SPTI MA
MATHEMATICAL PROGRAMMING SOCIETY NEWSLETTER
Testing the Theory of Evolution
An Application of Combinatorial Optimization
L. R. Foulds
University of Florida
Implications of C ,. I. Darwin's
theory of evolution have recently been
rigorously tested using combinatorial opti
mization techniques. It 11 come as no
surprise to many academics that the results
of the test in no way invalidate the theory.
Dr. David I'. uI, and Dr. Michael Ilendy of
Massey University, New Zealand, and I
have, over the last eight years, been investi
gating claims that Darwin's theory is not
respectable science but mere belief. A
graph theoretic model of evolution has been
constructed and solution techniques for the
model have been I1.. 1ti., i. using combina
torial optimization techniques. When the
techniques are applied to data gained from
modern biochemical methods, they produce
solutions which do not support these
claims.'
It would require millions of years to
verify by observation whether or not the
process of evolution actually takes place.
Thus all that can be done is to look for
evidence which suggests that evolution has
or has not taken place in the past. This is in
some ways unsatisfactory as it means that
the theory cannot be tested by the scientific
method in the same way as other scientific
theories, such as Einstein's photoelectric
effect.
But there is evidence to suggest that
each living species possesses a very detailed
description of itself in the form of a DNA
code, which is apparently used to build
genes. There is the I. i1 il' that, if the
theory of evolution is . I,' when species
evolve their DNA codes II evolve too. So
an analysis of the protein codes from diffe
rent species may be quite revealing.
What was done was to search for
ancestral relationships among five different
proteins from 11 vertebrate species. The
proteins can each be characterized by a
string of fixed length from an alphabet of
four symbols called nucleotides. For each
protein, the 11 strings (one for each species)
can be aligned and analyzed comparatively.
It turns out that any 11 species could
be related in over 34 million ways. But it
was found in this investigation that for each
independent protein, the simplest possible
pattern of relationship was remarkably
similar.
The five different proteins are inde
pendent of each other so it seems that the
ancestral relationships of the species which
can be deduced by these patterns support
the idea of some kind of evolutionary
process. I1, is because without such a
process it is not clear why the patterns
would be so .,iiiI .11 Also it is interesting to
note that the pattern tends to conform . II
with evolutionary trees produced by numer
ical taxonomists when studying bone struc
ture and fossil records.
Having read thus far, you are probably
thinking this is all very well, but what has it
to do with mathematical programming? To
justify the publication of this article in
OPTIMA we shall now briefly outline a
graph theoretic model of evolution and some
combinatorial optimization techniques
which can be used to analyze it.
The theory of evolution predicts that
existing biological species have been linked
in the past by common ancestors. A dia
gram showing these links is i;: .1 a phylo
genetic tree or '. i '. i'. A typical phylo
geny is given in Figure 1. Since the time of
Charles Darwin, many scientists have con
structed phylogenies which link both
existing and extinct species in the fossil
record. When the fossil record is inadequate
for a set of species, it may be possible to
determine its evolutionary history from a
knowledge of existing species in the set.
Suppose we wish to construct a
phylogeny for a given set of species with
little fossil evidence. We must use only
Alligator Turtle Man Monkey Sheep Cow Pig Horse
Figure 1: A Phylogenetic Tree
existing species. It is useful to use informa
tion about the versions of a protein called
cytochrome c. It seems that every iP ii.
species possesses a version of this protein as
it appears to be essential for respiration.
Each version can be represented by a string
of 312 letters chosen from the following
four: A, C, G, and U (nucleotides). It is
interesting to compare strings: Man and
Rhesus Monkey differ at only one site of the
312. However, the reader will be relieved to
know that man and bread mold differ at 129
sites!
Our model is concerned with con
structing a phylogeny which can be thought
of as a weighted tree in the graph theoretic
(Continued next page)
Next MP Symposium at MIT
I I, .11' Council is pleased to an
nounce that I'. .. . Jereim y I 1 ...
Professor of Operations Research and Co
director of the ).R. Center at M.l.T., .I11
be Chairman for organizing the 12th Inter
national Symposium on Mathematical Pro
grammingi The Symposium will take place
on the M.I.T. Campus in August, 1985.
AlIex Orden
August 1983
Number 10
TESTING, continued
sense. Each of the given species is repre
sented by a node in the phylogeny. The
other nodes represent other existing or
extinct species. The weight of each arc in
the phylogeny is defined to be the number
of sites at which the strings it connects
differ. The objective is to find a phylogeny
which spans the given set of species and
which is of minimum total weight as a sum
of the weights of its arcs. The objective is
S1 S4
AGUGUUAA AAUGUUCA
AfiliijIAA AAUGUUCA
2AG
Figure 2: The S1 S4 arc
RAIl
1CA
private substitutions. A possible ip.i i ,I, i.; tree
is given in Figure 3. The total number of
substitutions in this tree is 12. The objective
is to find a spanning tree which minimizes
the total number of substitutions.
We now iI!ii,.i ii. the ideas just
presented by returning to the example
problem. A natural first attempt at solution
is to use either the Prim or Kruskal method
for the minimal i iin; ,:. tree problem
Figure 3: A Spanning tree for the five species
(MSTP). Let's see how this approach turns
out for our example. The number of substi
tutions necessary for each pair of strings is
given in Table 2.
S2 S3 S4 S5
3 4 2 3
5 3 4
known to biologists by the somewhat Finding the phylogeny of maximum
contradictory title of maximum parsimony, parsimony for the species in a realistic study
To make this clearer, we now furnish a is no easy task. The number of possible
small artificial example which has been phylogenies has been enumerated assuming
constructed for expository purposes. Sup various restrictions.2'3 As the number of
pose we have five species: S1, S2,.... S5, possibilities grows astronomically, complete
with strings that are only eight characters enumeration is out of the question. The
long rather than the usual 312 characters, model appears on the surface to be a mini
The strings are given in Table 1, where each mal spanning tree problem which is, of
row is the string for a species: course, straightforward to solve. However,
the introduction of additional sequences
1 2 3 4 5 6 7 8 representing extra species (either existing or
extinct) often leads to a phylogeny with
S1 A G U G U U A A greater parsimony (a tree of less weight).
S C A A G U U A A Therefore, the model can be seen to be a
S3 C G U C C G A A variation of the Steiner tree problem on the
3 ncube. This problem is notorious for its
S4 A A U G U U C A intractability and indeed the phylogeny
S5 C G U G U U C U model has been shown to be NPcomplete.4
Given this gloomy state of affairs, it seems
Table 1 unlikely that a polynomial time algorithm
exists for the general problem. Yet all is not
As the sequences have been aligned, lost. It has been possible to modify existing
each column corresponds to a site. minimal spanning tree algorithms by using a
What we want to do is to connect the clustering algorithm to produce a phylogeny
five species by a ..inTiii:. tree where each of relatively low weight. What are clustered
species is represented by a node of the tree. are collections of columns of the matrix of
Suppose we join S1 and S4 directly by an data. This phylogeny of low weight is then
arc in the tree as shown in Figure 2. If we analyzed with view to either proving that it
compare S1 and S4 we see that the two is of maximum parsimony or modifying it so
strings differ at sites 2 and 7 only. At site 2 that it becomes one of maximum parsimony.
S1 has a G and S4 has an A. We label the S1 The approach is to use the clustering algo
 S4 arc with the symbols 2GA to denote rithm to partition the matrix of data verti
one difference between the strings. These cally. This breaks up each original string
three symbols are together called asubstitu into a number of smaller substrings. Each
tion. We also add the substitution 7AC to set of substrings represents a separate,
the arc to denote the difference at site 7. As smaller problem. These smaller problems are
there are no other differences there are no then analyzed and if necessary a further
other substitutions associated with the S1 decomposition is made. Eventually the sub.
S4 arc. Given one of the two strings and the problem can be solved directly. The weights
substitutions, we can deduce the other of their solutions provide valid lower bounds
string. \\ i.1 we wish to do is to construct a which are used to prove the minimality of
spanning tree for (S1, S2, S3, S4, S,) and either the original phylogeny or a modifica
label the arcs of the tree with the annro tion thereof.
Applying a MSTP algorithm we can
obtain (among other minimal trees) the
tree given in Figure 3. There is the possi
bility of the introduction of new strings
in order to reduce the total number of
substitutions. This is brought about by
using a process called coalescement. This
process can be explained as follows. Suppose
we have a substitution at site n between
nucleotides X and Y which appears on two
adjacent arcs in a spanning tree, as shown in
Figure 4(a). The tree can be modified to
produce a spanning tree with one less
substitution by the introduction of a new
string Sm, (called a Steiner point) which
differs from the Si string only at site n. It
has a Y rather than an X there. This process
can be used repeatedly to reduce the number
of substitutions. In applying it to our tree
we can remove one instance of duplication
in both 2GA (on arcs S1 S4 and S4 S5 )
and 1CA (on arcs S3 S1 and S S2 )
substitution. This produces the tree shown
in Figure 5, which has 10 substitutions. We
have introduced two new strings: S6 and
S7'
To discover whether this tree is a
minimal . i,,. r, tree we analyze Table 1,
Consider the three sites: 1,2, and 7. If
substitutions corresponding to the other
Table 2
TESTING, continued
sites are removed, the tree collapses to the
one in Figure 6. It has one Steiner point.
The introduction of more than one Steiner
point would lead to a tree with more substi
tutions. Can we create a tree with fewer
substitutions by having no Steiner points?
This is a normal MSTP. The substrings based
on sites 1, 2, and 7 are given in Table 3. Any
minimal solution to the MSTP for this data
has five substitutions. Therefore, we know
that sites 1, 2, and 7 jointly require at least
five substitutions in any phylogeny. The
other sites: 3, 4, 5, 6 and 8 require at least
one substitution each as their columns in
Table 1 each have two nucleotides. Thus,
any phylogeny must have at least 5 + 1 + 1 +
1 + 1 + 1 = 10 substitutions. As the phylo
geny constructed in Figure 6 has exactly 10
substitutions, it must be minimal. It is not
uniquely minimal.
We now return to the general, large
problems. Sometimes the site decomposi
tion process produces subproblems which
are small enough to be solved by a branch
and bound algorithm. This branch and
bound algorithm guarantees to produce all
phylogenies of maximum parsimony on any
data set with no more than 12 species in
reasonable computing time. So it was used
to calculate directly the phylogenies for the
11 species problems mentioned earlier.
2 7
G A
Table 3
There is no guarantee that the decom
position technique just described will
converge to a phylogeny of maximum
parsimony when applied to a large data set.
However, it has been used interactively with
a computer and has uncovered minimal
solutions to all data sets to which it has been
applied (up to 23 species).
The theory of evolution predicts that
similar phylogenies should be obtained from
different protein sequences. We have tested
this specific prediction from the theory,
rather than the general theory itself. Our
results are consistent with the theory of
evolution.
nxy
Firgure 4: The Coalescement Process
S3 ,4
4CG
sCU 2GA
6GU
ICA 7AC
S7 S^, Is AC
3AU 8AU
S 2AG s,
Figure 5: The Minimal Phylogeny
S4
2GA
2AG ICA 7AC
S23 S, S
1AC
S5
This idea of testing specific predictions
from an hypothesis, rather than the hypo
thesis itself, is inherent in the writing of
one of the leading scientific philosophers of
the day, Karl Popper.s It is more clearly
expressed by Lakatos.6
There may be exceptions when dif
ferent protein sequences will lead to dif
ferent trees as, for example, in the serial
symbiosis theory. Also in the precellular
evolution a network with circuits may be a
far better model than a tree. An interesting
philosophical question would arise if the
results of this work had falsified the predic
tion that the phylogenies would be similar.
This would not necessarily disprove the
theory of evolution as it may be that the
strings simply changed so rapidly that they
lost all information about their early history.
It could be argued that because protein
strings from different species can be aligned
so readily that this, in itself, is independent
evidence that the proteins retain evolu
tionary information.
This article has indicated a new
application for various techniques of com
binatorial optimization. Minimal spanning
tree methods, matching, and branch and
bound enumeration are useful in the con
struction of phylogenies. The next challenge
is to sharpen the decomposition technique
so as to be able to guarantee success with
larger data sets.
Figure 6: A Phylogeny for Sites: 1, 2 and 7
REFERENCES
1) Penny, David, L. R. Foulds, and M. D.
Hendy, "Testing the Theory of Evolution
by Comparing Phylogenetic Trees Con
structed from Five Different Protein Se
quences," Nature 297 (1982) 197207.
2) Foulds, L. R. and R. W. Robinson,
"Determining the Asymptotic Numbers of
Phylogenetic Trees," Lecture Notes in
Mathematics, No. 829, Combinatorial Math
ematics VII, SpringerVerlag, Berlin (1980)
110126.
3) Foulds, L. R. and R. W. Robinson,
"Enuneration of Binary Phylogenetic
Trees," Lecture Notes in Mathematics, No.
884, Combinatorial Mathematics VIII,
SpringerVerlag, Berlin (1981) 187202.
4) Foulds, L. R. and R. L. Graham, "The
Steiner Problem in Phylogeny is NP
Complete," Adv. in Appl. Math., 3 (1982)
4349.
5) Popper, K., Unended Quest: An Intel
lectual Autobiography, Fontana, London
(1976).
6) Lakatos, I., in Method and Appraisal in
Physical Science (ed. C. Howsen). pp. 140,
Cambridge, University Press, (1976).
Technical Reports & Working Papers
Stichting Mathematisdc Centrum
Kruislaan 413 1098 SJ Amsterdam
Postbus 4079 1009 AB
Amsterdam
THE NETHERLANDS
M.H. van Hoom and H.C. Tijms, "Approximations for the Waiting
Time Distribution of the M G/c Queue, BW 144.
M. L. Fisher, B. J. Lageweg, J. K. Lenstra and A. H. G. Rinnooy
Kan, "Surrogate Duality Relaxation for Job Shop Scheduling," BW 145.
E. L. Lawler, J. K. Lenstra and A. H. G. Rinnooy Kan, "Recent
Developments in Deterministic Sequencing and Scheduling: A Survey,"
BW 146.
A. W. J. Kolen, "A PolynomialTime Algorithm for Solving the Set
Covering Problem on a TotallyBalanced Matrix, BW 147.
E. L. Lawler and J. K. Lenstra, "Machine Scheduling with Prece
dence Contraints," BW 148.
C. Van Putten and J. H. Van Scluppen, "The Weak and Strong
Gaussian Probabilistic Realization Problem, BW 149.
C. N. Potts, "Analysis of Heuristics for TwoMachine FlowShop
Sequencing Subject to Release Dates," BW 150.
11. ". ii.. . and A. van der Schaft, "Controlled Invariance by Static
Output Feedback for Nonlinear Systems," BW 151.
I. C. l ,,i .... and H. Groenevelt, "Approximations for (s,S) Inven
tory Systems With Stochastic Lead Times and a Service Level Constraint, "
BW 152.
J. van der Bij and J. H. van Schuppen, "Short Term Adaptive
Prediction of Railway Power Demand, BW 153.
H. :. ii. i. r. "Observability of a Class of Nonlinear Systems:
A Geometric Approach, BW 154.
F. J. Burger and J. C. P. Bus, "An ALGOL68 Package for the
Solution of Systems on Nonlinear Equations; User Manual," BW 155.
II. "'., i. .r and A. J. van der Schaft, "Controlled Invariance for
Nonlinear Systems: Two Worked Examples," BW 156.
It. '" ,,ii. r and A. J. van der Schaft, "The Disturbance Decoupling
Problem for Nonlinear Control Systems, BW 157.
E. L. Lawler, "Generalizations for the Polymatroidal Network
Flow Model," BW 158.
Systems Optimization Laboratory
Department of Operations Research
Stanford University
Stanford, CA 94305
Philip E. Gill, Nicholas I. M. Gould, Walter Murray, Michael A.
Saunders. and Margaret H, Wright, 'i.. ... Space Methods for Convex
Quadratic Program ming, SOL 8214.
Philip E, Gill, Nicholas I. M. Gould, Walter Murray, Michael A.
Saunders, and '1 ,._ i. I H. Wright, "A RangeSpace Method for Quadratic
Programming Problems with Hounds and General Constraints," SOL 8215.
Khosrow Moshirvaziri, "Designing a Reporting System for the
Pilot Model," SOL 8216.
Philip E. Gill, Walter Murray, Michael A. Saunders, and Margaret IH.
I'rI1 "Sparse Matrix Methods in Optimization," SOL 8217.
Patrick II. McAllister and John Stone, "Economic Growth and
Electric (Capacity Availability, SOL 8218.
Lars Mathiesen, "Computational Experience in Solving Equilibrium
Models by a Sequence of Linear Complementarity Problems," SOL 831.
Robert James Wittrock, "Advances in a Nested Decomposition
Algorithm for Solving Staircase Linear Programs, SOL 832,
RIoy. L. Streit, "Solution of Systems of Complex Linear Equations
in the I, Norm with Constraints on the Unknowns," SOL 833.
University of Maryland
College of Business and Management
College Park, Maryland 20742
Gass, "A LifeCycle Approach to Documenting a ComputerBased
Model, 83008.
Bodin, Sexton, "The MultiVehicle Subscriber DialARide Prob
lem," 83009.
Resnick, Trader, "Determining Market Efficiency in the Presence of
New Information," 83010.
Wasil, Golden, Assad, "Alternative Methods for Comparing Heuristic
Procedures on the Basis of Accuracy," 83012.
Shao, "An Investigation Into the Efficacy of the Duaplex Method
for Large Scale and Special Case Linear Programs," 83014.
Assad, Golden, Wasil, "Statistics in Management Science," 83015.
Golden, Stewart, "The Empirical Analysis of TSP Heuristics,"
83016.
Bodin, Duguid, Mitchell, Ball, "The Vehicle Scheduling Problem
With Interlining, 83017.
Gass, Dror, "An Interactive Approach to MultipleObjective Linear
Programming Involving Key Decision Variables, 83021.
Ball, Bodin, Greenberg, "Enhancements to the RUCUS2 Crew
Scheduling System," 83022.
Wasil, Golden, Assad, "Using the Analytical Hierarchy Process to
Rank IIeuristics on the Basis of Accuracy," 83023.
THE JOHNS HOPKINS UNIVERSITY
Electrical Engineering and Computer Science Department
Baltimore, Maryland 21218
J. O'Rourke, "Minimum Convex Covers for Polygons: Some
Counterexamples," 82/1.
W. J. Rugh, "Construction of Minimal LinearAnalytic Realizations
for Polynomial Systems," 82/2.
J. O'Rourke, "A Note on Minimum Convex Covers for Polygons,"
82/3.
J. O'Rourke, "The Decidability of Covering Polygons," 82/3.
G. G. L. Meyer, "Synthesis of Optimization It.. ,in,., by Con
catenating Deterministic and Autonomous Algorithm Components,"
82/7.
G. G. L. Meyer and HI. L. Weinert, "Parallel Algorithms and Com
putational Structures for Linear Estimation Problems, 82/8.
J. O'Rourke, "The .... ir,.' of a Curve and its Applications to
Pattern Recognition, 82/9.
M. Kouta and J. O'Rourke, "East Algorithms for Polygons Decom
position," 82/10.
J. O'Rourke, "Galleries Need Fewer Mobile Guards: A Variation
on Chvatal's Theorem," 82/11.
R. Lejeune and W. J. Rugh, "A Note on Constructing Minimal
LinearAnalytic Realizations for Polynominal Systems, 82/12.
R. Lejeune and W. J. Rugh, "Linearization of Nonlinear Systems
About Constant Operating Points, 82/13.
G. G. L. Meyer, "One Step Dr. ..... It_..,i,,, for the BGM
SystemLevel Fault Model," 82/14.
J. O'Rourke, "An Alternate Proof of thie Rectilinear Art Gallery
Theorem," 82/15.
University of Edinburgh
Department of Business Studies
50 George Square
Edinburgh EH8 9JY
SCOTLAND
R. E. Day and H. P. Williams, "Magic: The Design and Use of an
Interactive Modelling Language for Mathematical Programming," 1982.
M. Cheshire, K. I. M. McKinnon and H. P. Williams, "The Efficient
Allocation of Private Contractors to Public Works," 1983.
H. P. Williams, "A Duality Relationship for Integer Programmes,"
1983.
Econometric Institute
Erasmus Universiteit Rotterdam
P. 0. Box 1738
3000 DR Rotterdam
THE NETHERLANDS
J. K. Lenstra, A. H. G. Rinnooy Kan and P. van Emde Boas, "An
Appraisal of Computational Complexity for Operations Research," 8211/0.
E. L. Lawler, "Generalization of the Polymatroidal Network Flow
Model, 8212/0.
F. J. A. Rijk and A. C. F. Vorst, "Equilibrium Points in an Urban
Retail Model and Their Connection with Dynamical Systems," 8214/M.
A. W. J. Kolen, "Solving Covering Problems and the Uncapacitated
Plant Location Problem on Trees," 8219/0.
A. H. G. Rinnooy Kan, J. R. de Wit and R. Th. Wijmenga, "Non
orthogonal TwoDimensional Cutting Patterns," 8227/0.
W. van Donselaar and G. van der Hoek, "Decision Support to the
Design and Operation of Trailer Dredgers an Application of Lasdon's
GRCCode for NLP," 8229/0.
R. Th. "\ iiir. I._ and G. van der Hoek, "User Manual to Subroutine
Package Wiem for Constrained Nonlinear Programming," 8230/1.
W. Kribbe, "Documentation of the FortranSubroutines for Quad
ratic Programming Conqua and Start," 8231/1.
C. G. E. Boender and A. H. G. Rinnooy Kan, "A Bayesian Analysis
of the Number of Cells of a Multinomial Distribution," 8232/0.
J. B. G. Frenk, M. van Houweninge and A. H. G. Rinnooy Kan,
"Asymptotic Properties of Assignment Problems," 8233/0.
V. Stern, "Minimizing Costs in the Dimensioning of a Telecommu
nication Network with Blocking Probability Constraints an Integer
Nonlinear Programming Approach," 8235/0.
A. Marchetti Spaccamela, A. H. G. Rinnooy Kan and L. Stougie,
"Hierarchical Vehicle Routing Problem." 8236/0.
J. K. Lenstra and A. H. G. Rinnooy Kan, "Two Open Problems in
Precedence Constrained Scheduling," 8237/0.
J. M. Schumacher, "On the Structure of .rr. ...i Controllable
Systems," 8238/M.
A. J. Hoffman, A. W. J. Kolen and M. Sakarovitch, "Totally
Balanced and Greedy Matrices," 8239/0.
W. B. van Dam, J. B. G. Frenk and A. H. G. Rinnooy Kan, "The
Asymptotic Behaviour of a Distributive Sorting Method, 8240/0.
A. H. G. Rinnooy Kan, "The Aggregation of Functional Models:
Technical Aspects, 8242/0.
J. M. Schumacher, "FiniteDimensional Regulators for a Class of
I.,,i,,,i ? ..ii........., System s," 8245/M.
R. E. D. Hcnderiks, "A Sufficient Condition in Optimal Control
Theory," 8246/M.
J. B. G. Frenk, A. II. G. Rinnooy Kan and L. :.... "A Hierar
chical Scheduling Problem with a WellSolvable Second Stage," 8307/0.
Institute of Actuarial Science and Econometrics
Jodenbreestraat 23
1011 NH Amsterdam
THE NETHERLANDS
W. Cook, L. Lovasz and A. Schrijver, "A PolynomialTime Test for
Total Dual Integrality in Fixed Dimension," AE 16/82.
A. Volgenant, R. Jonker and G. Kindervater, "A Note on Finding a
Shortest Complete Cycle in an Undirected Graph," AE 17/82.
A. Schrijver, "MinMax Results in Combinatorial Optimization," AE
18/82.
A. Schrijver, "Total Dual Integrality from Directed Graphs, Crossing
Families, and Sub and Supermodular Functions," AE 19/82.
A. Schrijver, "Bounds on Permanents, and the Number on 1Factors
and 1Factorizations of Bipartite Graphs," AE 22/82.
W. H. M. Zijm, "Generalized Eigenvectors and Sets of Nonnegative
Matrices," AE 26/82.
A. Schrijver, "Packing and Covering of Crossing Families of Cuts,"
AE 1/83.
M. Savelsbergh and A. Volgenant, "Edge Exchanges in the Degree
Constrained Minimum Spanning Tree Problem," AE 2/83.
Universtiy of Wisconsin
Mathematics Research Center
610 Walnut Street
Madison, Wisconsin 53706
O. L. Mangasarian, "A Condition Number for 1',r . ,ii..
Convex Inequalities, "No. 2476.
L. C. Evans and P. E. Souganidis, 1',i. r 'i.,l Games and Repre
sentation Formulas for Solutions of HamiltonJacobiIsaacs Equations," No.
2492.
O. L. Mangasarian, "Normal Solutions of Linear Programs," No.
2498.
Department of Industrial Engineering and Management Sciences
Northwestern University
Evanston, Illinois 60202
R. E. Bixby, W. H. Cunninghan and D. M. Topkis, "The Partial
Order of a Polymatroid Extreme Point."
R. E. Bixby, "Recent Algorithms for Two Versions of Graph
Realization and Remarks on Applications to Linear Programming. "
R. Fourer, "Staircase Matrices and Systems."
R. Fourer, "PiecewiseLinear Programming."
R. Giese and P. C. Jones, "An Economic Model of ShortRotation
Forestry."
W. L. Hsu, "Coloring Planar Perfect Graphs by Decompositions."
W.L. Hsu, "Decompositions of Perfect Graphs."
A. P. Hurter and P. Lederer, "LocationPrice Equilibria for Two
Firms in a Market with Inelastic Demand."
A. P. Hurter and J. Martinich, "Network ProductionLocation
Problems under Price Uncertainty."
P. C. Jones, "A Polynomial Time Market Mechanism."
P. C. Jones, R. Saigal and M. Schneider, "Computing Nonlinear
Network Equilibria."
P. C. Jones, R. Saigal and M. Schneider, "A Variable Dimension
Honmotopy for Computing Spatial Equilibria."
300K REVIEWS
STUDIES ON GRAPHS AND DISCRETE PROGRAMMING
edited by Pierre Hansen
Annals of Discrete Mathematics, Vol. 11
NorthHolland, Amsterdam, 1981
This volume contains 26 papers, which were presented partly at th<
Workshop on Applications of Graph Theory to Management, Brussels
March 1979; partly at the Xth Mathematical Programming Symposium,
Montreal, August 1979.
These Xrays of discrete optimization reveal a particularly healthy
field. There is still room for innovation and for fresh approaches to
traditional applications.
The book shows a good balance between theory and applications:
14 papers are theoretical (even though many of them are algorithmically
oriented); 8 deal with specific applications, including network design,
cuttingstock, location, school timetabling, fault detection, maintenance,
traffic assignment and even sports scheduling, while the four remaining
ones are computational.
Another distinctive feature of the book is its breadth: 8 papers deal
with graphs, while the other ones cover many important areas of discrete
optimization, ranging from submodular functions to totally dual integral
systems, from the geometry of numbers to ordered semigroups (matroids
are the only noticeable absence). As for current research trends, 8 papers
deal with complexity aspects, 5 with polyhedral combinatorics, 2 with
heuristics, 2 with threshold functions.
The basic ingredient of a proceedings volume is the quality of its
individual papers. Here the quality is indeed excellent This, perhaps,
ultimately explains the large success the volume is having on the market
Bruno Simeone, University of Rome
The Theory of Games and Markets, VIII, 554 p
Joachim Rosenmiiller
NorthHolland, Amsterdam, 1981
Rosenmilller's book contains a careful and rigorous treatment of
many parts of modern game theory. Omitted are multiplestage games, the
analysis of the role of information in games, and a treatment of games
modelled as atomless measure spaces of players. All other parts of game
theory are represented. As the title promises some room is given to the
presentation of relations between games and markets.
Although cooperative game theory and characteristic function forms
of games dominate the book, noncooperative games and the normal and
extensive forms are not neglected. On the contrary, a detailed anaylsis of
dynamic game models is given, leading from game trees via dynamic games
to differential games. Though differential games are not represented in the
most general possible framework the use of differential manifolds,
HamiltonJacobiSystems, partial differential equations and measurable
mappings requires a good mathematical background of the reader. The
chapter on twopersonzerosum and nonzerosum games reveals the
connection between game theory and optimization theory. In this context
I consider it useful that a presentation of the LemkeHowson algorithm,
which cannot be found in most treatments of game theory, is included.
Also it is nice that a thorough treatment of those structures needed in the
riii,.l, 11. of individual decision making precedes the proper game theoret
ical part of the book.
For some economists a more extensive economic motivation and
interpretation would have perhaps been helpful. Yet, I recommend this
book as a very good and useful one for students of mathematics as well as
of economics. It provides enough material for several courses in game
theory. I found it useful to partially base my lectures on games and
decision theory at Bonn University on this book.
W. Trockel, Bonn University
Books for review should be sent to the Book Review Editor, Prof Dr.
Achim Bachem, Institut fiir Operations Research, Nassestrasse 2, D5300
Bonn 1, W. Germany.
Algebraic Methods in Graph Theory
Vol. I and Vol. II
edited by L. Lovasz and Vera T. Sos
NorthHolland
AmsterdamOxfordNew York, 1981
The two volumes "Algebraic Methods in Graph Theory" are the
proceedings of a conference of the same title organized by the Attila Jozsef
University and the JSnos Bolyai Mathematical Society in Szeged, Hungary,
August 2431, 1978. The conference intended to give an account of the
stateoftheart in the field. This is reflected in the 850 pages of these
proceedings.
The title chosen for the conference only loosely ties together the
many combinatorial disciplines that are the topics of the various papers.
Since it is impossible here to treat the papers individually (there are 47 of
them!), I will just attempt a general outline of the main themes.
There are two classical approaches to study graphs algebraically.
One may try to study graphs through their automorphism groups and thus
draw from the powerful methods of group theory. On the other hand, the
representation of a graph by its adjacency matrix yields a spectral theory
for graphs. Furthermore, the vertexedge incidence matrix leads to the
investigation of the matroidal properties of a graph. Stopping short before
matroids, chaingroups provide a means to attack, for example factoriza
tion problems of graphs.
Thus the topics include more general combinatorial structures
ranging from relational structures, hypergraphs and matroids over partial
geometries and association schemes (of interest for the Shannon capacity
of a graph) to twographs. In particular, the interplay between graphs and
matroids receives much attention (arboresences, matching, matroid union,
decomposition, etc.) and a study of matroid constructions from the
point of view of exterior algebra is presented.
Other investigations go in the opposite direction and ask for the
realizability of given structures by certain graphs (this problem is com
pletely solved for graphical regular representations of finite groups).
Moreover, the methods of the theory of graphs and related combinatorial
structures are applied to other fields, e.g., number theory and geometry.
In recent years, the algorithmic aspect of combinatorics has become
increasingly important. Several papers deal with the problem to efficiently
construct a combinatorial object and/or to efficiently verify a certain
property (graphicness of matroids, listing the circuits of planar graphs, the
general matroid matching problem).
Of course, in the more than four years since the conference, much
progress has been achieved and many of the results in the papers of the
proceedings have been complemented and extended (to give just one
example: Seymour's decomposition of regular matroids). Some of the
papers mention pertinent recent developments "added in proof." Never
theless, the two volumes will be of interest to every researcher in the field
of graph theory and related areas since they provide survey articles as well
as special research results covering a wide range of theory and applications
and still open problems. If this review whets the appetite of the reader to
take a close look at these proceedings, it has fulfilled its purpose.
Last but not least, the reader should not be too disappointed if he
cannot learn about "the minimization of truth" as promised in the table of
contents. The volumes offer other rewards.
Ulrich Faigle, University of Bonn
The ILLIAC IV
by R. Michael Hord
Springer Verlag
BerlinHeidelbergNew York, 1982
Dr. Slotnik's conception of the first "supercomputer" was origi
nated in the midsixties. The ILLIAC IV was put into operation 1975.
In the shortlived world of computers this belongs to history today. How
ever, the present trend to more parallelism in hardware and software
structures creates new interest in this unique computer. The book is a
P~ 3~
~
comprehensive report of the in'.. ,a_.il,' and implementation activities
presenting a lot of detailed information, analysing hardware and software
problems and explaining success and failures. Computer experts of many
special fields may derive great benefit from it for design and development
of hardware, operation systems, programming environment, user's soft
ware, etc., for unconventional multiprocessor structures. More than
half of the book is dedicated to application problems which have a demand
for high processing power, as for example, in the field of hydrodynamics,
picture processing, seismology, and astronomy. This selection of applica
tion fields, however, is hardly representative for potential users of large
future parallel computers.
G. Fritsch, ErlangenNiirnberg
Modern Applied Mathematics Optimization and Operations Research
edited by Bernhard Korte
NorthHolland, Amsterdam New York, 1982
"Optimization and Operations Research" was the title of a summer
school organized by the Institute of Operations Research of the University
of Bonn, and the 17 StateoftheArt articles of this volume are basically
the written versions of the lectures. The volume is divided into seven parts
1. !o.. with the following topics: Foundations, Convex Analysis, Poly
hedral Theory, Complexity, Nonlinear Programming, Control and Approx
imation Theory and Numerical Analysis, Combinatorial Optimization,
Game Theory, Statistics, and Economics.
All articles are of high quality some are excellent from the didac
tical point of view. However most of the monographic articles address
readers who have at least a basic knowledge of the area. This book is
1. i ..Ii. I a welcome contribution to the literature on "applicable" mathe
matics and should be of utmost interest and use to everybody who is
working in fields like mathematical economics or operations research, and
who is willing to discuss and apply modern mathematical methods. More
over the book is ideally suited to serve as background text for teaching or
the basic reference for a seminar on modern applied mathematics.
l.Derigs, University of Bonn
JOURNALS & STUDIES
Volume 27, No.1
D. Goldfarb and A. Idnani, "A Numerically Stable Dual Method for
Solving Strictly Convex Quadratic Programs."
M. L. Overton, "A Quadratically Convergent Method for Minimizing
a Sum of Euclidean Norms."
R.D. Armstrong and P.O. Beck, "The Best Parameter Subset Using
the Chebychev Curve Fitting Criterion."
D. F. Karney, "A Duality Theorem for SemiInfinite Convex
Programs and Their Subprograms."
J. G. Ecker and M. Kupferschmid, "An Ellipsoid Algorithm for
Nonlinear Programming."
D. P. Bertsekas, "Distributed Asynchronous Computation of Fixed
Points."
Volume 27, No. 2
Ge Renpu and M. J. D. Powell, "The Convergence of Variable
Metric Matrices in Unconstrained Optimization. "
R. G. Jeroslow, 't nti. r,, Duality in SemiInfinite Convex Optimi
zation."
A. I....I i and A. LeNir, "QNLike Variable ti...... Conjugate
Gradients."
T. i.,Ihi... "Local and Superlinear Convergence for Truncated
Iterated Projection Methods."
R. W. Cottle and R. E. Stone, "On the Uniqueness of Solutions to
Linear Complernentarity Problems."
J. B. Orlin, "MaximumThroughput Dynamic Network Flows."
(Journal contents are subject to change by publisher)
CONFERENCE NOTES
NATO Advanced Study Institute on
Computational Mathematical Programming
July 23 to August 2, 1984
Bad Windsheim, Germany F.R.
The Committee on 'i....il.... (COAl,) of the Mathematical Pro
gramming Society announces a summer school on Computational Mathe
matical .... .......... to be held in Bad Windsheim, West Germany, from
July 23 to August 2, 1984, under the sponsorship of NATO (Advanced
Study Institute). The ASI consists of tutorials with the emphasis on new
mathematical programming algorithms, software products, computational
experiments, numerical test results, and practical optimization models.
The following topics will be treated: large linear systems; integer pro
gramming; model building in linear and integer programming; networks;
nonlinear programming; model building and computational aspects in
nonlinear programming: large scale nonlinear programming; geometric
programming; nondifferentiable optimization; global optimization;genera
tion of test problems. experimental design and comparative performance
evaluation; optimal control; stochastic optimization: parallel ..... ** I;,
Some time will be reserved for participants of the ASI to lecture about
their own research activities. A limited fund is available for participants
from NATOcounlries to cover a part of tie travelling and accommodation
costs. Participation is possible only by invitation. I'or more information
and an application fornn. write to Klaus Schittkowski, Institut fiir Iifor
matik, Univcrsitat Stuttgart. Azenbergstr. 12, I)7000 Stuttgart 1, Germany
I1.i.
The Fourth Mathematical
Programming Symposium, Japan
IThis annual symposium will be held on November 1415, 1983 at
the International Conference Center Kobe, Kobe, Japan. The symposium
will consist of the ii.. ,,.. three sessions:
1. Mathematical programming, general. Chairman: M. Kojima. 2.
Markov Decision Process. Chairman: K. Sawaki. 3. Applications. Chair
mal: T. Morildyo.
The first two sessions will consist of three or four talks of expos
itory nature and those presenting original development. No contributed
paper will be called for, and only invited papers will be presented.
Participation from abroad will be welcomed. The conference
language is Japanese but nonJapanese participants may use English.
For further information contact Organizing Chairmnu, Professor
Masao Iri, Faculty of Engineering, University of Tokyo, Bunkyoku, Tokyo
113, Japan, or Program Chairman, Professor R. Manabe, Kobe University
of Commerce, Tarumi, Kobe 655, Japan.
R. Manabe
GALLIMAUFRY
Herbert E. Scarf (Yale) has been awarded the John von Neumann Theory Prize by
ORSA/TIMS. The award cited Professor Scarf's contributions in the computation of fixed
points, in (s,S) policies in inventory theory, in balanced games, and in general equilibrium
theory in economics. . .Haverly Systems, Inc. of Denville, N.J. has received an award for
"excellence in exporting" from U.S. President Reagan. The company's major software
products include linear :.i. ... iiiiiinI. systems and other packages designed to solve optimiza
tion problems in a variety of applications. . Ubaldo Garcia Palomares (Univ. Simor'
Bolivar) is visiting Argonne National Laboratory from June 15 to December 15, 1983. ..A
workshop on "Algorithms and Software for Nonlinear Optimization" will be held Sept.
2123, 1983 in Cetraro, Italy. Contact Mrs. Chiara Zanini, Via Bernini, n.5, 87036
Qi ali..n.;.Ii., di Rende (Cosenza) Italy, Tel. i11':: 1,, 839711839738 ... New MPS dues for
I'!: 1 have been announced: 32 US dollars, 20 UK pounds, 68 Swiss francs, 250 French
francs, 84 German marks, or 94 Dutch guilders.
The Discrete Applied Mathematics journal is preparing a special issue on network
algorithms and applications. I'1. I submit papers (deadline was August 15, I'1 : .), to Profes
sor Darwin Klingman, David Burton Jr. Centennial Chair in Business Decision Analysis,
Department of General Business, (SB 4.138, University of Texas, Austin, Texas 78712. All
contributions will be tl.. ...i. .1 refereed.
Deadline for the next (: I' I i is December 1, 1983.
This public document was promulgated at a cost of $426.1.5 or
$0.61 per copy to inform researchers in mathematical programming of
recent research results.
Ci VI N
303 Weil Hall
College of Engineering
University of Florida
Gainesville, Florida 2''611 FIRST CLASS MAIL
Princeton University
School of Engineering & Applied
Science
Princeton, NJ 08544
CALE N DAR
Maintained by the Mathematical Programming Society (MPS)
This Calendar lists noncommercial meetings specializing in mathematical programming or one of its
subfields in the general area of optimization and applications, whether or not the Society is involved.
(The meetings are not necessarily 'open'.) Any one knowing of a meeting that should be listed here is
urged to inform Dr. Philip Wolfe, IBM Research 33221, POB 218, Yorktown Heights, NY 10598,
U.S.A; Telephone 9149451642, Telex 137456.
Some of these meetings are sponsored by the Society as part of its worldwide support of activity
in mathematical programming. Under certain guidelines the Society can offer publicity, mailing lists and
labels, and the loan of money to the organizers of a qualified meeting.
Substantial portions of meetings of other societies such as SIAM, TIMS, and the many national OR
societies are devoted to mathematical pr. :!i.nmmiii and their schedules should be consulted.
1983
September 2123: International Workshop on Algorithms and software for nonlinear optimization,
Cetraro (Cosenza), Italy. Contact: Mrs. Chiara Zanini, CRAI, Via Bernini No. 5, 87036
Quattromiglia di Rende (Cosenza), Italy. Telephone (0984) 938711.
1984
June 1214: SIAM Conference on Numerical Optimization, Boulder, Colorado, U.S.A. Contact: Hugh
B. Hair, SIAM Services Manager, 1405 Architects Iltiildi., 117 South 17 Street, Philadelphia,
Pennsylvania 19103, U.S.A. Telephone 2155642929.
July 23 August 2: NATO Advanced Study Institute on Computational Mathematical P i i.IIniin.:
Bad Windsheim, Federal Republic of Germany. Contact: Dr. Klaus Schittkowski, Institut fir
Informatik, Azenbergerst. 12, 7000 Stuttgart 1, Federal Republic of Germany. Telephone
0711 2078 335. Sponsored by the Society through the Committee on Algorithms.
1985
August 59: Twelfth International Symposium on Mathematical Programming in Cambridge, Massachu
setts, U.S.A. Contact: Professor Jeremy Shapiro, Sloan School of Management, Massachusetts
Institute of Technology, Cambridge, MA 02139, U.S.A. Telephone 6172537165. Official
triennial meeting of the MPS.
