Upload
agusdn21
View
215
Download
0
Embed Size (px)
Citation preview
8/8/2019 Optima 52
1/16
GOMATHEMATICALPROGRAMMINGSOCIETYNEWSLETTER
NO52
5 2
EMBER1996
N U M B E R
SEE PAGE TWO
1. Global Optimization Models and Solution Approaches
A large variety of qu antitative d ecision issues, arising in the sciences, engineering an d
economics, can be perceived and m odelled as a constrained op timization pr oblem. Ac-
cording to t his generic description, the best decision often expressed by a real vector
is sought wh ich satisfies all stated feasibility constraints an d m inimizes (or maximizes)
the value of an objective function. Applying standard mathematical programming nota-
tion, we shall consider problems in th e general form
(1) minf(x) subject to x D Rn.
The functionfsymbolizes the objective(s) in th e decision prob lem, and D denotes the
(non-empty ) set of feasible decisions.D is usually defined by a finite nu mber of func-
tions; for the pu rposes of the present discussion, we shall assume that
(2) D ={ x D Rn: l xu; g
j(x) 0 j =1,...,J}.
In (2) l an d u are explicit (finite) boun ds, and gj are given constraint functions. Postulat-
ing now that all fun ctions defined abov e are continu ous, the optimal solution set to
problem (1)-(2) is non -empty.
Most typically, it is assumed th at the d ecision problem m odelled by (1)-(2) has a u nique
locally and , at the sam e time, also globally optimal solution. Un iextremality is often
implied by t he math ematical model structur e (for example, by the strict convexity off,
and the convexity ofD). This paradigm correspond s to the situation in which one, sup-
posed ly, has a su fficiently close initial guess of the feasible region wh ere the op timal
solution x* is located. Hence, the global optim ality of the solution d irectly follows, hav-
ing found the single local optimum offon D. For examp le, linear and convex nonlinear
progr amm ing mod els both, in essence, satisfying the men tioned
Continuous
Global OptimizationSoftware:A Brief Review
AbstractFollowing a concise introduction to multiextremal mathematical programming
problems and global optimization (GO) strategies, a commented list of software
products for analyzing and solving continuous GO problems is presented.
Keywords: Multiextremaloptimization models; continuousglobal optimization; solutionapproaches; software review.
AMS Subject Classification:65K30, 90C05.
conference notes 10
book reviews 12
journals 15
gallimaufry 16
Jnos D. PintrPintr Consulting Services(PCS), and Dalhousie Uni-versity
PCS address: 129Glenforest Drive, Halifax,
NS, Canada B3M 1J2
e-mail: [email protected]://www.tuns.ca/~pinter/
O P T I M A
8/8/2019 Optima 52
2/16
O P T I M A
NO52PAGE 2 DECEMBER 1
un iextremality assump tion in most p ractical cases have been exten-
sively applied in the past decades to formulate and solve an impressive
range of decision problems.
Although very important classes of models naturally belong to theabove category, there is also a broad variety of problems in w hich the
property of uniextremality cannot be simply postulated or verified.
Consider, for instance, the following general problem ty pes:
nonlinear app roximation, including the solution of systems of nonlin-
ear equations and inequalities
mod el fitting to empirical data (calibration, param eterization)
optimized design an d operation of complex black box (oracle) sys-
tems, e.g., in d iverse engineering contexts
configurat ion/ arrangem ent design (e.g., in various data classification,
facility location, resource allocation, or scientific modelling contexts)
Such problems together with numerous other prospective application
areas are d iscussed by Pint r (1996) and in th e extensive list of related
references therein. For furth er ap plications consult, e.g., Pardalos an d
Rosen (1987), Trn an dv
Zilinskas (1989), Floudas and Pardalos (1990,
1992), Grossmann (1996), Bomze, Csendes, Horst and Pardalos (1996),
or sp ecial app lication-related issues of the Journal of Global Op timiza-
tion.
The emerging field of Global Optimization (GO) deals with m athemat i-
cal programm ing pr oblems, in the (possible) presence of multiple local
optim a. Observe that, typically, the nu mber of local (pseud o)solutions is
un know n and it can be quite large. Furtherm ore, the quality of the vari-
ous local and global solutions may d iffer significantly. In the presence
of such structu re often visualized by hilly land scapes correspond ing
to p rojections of the objective function into selected sub spaces (given by
coordinate-pairs of the d ecision variablex) GO problems can be ex-
tremely d ifficult. Hence, most classical nu merical app roaches are, gen-
erally speaking, not d irectly app licable to solve them. For illustra-tion, see Figure 1 wh ich displays a relatively simple compo-
sition of trigonometric functions with imbedded poly-
nom ial argument s, in just tw o variables (denot ed
by x and y).
general mod el form (1)-(2), and note th at the p roblem-classes listed be
low are not n ecessarily distinct; in fact, several of them are h ierarchi-
cally contained by m ore general problem-typ es listed.)
Bilinear and biconvex programming (fis bilinear or biconvex,D isconvex)
Combinator ial optimization (problems w hich have discrete decision
variables infand/ or in gj
can be equivalently reformu lated as GO prob
lems in continu ous variables)
Concave minimization (fis concave,D is convex)
Continuous global optimization (fis continuous,D is compact)
Differential convex (D.C.) optimization (fan d gj
can all be represente
as the d ifference of two correspond ing convex functions)
Fractional programming (fis the ratio of two real functions, and gjar
convex)
Linear and nonlinear complementarity problems (fis the scalar prod
uct of two v ector fun ctions,D is typ ically convex)
Lipschitz optimization (fan d gj
are arbitrary Lipschitz-continuou s
functions) Minimax problems (fis some minim ax objective, the maximu m is co
sidered ov er a discrete set or a convex set,D is convex)
Multilevel optim ization (mod els non-cooperative games, involving h
erarchies of decision-makers, their conflicting criteria are agg regated b
f;D is typically assum ed to be convex)
Multiobjective program ming (e.g., wh en several conflicting linear ob
jectives are to be optimized ov er a polyhed ron)
Multiplicative programming (fis the prod uct of several convex func-
tions, and gj
are convex, or more generally m ultiplicative functions
GOCONTINUED
Figure 1. A multiextremal function in two variables
Naturally, under su ch circum-
stances, it is essential to use a
prop er global search strategy. Fur-
thermore, instead of exact solu-
tions, most typ ically one has to a c-
cept diverse numerical approxi-
mations to the globally optimal so-
lution (set) and optimu m value.
Following early sporad ic work re-
lated to GO (since the late fifties),
the p resent state-of-the-art is char-
acterized by several dozen mono-
grap hs, a professional journ al and
at least a few thousan d research
articles devoted p rimarily to the
subject. A few illustrative refer-
ences are provided at the end of
this brief review.
The most important GO model-
classes wh ich have been exten-
sively studied includ e the follow-
ing exam ples. (Please recall the
8/8/2019 Optima 52
3/16
O P T I M A
NO52PAGE 3 DECEMBER 1
Network problems (fcan be taken from several function-classes in-
cluding nonconvex ones, and gj
are typ ically linear or convex)
Parametric nonconvex program ming (the feasible regionD a nd / o r
the objectivefmay also depend on a param eter vector)
Quadratic optimization (fis an arbitrary ind efinite quad ratic fun c-
tion; gj
are linear or, in the more g eneral case, can be arbitrary qu adra tic
functions)
Reverse convex programm ing (at least one of the functions gj
ex-
presses a reverse convex constraint)
Separable global optimization (fis an arbitrary non linear in general,
nonconv ex separable function,D is typ ically convex)
Various other nonlinear programming problems, includ ing, e.g.,
nonconv ex stochastic models (in which the d efining fun ctionsf, gj
de-
pend on ran dom factors, possibly in an imp licit, black box man ner)
For detailed descriptions of most of these mod el-types an d th eir connec-
tions consult, e.g., Horst and Pardalos (1995), with nu merou s further
references.
There are several main classes of algorithmic GO app roaches wh ichpossess strong th eoretical convergence properties, and at least in prin-
ciple are straightforward to implement and apply. All such rigorous
GO approaches have an inherent compu tational demand which in-
creases non-polynom ially, as a function of problem-size, even in th e
simplest GO instances. It should be emp hasized at this point th at GO
app roaches are (should b e) typically completed by a traditional local
optim ization phase at least when consid ering also num erical effi-
ciency issues. Global convergence, however, need s to be gu aranteed by
the global-scope algorithm compon ent w hich theoretically should be
used in a comp lete, exhaustive fashion. These remarks in dicate the sig-
nificant d ifficulty of developing robu st and efficient GO software.
Withou t aiming at completeness, several of the most imp ortant GO
strategies are listed below ; for details, consult, for instan ce, the corre-
spon ding w orks from the list of references. (Note that the listing is not
complete, and its items are no t necessarily mutally exclusive; some soft-
ware implementations combine ideas from several approaches.)
Adap tive partition and search strategies (includ ing, e.g., branch-and -
bound algorithms, Bayesian approaches and interval arithmetic based
method s) (Forg , 1988; Ratschek an d Rokn e, 1988; Mockus, 1989;
Neu maier, 1990; Zhigljavsky, 1991; Han sen, 1992; Horst an d Pardalos,
1995; Hor st an d Tuy, 1996; Pintr, 1996; Kearfott, 1996)
Adap tive stochastic search algorithm s (including rand om search,
simulated a nnealing, evolution and genetic algorithm s) (van Laarhoven
and Aarts, 1987; Zhigljavsky, 1991; Horst and Pardalos, 1995;
Michalew icz, 1996; Pintr , 1996)
Enum erative strategies (for solving combinatorial pr oblems, or certain
structured e.g., concave optimization problems) (Forg, 1988; Horstand Pardalos, 1995; Horst and Tuy, 1996)
Globalized local search method s (applying a grid search or rand om
search type global pha se, and a local search algorithm) (Horst and
Pard alos, 1995; Pintr, 1996)
Heu ristic strategies (deflation, tun neling, filled fun ction methods,
app roximate convex global un derestimation, tabu search, etc.) (Horst
and Pardalos, 1995; Pintr, 1996)
Homotopy (parameter continuation) methods and related approaches
(includ ing fixed p oint meth ods, pivoting algorithm s, etc.) (Horst and
Pardalos, 1995)
Passive (simultaneou s) strategies (un iform grid search, pure rand om
search) (Zhig ljavsky, 1991; Horst an d Pa rd alos, 1995; Pintr , 1996)
Successive approximation (relaxation) method s (cutting p lane, more
general cuts, minoran t construction app roaches, certain nested optimi-
zation and decomp osition strategies) (Forg, 1988; Horst a nd Pardalos
1995; Pintr, 1996)
Trajectory methods (differential equat ion mod el based, path-
following search strategies) (Horst and Pardalos, 1995)
In spite of a considerable pro gress related to the rigorous th eoretical
foundations of GO, software development and standardized use lag
behind . The m ain reason for this is, of course, the inheren t nu merical
difficulty of GO, even in the case of simpler specific instances (such a
the ind efinite quad ratic programm ing pro blem). In general, the diffi-
culty of a global optimization p roblem (GOP) can be expected to in -
crease as some exponen tial fun ction of the prob lem dim ension n. Con
sequently, d imensions 100, 50 or even 10 can be considered as large,
depend ing on the GOP typ e investigated. In the remaind er of this pa-
per, an illustrative list of software prod ucts to solve GOPs is reviewed
2. GO Software:Information Sources and Some General Remarks
For the pu rposes of collecting informat ion for this survey, GO softwar
auth ors have been asked (mainly by sendin g e-mail messages, and by
placing electronic ads at several prominen t math ematical program -
ming sites on the WWW) to submit docum entation related to their
work . The information or lack thereof sum marized below is largely
based on the respon ses received. Add itional information has been col
lected from th e Internet, from several GO books, and from the Journal
of Global Optim ization. Note that thou gh in many research publica-
tions reference is mad e to nu merical examp les, or even to sop histicated
specific app lications, only such work is reported below w hich is under
stood to be a general purp ose and legally distributable program system
For obvious reason s, the pr esent surv ey is far from being complete in
any p ossible sense; rather, it is an attempt to p rovide a realistic picture
of the state-of-the-art, supp orted by in stances of existing software. Th
short review is not inten ded to be either comparative or jud gemental
one simp le reason being that the information r eceived from GO soft-
ware d evelopers is used as is, mostly withou t the p ossibility of actual
software testing. By the sam e token, the accuracy of all information
cannot be guarant eed either. Further research in this direction
includ ing the p reparation of a more comprehensive and d etailed surve
is currently in p rogress.
The software list provided in the next section is simply alph abetical,
withou t categorization. For a more un iform presentation style, abbre-
viations are associated with all software prod ucts listed, even wh ensuch nam es were not given in the docum entation available for this sur
vey (existing names were not chang ed, of course). The descriptions ar
almost formu la-free and extremely concise du e to spa ce restrictions.
For the latter reason, we decided not t o include imp ortant classes of
more specific GO appr oaches and related meth odology. In particular
as reflected by the title pure or mixed integer prog ramm ing and mor
general combinatorial optimization algorithms are not discussed here.
Furtherm ore, althou gh m ost of the available top-of-the-line continu ous
nonlinear (convex) optim ization software can be applied with good
taste and som e luck to analyze GOPs, even the most p rominen t such
systems are excluded from this review. Again, a more detailed sur vey
is planned, app ropriately d iscussing also the program system types
mentioned.
GOCONTINUED
8/8/2019 Optima 52
4/16
O P T I M A
NO52PAGE 4 DECEMBER 1
The hardw are and softwar e platform of the systems reviewed is also
show n wh en such information is available. In order to assist in obtain-
ing ad ditional information, contact person(s), their e-mail add resses, ftp
and/ or WWW sites are listed, whenever known to me. (For brevity,
only a few such p ointers are provid ed in each case.)
The reader is assumed to have at least some basic familiarity with the
GO app roaches mentioned ; for related d iscussions, please consu lt the
references.
3. Short Software Descriptions
BB A GO Algorithm for General Nonconvex Problems
An im plemen tation of a Branch-and-Boun d (B&B) algorithm w hich is
based o n th e difference of convex functions (D.C.) transformation.
Non convexities are identified and categorized as of either special or ge-
neric structure. Special nonconv ex (such as bilinear or u nivariate con-
cave) terms are convex lower bounded using customized bound ingfunctions. For generic nonconvex terms, convex lower bound ing func-
tions are derived by utilizing the parameter (specified by the u ser or
derived based on theory). BB solves general unconstrain ed an d con-
strained p roblems; it requires MINOS and/ or NPSOL for the solution of
linear or convex optimization subp roblems. (Langu ages: C and Fortran.)
Contact: I.P. And roulakis , C.A.
Floudas , C.D. Maranas
, http:/ / titan.princeton.edu / .
ANNEAL Simulated Annealing
ANN EAL is based on th e core SA app roach, includ ing several possibili-
ties for parameter adjustment and a deterministic solution refinement
ph ase. It has been app lied to predict comp lex crystal structur es. Work-
station imp lementation. Contact: W. Bollweg , H. Maurer , H. Kroll
.
ASA CalTech Adapt ive Simulated Annealing
ASA was developed to find the global optimum of a continuous n on-
convex function over a m ultidim ensional interval (box). This algorithm
perm its an annealing schedule for temp erature decreasing exponen -
tially in ann ealing time. The introd uction of re-ann ealing also permits
adap tation to changing sensitivities in the param eter-space. Some other
adap tive options in ASA includ e self-optimize (to find op timal starting
conditions) and qu enching (to method ically find faster performan ce
that m ight be u seful for large param eter-spaces). (Language: C.) Con-
tact: L. Ingber , http:/ / ww w.ingber.com/
#ASA-CODE.
B&BA Family of B&B Algorithms
This obvious acronym (by the p resent au thor) attempts to sum marize
several B&B type algorith ms d eveloped to solve certain stru ctured GO P
classes. These includ e (among others) indefinite quad ratic, quasiconvex-
concave, and general Lipschitz problems. Workstation imp lementa-
tions . (Langu age: C.) Contact: R. H orst , M.
Nast , N . Thoai .
BARON Branch-And-Reduce Optimization Navigator
Combines interval analysis and du ality with enh anced B&B concepts.
The BARON mod ules can han dle structured nonconvex problems u p to
thou sand s of constraints and variables. The library of specialized mod -
ules includes solvers for nu merou s specific GOP-classes. (For other,
more general problems, underestimation routines need to be p rovided
by the u ser.) All modu les can solve also such problems in w hich some
or all of the variables are restricted to integer values. The specialized
mod ules use OSL or MINOS to solve interim subp roblems. Worksta-
tions, UNIX type op erating systems. (Languages: Fortran and GAMS.)
Contact: N .V. Sahinidis, http:/ / archimedes.me.uiuc.edu/ sigma/
baron.html, ftp:/ / aristotle.uiuc.edu .
BGOBayesian Global Opt imizat ion
This program system includ es four versions of Bayesian search, cluster
ing, uniform determ inistic grid, and p ure Monte Carlo search. Boun d
constraints and m ore general constraints can be hand led. Interactive
DOS and UNIX versions are available. (Languag es: Fortran and C.)
Contact: J. Mockus , L. Mockus
.
cGOP Global Opt imizat ion Program
Solves structu red GOPs w hich have an objective fun ction of the form
a
T
x+b
T
y+x
T
Ay+f1(x)+f2(y) with convexf1,f2, and linear constraints. Re-quires the p resence of the commercial codes MINOS and / or CPLEX to
solve linear, mixed-integer linear and convex subp roblems. cGOP has
been used to solve problems involving several hundred variables and
constraints. Versions are available for work stations. (Langu age: C.)
Contact: V. Viswesw aran , C.A. Floudas
, http:/ / titan.princeton.edu / .
CGU Convex Global Underestimator
This approa ch is designed to generate efficient ap proximations to th e
global minimu m of a mu ltiextremal function, by fitting a convex func-
tion to th e set of all know n (calculated ) local minima. This heuristicall
attractive strategy requires only th e sequential solution of au xiliary LP
and som e rather elementar y calculations. CGU has been applied to ca
culate molecular structure pred ictions, up to several dozen var iables.
Implemented on parallel workstations and supercomputers. Contact:
K.A. Dill, A.T. Phillips , J.B. Rosen
.
CRS Controlled Random Search
This is a recently developed v ariant of a pop ular class of rand om searc
based methods which can be applied un der very m ild analytical condi
tions imposed on the GOP. Several other related stochastic search
methods h ave also been d eveloped by this group . Workstation imple-
men tation s. Conta ct: M.M. Ali, A. Trn , S.
Viitanen .
CURVI Bound-Constrained Global O ptimiz ation
Windward Technologies (WTI) develops advanced numerical and visu
alization software, for solving constrained and unconstrained nonlinea
optim ization problems. On e of their solvers, CURVI is aimed at solvin
bound -constrained nonlinear programs w hich h ave a complicated
possibly m ultiextremal objective fun ction. (Languag e: Fortran.) Con-
tact: T. Aird , http:/ / users.aol.com/ WTI/ .
DE Differentia l Evolut ion Genetic Algorithm for Bound Con-strained GO
DE won third place at the 1st International Contest on Evolutionary
Comp utation on a real-valued function test set. It was the best genetic
algorithm ap proach (the first two places of the contest were w on by
non-GA algorithm s). (Langu ages: Matlab and C.) Contact: R. Storn
, http :/ / http .icsi.berkeley.edu / ~storn/
code.html.
GOCONTINUED
8/8/2019 Optima 52
5/16
O P T I M A
NO52PAGE 5 DECEMBER 1
ESA Edge Searching Algorit hm
An imp lementation of an edge search algorithm for finding th e global
solution of linear reverse convex p rogram s. ESA is based on an efficient
search technique and the use of fathoming criteria on the edges of thepolytop e representing the linear constraints. In addition, the method
incorporates several heuristics, including a cutt ing plane techniqu e
wh ich impro ves the overall performance. Implement ed for several
UNIX platforms; the TPG Test Problem Generator is also a vailable.
(Langu age: Fortran.) Contact: K. Moshirvaziri
, .
GA Geneti c Algorithms
Genetic algorithms as a rule can be applied to GOPs u nd er mild
structur al requirements. Both general and sp ecific information related
to this p opu lar solver class is available from the following sources: A
Comm ented List of Genetic Algorithm Cod es, ftp:/ /
ftp.germany.eu.net/ pub/ research/ softcomp/ ec/ faq/ www / q20_1.htm
GA Archive, http:/ / ww w.aic.nrl.navy.mil/ galist/ src/ . Only a few il-
lustrative examples are listed in the pr esent review.
GAS Geneti c Algorithm
Unconstrained and bound-constrained versions are available. For DOS
and UNIX operating systems. (Language: C++.) Contact: M. Jelasity
, J. Dombi , ftp:/ /
ftp.jate.u-szeged.hu/ pub/ math/ optimization/ GAS/ .
GAucsd Geneti c Algorithm
Developed an d m aintained at t he University of California, San Diego.
GAucsd was w ritten in C u nder Unix but should be easy to port to
other platforms. The package is accompan ied by brief information and
a Users Guide. (Langu age: C.) Contact: nici@ucsd .edu , GAucsd-
[email protected], ftp:/ / cs.ucsd.edu/ pub/ GAucsd/ .
GENERATOR Geneti c Algorithm SolverThis method is aimed at solving a variety of (combinator ial and con-
tinuou s multiextremal) scientific and engineering op timization prob-
lems. It is designed to interact with Excel which serves as a user inter-
face. (Platform: Excel.) Contact: New Ligh t Ind ust ries
, http:/ / ww w.iea.com/ ~nli/ .
GC Global Continuation
GC is a continuation ap proach to GO app lying global smoothing in or-
der to d erive a simpler app roximation to the original objective function.
GC is app lied by the author s to distance geometry problems, in the con-
text of molecular chemistry mod elling. IBM SP parallel system imple-
men tation . Contact: J.J.Mor , Z. Wu.
GENOCOP III Genetic A lgorithm for Constrained Problems
Solves general GOPs, in the p resence of add itional constraints and
boun ds (using quad ratic pena lty terms). System param eters, dom ains,
and linear inequalities are inp ut via a d ata file. The objective function
and any non linear constraints are to be given in approp riate C files.
(Langu age: C.) Contact: Z. Michalewicz, http :/ / ww w.coe.un cc.edu/
~zbyszek/ gcreadme.html, ftp:/ / ftp.uncc.edu/ coe/ evol/
genocopIII.tar.Z.
GEODES Minimum-Length Geodesic Computing
Approximating a m inimum-length geodesic on a multidimensional
man ifold, GEODES is differential geometry software. How ever, it has
poten tial also in th e GO context. GEODES includes examp le manifolds
and metrics; it is imp lemented in Elements (a matrix and function ori-
ented scientific modelling environmen t) to comp ute and visualize geo
desics on 2D surfaces plotted in 3-space. Portable to various hard ware
platforms. (Langua ges: C, C++.) Contact: W.L. Anderson
, http:/ / ww w.netcom.com/ ~elements/ ,
http:/ / www.netlib.org/ ode/ geodesic/ .
GLO Global and Local Optimiz er
GLO is a modu lar optimization system developed for black box prob
lems in w hich objective function calculations may ta ke a long time. Its
method ology is based on the coup ling of global (genetic) and local
(variable metric) nonlinear op timization software with scientific app li-
cations software. It has been applied to automated en gineering design
Besides the mod ular op timization control system, GLO also has a
grap hical user interface and includ es a pre-processor. Contact: M.J.
Murph y, http:/ / ww w.llnl.gov/ glo/ 09glo.html, M. Brosius
.
GLOBAL Multist art w ith Stochastic Clustering
GLOBAL can be used for th e solution of the general bou nd -constrainedGOP wh ich has a (measurable) real objective fun ction. The algorithm
a derivative-free imp lementation of the clustering stochastic mu ltistart
method of Boender et al., supplemented with a quasi-Newton local
search routine and with a robust rand om local search method . Avail-
able for UNIX machines, IBM-comp atible mainframes an d PCs. (Lan-
guag es: Fortran and C.) Contact: T. Csendes , http:/ / www .inf.u-szeged.hu/ ~csendes/ , ftp:/ / ftp.jate.u
szeged.hu/ pub/ math/ optimization/ index.html.
GLOBALIZER An Educational Program Syst em for Global O ptmization
Serves for solving un ivariate GOPs. After stating th e problem, the u ser
can choose amon g variou s (rand om search, B&B based, or Bayesian pa
tition based) solver techniques. The softwar e has interactive tutoring
capabilities, prov ides textual and grap hical information. Works on PC
un der MS-DOS. Contact: R.G. Strongin ,
V.P. Gergel, A.V. Tropichev .
GLOPT Constrained Global Opt imizat ion
Solves GOPs w ith a block-separable objective fun ction subject to boun
constraints and block-separable constraints; it find s a nearly globally
optim al point that is near to a tru e local minimizer. GLOPT uses a B&
technique to split the problem recursively into subp roblems that are ei
ther eliminated or redu ced in their size. It includes a new red uction
technique for boxes and n ew w ays for generating feasible points of con
strained nonlinear progra ms. The current implemen tation of GLOPT
uses neither derivatives nor simultaneous information about several
constraints. (Langu age: Fortran .) Contact: A. Neu maier
, S. Dallw ig and H. Schichl.GOPP Global Opt imizat ion of Polynomial Problems usingGrbner Bases
The (local) optim ality cond itions to polynom ial optimization pr oblems
lead to polyn omial equations, und er inequality constraints. App lying
recent Grbner basis techniques, this approach is aimed at finding all
solutions to such system s, hence also finding global optima. (Language
Map le.) Cont act: K. Ha gglof , P.O. Lind berg
, L. Svensson , http:/ /
www.optsyst.math.kth.se.
GOCONTINUED
8/8/2019 Optima 52
6/16
O P T I M A
NO52PAGE 6 DECEMBER 1
GOT Global Optimiz ation Toolbox
GOT combines rand om search and local (convex) optimization. DOS
and HP UX versions are available. (Langu age: Fortran.) Contact: A.V.
Kun tsevich .GSA Generalized Simulated Annealing
GSA is based on the generalized en tropy by Tsallis. The algorithm
obeys detailed balance conditions and , at low temp eratures, it redu ces
to steepest descent. (Note th at the mem bers of the same research grou p
have been involved in th e developm ent of several SA type algorithms.)
Cont act: J.E. Strau b , P.Ama ra
, J. Ma .
IHRImproving Hit- and-Run
IHR is a rand om search based GO algorithm th at can be used to solve
both continuou s and discrete optimization prob lems. IHR generates
random points in the search d omain by choosing a random direction
and selecting a point in that d irection. Versions have been imple-
mented , using different distribu tions for the random d irection, as wellas several ways to r and omly select points along th e search line. The al-
gorithm can also hand le inequ ality constraints an d a h ierarchy of objec-
tive functions. IHR has been u sed to solve GOPs in various d isciplines
such as in en gineering d esign. Contact: Z. Zabinsky
, ftp:/ / ftp.bart.ieng.washington.edu .
IMINBIS Interval Arithmeti c Based GO
This method app lies interval arithmetic techniques to isolate the station-
ary p oints of th e objective function. Next, a to pological characterization
is used to separate minima from m axima and saddle points, followed by
local minimization (sub)searches to select the global solu tion. The
method has been app lied also to noisy problems. Workstation and PC
implem entations, extensive related research. (Langu age: Fortran.) Con-
tact: M.N. Vrahatis , D.G.
Sotiropou los , E.C. Triantaph yllou.
INTBIS Global Solver for Polynomial Systems of Equations
Finds all solutions of polynomial systems of equations, with rigorou sly
guar anteed r esults. The software package INTBIS is ACM-TOMS Algo-
rithm 681; it is available throug h N ETLIB. Distributed w ith the package
are four source code files, samp le input an d ou tpu t files, and a brief
docum entation file. The source files consist of the following : interval
arithmetic, stack managem ent, core INTBIS routin es, and m achine con-
stan ts. (Langu age: Fortran .) Cont act: R.B. Kearfott ,
http:/ / interval.usl.edu/ kearfott.html, ftp:/ / interval.usl.edu/ pub/
interval_math/ intbis/ .
INTOPT_90 Verified (Interval) Global Optimization
Serves to the verified solution of nonlinear systems of equations and un -constrained and bound-and-equality-constrained global optimization.
Based on exhaustive search, driven by a local optimizer, epsilon-infla-
tion, interval Newton m ethod s, and interval exclusion prin ciples; uses
autom atic differentiation. Test results with hu nd reds of test examp les.
The un derlying interv al arithmetic package (ACM TOMS Algorithm
737) is also distribu ted. Workstation and PC implemen tations. (Lan-
guag e: Fortran.) Contact: R.B. Kearfott , http :/ /
interval.usl.edu / kearfott.html, ftp:/ / interval.usl.edu / pub/
interval_math/ intbis/ .
INTGLO, INTGLOB Integral Global Opt imizat ion
These methods solve un constrained and constrained, as well as discret
GOPs by the integral method . They also includ e a discontinu ous pen-
alty fun ction approach for constrained problem s. Problems up to onehu nd red variables have been solved. A set of test problems is also
available, including box or u nconstrained , constrained , concave mini-
mization, discrete variable programs and m ulticriteria program s. For
IBM PCs. (Langu age: Fortra n.) Contact: Q. Zheng
, D. Zhuang .
ISA Inductive Search Algorithm
ISA wo n first place at the 1st International Contest in Evolutionar y
Comp utation o n a real-valued function test-suite. (Languag e: C++.)
Contact: G. Bilchev, information available at http :/ /
solon.cma.univie.ac.at/ ~neum/ glopt/ test_results.html#bilchev.
LGO Continuous and Lipschitz Opt imizat ion
Solves bound -constrained and more general GOPs und er mild struc-
tural requ irements; it can be ap plied also to black box problem s. LGOintegrates several global (adaptive partition and random search based)
and local (derivative-free conjugat e directions type) strategies: these
can be activated in interactive or automatic execution mod es. The PC
version has a menu interface to assist the application developmen t pro
cess, includes a concise information / tutor ing session, and h as visual-
ization capabilities. Available also for w orkstations. LGO has been ap-
plied to p roblems with up to 100 variables (can be configured to encom
pass larger sizes). Accomp anied by a Users Guide and samp le prob-
lems. (Langu age: Fortran .) Cont act: J.D. Pintr ,
http:/ / www.tuns.ca/ ~pinter/ .
LOPS Lipschitz Opt imizat ion Program System
In all appro aches listed below , the objective fun ction is defined over n-
intervals. The Lipschitz-continuity of f or f is also assum ed. Problem-
classes and correspon ding ava ilable versions include: one-dim ensiona
GOPs (sequen tial methods w ith local tun ing, PC version (Language:
C++) one-dimen sional GOPs, parallel solver imp lementations (Lan-
guag e: Alliant FX/ 80, parallel Fortran) mu lti-dim ensional GOPs:
sequential and parallel algorithms using Peano curves (Language:
Alliant FX/ 80, pa rallel Fortran ) Contact: Y.D. Sergeyev
.
MAGESTIC Data Fitt ing by Global Opt imizat ion
Automatic global optimization based on a fast modified Gauss-Newto
app roach combined w ith Monte Carlo search. MAGESTIC handles cal
bration m odel variants (e.g., parameter and error masks for restricted
sub-fitting, implicit equ ation fitting w ithout solving, etc.). Suitable for
use also with Lagrange mu ltipliers for constrained optimization. Uses
Excel as an interface (und er Window s) and for generating grap hics.(Platform: Excel.) Contact: Logix Consu lting ,
http:/ / www .lgx.com/ magestic.html.
MULTISTART Clustering Algorithm
This widely used app roach is based on random search or some other
initial sampling in th e feasible set combined with clustering a nd loca
optim ization launched from the most promising point(s). Imple-
mented on SUN workstations. Several interesting app lications in
combination w ith simulation m odels are related to the analysis of oil
resources. (Language: Fortran.) Contact: S. Buitrago
.
GOCONTINUED
8/8/2019 Optima 52
7/16
O P T I M A
NO52PAGE 7 DECEMBER 1
NETSPEAK General Network Opt imizat ion
This is an algebraic mod elling langu age used to specify, solve, and ana-
lyze general linear, but also possibly nonconvex minimu m cost net-
work flow problems. A wide variety of netw ork and netw ork-relatedtopologies (pure networks, networks with side-constraints and / or side
variables, generalized n etworks) can be mo delled u sing N ETSPEAK.
The language is being developed as a Wind ows ap plication; it features
flexible I/ O, robust program control, and intu itive comman ds. Contact:
B.W. Lamar .
PA Packet Annealing
In PA, the Gibbs d istribution of the ob jective fun ction is deterministi-
cally annealed by tracing the evolu tion of a mu ltiple Gaussian p acket
app roximation. The approach has been applied to analyze complex mo-
lecular conformation models. IBM PC implem entation. Contact: D.
Shallowa y , B.W. Chu rch , M.
Oresic .
PROFIL Interval Branch and Bound MethodBound constrained interval global optimization, with rigorously guar-
anteed resu lts. PROFIL is based on BIAS (Basic Interv al Arithm etic
Subroutines) which pro vides an interface for interval operations. For
PC and a nu mber of UN IX systems. (Langu age: C.) Contact: C. Jansson,
O. Knueppel , http:/ / ww w.ti3.tu-
harburg.de/ Software/ PROFILEnglisch.html, ftp:/ / ti3sun.ti3.tu-
harburg.de/ pub/ profil/ unix/ profopt.tar.Z, ftp:/ / ti3sun.ti3.tu-
harburg.de/ pub/ profil/ pc/ profopt.tgz.
PVGO Parallel Verified Global O ptimiz ation
PVGO is a new parallel method for interval global optim ization. Imp le-
mented on a Con nection Machine CM5. (Language: Pascal-XSC.) Con-
tact: S. Berner .
RSBB Reduced Space Branch-and-Bound MethodRSBB applies variable domain reductions and an underestimating mod-
ule. Two versions run u nd er Unix; one of these algorithms is described
in Chap ters 1 and 2 of Grossmann (1996). (Language: both versions in
C++, they use external Fortran p rocedures.) Contact: T. Epperly,
[email protected], h ttp:/ / ww w.ps.ic.ac.uk/ ~epperly/ index.html.
SA Simulated Annealing
In add ition to the algorithm itself, this includes on-line interactive dem -
onstration, and additional information on C++ classes , random number
generation, Monte Carlo method s and Forth. A Nelder-Mead simplex
method implem entation is also available. (Langu ages: C, C++, Ada
and Fort h.) Contact: E. Carter , Taygeta Scientific Inc.
, http:/ / ww w.taygeta.com/ annealing/
simanneal.html.SAT Global Opt imizat ion for Satisfiability Problems
Boolean satisfiability (SAT) problems can be d irectly tran sformed into
un constrained GOPs. These, in tu rn, can be solved by sp ecifically tai-
lored solvers. Workstation im plementa tions. Contact: J. Gu
.
SIGMA Stochastic Integration Global Minimization Algorithm
The software package SIGMA is Algorithm 667; app eared in ACM-
TOMS 14 (1988) 366-380. Includ es also th e code of sever al test GOP s.
(Langu age: Fortr an.) Contact: F. Aluffi-Pentin i, V. Parisi and F. Zirilli
, http:/ / ww w.netlib.no/ netlib/ toms/ 667.
SIMANN Simulated Annealing
This program implements the continuous simulated annealing global
optim ization algorithm d escribed in Corana et al., ACM TOMS 13
(1987) No. 3, 262-280. Algorithm mod ifications and man y d etails on it
use can be fou nd in Goffe et al., J. of Econom etrics 60 (1994) No. 1-2, 65
100. (Langu age: Fortra n.) Con tact: B. Goffe,
, http:/ / ww w.netlib.no/ netlib/ opt/
simann.f.
SOLVEX Solver for Nonlinear Optimization Problems
For solving constrained an d u nconstrained n onlinear, multiobjective
and GOPs. SOLVEX algorithm libraries include th e method s listed be-
low: unconstrained m inimization: Hooke-Jeeves direct search, conju-
gate grad ient method , Shor R-algorithm , Powell-Brent meth od; genera
nonlinear programming problem: penalty functions, Lagrange function
method, parameterization method; global optimization: sequential set
covering techniqu e, simulated an nealing, clustering algorithm;
mu lticriteria optim ization: convolution method s, including goal pro-
gramm ing, direct Pareto app roximation. Interactive use enhanced by a
built-in problem ed itor and graph ics capabilities. Contact: M.A.Potapov .
TORUS Stochastic Algorithm for Global Minimization w ithConstraints
It is a Monte Carlo algorithm, combined w ith annealing search pr in-
ciples. The softwar e package TORUS is Algorithm 774; appeared in
ACM-TOMS 21 (1995) 194-213. Includ es also th e code of sever al test
GOPs. Contact: F. M. Rabinowitz, http:/ / ww w.netlib.no/ netlib/ toms
744.
TRUST Terminal Repeller Unconstrained Subenergy Tunneling
This method formu lates the GOP as the solution of a deterministic dy-
namical system incorporating terminal repellers and a subenergy tun-
neling function. Benchmark tests comparing th is method to other glo-
bal optimization procedures are presented, with favourable results.
The TRUST formulation leads to a simple stopp ing criterion. In add i-
tion, the structure of the equation s enables an implemen tation of the a
gorithm in analog VLSI hard ware (in the sp irit of artificial neural n et-
work s) for further speed enhan cements. Contact: B.C. Cetin, J. Barhen
and W. Bur dick; TRUST is described in JOTA 77 (1993) No. 1.
TVC Toolbox for Verified Computing
Can be app lied to the rigorous solution of nonlinear systems of equa-
tions and to general unconstrained and bound-constrained GOPs. TVC
is based on interval B&B and interval New ton meth ods; it also has au-
tomatic differentiation capabilities. The toolbox can be u sed on PCs,
work stations an d parallel compu ters. (Langu ages: PASCAL-XSC, C++
Test problems have been solved up to one hun dred v ariables. A drive
program an d hund reds of test examples are available from the au thor.Contact: D. Ratz , http :/ /
ww w.uni-karlsruhe.de/ ~iam, http:/ / ourworld.compu serve.com/
homepages/ num erik_software.
UFO Universal Functional O ptimiz ation
Interactive modular system for solving problems and for algorithm de
velopm ent. Several types of GO method s rand om search, continu a-
tion, clustering, and rand om p lus local search can be ap plied. For PC
(Langu age: For tran .) Cont act: L. Luksan , M.
Tuma, M. Siska, J. Vlcek and N. Ramesova, ftp: uivt.cas.cz/ pu b/
msdos/ ufo.
GOCONTINUED
8/8/2019 Optima 52
8/16
O P T I M A
NO52PAGE 8 DECEMBER 1
UNICALC Interval Branch and Bound Algorithm
UNICALC serves for bound -constrained GO; accepts also inequality
and / or equality constraints and decision variables. Contact: A.
Semenov, information available at ftp:/ / ftp.iis.nsk.su/ pub/ ai/ unicalc.VerGO Verified Global Opt imizat ion
VerGO is designed for rigorous bound (and approximate general con-
strained) GO of a twice continuou sly differentiable objective function.
VerGO features include interv al arithmetic, automa tic differentiation,
non-convexity test, mon otonicity test, and local optimization. Tested on
problem s up to over 30 variables. DOS, OS/ 2, Linux and w orkstation
versions. (Langu age: C++.) Contact: R. v an Iwaar den
, http:/ / ww w.cs.hope.edu/ ~rvaniwaa/
VerGO/ VerGO.html.
VTT Interval Arithmeti c Research
The goals of the Interval Arith metic, Constraint Satisfaction and Prob-
ability Project are summ arized as follows: developm ent of portable C++
libraries for interval p rogram ming task s; integration o f the libraries toMicrosoft Excel; application in financial planning softw are pr odu cts
(Platform s: C++, Excel.) Contact: S. De Pascale
, http:/ / ww w.vtt.fi/ tte/ .
GOCONTINUEDHansen , E.R. (1992) Global
Optimization Usin g Interval
Analysis. Marcel D ekker, New
York.
Horst, R. and Tuy, H . (1996) Glob al
Optimization - D eterministic
Approaches. Springer, Berlin /
Heidelberg / New York. (3rd Edn.)
Horst, R. and Pardalos , P.M., eds.
(1995) Handbook of G lobal
Optimization. Kluwer Academic
Publishers, Dordrecht / Boston /
London.
Journal of Global Optimization
(published since 1991, by Kluwer
Academic Publis hers).
Kearfott, R.B. (1996) Rigorous
Global Search: ContinuousProblems. Kluwer Academic
Publishers, Dordrecht / Boston /
London.
Michalewicz, Z. (1996) Genetic
Algorithms + Data Structures =
Evolution Programs. Springer,
Berlin / Heidelberg / N ew York.
(3rd Edn.)
Mockus , J. (1989) Bayesian
Approach to Global Op timization.
Kluwer Academic Publishers,
Dordrecht / Boston / London.
Ne umaier, A. (1990) Interval
Methods for Systems of Equations.
Cambridge University Press,
Cambridge.
4. Acknowledgements
The software review presented hereis based to a significant extent oninformation kindly provided bycolleagues working on GO and/orclosely related areas. I would liketo especially thank ArnoldNeumaier and Simon Streltsov forthe information collected on theirWWW Global Optimization Pages(respectively, http://solon.cma.univie.ac.at/~neum/glopt/, and http://cad.bu.edu/go/). I also wish to thank Faiz Al-Khayyal for his valuable commentson the manuscript.
The space (and time) limitations ofthis review certainly have made itillusory to include all existingsoftware on this rapidly changingarea; omissions are entirelypossible but absolutelyunintentional. It is planned tocontinue this work and to provide amore comprehensive andinformative picture of thestate-of-the-art for the mathematicalprogramming community.Comments and suggestions aremost welcome; they will contributeto an unabridged GO softwarereview in the near future.
References
To avoid a superfluou sly long list-
ing, the reference list is redu ced to
the most top ical journal, and to
several GO monograph s and
handbooks pu blished in the past
ten years.Bomze, I.M., Csend es, T., Horst, R.,
and Pardalos, P.M., eds. (1996)
Developments in Global Optimiza-
tion. Kluwer Academic Publishers,
Dordrecht / Boston / London.
Floudas, C.A. and Pardalos , P.M.
(1990) A Collection of Test
Problems for Constrained Global
Optimization Algorithms. Lecture
Notes in Computer Science 455,
Springer, Berlin / Heidelberg / N ew
York.
Floudas, C.A. and Pardalos , P.M.,
eds. (1992) Recent Advances inGlobal Optimization. Princeton
University Press, Princeton.
Forg, F. (1988) Nonconvex
Programming. Akadmiai Kiad,
Budapest. Grossmann, I.E., ed.
(1996) Global Optimization in
Engineering Design. Kluwer
Academic Publishers, Dordrecht /
Boston / London.
Pardalos , P.M. and Rosen, J.B.
(1987) Constrained Global
Optimization: Algorithms and
Applications. Lecture N otes i n
Computer Science 268, Springer,
Berlin / Heidelberg / N ew York.
Pintr, J.D. (1996) Global Optimi
tion in A ction. Kluw er Academic
Publishers, Dordrecht / Boston /
London.
Ratschek, H. and Rok ne, J.G. (19
New Computer Methods for Glo
Optimization. Ellis Horwood,
Chichester.
Trn, A.A. andv
Zilinskas, A. (198
Global Optimization. Lecture No
in Computer Science 350, Spring
Berlin / Heidelberg / N ew York.van Laarhoven, P.J.M. and Aarts
E.H.L. (1987) Simulated Anneali
Theory and App lications. Kluw e
Academic Publishers, Dordrecht
Boston / London.
Zhigljavsky, A.A. (1991) Theory
Global Random Search. Kluwer
Academic Publishers, Dordrecht
Boston / London.
8/8/2019 Optima 52
9/16
O P T I M A
NO52PAGE 9 DECEMBER 1
The Mathematical Programming Soci-
ety invites nominations for the A.W.Tucker Prize for an outstanding paperauthored by a student. The award willbe presented at the International Sym-posium on Mathematical Program-ming in Lausanne (August 24-29,1997). All students, graduate and un-dergraduate, are eligible. Nominationsof students who have not yet receivedthe first university degree are especiallywelcome. In advance of the Sympo-sium an award committee will screenthe nominations and select at mostthree finalists. The finalists will be in-vited, but not required, to give oralpresentations at a special session of the
Symposium. The award committeewill select the winner and present theaward prior to the conclusion of theSymposium. The members of thecommittee for the 1997 A.W. TuckerPrize are Kurt Anstreicher, Depart-ment of Management Sciences, Uni-versity of Iowa; Rolf Moehring,Fachbereich Mathematik, TechnicalUniversity of Berlin; Jorge Nocedal,EECS Department, NorthwesternUniversity; Jean-Philippe Vial (Chair-man), HEC/Management Studies,University of Geneva; David
Williamson, IBM T.J. Watson Re-search Center, Yorktown Heights.
Eligibility The paper may concern anyaspect of mathematical programming;it may be original research, an exposi-tion or survey, a report on computerroutines and computing experiments,or a presentation of a new and inter-esting application. The paper must besolely authored and completed afterJanuary 1994. The paper and thework on which it is based should havebeen undertaken and completed inconjunction with a degree program.
NominationsNominations must bemade in writing to the chairman ofthe award committee as follows:
Jean-Philippe VialHEC/Management StudiesUniversity of Geneva102, Bd Carl-VogtCH-1211 Geneva 4SwitzerlandFax: 41 22 705 81 04
E-mail: [email protected]
Call for Nominations: Nominationsare being sought for the Mathemati-cal Programming Society Beale-Orchard-Hays Prize for Excellencein Computational MathematicalProgramming.
Purpose: This award is dedicated tothe memory of Martin Beale andWilliam Orchard-Hays, pioneers incomputational mathematical pro-gramming. To be eligible a paper ora book must meet the following re-quirements:
1) It must be on computationalmathematical programming. Thetopics to be considered include:
a) experimental evaluations of one or
more mathematical algorithms,
b) the development of quality math-ematical programming software (i.e.well-documented code capable of ob-taining solutions to some importantclass of MP problems) coupled withdocumentation of the applications ofthe software to this class of problems(note: the award would be presentedfor the paper which describes thiswork and not for the software itself),
c) the development of a new compu-tational method that improves the
state-of-the-art in computer imple-mentations of MP algorithmscoupled with documentation of theexperiment which showed the im-provement, or
d) the development of new methodsfor empirical testing of mathematicalprogramming techniques (e.g., devel-opment of a new design for compu-tational experiments, identificationof new performance measures, meth-ods for reducing the cost of empiricaltesting).
2) It must have appeared in the openliterature.
3) If the paper or book is written in alanguage other than English, then anEnglish translation must also be in-cluded.
4) Papers eligible for the 1997 awardmust have been published within theyears 1993 through 1996.
They must be submitted by a facultymember at the institution where thenominee was studying for a degreewhen the paper was completed. Let-ters of nomination must be accompa-nied by a statement that each mem-ber of the committee (including thechairman) was sent the followingdocuments: the students paper; aseparate summary of the papers con-tributions, written by the nominee,and no more than two pages inlength; and a brief biographicalsketch of the nominee.
Deadline Nominations must be sentto the chairman and postmarked nolater than February 15, 1997.
Addresses of the other members ofthe committee:
Prof. Kurt M. AnstreicherDepartment of Management SciencesUniversity of IowaIowa City, IA 52242 USA
E-mail: [email protected]
Prof. Dr. Rolf H. MoehringFachbereich Mathematik, Sekr. 6-1Technische Universitat BerlinStrasse des 17. Juni 13610623 Berlin Germany
Fax: 49 30 314-25191E-mail: [email protected]
Prof. Jorge NocedalElectrical Engineering and ComputerScienceNorthwestern UniversityEvanston, IL 60208-3118 USAFax: (847) 467-4144
E-mail address:[email protected]
Dr. David P. WilliamsonIBM T.J. Watson Research LabsP.O. Box 218
Yorktown Heights, NY 10598 USAFax: (914) 945-3434
E-mail: [email protected]
The above information is reproduced
on the web at the address:
http://dmawww.epfl.ch/roso.mosaic/ismp97/tucker.html
These requirements are intended asguidelines to the screening committbut are not to be viewed as bindingwhen work of exceptional meritcomes close to satisfying them.
Frequency and Amount of the Awar
The prize is awarded every threeyears. The 1997 prize of $1500 andplaque will be presented in August1997, at the Swiss Federal Institute Technology (EPFL), Lausanne Switzerland, at the Awards Session of thInternational Symposium on Math-ematical Programming sponsored bthe Mathematical Programming Soety.
Judgement criteria: Nominations w
be judged on the following criteria:1) Magnitude of the contribution tothe advancement of computationaland experimental mathematical programming.
2) Originality of ideas and methods
3) Clarity and excellence of exposi-tion.
Nominations: Nominations must bin writing and include the title(s) ofthe paper(s) or book, the author(s),the place and date of publication anfour copies of the material. Support
ing justification and any supplementary materials are welcome but notmandatory. The awards committeereserves the right to request furthersupporting materials from the nominees.
Nominations should be mailed to:
Professor Robert J. VanderbeiDept. of Civ. Eng. and OperationsResearchACE-42 Engineering QuadPrinceton UniversityPrinceton, NJ 08544 USA
The deadline for submission of nomnations is January 1, 1997.
This call-for-nomination can beviewed online by visiting:
http://www.sor.princeton.edu/~rvdb/BOH97.html
Nominations for the A.W. Tucker PrizeExtended Deadline
Beale-Orchard-Hays Prize
8/8/2019 Optima 52
10/16
8/8/2019 Optima 52
11/16
O P T I M A
NO52PAGE 11 DECEMBER 1
call
MPDP-19 Nineteenth Symposium on Mathematical Programming with Data Perturbations
May 22-23, 1997, The George Washington University, Washington, DC
The NINETEENTH Symposium on Mathematical Programming with DataPerturbations will be held at George Washington Universitys Marvin Center on22-23 May 1997. The objective is to bring together practitioners who usemathematical programming optimization models, and deal with questions ofsensitivity analysis, with researchers who are developing techniques applicable tothese problems.
The symposium webpage is :http://rutcor.rutgers.edu:80/~bisrael/MPDP-19.html
CONTRIBUTED papers in mathematical programming are solicited in the followingareas: 1. Sensitivity and stability analysis and their applications; 2. Solution
methods for problems involving implicitly defined functions; 3. Solutionapproximation techniques and error analysis. CLINICAL presentations thatdescribe problems in sensitivity analysis encountered in applications are also
invited.
DEADLINES: 15 March 1997 Registration and submission of tentative title andabstract;1 May 1997 Submission of final abstract for inclusion in the Symposium Program
REGISTRATION FEE: $50 USD payable at the meeting.
To REGISTER and/or SUBMIT ABSTRACT please use the electronic form in the URL:
http://rutcor.rutgers.edu:80/~bisrael/MPDP-19.html#form
Or mail to:MPDP-19c/o Adi Ben-IsraelRUTCOR - Rutgers Center forOperations ResearchP.O. Box 5062Rutgers UniversityNew Brunswick, NJ 08903-5062, USA
Or email/fax to one of the organizerslisted below:
Adi Ben-IsraelRutgers University
Tel: +1-908-445-5631Fax: +1-908-445-5472
Hubertus Th. JongenRWTH-AachenTel: +49-241-804540Fax: +49-241-8888390
Special Issue of Computers &Operations Research:
Travelling Salesman Problem
Computers & Operations Research will publish a sp
cial issue on the Travelling Salesman Problem.P
pers are sought in the broad area of the travelling sale
man problem and its variations which discuss comp
tational and/or algorithmic aspects. In particular,
are soliciting papers on computational study of exact an
or heuristic algorithms, analysis of heuristics, domin
tion analysis and exponential neighbourhoods, problewith special structures, new applications, etc.
All papers submitted for consideration will undergo sta
dard review process. Four copies of the paper, followi
the standard guidelines for Computers & Operations R
search, should be sent by March 1997 to:
Dr. Abraham Punnen
Dept. of Mathematics, Statistics & Computer Scien
University of Brunswick
Saint John, New Brunswick
CANADA E2L 4L5
Financial Support forINTERNATIONAL CONGRESS OF MATHEMATICIANSBerlin, 1998
The ICM98 Organizing Committee has already re-ceived quite a number of requests concerning financial
support for participation at the International Congress
of Mathematicians 1998 in Berlin. The Circular Let-
ter ICM98-CL6 describes how mathematicians from
developing countries can apply for financial help. The
local Organizing Committee is currently making ef-
forts to obtain donations from German industry, gov-
ernment, foundations and individuals to be able to
partially support mathematicians from Eastern Europe
and the independent states of the former Soviet Union.
To secure participation of as many persons as possible,
the local Organizing Committee will only support lo-
cal costs in Berlin. Berlin is very close to Eastern Eu-
rope, and it is expected that applicants will find other
means to cover their travel costs.To handle the applications and manage the financial
support the ICM98 Organizing Committee has set up
a subcommittee, Committee for Support of Mathema-
ticians from Eastern Europe (CSMEE).
CSMEE will distribute application forms for grants,
as described above, for mathematicians from Eastern
Europe in late summer 1997. T hese forms will be
made available through the ICM98 server (http://
elib.zib.de/ICM98) and by e-mail. Applicants will be
asked to provide a brief curriculum vitae (including
academic education, degree, professional employment,
and a list of publications).
Applicants should submit their application form
before January 1, 1998 to CSMEE to one of the following addresses:
Prof. Dr. H. KurkeHumboldt-Universitaet Institut fuer MathematikUnter den Linden 6 D-10099 BerlinGermany
e-mail: [email protected]
Prof. Dr. W. RoemischHumboldt-UniversitaetInstitut fuer Mathematik Ziegelstrasse 13AD-10099 BerlinGermany
e-mail: [email protected]
All applications will be reviewed.
Further questions concerning financial support for
mathematicians from Eastern Europe to attend
ICM98 should be directed to Professors Kurke orRoemisch.
More information about ICM98 can be found inthe ICM98 WWW-server (URL: http://elib.zib-berlin.de/ICM98). This WWW-server also offersan electronic preregistration form. If you do nothave access to the World Wide Web and wouldlike to subscribe to the ICM98 circular letters,just send an e-mail to
[email protected] PREREGISTRATIONin the SUBJECT line.
MARTIN GRETSCHEL, PRESIDENT OF THE ICM98
ORGANIZING COMMITTEE
Riethard KlatteUniversity of ZurichTel: +41-1-257 3772Fax: +41-1-252 1162
Doug WardMiami University
Tel: +1-513-529-3534Fax:+1-513-529-1493
Anthony V. Fiacco, General Chairman
Sponsored by the Dept. of OperationsResearch & the Institute forManagement Science & Engineering,School of Engineering & AppliedScience The George WashingtonUniversity Washington, DC 20052,USA
FOR
PAPERS
8/8/2019 Optima 52
12/16
O P T I M A
NO52PAGE 12 DECEMBER 1
B O O K r e v i e w sCombinatorial Network TheoryEdited by Ding-Zhu Du andD. Frank HsuSeries of Applied Optimization
Kluwer Academic Publishers
Dordrecht, 1996ISBN 0-7923-3777-8
The design of interconnection networks and related theo-
retical research has received a lot of attention in recent
years. A network suitable for parallel computation should
satisfy two major requirements. First, a network should
have computational efficiency to run parallel algorithms.
One of the main overheads incurred during parallel ex-ecution is the time spent on communication between
processors. Thus, the ability of an interconnection net-
work to disseminate information efficiently is one of the
most important properties of a network. Secondly, reli-
ability of a network is another important requirement.
An interconnection network should be fault-tolerant so
that it can work even if some of its links or nodes fail. This
book is devoted to the theoret ical study of dissemination
of information in interconnection networks and their re-
liability.
Caley graphs provide suitable candidates for interconnec-
tion networks. One of their advantages is their symmet-
ric structure, which makes them easy to build and sim-
plifies routing and communication algorithms. Examples
of prominent interconnection networks which are Caley
graphs include the H ypercube, the Butterfly and the
Cube-Connected-Cycles network. In the first two chap-
ters of the book Caley graphs are considered. Since Caley
graphs are defined on groups, group theory provides a
powerful tool in the study of Caley graphs.
An extensive study of the connectivity of Caley graphs
defined on finite Abelian groups is presented in Chapter
1. Connectivity can be viewed as a measure of the reli-
ability of a network. Along with the presentation of new
results, it is shown how some well-known results from the
additive group theory were rediscovered and applied to
graph-theoretical problems such as that of network con-
nectivity. Unfortunately, this chapter contains many
textual mistakes.
State of the Art in GlobalOptimization
Computational Methods andApplicationsEdited by C.A. Floudas andP.M. PardalosNonconvex Optimization and Its
Applications 7
Kluwer Academic Publishers
Dordrecht 1996ISBN 0-7923-3838-3
This is the seventh volume of an excellent and muneeded series in Nonconvex Optimization and Its A
plications put together by leaders in this field. The vo
ume contains 36 invited and thoroughly refereed pape
that were presented at the conference on State of the A
in Global Optimization: Computational Methods a
Applications. The conference was organized by C.
Floudas and P.M. Pardalos and held at Princeton Un
versity, April 28-30, 1995. The conference pape
spanned the gamut of theory, computational impleme
tations and applications of global optimization.
Among other fine contributions, one finds the followin
R. Horst and N. van Thoai propose two types of n
finite branch and bound algorithms for global minim
zation of separable concave functions under linear co
straints with totally unimodular matrices. The key obs
vation is that the underlying problems can be viewed
integer programs. Finiteness can, therefore, be achiev
by integral branching.
K.G. Ramakrishnan, M.G.C. Resende and P.M
Pardalos report on computational experience with
branch and bound algorithm for Quadratic Assignme
Problems. All problems of dimension n 15 of QAPL
are solved.
R.B. Kearfott reports practical experience with an i
terval branch and bound algorithm for equality-costrained optimization.
T . Van Voorhis and F. Al-Khayyal use range reducti
techniques to accelerate a branch and bound algorith
for quadratically constrained quadratic programs.
J.P. Shectman and N.V. Sahinidis prove that exhau
tiveness and branching on the incumbent whenev
possible in tandem ensure finiteness of rectangular-su
division-based branch and bound algorithms for glob
minimization if separable concave programs.
The study of edge and vertex connectivity in Caley graphscontinues in Chapter 2. In contrast to Chapter 1, most
of the proofs in this chapter are obtained by using atoms
of graphs.
De Bruijn and Kautz digraphs and their generalizations
are other popular candidates for interconnection net-
works. Their popularity is explained by the fact that these
networks have a highly symmetric structure and give
optimal or nearly optimal solutions of the problem of
minimizing the diameter and maximizing connectivity for
a graph with a given number of nodes and degree. Con-
nectivity and diameter are important parameters for both
reliability and the ability of a network to disseminate
information efficiently. Chapter 3 presents an extensivestudy of diameter, connectivity, line-connectivity, super-
line connectivity and the Hamiltonian property of de
Bruijn and Kautz digraphs and their generalizations.
In Chapter 4 various properties (in particular, link con-
nectivity) of extended double loop networks (EDLN) are
studied. A large class of EDLN includes such well-known
networks as generalized de Bruijn networks, the Imase-
Itoh networks and double loop networks. This study
shows that certain EDLN networks are suitable candidates
for good interconnection networks.
Dissemination of information in interconnection net-
works is considered in Chapter 5. The three main prob-
lems of dissemination of information are broadcasting,
accumulation and gossiping. These problems are consid-
ered for all well-known interconnection networks under
several communication modes. The solutions for these
problems are given mainly in terms of lower and upper
bounds on communication t ime. This chapter provides
the reader with many basic proof techniques and ideas in
this area, gives a good survey of known results and for-
mulates many open problems.
The book is aimed at graduate students and researchers
and provides the reader with a deep insight into combi-
natorial network theory. Some of its chapters (especially
Chapter 5) would also be suitable for undergraduate stu-dents.
-E.A. STHR
B O O K
8/8/2019 Optima 52
13/16
O P T I M A
NO52PAGE 13 DECEMBER 1
H . Tuy shows that many important location problems(Webers problem with attraction and repulsion, con-
strained multisource and multifacility problems and oth-
ers) can be formulated as d.c. optimization problems in
low-dimensional spaces and describes algorithms for their
solution.
Additional articles in the volume are authored by S.
Zlobec; E. Novak and K. Ritter; S. Shi, Q. Zheng and D.
Zhuang; B. Ramachandran and J.F. Pekny; G. Isac; P.
Maponi, M.C. Recchioni and F. Zirilli; G.H. Staus, L.T.
Biegler and B.E. Ydstie; V. Visweswaran, C.A. Floudas,
M.G. Ierapetritou and E.N. Pistikopoulos; J. Barhen and
V. Protopopescu; D. MacLagan, T. Sturge and W.
Baritompa; K. Holmqvist and A. Migdalas; D.W. Bulgerand G.R. Wood; W. Edmonson, K. Srinivasan, C. Wang
and J. Principe; E. Falkenauer; I.P. Androulakeis, V.
Visweswaran and C.A. Floudas; T. Q ian, Y. Ye and P.M.
Pardalos; A. Torn and S. Viitanen; K.I.M. McKinnon,
C. Millar and M. Mongeau; E. Haddad; J. Shi and Y.
Yoshitsugu; I. Garcia and G.T. H erman; P. Sussner, P.M.
Pardalos and G.X. Ritter; J.A. Filar, P.S. Gaertner and
M.A. Janssen; W.F. Eddy and A. Mockus; L. Mockus and
G.V. Reklaitis; A. Lucia and J. Xu; J.R. Banga and W.D .
Seider; M. Turkay and I.E. Grossmann; F. Friedler, J.B.
Varga, E. Feher and L.T. Fan; E.S. Fraga.
The above papers cover the theory and algorithms of
deterministic global optimization, stochastic global op-
timization, branch and bound methods, interval arith-
metic methods, d.c. programming, duality, concave pro-
gramming, bilevel programming, integral optimization,
decomposition methods, logic-based algorithms, and
trust algorithms.
The book is also very rich in applications in resource
allocation, computer vision, chemical process design,
control and optimization, chemical and phase equilib-
rium, facility location, climate change dynamic visualiza-
tion, batch process scheduling, and process synthesis.
One, therefore, cannot help but to fully agree with the
editors that the book will be a valuable source of infor-mation to faculty, students and researchers in optimiza-
tion, engineering, mathematics, computer sciences and
related areas.
-NIKOLAOS SAHIN IDIS
Postoptimal Analyses,Parametric Programming andRelated Topics
Tomas Galde Gruyter
Berlin, 1995ISBN 3-11-014060-8.
This book is the second edition of a monograph first writ-
ten in the years 1968-1969 in Czech, translated in 1973
to German and in 1979 to English. Each chapter is divided
into two parts. The first is based on illustrative examples;
the second part is an abridged mathematical presentation.
Each chapter ends by a selected bibliography.
This presentation in two parts has the great advantage that
the book is easy to read and understand for a wide audi-
ence in the first part and is mathematically complete in the
second part. It is also a good, quite complete reference on
the subject. H owever, it has two main disadvantages. The
first is probably linked to the origin of the book: there
remain some typographical errors and inconsistencies in the
notations. Furthermore, the references at the end of each
chapter often date before 1970. The second disadvantage
is an omission: the consideration of the sensitivity analysis
results in case of degeneracy. This part is quite poor and
recently proposed new results should be introduced.
Chapter one is devoted to fixed basic concepts and nota-
tions in linear programming. It recalls all the very well-
known concepts in linear programming such as basic vari-
ables, reduced cost, dual value, etc. It also recalls the basis
of the Simplex method for solving linear problems. The dual
problem is also presented with the dual Simplex method
and finally the concepts of primal or dual degenerate so-
lutions are presented.
Chapter two is devoted to suboptimal solutions, redundant
constraints and degeneracy. For suboptimal solution, the
effect on the objective level and on the activities level of
producing a nonoptimal activity level is derived from the
optimal Simplex tableau. A redundant constraint is definedas a constraint that does not influence the feasible region.
A weakly redundant constraint has a point in common with
the feasible region. The strongly redundant one does not.
This simply means that the associate slack cannot be
equalised to zero. This can be easily detected and the cor-
responding constraint can be deleted. A primal degenerate
solution is obtained when some basic variables are equal to
zero. This can lead to a degenerate step in the Simplex
algorithm: the base changes but the algorithm remains at
the same vertex.
Chapter three is devoted to sensitivity analysis with respeto changing right-hand side without basis exchange. T
classical sensitivity analysis with a single component of t
right-hand side is introduced through a simple examp
The same basis remains optimal if the basic variables rem
nonnegative. One can deduce a range for the maxim
variation of this component. The effect on the optimal ba
variables is given by the corresponding column of the
verse of the basic matrix. The effect on the objective fun
tion is given by the value of the corresponding dual va
able. Then sensitivity analysis with respect to several co
ponents of the right-hand side depending on a scalar p
rameter is examined. Finally, the multiparametric sensit
ity analysis case is considered, i.e. changing several compnents of the right-hand side depending on several para
eters.
Chapter four concerns linear parametric programming w
respect to change in the right-hand side. This implies
general a basis exchange. The critical values of the param
eters are defined as the values for which the optimal ba
changes. First, the case of a single change in the right-ha
side is considered. Then, a change in several componen
of the right-hand side is considered. The case where the
components change linearly with a single parameter a
the case where they change multilinearly with several p
rameters are both considered.
At the end of this chapter, the problem of sensitivity ana
sis under primal degeneracy is briefly considered. In fa
the sensitivity analysis with the right-hand side consists
determining the critical interval where the same basis
mains optimal. If the optimal solution is degenerate, se
eral optimal bases exist. The critical interval is defined he
as the union of all the critical intervals associated with t
different optimal bases. Note that the analysis concerni
the shadow price that can be different left and right is n
very developed here. For a more complete analysis, see,
example, M.P. Williams, Model Building in Mathema
cal Programming, John Wiley, 1990. Also, the assertion
Gal (see page 175) that commercial LP software offeri
sensitivity analysis and shadow prices yield false results wh
primal degeneracy occurs, is not correct. There is o
exception that I know: XPRESS-MP of Dash Associat
which gives the two (left and right) correct values.
Chapter five concerns sensitivity analysis with respect
changing cost coefficients without basis exchange. Note t
very bad choice made for notations: the market price
denoted by c and the unit cost by p (see page 211). The ca
of a single change for a nonbasic objective coefficient a
for a basic variable objective coefficient are considere
Then the case of changing several cost coefficients depen
8/8/2019 Optima 52
14/16
O P T I M A
NO52PAGE 14 DECEMBER 1
ing on a scalar or multiple parameters is considered. In each
case the critical region of the parameter (i.e. the region in
which the same basis is optimal) is computed.
Chapter six is devoted to linear parametric programming
with respect to change in cost coefficients with basis ex-
change. Since the partial derivative of the optimal value of
the objective function with cost coefficient is the value ofthe primal variables, it is sufficient to compute the succes-
sive optimal solutions when the objective coefficients vary.
This can be done by changing from basis to adjacent basis,
simply by a primal Simplex step. Then the problem of
finding the minimal and maximal value of the parameter
such that the problem has a bounded optimal solution is
considered. It is asserted (see example 6-2, page 239) that
the example has no solution for zero value of the param-
eter. This is false. There is not finite optimal solution. The
geometric interpretation (slope change) of cost coefficient
change is then given. Then the case of changes depending
on several parameters is considered. The chapter ends with
the problem of determining the optimal objective costcoefficients where these coefficients can vary homoge-
neously with a parameter. The consideration of degeneracy
is omitted in this chapter.
Chapter seven is dedicated to sensitivity analysis to simul-
taneous changing of the right-hand side and of the cost
coefficients. The objective coefficients and the right-hand
side are written here as linear functions of the same param-
eter. First, the case of changing of the right-hand side and
cost coefficient with a scalar parameter is considered. Then
the case of the dependence on a vector of parameters is
considered.
Chapter eight is dedicated to sensitivity analysis with re-
spect to the elements of the technological matrix often noted
A in linear problems. The coefficient at the intersection of
line i and column j denotes the consumption of the ith pro-
duction factor per unit of the jth production and can be
affected by a change in technology. Two small examples
of one column of A depending on a single parameter illus-
trate the complications which arise from such changes: the
optimal value of the objective function varies nonlinearly
with this parameter! This chapter is certainly one of the most
original as such results are rarely presented in books and
most of the time not published. I agree totally with the
author when he says that, despite the importance of this
sensitivity analysis with respect to matrix coefficients, papers
have been rarely published. He cites as examples his CORE
discussion papers 7013 and 7018 dating from 1970!
However, I regret that the author directly goes to the case
of one column that entirely varies and does not consider
the partial derivatives of the optimal value of the objective
function with respect to a single coefficient of matrix A. One
can derive from the (more complicated) formula of the
abridged mathematical presentation of chapter 8 the pretty
nice result: this partial derivative is equal to the opposite
of the product of the dual optimal value associated with the
line of the coefficient by the primal optimal value associ-
ated with the column of the coefficient. The degenerate case
is also totally omitted. The result however exists: it is forth-
coming in Mathematical Programming under the title
Generalised derivative of the optimal solution of the ob-
jective function of a linear problem with respect to matrix
coefficient by D. De Wolf and Y. Smeers.
Chapter nine is dedicated to multicriteria linear program-ming, i.e. the problem of maximising several conflicting
linear goals over a linear feasible region. One can define in
this framework an efficient solution as a nondominated
solution in Pareto sense: i.e. when trying to improve the
value of one goal, the value of at least one of the remaining
objective functions becomes worse. A method for determin-
ing the set of all efficient solutions is proposed. The reason
why multicriteria problems are considered in this book is
the following: the so-called Efficiency Theorem states that
there is a one-to-one correspondence between efficient
solutions and optimal solutions for the homogeneous
multiparametric problem that is defined in chapter six.
Therefore parametric programming can help to computeefficient solutions for the mult icriteria linear problem.
Finally, the non-essential objective functions are defined
as objective functions that do not affect the set of efficient
solutions. By an efficiency test, one can identify these non-
essential objectives and thus reduce the number of objec-
tive functions to consider.
Chapter ten indicates possible applications of sensitivity
analysis and linear parametric programming in decision
making. Firstly, if for some reason, one wants to change the
value of some basic variables, it is sufficient to change the
corresponding right-hand side. Changing the value of some
nonbasic value is equivalent to considering a suboptimal
solution. Secondly, the problem of the inconsistency of the
constraint is considered. For practical applications with
thousands of constraints, it often occurs that the solution
set is empty (one artificial variable remains positive in phase
one). A simple way of removing inconsistency is to change
the value of the right-hand side. Thirdly, the problem of
redundancy of some linear inequalities is considered. Re-
call that a redundant constraint is a constraint that does not
affect the feasible region. One way for detecting such con-
straints using sensitivity analysis is to notice that the range
of variation of the right-hand side of such constraints var-
ies from minus infinity to plus infinity. In conclusion, sen-
sitivity analysis helps to compute efficient solutions in
multicriteria programming (first application), to remove
inconsistency (second application) and to detect redundant
constraints (third application).
-DANIEL DE WOLF
8/8/2019 Optima 52
15/16
O P T I M A
NO52PAGE 15 DECEMBER 1
CREDITCARDNUMBER: EXPIRY DATE:
FAMILY NAME:
MAILING ADD RESS:
TEL.NO.: TELEFAX:
E-MAIL:
SIGNATURE
I w ish to enroll as a member of the Society.
My subscription is for my personal use and not for the benefit of any library or institut ion.
I w ill pay my membership dues on receipt of your inv oice.
I w ish to pay by creditcard (Master/Euro or Visa).
Mail to:The Mathematical Programming Society, Inc.
c/o International Statistical Institute
428 Prinses Beatrixlaan
2270 AZ Voorburg
The Netherlands
Cheques or money orders should be made payable tThe Mathematical Programming Society, Inc., inone of the currencies listed below.Dues for 1996, including subscription to the journa
Mathematical Programming, are Dfl.105.00 (or$60.00 or DM94.00 or 39.00 or FF326.00 orSw.Fr.80.00).Student applications: Dues are one-half the aboverates. Have a faculty member verify your student status and send application with dues to above address
Faculty verifying status
institution
A P P L I C A T I O N F O R M E M B E R S H I P
JOU
RNA
LVolume 75, No. 1
A. von Arnim, R. Schrader and
Y. Wang, The permutahedron of
N-sparse posets
Z.-Q. Luo, J.-S. Pang, D. Ralph
and S.-Q. Wu,Exact penaliz a-
tion and stat ionarity conditions
of mathematical programs w ith
equilibrium constraints
Tsuchiya, T. and R.D.C.
Monteiro, Superlinear conver-
gence of the a ffine scaling
algorithm
Volume 75, No. 2
R.J.B. Wets, Challenges in
stochastic programming
E.L. Plambe ck, B.-R. Fu, S.M.
Robinson and R. Suri, Sample-
path optimiz ation of convex
stochastic performance func-
tions
S.S. Nielsen and S.A. Zenios, A
stochast ic programming model
for funding si ngle premium
deferred annuities
K. Marti,Differentiation
formulas for probability
functions: The transformat ion
method
P. Kall and J. Maye r, SLP-IOR:
An interacti ve model ma nage-
ment system for stochastic linear
programs
G. Infanger and D .P. Morton,
Cut sharing for mult ist age
stochastic l inear programs w ith
interst age d ependency
J.L. Higl e and S. Sen ,Duality
and st atistical tests of optimality
for tw o st age stochastic pro-
grams
K. Frauendorfer,Barycentric
scenario t rees in convex mult i-
stage stochastic programming
N.C.P. Edirising he and W.T.
Ziemba,Implementing bounds-
based approximations in convex-
concave tw o-stage stochastic
programming
J.R. Birge, C.J. Do noh ue, D .F.Holmes and O.G. Svintsitski,A
parallel implementation of the
nested decomposition algorithm
for multistage stochastic linear
programs
a t h e m a t i c a l
rogramming
Volume 75, No. 3
M. Constantino,A cutt ing plane
approach to capacitated lot -
sizing w ith start-up costs
H. Yamashita and H. Yabe,
Superlinear and quadratic
convergence of some primalduainterior point methods for
constrained opt imizat ion
J.-P. Crouz eix and J.A. Ferland,
Criteria for d ifferenti able
generaliz ed monot one maps
T. De Luca, F. Facchinei and C.
Kanzow,A semismooth equatio
approach to the solution of
nonlinear complementarit y
problems
D. Goeleven, G.E. Stavroulakis
and P.D. Panagiotopoulos,
Solvability theory for a class of
hemivariational inequalitiesinvolving copositive plus
matrices.Applications in robotic
E.C. Sew ell, Binary integer
programs w ith tw o variables pe
inequality
M.D. Grigoriadis and L.G.
Khachiyan,Approximate
minimum-cost multicommodity
flow s in Oe -2 KNM time
8/8/2019 Optima 52
16/16
FIRST CLASS MAIL
O P T I M AMATHEMATICAL PROGRAMMING SOCIETY
Center for Applied Optimization
371 Weil Hall
PO Box 116595
Gaines vill e FL 32611-6595 USA
UN IVERSITY OF
FLORIDA
Do nald W. Hearn, EDITOR
Karen Aardal, FEATURESEDITOR
Utrecht University
Department of Computer Science
P.O. Box 80089
3508 TB Utr echt
The Netherlands
Faiz Al-Khayyal, SOFTWARE & COMPUTATIONEDITOR
Georgia Tech
Industrial and Systems Engineering
Atlan ta, GA 30332-0205
Dolf Talman, BOOKREVIEWEDITOR
Departm ent of Econom etrics
Tilburg University
P.O. Box 90153
5000 LE Tilbur g
The Netherlands
Elsa Drake, DESIGNER
PUBLISHEDBYTHE
MATHEMATICALPROGRAMMINGSOCIETY &
GATOREngineeringPUBLICATIONSERVICES
UNIVERSITY OF FLORIDA
Journal contents are subject t o change by t he publisher.
gallimaufry
sHarvey Greenberg has developed a Mathem
Programming Glossary on the Web. URL is http://w
math.cudenver.edu/cudenver.edu/~hgreenbe/glo
glossary.html and it includes links to several bibliogHe invites comments and suggestions via a link to h
home page. sOther useful URLs are the MPS home
http://www.caam.rice.edu/~mathprog/ and the ho
page for the next Symposium: http://dmawww.epf
roso.mosaic/ismp97/welcome.html . sDeadline
for the next OPTIMA is Feb. 15, 1997.