1986b, coauthors G. B. Dantzig and R. D. Watson) to cases where some

of the coefficients of the 'linear programming problem' (for example the

cost of manufacturing one unit of some of the commodities in the current

time period) are also stochastic.

Integer programming is combined with statistics in an efficient algo-

rithm for discarding variables in linear regression (1967b, coauthors

M. G. Kendall & D. W. Mann). Here a random variable is to be

Evelyn Martin Lansdowne Beale 41

out of p random variables, where p and r are given with r < p. This

calculation would be ordinary linear regression if the set of non-zero

coefficients were known, but it has to be chosen automatically to

minimize the residual sum of squares. The algorithm solves a sequence

of linear regression problems with various numbers of parameters,

because it takes advantage of the fact that it is easy to update the solution

when there is an addition to or a deletion from the set of coefficients that

are allowed to be non-zero (see also 1970d and 1974f). These additions

and deletions are controlled by a branch and bound strategy: for example

a coefficient must be non-zero finally if, without it, the residual sum of

squares is always larger than a value that has already been calculated by

using at most r non-zero coefficients. Details of a general form of this

integer programming procedure are given in reference 1970e. In two

papers (1970f, 1982 a) he compares the algorithm in reference 1967 b with

some other methods for the discarding of variables problem. Moreover,

it is shown (1974d, coauthor P. C. Hutchinson) that it is futile to increase

r if, at the solution, there are fewer than r non-zero coefficients.

His other contributions to statistics are mainly academic. In 1959

(1959b, coauthor C. L. Mallows) a necessary condition was found for

the existence of an independent random variable Z such that X = Y+ Z,

where X and Y are given random variables. Reference 1960 is a tour-de-

force on confidence regions in nonlinear regression. Assuming that errors

of observations are normal and unbiased it considers whether level sets of

the maximum likelihood function are suitable estimates of confidence

regions. Letting x be a given observation of an unknown random vector

that could have the true value y, the key question is whether the

probability of observing y when x is true is a good approximation to the

probability of observing x when y is true. The answer depends on

departures from linearity that cannot be corrected by reparameterization,

and some ways of estimating the magnitude of this effect are suggested.

Some approaches to the missing values problem of linear regression are

studied in reference 1975c (coauthor R. J. A. Little): crudely speaking

this is a least squares calculation with some unknown matrix elements.

From the statistical point of view, however, these elements are

observations of correlated random variables, so one can find the most

likely values of coefficients even if there are no complete rows of data. It

is pointed out that some algorithms for this calculation give biased

estimates, and there are some illuminating numerical results. In reference

1979 a (coauthor R. G. Seeley) it is shown that algorithms for linear least

squares calculations can be used to solve a problem that has serial

correlation in the errors of successive observations. The 1984 Kendall

Memorial Lecture (1985c) begins with a view of Sir Maurice's five

careers and reminisces a little on reference 1967b. The main subject is

control variables and Martingales. It is explained that when analysing

42 Biographical Memoirs

simpler models. The idea is to express an outcome of the original model

as the sum of an outcome of a simpler model, and of the difference

between the two outcomes, the intention being that the first term is

relatively easy to analyse and the second term has a much smaller variance

than the outcome of the original model. Two examples of this technique

are given: one demonstrates a way of avoiding bias and the other one is

a queueing problem. As usual Martin concentrates on results that are

valuable to practical calculations.

In reference 1963 a he wrote 'For the benefit of those who, like the

present author, prefer numbers to formulas, I now present a miniature

scale numerical example.' This inclination towards actual calculations,

and this intention to communicate his work in simple terms are typical of

his illustrious career. No one else has done so much to advance the

successful use of mathematical programming algorithms, and his

publications are a fine memorial of his achievements.

ACKNOWLEDGEMENTS

(Martin's mother), Mrs E. M. L. Beale (Martin's wife) and S. Vajda, and

for permission to quote from articles by K. Bowen, G. B. Dantzig, A. S.

Douglas and J. A. Tomlin. Further, I wish to thank not only the other

contributors to this memoir but also the larger number of people who

kindly offered material that has not been included.

BIBLIOGRAPHY

1954 An alternative method for linear programming. Proc. Camb. phil. Soc. 50, 513-523.

1955a On minimizing a convex function subject to linear inequalities. Jl R. statist. Soc. B 17, 173-

184.

b Cycling in the dual simplex algorithm. Nav. Res. Logist. Q. 2, 269-275.

1956 (With M. DRAZIN)Sur une note de Farquharson. C. r. Acad. Sci. Paris 243, 123-125.

1958 (With G. MORTON)Solution of a purchase-storage programme. 1. Opl Res. Q. 9, 174-187.

1959a An algorithm for solving the transportation problem when the shipping cost over each route

is convex. Nav. Res. Logist. Q. 6, 43-56.

b (With C. L. MALLOWS) Scale mixing of symmetric distributions with zero means. Ann. math.

Statist. 30, 1145-1151.

c On quadratic programming. Nav. Res. Logist. Q. 6, 227-243.

1960 Confidence regions in nonlinear estimation. Jl R. statist. Soc. B 22, 41-88.

1961 a Brooke variance classification system for DF bearings. J. Res. natn. Bur. Stand. D 65, 255-

261.

b Estimation of variances of position lines from fixes with unknown target positions. J. Res.

natn. Bur. Stand. D 65, 263-273.

Evelyn Martin Lansdowne Beale 43

1961 c Some uses of computers in operational research. In Proceedingsof the Inaugural Meeting of the

Swiss Operational Research Society, pp. 51-52. Zurich: Industrielle Organisation.

1962 (With G. P. M. HESELDEN) An approximate method of solving Blotto games. Nav. Res. Logist.

Q. 9, 65-79.

1963a The simplex method using pseudo-basic variables for structured linear programming

problems. In Recent advances in mathematical programming(ed. R. L. Graves & P. Wolfe),

pp. 133-148. New York: McGraw-Hill.

b Two transportation problems. In Proc. Third Internat. Conf. Operational Research (ed. G.

Kreweras & G. Morlat), pp. 780-788. London: English Universities Press.

1965a Survey of integer programming. J. oper. Res. Soc. 16, 219-228.

b (With P. J. COEN & A. D. J. FLOWERDEW) Separable programming applied to an ore

purchasing problem. Appl. Statist. 14, 89-101.

c (With R. E. SMALL)Mixed integer programming by a branch and bound technique. In Proc.

1965 IFIP Congr. (ed. W. A. Kalenich), vol. 2, pp. 450-451. Washington, D.C.: Spartan

Books.

d (With P. A. B. HUGHES& R. E. SMALL)Experience in using a decomposition program.

Comput. J. 8, 13-18.

1966 a Blotto games and the decomposition principle. In Theory of games techniquesand applications

(ed. A. Mensch), pp. 64-84. London: English Universities Press.

b (With P. A. B. HUGHES& S. R. BROADBENT) A computer assessment of media schedules. Opl

Res. Q. 17, 381-411.

1967a Numerical methods. In Nonlinear programming (ed. J. Abadie), pp. 135-205. Amsterdam:

North Holland.

b (With M. G. KENDALL & D. W. MANN)The discarding of variables in multivariate analysis.

Biometrika 54, 357-366.

1968 Mathematical programmingin practice. London: Pitmans.

1969a Mathematical programming: algorithms. In Progress in operations research (ed. J. S.

Aronofsky), vol. 3, pp. 135-173. New York: John Wiley.

b Euclidean cluster analysis. In Proc. 37th Session of the ISI, Bull. Internat. Statist. Inst. 43 (2),

99-101.

c Nonlinear optimization by simplex-like methods. In Optimization (ed. R. Fletcher), pp. 21-

36. London: Academic Press.

1970a A management advisory system using computerized optimization techniques. Outl. Agric. 6,

143-147.

b Matrix generators and output analysers. In Proc. Princeton Symp. Mathematical Programming

(ed. H. W. Kuhn), pp. 25-36. Princeton University Press.

c Advanced algorithmic features for general mathematical programming systems. In Integer and

nonlinear programming(ed. J. Abadie), pp. 119-137. Amsterdam: North Holland.

d Computational methods for least squares. In Integer and nonlinear programming (ed. J.

Abadie), pp. 213-227. Amsterdam: North Holland.

e Selecting an optimum subset. In Integer and nonlinear programming (ed. J. Abadie), pp.

451-462. Amsterdam: North Holland.

f Note on procedures for variable selection in multiple regression. Technometrics 12, 909-

914.

g (With J. A. TOMLIN)Special facilities in a general mathematical programming system for

non-convex problems using ordered sets of variables. In Proc. Fifth Internat. Conf.

Operational Research (ed. J. Lawrence), pp. 447-454. London: Tavistock Publications.

h (Editor) Applications of mathematical programmingtechniques. New York: Elsevier.

1971 Sparseness in linear programming. In Large sparse sets of linear equations(ed. J. K. Reid), pp.

1-15. London: Academic Press.

1972 a (With J. A. TOMLIN)An integer programming approach to a class of combinatorial problems.

Math. Prog. 3, 339-344.

b A derivation of conjugate gradients. In Numerical methodsfor nonlinear optimization (ed. F.

A. Lootsma), pp. 39-43. London: Academic Press.

1974 a (With A. G. MCDONALD & G. C. CUDDEFORD) Balance of care: some mathematical models of

the National Health Service. Br. med. Bull. 30, 262-271.

b The significance of recent developments in mathematical programming systems. In

Mathematical programmingin theory and practice (ed. P. L. Hammer & G. Zoutendijk), pp.

11-29. Amsterdam: North Holland.

44 Biographical Memoirs

study: a case study in mathematical programming. In Mathematical programmingin theory

and practice (ed. P. L. Hammer & G. Zoutendijk), pp. 417-442. Amsterdam: North

Holland.

d (With P. C. HUTCHINSON) Note on constrained optimum regression. Appl. Statist. 23, 208-

210.

e A conjugate gradient method of approximation programming. In Optimization methodsfor

resourceallocation (ed. R. Cottle & J. Krarup), pp. 261-277. London: English Universities

Press.

f The scope of Jordan elimination in statistical computing. I.M.A. Bull. 10, 138-140.

1975 a Some uses of mathematical programming systems to solve problems that are not linear. Opl

Res. Q. 26, 609-618.

b The current algorithmic scope of mathematical programming systems. Math. Prog. Stud. 4,

1-11.

c (With R. J. A. LITTLE) Missing values in multivariate analysis. Jl R. statist. Soc. B 37,129-145.

1976a (With J. J. H. FORREST) Global optimization using special ordered sets. Math. Prog. 10,

52-69.

b Optimization techniques based on linear programming. In Optimization in action (ed. L. C.

W. Dixon), pp. 447-466. London: Academic Press.

1977 a Integer programming. In The state of the art in numerical analysis (ed. D. A. H. Jacobs), pp.

409-448. London: Academic Press.

b (With J. SULLIVAN) Mathematical and computational concepts of a macro-economic model of

the balance of health care. In Systems science in health care (ed. A. M. Coblentz & J. R.

Walter), pp. 313-319. London: Taylor and Francis.

1978a Mathematical programming systems. In Numerical software-needs and availability (ed.

D. A. H. Jacobs), pp. 363-376. London: Academic Press.

b (With P. S. AYLES,R. C. BLUES& S. J. WILD) Mathematical models for the location of

government. Math. Prog. Stud. 9, 59-74.

c (With J. J. H. FORREST) Global optimization as an extension of integer programming. In

Towards global optimization, II (ed. L. C. W. Dixon & G. P. Szego), pp. 131-149.

Amsterdam: North Holland.

d Nonlinear programming using a general mathematical programming system. In Design and

implementationof optimization software (ed. H. J. Greenberg), pp. 259-279. Alphen aan den

Rijn: Sijthoff & Noordhoff.

e (With R. BENVENISTE) Quadratic programming. In Design and implementationof optimization

software (ed. H. J. Greenberg), pp. 249-258. Alphen aan den Rijn: Sijthoff & Noordhoff.

1979 a (With R. G. SEELEY)First order auto-regressive regression analysis. In Forecasting (ed. 0. D.

Anderson), pp. 167-175. Amsterdam: North Holland.

b Branch and bound methods for mathematical programming systems. In Discrete optimization,

II (ed. P. L. Hammer, E. L. Johnson & B. H. Korte). Amsterdam: North Holland. (Ann.

Discrete Math. 5, 201-219.)

c The optimisation of gas gathering pipeline networks. I.M.A. Bull. 15, 126-128.

1980a (With J. J. H. FORREST & C. J. TAYLOR)Multi-time-period stochastic programming. In

Stochastic programming (ed. M. A. H. Dempster), pp. 387-402. London: Academic

Press.

b (With A. S. J. BATCHELOR) A revised method of conjugate gradient approximation pro-

gramming. In Survey of mathematical programming (ed. A. Prekopa), pp. 329-346.

Budapest: Akademie Kiado.

c Pitfalls in optimization methods. In Pitfalls of analysis (ed. G. Majone & E. S. Quade), pp.

70-88. Chichester: John Wiley.

d Branch and bound methods for numerical optimization of non-convex functions. In

COMPSTAT 1980 (Proc. Fourth Symp. Comp. Stat.) (ed. M. M. Barritt & D. Wishart),

pp. 11-20. Vienna: Physica Verlag.

e Fractional programming with zero-one variables. In Extremal methods and systems analysis,

lecturenotes in economicsand mathematicalsystems, 174 (ed. A. V. Fiacco & K. 0. Kortanek),

pp. 430-432. Berlin: Springer-Verlag.

f The Blackett Memorial Lecture 1980. Operational research and computers: a personal view.

J. oper. Res. Soc. 31, 761-767.

1982 a Elimination of variables. In Encyclopediaof statistical sciences,vol. 2, pp. 482-485. New York:

John Wiley.

Evelyn Martin Lansdowne Beale 45

1982b Algorithms for very large nonlinear optimization problems. In Nonlinear optimization 1981

(ed. M. J. D. Powell), pp. 281-292. London: Academic Press.

1983 A mathematical programming model for the long-term development of an off-shore gas field.

Discrete Appl. Math. 5, 1-9.

1984 a Note on 'A special multi-objective assignment problem' by D. J. White. J. oper. Res. Soc. 35,

769-770.

b Mathematical programming. In Developments in operational research (ed. R. W. Eglese &

G. K. Rand), pp. 1-10. Oxford: Pergamon Press.

1985a The evolution of mathematical programming systems. J. oper. Res. Soc. 36, 357--366.

b Integer programming. In Computational mathematical programming (ed. K. Schittkowski),

pp. 1-24. Berlin: Springer-Verlag.

c The Kendall Memorial Lecture. Regression: a bridge between analysis and simultation.

Statistician 34, 141-154.

d (With O. S. BROOKER) The use of hypothetical points in numerical optimization. Math. Prog.

Stud. 25, 28-45.

1986a Optimization methods in oil and gas exploration. I.M.A. J. appl. Math. 36, 1-10.

b (With G. B. DANTZIG & R. D. WATSON) A first order approach to a class of multi-time period

stochastic programming problems. Math. Prog. Stud. 27, 103-117.

Bowen, K. 1986 Professor E. M. L. Beale-a personal tribute. O.R. Newsletter (February), pp. 8-9.

Dantzig, G. B. & Tomlin, J. A. 1987 (ed.) E. M. L. Beale, FRS; friend and colleague. Math. Prog.

38. 117-131.

Douglas, A. S. 1986 Obituary. Professor Evelyn Martin Lansdowne Beale, FRS, FIMA. I.M.A.

Bull. 22, 120-122.

Obituary: Evelyn Stewart Lansdowne Beale. The Times, 26 January 1972.

- lpSolveAPIUploaded bycp0000
- Optimization HandoutUploaded byKaseh Dhaah
- Optimization Scheduling LKABUploaded byDavid Victor Halomoan Simatupang
- OptimizationUploaded byMohamed Hussainsha
- Mar-31-2011-3-40-46-yehiag2002@yahoo.co.uk-DrYaheia (5)Uploaded byalex240574
- OT I Quiz 2Uploaded byRobertBellarmine
- MILP Unit Commitment FormulationUploaded byAhmed Asim
- Solving a Multi Objective Fuzzy Variable Linear Programming Problem Using Ranking FunctionsUploaded byIjsrnet Editorial
- HTC PaperUploaded bySalman Zafar
- 6. 81-88Uploaded byiiste
- MILP model for HEN Retrofit(Barbaro et al)-05.pdfUploaded byAnonymous uCYIu1
- QCh1Uploaded bySaud Hussain
- MATH 415 (Operational Research II)Uploaded bysaeed
- Restrepo J. MastersUploaded byZBojanic
- A Disassembly Line Balancing Problem With Fixed Number of WorkstationsUploaded byamalia permata
- BI_09_Optimiz.pptxUploaded byThiran Vinhar
- render_07Uploaded byillustra7
- Simplex Method (2-Stage)Uploaded byNoor Asiah
- 1-s2.0-S0378779612000405-main_2Uploaded byBodoShow
- Ch11Uploaded byetayhailu
- Lecture2 EnUploaded byclifford
- 06190757Uploaded byJiadong Wang
- Linear ProgrammingUploaded byarchana_bharti_23
- Linear Programming.pptx 1Uploaded byAlisa Conley
- Lecture 01Uploaded bynarendra
- OptimalUploaded byThaiHuynhNgoc
- bfm-978-3-319-62350-42F1.pdfUploaded byGauna Anuga
- Kelompok 1_RPP Pembelajaran LangsungUploaded byarini
- evolucion diferencialUploaded byJhonatan Soto
- ejercicio-2 (1).pdfUploaded byOscarcito Mas Naa

- Ethics and Social ResponsibilityUploaded byMaria Gabriela Popa
- Introduction to Dojo ToolkitUploaded byharishmce2010
- Allelopathic potential of Notholirion thomsonianum (D.Don) stapfUploaded byInternational Network For Natural Sciences
- Assessment of seismic damage of multistory structures using fragility curvesUploaded byAnonymous 7VPPkWS8O
- Medical Ethics Basic PrinsipUploaded byAdeLia Nur Fitriana
- The Relevance of Garbage- Can ModelUploaded byPersephona13
- TrimbleProXT&XHRecieverUserGuide.pdfUploaded bynicoparty29
- 1 Land Management and Democratic Governance InCoJ 02Uploaded byNatália Adriele
- S2.pdfUploaded bydaniel
- Abap Program Tips v3Uploaded byRoberto Martínez
- Accenture Life Sciences OverviewUploaded byBobAdler
- Fire in the sky de walton travisUploaded byMaxwell Demon
- acetobacterUploaded byJhonatan Cáceres
- Cox-Murray - Apollo the Race of the Moon Part 1Uploaded byHipatia
- 11595133421Excel Guidebook V7Uploaded bydragon_287
- supplier_auditUploaded byAmilaLakmal
- Simulated AnnealingUploaded byLokesh Singanaboina
- Reporty Psycho.Uploaded byReedon Quinan
- Group2-Offenbacher-Hafen.pptxUploaded byDyezika Perez
- Introduction to RetailingUploaded byMadhan Krishnamurthy
- Admission Form UOG 2010Uploaded byadnan_westridgian
- e-unit format edu512a fall 18 gabrielagarciaUploaded byapi-438371615
- Final Year ECE Project in Bangalore ( http://www.modainnovations.com)Uploaded bymodainnovations
- Dragonframe User Guide - WinUploaded byDeepSixitnow
- Current Affairs Study PDF - July 2017 by AffairsCloudUploaded byabcd_123425
- IEC 62337 - ComisionamientoUploaded byRichard Julan G.
- God, Creation and ExistenceUploaded byMark Ó Duibhdábhoireann
- Theory X and Theory YUploaded byhanizzz
- Excel ToolsUploaded byPriyank Gaba
- Tribology Lesson PlanUploaded bykumar