(Vol. 115: Lecture Notes in Operations Research and Mathematical Economics, Vol. 1659: Lecture
Notes in Operations Research and Mathematical Systems)
Vol. 1; H. BUhlmann, H. Loeffel, E. Nievergelt, EinfUhrung in die Vol. 30; H. Noltemeier, Sensitivitatsanalyse bei diskreten linearen
Theorie und Praxis der Entscheidung bei Unsicherheit. 2. Auflage, Optimierungsproblemen. VI, 102 Seiten. 4°. 1970. DM 16,
IV, 125 Seiten. 4°. 1969. DM 16, Vol. 31; M. KOhlmeyer, Die nichtzentrale tVerteilung. II, 106 Sei
Vol. 2; U. N. Bhat, A Study of the Queueing Systems M/G/1 and ten. 4°. 1970. DM 16,
GIIM1. VIII, 78 pages. 4°.1968. DM 16, Vol. 32; F. Bartholomes und G. Hotz, Homomorphismen und Re
Vol. 3; A. Strauss, An Introduction to Optimal Control Theory. VI, duktionen linearer Sprachen. XII, 143 Seiten. 4°.1970. DM 16,
153 pages. 4°.1968. DM 16, Vol. 33; K Hinderer, Foundations of Nonstationary Dynamic Pro
Vol. 4; Branch and Bound; Eine EinfOhrung. 2., geanderte Auflage. gramming with Discrete Time Parameter. VI, 160 pages. 4°.1970.
Herausgegeben von F. Weinberg. VII, 174 Seiten. 4°.1972. DM 18, DM 16,
Vol. 5; Hyvarinen, Information Theory for Systems En9ineers. VIII, Vol. 34; H. Stormer, SemiMarkoffProzesse mit endlich vielen Zu
205 pages. 4°. 1968. DM 16, standen. Theorie und Anwendungen. VII, 128 Seiten. 4°.1970.
DM 16,
Vol. 6; H. P. KOnzi, O. MOiler, E. Nievergelt, EinfOhrungskursus in
die dynamische Programmierung. IV, 103 Seiten. 4°.1968. DM 16, Vol. 35; F. Ferschl, Markovkenen. VI, 168 Seiten. 4°.1970. DM 16,
Vol. 7; W. Popp, EinfUhrung in die Theorie der Lagerhaltung. VI, Vol. 36; M. P. J. Magill, On a General Economic Theory of Motion.
1 73 Seiten. 4°. 1968. DM 16, VI, 95 pages. 4°. 1970. DM 16,
Vol. 8; J. Teghem, J. LorisTeghem, J. P. Lambone, Modeles Vol. 37; H. MUlierMerbach, On RoundOff Errors in Linear Pro
d'Anente M/G/1 et GI/M/1 il Arrivees et Services en Groupes. IV, gramming. VI, 48 pages. 4°. 1970. DM 16,
53 pages. 4°. 1969. DM 16, Vol. 38; Statistische Methoden I. Herausgegeben von E. Walter.
Vol. 9; E. Schultze, EinfUhrung in die mathematischen Grundlagen VIII, 338 Seiten. 4°. 1970. DM 22,
der Informationstheorie. VI, 116 Seiten. 4°. 1969. DM 16, Vol. 39; Statistische Methoden II. Herausgegeben von E. Walter.
Vol. 10; D. Hochstadter, Stochastische Lagerhaltungsmodelle. VI, IV, 155 Seiten. 4°. 1970. DM 16,
269 Seiten. 4°. 1969. DM 18, Vol. 40; H. Drygas, The CoordinateFree Approach to Gauss
Vol. 11/12; Mathematical Systems Theory and Economics. Edited Markov Estimation. VIII, 113 pages. 4°.1970. DM 16,
by H. W. Kuhn and G. P. Szego. VIII, IV, 486 pages. 4°. 1969. Vol. 41; U. Ueing, Zwei Losungsmethoden fUr nichtkonvexe Pro
DM 34, grammierungsprobleme. VI, 92 Seiten. 4°.1971. DM 16,
Vol. 13; Heuristische Planungsmethoden. Herausgegeben von Vol. 42; A. V. Balakrishnan, Introduction to Optimization Theory in
F. Weinberg und C. A. Zehnder. II, 93 Seiten. 4°. 1969. DM 16, a Hilbert Space. IV, 153 pages. 4°.1971. DM 16,
Vol. 14; Computing Methods in Optimization Problems. Edited Vol. 43; J. A. Morales, Bayesian Full Information Structural Analy
by A. V. Balakrishnan. V, 191 pages. 4°.1969. DM 16, sis. VI, 154 pages. 4°. 1971. DM 16,
Vol. 15; Economic Models, Estimation and Risk Programming; Vol. 44; G. Feichtinger, Stochastische Modelle demographischer
Essays in Honor of Gerhard Tintner. Edited by K. A. Fox, G. V. L. Prozesse. XIII, 404 Seiten. 4°. 1971. DM 28,
Narasimham and J. K. Sengupta. VIII, 461 pages. 4°.1969. DM 24,
Vol. 45; It Wendler, Hauptaustauschschrine (Principal Pivoting).
Vol. 16; H. P. KOnzi und W. Oettli, Nichtlineare Optimierung; 11,64 Seiten. 4°. 1971. DM 16,
Neuere Verlahren, Bibliographie. IV, 180 Seiten. 4°. 1969. DM 16,:
Vol. 46; C. Boucher, Levons sur la tMorie des automates ma
Vol. 17; H. Bauer und K Neumann, Berechnung optimaler Steue thematiques. VIII, 193 pages. 4°.1971. DM 18,
rungen, Maximumprinzip und dynamische OptijTlierung. VIII, 188
Seiten. 4°.1969. DM 16, . Vol. 47; H. A. Nour Eldin, Optimierung linearer Regelsysteme
mit quadratischer Zielfunktion. VIII, 163 Seiten. 4°. 1971. DM 16,
Vol. 18; M. Wolff, Optimale Instandhaltungspolitiken in einfachen
Systemen. V, 143 Seiten. 4°. 1970. DM 16, Vol. 48; M. Constam, FORTRAN fOr Anfanger. VI, 148 Seiten.
2., verbesserte Auflage. 4°.1973. DM 16,
Vol. 19; L. Hyviirinen Mathematical Modeling for Industrial Pro
Vol. 49; Ch. SchneeweiB, Regelungstechnische stochastische
cesses. VI, 122 pages. 4°. 1970. DM 16,
Optimierungsverfahren. XI, 254 Seiten. 4°.1971. DM 22,
Vol. 20; G. Uebe, Optimale Fahrplane. IX, 161 Seiten. 4°.1970.
Vol. 50; Unternehmensforschung Heute  Ubersichtsvortrage der
DM 16,
ZOricher Tagung von SVOR urrd DGU, September 1970. Heraus
Vol. 21; Th. Liebling, Graphentheorie in Planungs und Touren gegeben von M. Beckmann. VI, 133 Seiten. 4°. 1971. DM 16,
problemen am Beispiel des stadtischen StraBendienstes. IX,
Vol. 51; Digitale Simulation. Herausgegeben von K Bauknecht
118 Seiten. 4°.1970. DM 16,
und W. Nef.IV, 207 Seiten. 4°. 1971. DM 18,
Vol. 22; W. Eichhorn, Theorie der homogenen Produktionsfunk
Vol. 52; Invariant Imbedding. Proceedings of the Summer Work
tion. VIII, 119 Seiten. 4°.1970. DM 16,
shop on Invariant Imbedding Held at the University of Southern
Vol. 23; A. Ghosal, Some Aspects of Queueing and Storage California, JuneAugust 1970. Edited by R. E. Bellman and E. D.
Systems. IV, 93 pages. 4°. 1970. DM 16, Denman. IV,148 pages. 4°.1971. DM 16,
Vol. 24; Feichtinger, Lernprozesse in stochastischen Automaten. Vol. 53; J. RosenmOller, Kooperative Spiele und Markte. IV, 152
V, 66 Seiten. 4°.1970. DM 16, Seiten. 4°.1971. DM 16,
Vol. 25; R. Henn und O. Opitz, Konsum und Produktionstheorie. Vol. 54; C. C. von Weizsacker, Steady State Capital Theory. III,
I. II, 124 Seiten. 4°.1970. DM 16, 102 pages. 4°.1971. DM 16,
Vol. 26; D. Hochstadter und G. Uebe, Okonometrische Methoden. Vol. 55; P. A V. B. Swamy, Statistical Inference in Random Coef
XII, 250 Seiten. 4°. 1970. DM 18, ficient Regression Models. VIII, 209 pages. 4°. 1971. DM 20,
Vol. 27; I. H. Mufti, Computational Methods in Optimal Control Vol. 56; Mohamed A. EIHodiri, Constrained Extrema. Introduction
Problems. IV, 45 pages. 4°. 1970. DM 16, to the Differentiable Case with Economic Applications. III, 130
Vol. 28; Theoretical Approaches to NonNumerical Problem Sol pages. 4°.1971. DM 16,
ving. Edited by R. B. Banerji and M. D. Mesarovic. VI, 466 pages. Vol. 57; E. Freund, Zeitvariable MehrgroBensysteme. VII, 160 Sei
4°. 1970. DM 24, ten. 4°.1971. DM 18,
Vol. 29; S. E. Elmaghraby, Some Network Models in Management Vol. 58; P. B. Hagelschuer, Theorie der linearen Dekomposition.
Science. III, 177 pages. 4°. 1970. DM 16, VII, 191 Seiten. 4°. 1971. DM 18,
Operations Research
95
M. Zeleny
Linear Multiobjective
Programming
Spri ngerVerlag
Berl in' Heidel berg . New York
Editorial Board
H. Albach · A . V.Balakrishnan' P. Dh ry mes . J. G reen ' W . Hildenbrand
W . Krelle . K. Ritter' R. Sato . P . Schonfeld
Thi~ work to copyright. All rights are re~tved, wh ether th e whole or part of th e material is concerne.::l,
i~ ~ubje<:t
s~cifically tho~ of tunslation, reprinting, re' u~ of illustrations, broadcasting, reproduction by photocopying machine
or similar means, and storage in data hanks.
Under § )4 of the German Copyright Law where copies are made for other than private use, a ef e is payable to the
publisher, the amount of the fee to be determine.::l b)· agreement with the publisher.
C by Springer·Verlag Berlin· Heidelberg 1974. Library of Congress Catalog Card Number 73'22)77.
Softcover reprin t of the hardcover I st edition 1974
Offselprinting and bookbinding:) ulius Belu, HemsbachfBergstr.
Acknowledgment
Milan Zeleny
Columbia University,
New York, Autumn 1972
ABSTRACT
1. Introduction 1
Appendix: 186
Al. A Note on Elimination of Redundant
Constraints 187
which all objective functions are unchanged or improved and at least one
strictly improved.
John von NeumannY: " .... This [multiple objective situation, M.Z.]
is certainly no maximum problem, but a peculiar and discon
certing mixture of several conflicting maximum problems ••••
This kind of problem is nowhere dealt with in classical
mathematics. We emphasize at the risk of being pedantic
that this is no conditional maximum problem, no problem of
the calculus of variation, of functional analysis, etc. It
arises in full clarity, even in the most "elementary" situa
tions, e.g., when all variables can assume only a finite
number of values."
The problem of the formation of a single optimality criterion from
a number of essentially noncomparab1e elementary criteria first appeared
in [Pareto, 1896].
The concept of "Pareto optimality" (here nondominated solutions)
found its way into operations research in the pioneering work of [Koopmans,
1951]. This was, of course;' in connection with the activity analysis of
production and allocation. ,A more general approach, viewed as a vector
function maximization problem of mathematical programming, can be found
in [Kuhn ,Tucker, 1951]. We should also mention the work of [Markowitz,
Y von Neumann, J., and Morgenstern, 0.: "Theory of Games and Economic
Behavior", 3rd ed., Princeton University Press, Princeton, 1953,
pp. 1011.
4
analysis framework.
approached from the more general viewpoint of Kuhn and Tucker. Most
are essentially twofold. First, we have found that the linear case is
years, satisfactory algorithms have not been published yet. Also, the
12k
X = {x ,x , ... ,x },
where
i
x l, ... ,k.
12k
Also, given {x ,x , ... ,x } we shall define a convex hull of
1 k
x , .•• ,x as
k
I 2
[x ,x , ... , x ]
k
{xix ~ 0, I A.
i=l 1.
1 k
or, in abbreviated form, as C[X] or C[x , .•. ,x ].
Similarly,
1 i i I i
[x ,x ] = {x + AeX x ); 0 ~ A ~ l}
o< A < n.
Given a set X, its interior will be denoted by Int X and its
and interior points with respect to the relative topology. A fine dis
A.,S(x).
The approach works quite well for bicriterial cases [see Geoffrion,
tions as
l
e (x) (e I (x), ... , e (x))
that is, there exists no other x E X such that e(x) ~ e(x) and
e (x) ;t e (x) .
We may denote the set of all nondominated solutions as N and define
v X N as the set of all dominated solutions. It is seen that
 9
.e.
A {I
AA E E.e. ,A. ~o , L\ A. 1, i 1, ... ,D. C23)
l. i=l l.
Defini tion 2.2. Given A E A, let P CA) denote the following problem:
.e. .
Max L A.el.cX)
XEX i=l l.
(c) X closed and convex, Sex) concave for all x E X, one component
1967].
10
x E N n L.
contradiction. Q.E.D.
.e.
A E E and let the set of x E X which maximizes the function A.CX
Because of the above inclusions, we could call N> and N~ the inner and
Under the maximization, both approaches yield the same solution. The
Figure 2. 1. 1.
;\3
Remark 2.1.4. In general, the nonempty polyhedron X c En is defined by
X (211)
Ax* ~ b
~. (Ax*b) = O.
Notice that the definition (211) differs from that in (223). Also,
constraints.
R(x*) (213)
13
th
where Ar is the r row of A, and R = {l, •.• ,m+n}.
We may distinguish two cases:
(1) R(x*) ~. Then x* € Int X, P = 0, A.c = 0
(2) R(x*) ~~. Let PR(x*) and ~(x*) be the vectors derived
from P and A by deleting all components not in R(x*).
Then from (212), we get
or
(214 )
where
Let us define
Remark 2.1.6. H> and H;;' are the positive and semipositive cones generated
by the gradients {c 1 , •.. ,cl }, and G(x*) is the nonnegative cone defined
1lJ.
i l ..... (bl).
Then
l+l. ~ l+l l .
A.ex = l A.e 1 .x = (1 ,A.)e .x + l A.e~.x
i=l ~ i=l ~ i=l ~
i .e.+l
To simplify our notation. let (e e ) c i for i 1,2, ••. ,l. Then
f
(el+l +, X.C i ).x = L ().'.+l
~
i
+ L X.c.)x. ¥ f (221)
i=1 ~ j=1 J i=l 1 J J
.e.+l
Set (e + I X.c )
i
c(X). Then
i=l 1
15
n
c (A). x I c. (A)X. (222)
j=l J J
Let X be defined by
n
X { x Ix E En; I a .x . b . x. ~ 0; r
r' J
= 1, ... ,m;
j =1 rJ J
(223)
1, ... ,nl.
Let
0; I A. < l}.
i=l 1
where PA* is given by (221) and by (222) with A = A*. For simplicity
represent C(A*).
for XO is:
c *I ... c*
m
*
cm+l . .. c.*
J
... c*
n
r Basis c* bO xl ... x
m
xm+l .. . x.
J
" . x
n
c* 0
... .. . .
I xl I YI I
° Ylm +l Ylj " YIn
c*
0
... .. . ...
m x
m m Ym
° I Ymm+l Ymj Ymn
* 2* .. . 2.* ... 2*
:4
0
° ...
'
° m+l J n
Table 2.3.1.
o 0
If x is nondegenerate, then x = {xl, ... ,xm' O, ... ,O}, where
x.
J
Y~
J
> ° for j E J and x. = 0 for j
J
E J.
Looking at the tableau we see that
where
z.* I, ... ,n
J
17
and
m
z*
o
I * 0
r=lcr>'r
for PA* is
m
2. (A*) = I c (A*)Y •  c. (A*).
J r=l r . rJ J
!
2. (A *)  2.* ~l. (~!+l ~ +
~l. ,* i) y.
I\.C + I
*.
A.C:)
J J r=l r i=l 1 r rJ i=l 1 J
(233)
Substituting in (233)
m
Yj  L ./+1 Yrj  c..1:'.+1 and
r=l r J
m
c~  i
J L c!Yrj
r=l
c.
J
we get
.e. * .
z. (1.*)
J
z.*
J
y. + L A. 0: ~ 0, j E J (234)
J i=l 1 J
Max PA* .
XEX
Let
{AIAEE;y.+
.e. .e.LA.c.i >
O;jEJ}. (235)
J i=l 1 J
19
o 0 0
Then A* E A(x ), and the extreme point x solves Max P A for all A E A(x ).
XEX
further developments:
3) If y . ;;
rJ
° for all r and z . (A) =
J
° for I E oA(X°)' then
, 0
a polyhedron adjacent to A(x ) may be constructed such that
and A <f A(xo ). Let I E oA(xo ) and k E J be such thatZ k(I) = ° and 2(A) is
<
a nonzero function of A. If all the elements of the pivot column k, Yrk = 0,
Ak such that
(ii) A(x o ) n Ak ~ $
l
y. + l.\' '1\.
J
~l. >
\ u.
i=l l. J
.
°, J' (236)
·Recall that all the elements of the pivot column Yrk < 0,
l .
Yk + L A.O~ > 0,
i=l l.
l
Yk + L A.O~ < 0,
i=l l.
21
we have
 0 o
A E I1(x ) n Ilk' i.e. , II (x ) n Ilk '" cj>. Q.E.D.
i
{AlA E Ei. y. + L
A.O~ >
0; E J }.
o
(237)
, J i=l]. J
th
Let the k column represent the pivot column, and Ypk is the pivot element
,,,
 0 YljY lk  Yl
, , ,,
 1
, ~PjG  ,0
Yp
,
,, ,, ,, ,
,0
 0 YmjYmk Ym
,
0: z. (A)z (A) z
'J k 0
where zo is defined by (231) and 2j (A) by (234).
Remark 2.3.5. Notice that A(xo ) of (237) may be rewritten simply using
(234) as
(238)
1 1
After the simplex iteration we get x and A(x). Notice that p E Jo
is the leaving column vector. We shall denote the new basic and nonbasic
columns by:
1
21 (A) = y'Zk(A), p (2311)
p
E II
pk
Y .
and Z~ (A) z. (A)  ~Z (A) E II  {p} (2312)
J J Ypk k '
A(x 1 ) = {I
AAE E.t ; Zj1 (A) =
>
0; J. E
}
J1 . (2313)
Note ypk > O. Then (2311) and (2312) can be rewritten as they appear
in (2313):
23
y .
and Z.(A)  ~·z (A) ~ 0 for (2315)
J ypk k
(2316)
Int A(x0 10
) n A(x ) = A(x ) n Int A(x 1 ) = ~.
(2319)
(2320)
On the other hand, since Hk is the separating hyperplane for A(xo ) and
A(x l ) we have
Introducing the kth column into the basis according to Theorem 2.3.4.
two possibilities can be distinguished:
< o
(1) all Yrk = 0, r € Rand Ak is constructed. Then A(x ) n Ak
Hk n A(xo ).
(2) let Ypk > 0 be the pivot element. Then a new basis
~.
A(xO) n I nt Ak =, 'Y.
'"
Then,
A (xo) n Ak =
For (2). The parts (a), (b), (c) have been shown by (2317),
Remark 2.3.7. The following Figure 2.3.1. shows the relations of Theorem
In.U"k.
27 
the A(x 3) cannot be constructed in the adjacent sense because only two
3 0
hyperplanes pass through X. However, A(x ) can be constructed from A(x )
6.
J
Min{y~
r y rj
; y '. >
rJ
ol.J (2323)
In the case (i) we could construct A(J) such that A* E A(J) and J
they produce only a finite number of polyhedra A(J) and open halfspaces
covering of A.
££2Qi. Parts (1), (2) and (4) follow directly from the previous discus
may be expressed by
n{complement Ak }
k
shall state another theorem which will help us to decide the nondominance
t t <
Int A {AlA E E , Ai > 0; I A. I} (2324)
i=l 1
In view of the notation used in Theorem 2.3.6., let us state the following:
Theorem 2. 3.11. 1
.
x , constructed by intro
We have started with the fact that any nondominated extreme point
~. The proof follows directly from Theorem 2.3.9. and Theorem 2.3.1l.
Let XV and XW € Nex Then each of the corresponding A(xv ) and A(xw)
has at least one point in common with A.
v v 'w w
Choose A € A(x ) n A and A € A(x ) n A. Because of the convexity
of A we can write:
[AV.A W] € A.
We know also that the line segment [AV,AW] is contained in
the union of all polyhedra which are associated with bounded solu
tions. Because of a finite covering of A, we can select a finite
sequence of distinct polyhedra {A (xv). A (xV+1), ... , A (xV+~ }, such
v+k w v+i v+i+l
that A (x ) = A (x ). Furthemore, A (x ) and A (x ) are connect
ed in the sense of Theorem 2.3.6. Q.E.D.
Figure 2.4.1.
o o
Figure 2.4.2.
Shaded polyhedra are A(xo ). Crossed are the faces (and their correspond
ing nonbasic columns) which are not to be introduced in the basis. Notice
.e. i
Yk + I
i=l
A. Ok
l.
0 k E J
.e.
Yj + I A.Ot
>
0 j E Jk (241)
i=l l.
.e.
l A. < I
i=l l.
33
with A.
points.
and 2.4.5.
Remark 2.4.3. Notice that using the Strategy II we construct all adjacent
polyhedra to A(x i ) for which ' the intersection with ]KAis nonempty. For
that we do not need to find all the extreme points of A(x i ). We consider
only those which have required properties (the face intersects with1nt A,
Figure 2.4.3.
12 extreme points 13 extreme points
8 faces 13 faces
Figure 2.4.3.
From the previous dis 7ussion we may conclude that the available
STRATEGY I
( Start ')
1 !
For AiEl\. construct P •
Al
and solve M~ p Al•
•
xX
2
I
Find xi and construct I\. (xi)
I
3 ~ NO I
Is I\. (x i ) ' n Int xi E N I
r
A=~ ? I
s
i=i+l I xi E V I
lj.
1
Generate an adja~ent basic (
solution X1 +l
5
NO Is l1 [I\. (xi) n I\.J = A?
1 YES
~
Figure 2.4.4.
 36
STRATEGY I
(Comments)
i i
3. To show that A(x ) nlhtA = </> => x E V let us state the
following:
2.4.2.)
AE A(x o) and A(x o) rilhtA i </>,
A E A(x o ),
0
i.e. x E N. Thus, for all we have
L cx 0 ~ A.cx 1
 0  1
and Lcx A.cx
Since A(xl) nlhtA= </> there is no A E A such that
all A E A,
(iii) cx l cx o
(iv) cx l 'I> cx o
The possibility (i) must be excluded by definition of x € N.
1 0
Possibility (iv): cx 'I> cx implies there is at least one
i = I, ... ,!, with i I> i
c.x c .x0 • Choose A € A such that
all A. > i I> i
0 i f c.x c .x0 and A. = 0 otherwise. Then
J. J.
STRATEGY II
1
i
,
Start)
I i=i+1 I
5
YES
Is ~ [A (xi) n A ] = A titop )
?
NO
6
Find X~T.L E N by introducing
k E J into the basis
Figure 2.4.5.
39
STRATEGY II
(Comments)
The steps (1) and (2) are the same as for STRATEGy I.
nique.
all Npoints have already been discovered, the A may not be fully decom
posed.
x j and A(x j ) may be destroyed. That is, for all {x j } such that {x j }
are optimal solutions to PA for all A E A(x j ) the A(x j ) are identical,
independent of xj. This would imply that although A has been fully
<=>
are the bases associated with xj. then there is a possibility that
. (k)" > .
A(xJ ) c A(x J ). where A(xJ ) is a representation of H n G(xJ ) in A.
Figure 2.5.1.
t.J.2 
c,·
Figure 2.5.2.
;.
Notice that though X is threedimensional, H may be twodimensional.
1 2 1 ;. 2 ;.
Then, although G(x ) ~ G(x ), we may see that G(x ) n H G(x) n H
which implies that A(xl) = A(x 2) ~~. Also, A E A(x l ) n A(x 2) implies
1 2
that x ,x are two alternative solutions of PA'
1
c .X = x2 + x3 + 2x 4 + 3x S + x6
2
c .X = xl + x3  x4  x6  x 7
3
c .X = xl + 2x 2  x3 + SX 4 + 2xS + x7
subject to
<
xl + 2x 2 + x3 + x4 + 2xS + x6 + 2x 7 16
<
2x l  x2 + x4 + 2xS + x7 16
<
 xl + Xs + 2xS  2x 7 = 16
<
x2 + 2xS  x4 + Xs  2x6  x7 = 16
x. ~ 0, i
1.
= 1, ... ,7
Let
table:
lj.lj.
j 1 2 3 4 5 6 7
3
c. 1 2 1 3 2 0 1
J
1
c. 1 1 2 1 1 1 1
J
2
c. 0 2 2 4 2 1 2
J
xl x2 x3 x4 Xs x6 x7 Yl Y2 Y3 Y4
Yl 1 2 1 CD 2 1 2 1 0 0 0 16
Y2 2 1 0 11; 2 0 1 0 1 0 0 16
'
Y3 1 0 1 0 2 0 2 0 0 1 0 16
Y4 0 1 2 1 1 2 1 0 0 0 1 16 (25 11)
1 2 1 3 2 0 1 0 0 0 0 0
1 1 2 1 1 1 1 0 0 0 0 0
0 2 2 4 2 1 2 0 0 0 0 0
x4 1 2 1 1 2 1 2 1 0 0 0 16
Y2 3 3 1 0 0 1 1 1 1 0 0 0
Y3 1 0 1 0 2 0 2 0 0 1 0 16
Y4 1 3 3 0 3 1 1 1 0 0 1 32
(2512)
2 4 4 0 4 3 5 3 0 0 0 48
0 1 3 0 3 2 1 1 0 0 0 16
4 6 6 0 6 3 6 4 0 0 0 64
°
x (0,0,0,16,0,0,0,)
<
4"2 = 2 l. +
<
"1 + 6"2 4 2.
<
3"1 + 6"2 = 4 3.} +
A(xo)= 3"1 + 6" 2 ~ 4 5.
2"1 + 3" 2 ~ 3 6.
<
"1 + 6"2 = 5 7.
<
"1 + 4"2 = 3 8.
the cell (2,4) of Tableau (2 SolI) is the pivot element. We get:
46
Y1 CD 3 1 0 0 1 1 1 1 0 0 0
x4 2 1 0 1 2 0 1 0 1 0 0 16
Y3 1 0 1 0 2 0 2 0 0 1 0 16
Y4 2 0 2 0 3 2 0 0 1 0 1 32
(2513)
7 5 1 0 4 0 2 0 3 0 0 48
3 2 2 0 3 1 0 0 1 0 0 16
8 6 2 0 6 1 2 0 4 0 0 64
row indicates that we have to introduce the first column before the
1 1 1 1 1
xl 1 1 0 0 3 0 0 0
3 3 3 3
2 2 5 2 1
x4 0 1 3 1 2 3 3 3 0 0 16
3
4 1 5 1 1
Y3 0 1 3 0 2 3 3 3 1 0 16
3
8 4 2 2 1
Y4 0 2 0 3 3 0 1 32
3 3 3 3
(2514)
10 7 13 7 2
0 2 0 4 0 0 48
3 3 3 3 3
0 1 3 0 3 2 1 1 0 0 0 16
14 5 14 8 4 0
0 2 3 0 6 3 3 3 3 0 64
47
Al  2A2 ;l 2
14 < 10
3Al + 3"A2 =3"
3Al + 6A 2 ;; 4 +
14 < 13
Al + 3"A2 = 3""
Al + ~A ;; 7
3 2 '3
4 ;; 2 +
'3A2 '3
3Al + 6A 2 ;; 4
A ;;.!.
2 2
3Al + 6A 2 ~ 4 3.
3Al + 6A 2 ;; 4 5.
A(xo) A ;l.!. 1.
2 2
Al + A2 ;l 1
Al ,A 2 ~ 0
48 
Looking at original Tableau (2 512) we see that the first and the
into the basis. Notice the 2nd column is corresponding to the redundant
1 1 1 1 1
x2
Z
3
1
Z
1
Z
3
1
1
Z 1
1
° °
Z ° 8
Y2 Z
° Z Z 3
Z 2
Z 1 ° ° 24
° ° ° °1 ° °
Y3 1 1 2 '2 16
1 3 3 5 2
Y4 Z
° Z Z ° Z Z ° ° 1 8
(2515)
Z ° z I z ° I ° ° ° 16
1
°x =3 (0,8,0,0,0,0,0)
3
° ° ° 1 ° ° °
Actually x is a dominated extreme point and the corresponding Aex) is as
follows:
49 
1
7'1  A ~ 0
2 1.
5
IA1 + 3A 2 ~ 2 2.
1
IA1 + 3A 2 ~ 2 3.
A (';.)
<
ZA1 0 4.
3 <
IAI 1 5.
1 A :; 1
Yl +
2 6.
2.5.1.2 .
Notice that A(;) is an empty set as could be seen in Figure
i·
\. Figure 2.5.1. 2.
50
We get:
xl 1 2 1 1 2 1 2 1 0 0 0 16
Y2 0 3 2 3 "
',6,' 2 5 2 1 0 0 48
Y3 0 2 2 1 4 1 0 1 0 1 0 32
Y4 0 1 CD 1 1 2 1 0 0 0 1 16
(2516)
0 0 2 2 0 1 1 1 0 0 0 16
0 1 3 0 3 2 1 1 0 0 0 16
0 2 2 4 2 1 2 0 0 0 0 0
1
x (16,0,0,0,0,0,0)
straints we get:
31. 1 + 21. ~ 2 3.
2
41. ~ 2 4.
2
A(x1) 31. 1 + 21.2 ~ 0 5.
Al + 1.2 ~ 1
~ 0
1.1'1.2
See Figure 2.5.1.1.
51
1
Notice at Al = 3' A2 = 21 the constraints corresponding to the third,
fourth and fifth columns are active and may be introduced into the basis
according to Theorem 2.3.4. However, only the third and the fifth are
eligible since the fourth would lead us back to xo. Choose the third
column first:
Xl 1
3
2 0
CD CD 2 5
2
1 0 0
1
2 8
Y2 0 2 0 4 5 4 6 2 1 0 1 32
13 \
,~
Y3 0 1 0 2 ... / 3 1 1 0 1 1 16
.1 1 1 1 1
x3 0
2 1 2 2 1 2 0 0 0
2
8
(2517)
0 1 0 1 1 3 2 1 0 0 1 0
1 3 3 5 3
0
2 0 2 2 .5 2 1 0 0
2 8
0 3 0 3 3 1 1 0 0 0 1 16
X2 (8,0,8,0,0,0,0)
3 3A ~ 1 4.
YI + 2
3 >
iI + 3A 2 =1 5.
A(i) 3 >
construct 2AI + A2 =1 11.
<
Al + A2 =1
>
AI' AZ 0
difficulty discussed in section 2S. Notice that for the fourth and
is in agreement with the theory, since no two polyhedra may have a common
2 4 5 2 16
x4 1 0 1 1 0 0 3
3 3 3 3 3
8 4 2 2 1 32
Y2 3 2 0 0 1 3 3 3 0
3 3
4 1 7 1 1 16
3 1 0 0 1 3 3 0 1 3
Y3 3 3
1 1 1 1 1 32
x3 1 1 0 1 3 0 0
3 3 3 3 3
(25 18)
2 13 11 5 4 16
0 0 0 0 0 0 3
3 3 3 3 3
1 2 0 0 0 3 0 0 0 0 1 16
2 0 0 0 0 5 4 2 0 0 2 0
3 32 16
x (0,0'3'3,0,0,0)
2
 Al + 2A2 ;;; 3 1.
3>'1 + 5A 2 ~ 13 6.
3
 3 ~ 4 11.
construct A(X ) Al + 2A2 3
Al ;;; 1
+ A2
>
AI' A2 0
See Figure 2.5.1.1.
53
Notice in Figure 2.5.1.1. that we would never introduce the sixth column
since the associated constraint (better, the face) does not have a point
in common with A.
2 4 5 2 1 16
x5 3
10
1
° 1 1
3
8
3
7
3
4
° ° 3
2
3
16
Y2 3 3
° 1
° 3 3 3 1 ° 3 3
Y3 2
1
2
° 1
° 1
5
4
4
1
1
° 1 °
2
°
16
x3 3 ° 1 1
° 3 3 3 ° ° 3 3
(2 .'i.19)
2 13 11 5 4 16
3 ° ° ° ° 3 3 3 ° ° 3 3
1 2 ° ° ° 3
° 2° ° ° 1 16
2
° °
4
°
16
°
16
5 4
° ° 2 °
x (°'°'3'°'3,0,0)
< 2
 \ + 21.2 3 1.
< 13
31. 1 + 51. 2 =3 6.
section 2.3.
1
"2A1 + 3A 2 ~ 1 2.
3 >
"2A1 + 3A 2 1 4.
3 >
Y1 + 3A 2 1 5.
<
A (x 2)  5A 1 + A2 = 3 6.
5 <
"2A1 A2 = 2 7.
<
A2 = 1 8.
3 >
"2A1 + A2 1 11.
8.
Notice that the face corresponding to the sixth column (i. e. > constraint
2
6.) of A(x ) satisfies
H6 n A(x 2) rifutA = ~.
1 3 3 3 5 1 1
x6 2" 4 0 4 4 1 4 2" 0 0 4 4
Y2 2 2 0 1 2 0 1 0 1 0 0 16
1 3
Y3 z3 5
4 0 4 4 0
11
4 z1 0 1
1
4 4
1 5 1 5 3 1 1
1 0 0 0 12
x3 Z 4 4 4 4 Z 4
1
z3 13
4 0 13
4
13
4 0 4
7
z1 0 0 4 12
1 1 0 1 '1 0 2 1 0 0 0 16
1 1 1 1 1
1 0 0 0 0 0 Z 8
Z Z Z Z
8
x E V.
 1
Let us return to the Tableau (2516) and consider A(x ) again.
>
The constraint 3"1 + 2"2 = 0 has only one single point in common with
 1
A(x ), specifically 1 =
"
1/3,
" 2 = 1/2. Then, of course, the adjacent
5
polyhedron resulting from introducing the fifth column, say A(x ),
1 1 1 1 1
xl 1 1
1
3
1
°1 ° 3
1
3
5
3
1
3
° ° °
1
Xs
° 2 3
2
2
1
3
1
6
10
3 6
1 2
° ° 8
Y3
° °1 3
5
1
3
° 3 3'
7 11
3 3
1 1
1
° °
Y4
° 2 3 2
° 3 6" 3 6
° 1 8 (25110)
° °1 2 2
3 ° 1
3
1 1
°1 ° ° 168
° 2 2
8
2"
° 1
1
2
1
°2 2"1 ° °
° 1 3
5
3
° 3 3 3 3
° ° 16
X (0,0,0,0,8,0,0)
8 <
2Al + ?'2 =2 3.
3 + 3A ~
2 4.
¥1 2
2Al  k
construct A(x S) 1 ~
3 2
Al +
< °
A2 = 1
9.
>
Similarly introducing the fourth column would lead to xo. So, only the
covered with no "holes" and still not all nondominated extreme points
S  S 1 1
may be generated as it is demonstrated by x and A(x ) = (3'2).
transformation does not have to be performed. Also, from all the remain
polyhedra corresponding to the same extreme point (but two or more dif
ferent bases). However, this can always be resolved through the series
fifth column but this time yhoose 3 to be the pivot element. We get:
1 1 1 1
xl 1 1 0 0 2 0 2 0 0
2 2 2
1 2 13 1 S 2 16
0 0 0 1 1 3
Y2 3 "3 3 3 3 3
1 2 1 1 1 1 16
Xs 0
3 0
3 1 1 3 3 0
3 3 3
1 S 3 2 1 1 2 16
x3 0
3 1 "6 0 2 3" "6 0 "6 3 3
(25 111)
2 1 7 4 1 4 16
0 3" 0 3 0 4 0 3"
3 3 3 3
1 7 1 1
0 1 0 2" 0 2 2 2" 0 16
2"
0 2 0 1 0 4 0 1 0 1 2 0
4 16 16
x (0,0'3,0'3,0,0)
4
which is the same as x
 sa
.!.A + > 1
:\2 4.
2 1 3"
7 <
i l·1 + 4A Z 4 6.
4 < 1
Construct A(x ) .!.A + 10.
2 1 1..2 3"
> 4
Al + 2h Z 3" II.
<
Al + AZ = 1
>
AI' AZ 0
iterations.
Z 3
Considering only the calculated x O, xl, x , x ,
see that
A(i 4). We list all Npoints together with the values of the objective
functions.
0 1 Z 3 4 5
x x x x x x
c 1 .x 32 0 8 21. 33 21. 33 24
2
c .x 16 16 16 5.33 5.33 0
3 5.33
c .x 48 16 0 5.33 16
 59
\
\
\
\
\
\
\
\
\
\
\
\
,." ......................
..... .....
let us make the following remark. We have shown in Remark 2.s .1. 2.
subject to
<
2x l + x2 20
5 <
6"xl + x2 = 10
>
xl + x2 10
0
We get the following feasible solution, say x :
1 2 3 4 5
xl 1 0 0 6 6 0
x2 0 1 0 6 5 10
0
0 0 0 18 19 10 x (0,10)
0 0 0 6 5 10 + x0 € N
61
A(xo ) is given by
>
18A l + 6A 2 0 redundant
>
19A l + SA 2 0
24A2 > 19
7 5
Y2 0 0 L 1
"6 3"
1
xl 1 0 6 0 1 10 x (10,0)
x2 0 1 6. 0 2 0
0 0 18 0 2 40 nondominated solution
0 0 6 0 2 0
0
Introducing the fifth column in x (nonredundant) we obtain also a non
dominated solution:
1 6 10
0 0 1
Y3 7 7 7
6 6 60 2
xl 1 0
7 7 0 7 x = (60
7'720) wh'1Ch 1S
.
nondominated as can be
5 12 20
x2 0 1 7 ., 0 7" seen in Figure 2.5.1.4.
19 12 260
0 0 7 7 0 7
5 12 0 20
0 0 7 7
7
 62 
10
~
Figure 2 .5.1. 4.
63
cl
1
.. . 1
c I cm
m +l .. . 1
c. . .. c1
n
J
· · · · · · · · · · · ·
· · · · · · · · · · · ·
· · · · · · · · · · · ·
m xm 1
cm 0
Ym 0 ... 1 Ymm+l ... Ymj ... ~
zl
0
0 ... 0 z(l)
m+1 ... z~l)
J
. .. z(l)
n
Table 3.1.1.
64
where 1
1, .•. ,n and z
o
o
If all o for all j, then x
0 0
= (Y l ' Y2 ,
0
... , Ym' 0, ... ,0)
1 0
is a maximal feasible solution with c .x is an
extreme point of X.
2 J(.1) > 0 f
or'J = m+ 1 , ... , n. In the case of alternate solutions,
i
C.x i 1, ... ,t.
a column vector
65
2.
J
(3 11)
1
z0,
2
zo·
z
o
(312)
or
~l 1
z z
o o
(2)
2.
J
6.
J
.e.
z 2~.e.)
o J (313)
(b) 'f
1 2j ~ 0, t h en lntro
. d '
uClng t h e J.th co 1umn lnto
. t he
P
~.f For ()
a. ·
I ntro duClng th
e 'J th co 1umn lnto
. t h e b aSls,
. we get
1
a new adjacent extreme point, say x , for which zo ~ z0
because 6.2. ~ O.
J J
th
l'£QQi • Introducing the k column, we get zo; and introducing the
. th :::
J column, we get z
0
Then z
0
z
0
 9k 2k and z
0
z
0
 9.2 .•
J J
2~i)
]
~ 0 for all j E J then the i th objective function is at
Then the only columns eligible for an introduction are those which
x is dominated or nondominated.
n
Since x € X, i.e., I a .x. = b r , r = 1, ••• ,m, let us
j=l rJ J
i
c.x  E.1 + Yi
i
c.x, i 1, ••• ,.e.. ( 3..14)
Max v I E.
1
i=l
subject to
Ax b
i i
c.x, ,.e.
c.x E.1 + Yi i 1, ...
> 0,
E.1 = X ~ O. ( 315)
i i i i
Since c.x c.x + E. we have c.x > c.x for at least one i.
1
Thus cx ~ cx and x € V.
69
cx ~ cx, i.e., x E N.
The above discussion may be summarized in Theorem 3.1.6 •
.e.
Theorem 3.1.6. Solve the following LP problem: Maxv,v= Le:.
i=l 1
Max v = O.
In order to derive an etficient method for an application of
Theorem 3.1.6., let us consider incorporating.e. additional constraints
· ·· · ·· ·· · · ·· · ·· ·· ·· ·· ·· · ·
· · · · · · ·· ·· · · · · · · · 0
m 0X·..m
1
Ym(m+1) ... Ymj ... Ymn 0 ... 0 0 ... 0 Ym
              
· I
· · ......
o
·· ··
I
·· ·· ·· ·· · ·· ·· ·· ·· ··
· · · · · ··· · · · · · · · ·
m+.t 0 ... 0 ••• Y Cm+i)] ••• YCm+.t)n 0 ... 0 1
·.. 0 1
Y.t YCm+.t)Cm+1)
· · · · ·· · · ·
·
· ·· ··· ·· ·· ·· ··· ·· · ··
0 :lCi) ... :l~.t) ... :lC.t) 0 ... 0 0 ·.. 0 z.e. 0
·.. 0 m+1 ] n

Table 3.1.2.
~71
o 0 0
x (Yl' Y2' ••• , Ym' ~
l
where all Ei' i = l, •.. ,l are outside the new basis. If we are
level, then
l
I E.
1.
> 0 and X E v.
i=l
x  ndimensional vector
V Max ci.X .
~ = 1 , ••• ,<
D subject to Ax =b, x ~ 0
may have the associated initial simplex tableau written in the matrix
notation as follows:
Tableau 1.
[:~f;;~f~: ]
(1)
(2)
~ <XDl : <xl
point x and Bk1 ' its inverse. Then the simplex tableau associated
with Bk is given by:
Tableau 2.
(3)
(4)
at x, i.e.
73
Tableau 1:
Tableau 3.
omx.e.
I
(5) A
I
1mxm
I
: b
I
°
rrrr
I I
mx.e.
(6) C :I 0l"xm
0 :
I
ex
] no:
t. X ..{... I
1..{..x.(..
n n :
I
rrrr
I 1 I I
(7) C l
1
°.e.xm l
I
°.e.x.e. l
I
°.e.x.e. l
I
°.e.xl
Tableau 4.
I
(8) : Elb
: k
TTTT
1 I I I
I : 1: : :
(9) CsEk A+C : lsBk : l.e.x.e. : ].ex.e. : 0.e.Xl
TTTT
I I I I
1
Note. Row (8) 13 k . Row (5)
1
Row (9) Row (6)  CJ\ • Row (5)
1
Row (10) Row (7) + CBB k • Row (5).
74
Remark 3.1. 7.
(i) Notice the right hand side of the Row (9) is 0lxI
, 1
because fX = f/'k b.
(ii) Comparing the Rows (9) and (10), notice that because
of the second column, we can write y . = _2~rm) for
rJ J
r = m+l, ••• ,m+l and j = m+l, ••. ,n. (See Table 3.1.2.)
Since the values of z~rm) are already known as the
J
indicators in criterial rows for the basic solution X,
Tableau 5.
,
(i)
k
BlA
____________
B 1
, ___________
~ k
0
mxl ,
l, __________ l _________ l ______
k _
°mxl :: Blb
I I 1 I
of Tableau 6:
75
Tableau 6.
I
(iv)
I
13 k
I <.. I
0mxh
<.. I k
0mx n i Blb
~t~ r
(v)
1
tBBk AC i
I
tl)k
1
i
I
1f.xf. i
I
ttt ~
1 f.xf. i
I
°f.xl
1 1
k AC] i k] ! i i
I I I I
(vi) 1 lxf.[C BB 1 lx f.[C BB °lXf. 11Xf. o
This llumn
can be omi tted.
may be constructed directly from the latter one. The Row (vi) repre
sents new criterial indicators of the new problem. From this row
element of Row (vi) which is negative, say the jth , and all elements
of the jth column in Row (v) are negative, then for e. > 0 we may
J
conclude that Max v > 0 and the corresponding basis i" is dominated.
above property will be used in the subroutine for checking the non
to be appended.
(1)
2.
J
(2)
2.
J
2.
J
(i+1)
2.
J (316)
Let !+1
L
i=l
A.2~i)
1 J
(1 f
i=l
A.)
1
2~!+1)
J
+ f A.2~i)
i=l 1 J
2~f+1)
J
+
i=l
f A.
1
[2~i)
J
(i+1) (i)
From (234), we may see that y. = 2. and 8~ 2.
J J J J
methods, let us choose some basic solution and its tableau (for
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. II.
xl 1 2 1 1 2 1 2 1 0 0 0 16
Y2 0 3 2 3 6 2 5 2 1 0 0 48
Y3 0 2 2 1 4 1 0 1 0 1 0 32
Y4 0 1 2 1 1 2 1 0 0 0 1 16
0 0 2 2 0 1 1 1 0 0 0 16
0 1 1 2 3 1 0 0 0 0 0 0
0 2 0 2 2 2 3 1 0 0 0 16
We may see how we can read the constraints of A(xl) from the
last three rows of this tableau. (We consider the first row as rep
Nonbasic
Columns Constraint
Calculation
j
8 1 + Al (01) + A2 (11) Al a 1
78
Figure 3.1.1.
79
which gives
Remark 3.1.9. It is not the degeneracy which could make the subproblem
indecisive, but rather a possibility of e. = 0 ].'f teJ
h .th coumn].s
1 .
J
the one to be introduced. Notice that the degenerate solution itself
does not imply e. = O.
J
(Notice that two or more different v's may correspond to the same
Nex Basically, two approaches are considered; see e.g. [Hadley, 1961
including v.
Remark 3.3.1. Let us agree that the phrase "to calculate v" means to
k k k
identify the basic solution (or an extreme point) x = (xl, ... ,xn )
correspond to vo of r.
(i) Let Rl = {vol and WI = r(v o)  R1 • Discard those vertices
vertices in WI.
Transform W2 into W2 •
found, i.e., Rk = V.
(v) If W
k ~~, calculate all vertices from Wk and form
and
83
each step k not only all vertices in ~ but also all inverse matrices
adjacent vertices.
as follows:
and construct ~
k+l and Wk+l as above. Transform
and Rk = V .
3,_________ L_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ,
Let [vk } denote all vertices in WI
lj.
9rL_N_O_..
1
8
6: __________ L____________ .
NO
Figure 3.2.1.
86
since Ro = <1>.
construct WI +l which collects only those which were not yet explored.
01£ x
1+1
E N, then transfer vI+l
o Remove the decided vertex from WI +l and check 0 again. After all
be empty.
87
found.
If 1+ I ;,
W ~, we add W
1+ I to our current R1+ I formed in
2rL.
3
YES
Sto
D=D+l
4~LL~
NO Is there any other v I +1 E WI +1
with the distance D from vI ?
Figure 3.2.2.
89
of adding to RI +I all adjacent vertices WI' we add only one which was
Rl = vI and Wo = <p.
i.e. with the distance D+l. ' There is always at least one vertex vI+l
because WI +l # <p.
1+1 1+1
~ Check the nondominance of corresponding x If x E N,
continue with ~ .
Figure 3. 2.3.
91
Method A.
iii" = q, STOP.
4
Method B.
92
7) R
7
= {Yo' vI' v2 ' v 3' v4 '" V6 ' vS}
W7 = {v 7 ' vg , vIO' v 11 }
10) RIO = {Yo' vI' v Z' v 3' V4 ' v6 ' vS' vg , v 7 ' v 8}
WIO = {vIO' v11 ' VIS}
1 o Y1(m+1)
x
m
o 1
Ym(m+1)
j
o o 2(1)
m+1
••• 2(1)
n z~
o o /o
I I
(331)
(I) Check the (i+l)st row of the simplex tableau and find the
largest negative coefficient. If there is none  PRINT since
000
x (Yl' ••• , Ym) is nondominated.
(2) In the column chosen at (1), say the jth, we may find two cases:
1
Is there any j E J such
(ktl) < 0
that Zj ?
YES
5 ~
Choose z.(i) > 0 as the pivot
]
element
Figure 3.3.1.
97
3.1.6.)
2(i+l)
j ,
and if all 2~i) ~ 0 (Notice that 2. " 0 since 2~i+l) < 0)
0 .
J J J
(i)
then go to ~ . If there is at least one 2. > 0 go to
J
(DIf e.
J
> 0, ·
ob v~ous 1Y b
y '~ntro d '
uc~ng t h e J.th co 1umn, we get
(i)
~ Perform the simplex iteration on 2. > 0 fromeD and go
J
back to (2) .
~ If we added the rows giving ej = 0 in 0 then we may explore
them after each step for any y. > O. If there is none, then
rJ
actual e. > 0 since we assume bounded solution exists. If there
J
is Yrj >'0, then e. = 0 and we perform the next iteration around
J
Xl ·.. x
m Xm+l ·.. X
n Yl ... Y.e
xl 1 ·.. 0
Yl(m+l) ·.. YIn 0 ... 0 Yl
0
·· ·· ·· ·
··· ·· ··
· ··
··
· · · ·
xm 0 ... 1 Ym(m+l) ·.. Ymn 0 ... 0 Ym
0
r
·..
1
Yl 0 ... 0
2 (1)'
m+l
2(1)
n
1 . .. 0 0 (33 2)
··· ··· · ·
·· ··
·
··· ··· ··
· ·
Y£. 0 ·.. 0 2~:i ·.. 2(£.)
n
0 ... 1 0
J
J
·..
J
... .. .
J
0 0
J
J 2~:;1) 2(£.+1) 0 0 0 J
J n J
J
~ .J
scheme which had to be expanded in much more detail for actual pro
? rNO=f\.!?St~o~pu
',",
1
NO Is Zj(i) 2EO for at least Introduce the j
one i? all j E J
column to get xi+l
2
YES i
Is x unique?
NO
5r~~~L,
Is there any other j E J such that
~ 9k 3 k for all k E J ?
NO
6
Is there any j E J such that Zj ~ 0 ? o
YES
Xi EN? j"I.U_ _ _ _ _ _,
S
Store all those jth columns
which would lead to an unex
plored basis
Figure 3.3.2.
100
to 0. Otherwise, go to ®.
~ Use the nondominance subroutine as described in Figure 3.4.1.
all other points reachable from xi. If there is one yet unexplored,
we make the transformation. Otherwise, go to ~ •
some, but i E V, go to CD .
~ Select and store those jth columns (and their bases) which would
lead to an unexplored solution. These are bases which might potentially
be nondominated.
step by step.
1
c .X xl + 2x 2 x3 + 3X 4 + 2XS + x7
2
VMax c .X x2 + x3 + 2x4 + 3x S + x6
3
c .X xl + x3  x4  x6  x 7
<
subject to xl + 2x 2 + x3 + x 4 + 2xs + x6 + 2x 7 16
<
2x 1 x2 + x4 + 2xs + x7 16
<
 xl + x3 + 2Xs 2x 7 16
<
x 2 + 2x 3  x 4 + Xs 2x 6  x7 16
x. ;; 0, i 1,2, ... 7.
1
L = 2x 1 + 3x 2 + x3 + 4x4 + Sx S '
Y1 1 2 1 1 CD °1 ° °
1 2 1 16
Y2 2 1
°1 1 2
° 2 °1
°1 ° 16
Y3 1
°1 ° 1 2° 1 ° °
2
°1 16
Y4
° 2 1
° ° ° 16
1 2 1 3 2
° 1
° ° ° ° °
° 1 1 2 3 1
°1 ° ° ° ° °

1
° 1 1
° 1
° ° ° °
 °

L °
2 3 1 4 s
° ° ° ° ° °
102
Theorem 3. 1. 2.
1 1 1 1 1
xs 2" 1
2" 2" 1
2" 1
2" ° ° ° 8
Y2 3 3 1
° ° 1 1 1 1
° ° °
Y3 2 2
° 1
° 1 4 1
° 1
° °
1 3 3 5 1
Y4 2"
° 2" 2" '
° 2" 2 2"
° ° 1 8
° ° 2 2
° 1 1 1
° ° ° 16
. 3 1 1 1 3
2" 2 2" 2" ° 2" 3 2" ° ° ° 24
1
° ° °
1
° ° ° ° °
1 1
(334)
0 0 2 2 1 1 1 1 0 0 0
3 1 1 1 3
2 2 3 0 1 0 0
2 2 2 2
1 0 1
CD 1 1 0 0 0 1 0
1 3 3 5 5
2 2
2 2 2 5
2 0 0 0 0
2 0 0 Q 3 3 1 1 0 2 0
7
C0
3 1
2 0 0 1 0 1 0
2 2 2
1 0 1 1 1 1 0 0 0 1 0
13 5 3
1 2 0 0 4 0 0 0
2" 2 2
0 4 0 0 5 10 4 1 2 3 0
7 3 1
1 2 0 0 1
2 2 0 1
2 0 PRINT
9 3 3
0 2 1 1 2 0 1 0
2 2 2
0 4 0 0 5 10 4 0 1 2 0 =l> Max v = 0
104
x4 CD 2 1 1 2 1 2 1
° ° ° 16
Y2 3 3 1
° ° 1 1 1 1
° ° °
Y3 1
° 1
° 2 ° 2
° ° 1 ° 16
Y4 1 3 3
° 3 1 1 1
° ° 1 32
2 4 4
° 4 3 5 3
° ° ° 48
2 3 1
° 1 1 4 2
° ° ° 32
2 2 2
° 2
° 1 1
° ° °
 ~ 
16
I 2 5 3
° 3 4 8 4
° ° ° 64
(3 35)
example, the first and the second objective functions are at their
Xl 1 2 1 1 2 1 2 1 0 0 0 16
Y2 0 3 2 3 6 2 5 2 1 0 0 48
Y3 0 2 2 1 4 1 0 1 0 1 0 32
Y4 0 1
CD 1 1 2 1 0 0 0 1 16
0 0 2 2 0 1 1 1 0 0 0 16
0 1 1 2 3 1 0 0 0 0 0 0
0 2 0 2 2 2 3 1 0 0 0 16
(3 36)
4
Again, PRINT X = (16,0,0,0,0,0,0) since the third objective function
is at its maximum here. The alternate solution resulting from intro
ducing the third column gives us a vector of values of objective
functions:
o 16
8 o
16 16
24 32
106
xl 1
3
2 0
CD 2
3
2
2
5
1 0 0
1
2 8
Y2 0 2
° 4 5 4 6 2 I
° 1 32
Y3
° 1
° 2 3 3 1 1
° 1 1 16
1 1 1 1 1
x3
° '2 1 2 2 1 2
° ° ° 2 8
° 1
° 1 1 3 2 1 0
° 1
°
1 5 5 1 1
° 2
° 2 2 2 2
° ° ° 2 8
0
°2 2 2 2
° ° 3
° 1 16
(337)
x4
3
2
1 0 1
CD 3
4 5
3
2
3 0 0
1
3 3
16
8 4 2 2 1 32
3 2 0 0 1 3 3 3 1 0
Y2 3 3
4 1 7 1 1 16
Y3 3 1 0 0 1 3 3 0 1 3
3 3
1 1 1 1 1 32
1 1 0 1 3 0 0
x3 3 3 3 3 3
2 13 11 5 4 16
0 0 0 0 0 0 3
3 3 3 3 3
5 4 11 5 1 64
2 0 0 0 0 0 3
3 3 3 3 3
4 2 1 1 2 16
3 0 0 0 0 3 3 3 0 0 3 3
L 1 2 0 0 0 5 7 3 0 0 1 32
(33 8)
6 32 16
Here x = (0,0'3'3,0,0,0) and since no objective function
2 13 11 5 4
3 0
3" 3" 3 3 1 0 0 0
5 4 11 5 1
2 0 1 0 0
"3 3 3" 3 3
4
3" 0
2
3
1
3"
1
3" CD 0 0 1 0
1 2 5 7 3 1 0 0 0 0
2 0 3 3 1 0 1 0 2 0
0
21 3 1
2 1 (9 0 1 0
6" 2 2
1 1 3
2 0 1 2 1 0 0 0
2" 2
13 5 3
1 2 4 0 0 0 0
"2 2 2
0 4 5 10 4 0 1 2 3 0
21 3 1
1 2 1 0 0 1 0
6" 2 2
13 5 5
0 4 1 1 0 2 0
"2 2 2
0 4 5 10 4 0 0 1 2 0
109
Xs
2
"3 1 0 1 1
CD "3
5 2
"3 0 0
1
"3 3
16
10 8 7 4 2 16
3 3 0 1 0 "3 "3 "3 1 0
Y2 "3 3
Y3 2 2 0 1 0 1 4 1 0 1 0 0
1 5 4 1 2 16
x3 "3 0 1 1 0 "3 "3 "3 0 0
"3 3
2 13 11 5 4 16
0 0 0 0 0 0 "3
"3 3 3 "3 3
5 4 11 5 1 64
2 0 0 0 0 0 "3
"3 "3 3 "3 3
4 2 1 1 2 16
0 0 0 0 "3 "3 0 0
"3 "3 "3 3
  
L 1 2 0 0 0 5 7 3 0 0 1 32
(339)
110
7 16 16
PRINT x = (0,0':3,0':3,0,0).
x6
1
Z 4
3
° 4
3
4
3
1
CD 1
2
° °
1
4 4
Y2 2 2
° 1 2
° 1
° 1
° ° 16
3 5 1 3 11 1 1
Y3 2 4
° 4 4 ° ""4 z ° 1 4 4
1 5 1 5 3 1 1
x3 
2 4 1
4 4 ° 4 Z ° ° 4 12
3 13 13 13 7 1 1
Z ""4
° ""4 ""4
° 4 z
° 0 4 12
1 1
° 1 1
° 2 1
° ° ° 16
1 1 1 1 1
1
Z ° Z Z ° Z ° 0
° Z 8
3 7 15 15 3 1 1
l: ° °
Z 4
° ""4 ""4
° 4 Z 4 12
(3310)
8 8
So, x = (0,0,12,0,0,4,0). To check the nondominance of x ,
z3 13
4 4 4
13 13 7
4
1
2"
1
4 1 0 0 0
1 1 1 1 2 1 0 0 1 0 0
CD
1 1 1
1
Z 0 1 0 0 1 0
Z Z 2
15
z3 4
7
4 4
15 3
4
1
Z
1
4 0 0 0 0
,....,
3 1 13
8 0 0 0 Z 3 1 0 0
Z 2
1 2 0 0 3 1 1 0 1 2 0
2 1 1 1 1 0 1 0 0 2 0
9 1 15
9 2 0 0  4 0 0 0
2 2 2
''"
t
4 11 11 11 7 1 3
5 5 5 5 5 5 5 1 0
° °
1 1 11 11 8 1 2
5 5 5' 5 5 5 5 ° 1
° °
6
5
1
5 CD 1
5
2
5
1
5
3
5 ° ° 1
°
9 11 21 21 3 1 2
5 5 5 5 5 5 5 ° ° ° °
~
14
° ° ° 3 2 6 1
° 11
°
13 2
° 0 6 2 7
° 1 11
°
 6 1 1 1 2 1 3 0
° 5
°
107
27
'.../
2
° ° 9 4 13
° 0 5
°
or nondominance N:
2 4 5 6 / 8
x9
3 7
It X X X X X X
1
c .x 16 48 16 0 5.33 5.33 12 6.4
2
c .x 24 32 0 8 21.33 21.33 16 9.6
3
c .x 0 16 16 16 5.33 5.33 8 6.4
N N N N N N ,0 0
(3312)
9
Also, X is dominated by the same point, i.e.,
114
3. 4. Computer Analysis.
x2 x4 x5 , x6 x3 x7
c 1 .x 16 16 0 5.33 @) 5.33
c3 .x 0
® ® 5.33 16 5.33
~ 40 32 24 32 64 32
vMax SX l  2x 2 + SX 3 + 6x S + 7x6 + 2x 7 + 6x S
xl + x 2 + x3 + x4 + Xs + x6 + x 7 + Xs
subject to:
xl + 3x2  4x 3 + x4  Xs + x6 + x7 + Xs < 40
xl + x2 + ,x3 + x4 + Xs + x6 + x7 + Xs > 12
x. ;:
1
o. i = 1 ••••• 8
1 2 3
x x x
1
c .x 173.33 G6.8J) 170.55319
c 2 .x
3
178.66 176.33
€3
c .x 35.11 38.611 €.34~
number of Nex is only three for this example. The program handled
the situation quite effectively as it is reflected in total time
of 0.814 minutes. The next problem has been also done on IBM 360/91.
vMax
118
subject to:
4x 2  x3  x 4  3x S + Xg < 19
xl + x2 + x3 + x4 + Xs + x6 + x7 + Xg > 12
of results:
1 2 3 4 5
c .x c .x c .x c .x c .x
cant.
119
1 2 3 4 5
c .x c .x c .x c .x c .x
cant.
120
5
1
a .x
2
c .x
3
c .x c 4 .x c .x
~~~
48 19.55 46.46 66.00 24.44
cant.
121
1 2 3 4 5
c .x c .x c .x c .x c .x
()
a ·
S1nce x 1 ,x 2 E V, th ere eX1st
. some x1 ,x2 E X su ch t h at cx1 ~ cx I
1 2
and cx 2 ~ cx2. Let x = Axl + (I_A)x 2 . Notice, AX + (lA)x E X. Then
Acx l + (I_A)cx 2 ~ Acx l + (1_A)cx 2 = cx.
For 0 ~ A < 1 let x x2 A(xl_x2), i.e., x 1 2
(b) = + E (x ,x ]. Notice

that x

= x2 +
1 2
A(X x ) E X, and cx = cx2 + ACX
I
 ACX
2
=
2 1
(1A)cx 2 + Acxl ~ (lA)cx + ACX cx.
1 2
(c) Notice x ,x E V implies x E V, i.e. , contradiction.
2 1 2
x E N or x E N, x E V, this implies x E V and we have a contradiction
1 2
again. So x E N must imply xl,x2 E N. In order to see that [x ,x 1 c N,
123
 2
X E (x,x). According to part (b) of this lemma, (x 1 ,x]

c V, and
 2
[x,x ) c V which imply x E V.
(d) Proof is similar to that for part (c) since this is a partial
C[x l , ... ,xk] to be a convex hull of the set of points {xl, ... ,xk }. Also
recall from our notational agreement in section 1.4. the relative interior
I k
Theorem 4. 1. 2. Let H = C[x , ... ,x ] be the convex hUll of the set of
1 k
points {x , ... ,x }. Then
iiI k
(i) If at least one x E V, X E {x , ... ,x } then there is some
point ~ E H such that x E (x,~). It follows from Lemma 4.1.1. (a) and
x= {xix E En. A x ~ b }
'r r '
(411)
Let N = {xi}, i E K = {I, ... ,k}. Since any such extreme point
ex
i Each xi E N
of X is also a boundary point of X, then R(x ) ~ ~, i E K.
i <
divides the components Arx = b r , r E R into two subsets:
(i) if there is x E
pI such that x E N, then Pr c N.
r
(ii) if there is x E
pI such that x
r
E V, then pIr C V.
Remark 4.1.4. Two extreme points of a closed polyhedron are called adjacent
a single simplex transformation we can from one of the two points reach the
t}, (415)
(416)
and
(417)
*
Note Xt(r) is the r ili hyperplane defined by t linear equalities and Xt(r)
following:
Theorem 4. 1. 6.
By assumption
follows.
(b) Observe that G(x) = G~t) for each x E C[x~(r)]I. The conclusion
Corollary 4.1.7.
which associates Nexpoints with the Kt subsets of the set 1t' which has
exactly t active constraints.
Let, for example, t = 1, i.e., all constraints are considered
individually. Then KI ,. m+n and N (1) = [n~!)J may be written as follows:
x1 (1)
nIl
(1)
n I2 ... (1)
, n lm (1)
n i (m+I) ... nn~+n)
x2 (1)
n 21 (1)
n 22 ... (1)
n 2m n~~~+l) '" n~~~+n)
xk (1)
nk1 (1)
nk2 ... (1)
nkm (1) ... (1)
nk(m+l) nk(m+n)
(1)
111
(1)
112 .. . (1)
11m
(1)
I1m+1 ... (1)
I1m+n
The ~(l) indicates the number of points xi, i E K which make the
r
rth constraint active. Notice that 0 ~ ~(l) ~ k, r 1, ... ,K I .
r
Observe, if the rth column of N(l) contains only one or less non
plane XI(r) (see (416)) contains no Nex point, i.e., the rth constraint
is not active for any xi E Nex Since any Npoint is a convex combination
Then all other points x E Xl (r) will be dominated and the rth column may
1 2 ... K
t
x1 (t)
nIl
(t)
n l2 ... (t)
n lK
t
x
2 (t)
n 21 (t)
n 22 ... (t)
n 2K
t
x
k (t)
n kl
(t)
n k2 ... (t)
nkK
t
~l
(t)
~2
(t) ... (t)
nK
't
where
(t)
n ir .{ i
1 if x makes the r
in J t active
0 otherwise
th
subset of t constraints
(423)
129
Notice that n~t) indicates the number of points xi, i E K which make
t h e r th ttup I
ef0 . .
constral.nts actl.ve.
Remark 4.2.1.
Ca) Suppose the rth column of N(t) contains no nonzero element
(n~t) = 0). (See (417)). If n (t) =1
r '
*
(and Xt(r)) contains exactly one Nexpoint. *
Then C[Xt(r)] c N by
assumption. In both cases we may delete the rth column from N(t) .
(424)
Let us start with the matrix N(l). All columns must contain at
least two l's, since we have deleted those with n~l) =0 and n~l) = 1.
The number of such columns is Kl , i.e., r = 1,2, ... ,K l . We shall inves
(A) C[X~(1)]IcN
Case (A). In this case the Xl* (1) is a part of the resulting decomposition
X
i
co
*
Xl(l) belong to N, *
·i.e., C[Xl(l)] eN. Now we may delete the first
Xl* (r) c Xl* (1), r = 2,3, ... ,K l . This follows from Corollary 4.1.7. That
is, if the column N(l) of N(l) is such that
r
Case (B). In this case Xl* (1) is not part of the solution since
We repeat this step for all r = 1,2, ... ,K I , i.e., until all
such that C[X *I (r)] I ~ NI , will be the starting point for the next step.
Step 2. From N(l) with KI columns let us construct N(2) with K2 columns
(a) Let
C )2 active
otherwise
N(l) and N(l) be two different columns of N(l) • A
z w
column N(2) of N(2) could be derived as follows:
r
If 1(2) = 1(1) u 1(1) we define
r z w'
(425)
More precisely.
• n lw
(1) )
n (1)
K.w
constraints active. Notice that r is the index for the set {z.w}. Also.
1;2) contains exactly one more element than I~l). For generalization of
with K2 eolumns.
Step 3. In general. NCt ) is transformed into N(t+l) using the following
rule:
where N(t) and N(l) are columns of NCt ) and of N(l) respectively. and
z w
I(t+l) = I(t) ICI).
r z U w
The Kt +l columns of N(t+1) are reordered in such a way that
133
*
and those with Xt+l(k) *
c Xt+l(r), k ~ r, r = 1, •.• ,Kt +1 are deleted.
Resulting is N(t+l) with Kt +l columns.
i.e., for t = 1,2, ••. ,m+n. The routine ends whenever some N(t) becomes
a vacuous matrix. For summary see Figure 4.2.1.
Remark 4.2.5. As a matter of fact, the routine ends always before reach
ing N(n) because all n~n) < 2, r = 1,2, .•. ,Kn .
,
on simple examples.
134
NO 8 ._ _ _ _ _.1..:_,
Delete the column of
N(t) corresponding to
Has all columns of X*t (r) as well as all
YES
those corres onding to
subsets of X*t (l')
LJ..________________________ ~
*
Construct Xt(l') successively
Is
for all remaining columns in
N(t)
5
From all remaining columns
I~ of N(t) form N(t)
Figure 4. 2 . 1.
135
as in (422).
2. Calculate n~t) for all columns of N(t), r = 1, ... ,Kt . Delete all
go to 4.
4. *
Construct all xt(r) and go to 7.
5. *
All columns for which C[Xt(r)] f N are retained. All such columns
to 1.
dominated.
of Corollary 4.1.7. Go to 3.
136
i. e., K1 = 5.
renumbered
r
columns I 2 3 4 ($] Nonnegativity conditions
1(1) O} {2} {3} {4} {5} x2
r xl x3
x1 1 1 1 1 I
2
x 1 1 1 1 I
3
x 1 1 1 1
x4 1 1 1
x5 1 1 1 1
6
x 1 1 I 1
*
Xl(l) {x1 ,x2,x3 ,x4 }
Using the nondominance subroutine we would check now c[x~(r)]I for non
dominance. From Figure 4.3.1. we see that C[X*I (5)] I c N => C[X *I (5)] eN.
Then C[x3 ,x 4 ,x5] c N is a part of the resulting decomposition and the
fifth column may be deleted. No further columns may be deleted. Thus
the reduced matrix N(l) has'been obtained with the number of columns
1<1 4.
The next step is to construct the matrix N(2) = [nI;)] , r = 1, .•• ,K 2 .
The rth column of N(2) is obtained by (see (425)):
Repeating this for all possible pairs z and w we obtain all columns of
N(2). We shall consider only those columns r with n;2) ~ 2 and get
N(2) :
139
renumbered
r
columns
1(2)
r U,2} U,3} U,4} {2,3} {2,4} {3,4}
1
x 1 1 1
2
x 1 1 1
3
x 1
4
x 1
5
x 1
6
x 1 1 1
L 2 2 2 2 2 2
Checking C[X *2 (r)] I for non dominance we can see from Figure 4.3.1. that
only C[X *2 (l)], C[X *2 (3)] and C[X *2 (S)] are part of N, so the first, third
(2) 1 4 2 3
and fifth columns may be deleted in N . Then C[x ,x ], C[x ,x ], and
5 6
C[x ,x ] are part of the reSUlting decomposition.
We have obtained the reduced matrix N(2) with the number of columns
Remark 4.3.1. Notice also that at each step we may check the rows of
N(t) and delete those which have all zero elements, e.g., we may delete
To get the new columns for N(3) we multiply the columns N;2) 0 N~l) .
Notice that only N(2) G N(l) N(2) 0 N(l) and N(2) G N(l) are sufficient
22'23' 43
since all remaining combinations are identical. We obtain:
r
renumbered
columns 1 2 3
1(3) {l,2,3} {l,3,4} {2,3,4}
r
1
x 1 X3* (1) {xl}
2
x 1 X3* (2) {x2}
3
x 1 X3* (3) {x6}
I 1 1 1
2
But x 1 , x , and x 6 are nondominated by definition so the entire
step is unnecessary.
~
{X3'X4'X5}
1 4
X* {x I ,x 2 ,x 3 ,x 4 ,x56}
,x
{x ,x }
{x 2 ,x 3 }
{x 5 ,x 6 }
345 14 23 56
meaning that C[x ,x ,x ], C[x ,x ], C[x ,x ] and C[x ,x ] c N.
141
("
4x l + x 2 + 2x3
VMax c .x = xl + 3x 2  x3
3 x 2 + 4x 3
c .x = xl +
subject to
<
xl + x 2 + x3 3 1.
<
2x 1 + 2x 2 + x3 =4 2.
<
X
xl x2
° >
3.
xl
=
° >
4.
~2
° >
5.
x3
° 6.
1 xl 2 4 5
X (0,0,3,0,1,0) x x3 x x
2 1
X (0,1,2,0,0,1) c .x 6 5 6.5 2 5
3 1 1 2
x (2'2,2,0,0,0) c .X 3 1 0 6 4
4 3
x
X
5
(0,2,0,1,0,2)
(1,1,0,1,0,0)
c .x 12 9 8 2
°
We would like to decompose X* = {x1
,x2 345.
,X ,X ,x } 1nto all subsets whose
renumbered
r
columns Ci5 0 3 4 5 ~
1(1) O} {2} {3} {4} {5} {6}
r
x1 1 1 1 1
2
x 1 1 1
3
x 1 1 1
4
x 1 1 1
5
x 1 1 1
L 3 4 3 3 1
x
2
>
Excluding the fifth column, all 11 (1) 2, ... = 1, .•. ,6.
r
' >
First establish H> and W:
2~ = Al + 4A2  A3
2~ = 3Al + A2 + A3
>
~ = AI + 2h2 + 4A3 AI ,A 2 ,A 3 > 0, ~ = O.
143
"3
l · 0)
O}
:}
3Al Al 3A3 Al
" =
2A l + 3A 2  2A: =0 + 3A 2  8A3 A2 A3 =0
SAl  3A 2 + 7"3 =0  3A 2 + 8A3 = 0 0
deleted as well as the sixth column, since X*l (6) c X*l (2), i.e.,
C[x 4 ,x5 ] c N.
Next construct Gil) = {~(l,l,l) I~ > ol. We obtain the following
system:
Al + 4A2
'3}
~ 2Al+ 2A2
~ 3Al + ),2 + A3 + 2A l + 3A 2
:}
5 5
),1 + A2  '2A3 Al + A2  '2A3 =
5A 2 + 7>.3 OJ
=: +
7
A2  5),3 =
5A 2 + 7>.3 0
"3 5. Therefore
Le. ,
1 Z 3
C[x ,x ,x ] c N.
system:
5"z
4"1 +
' O}
~
o SAl + 3A Z 4A3 0
From the first equation of the last system we see that the solution does
not exist for Al > 0, "Z > O. Therefore, the third column of N(l) may
not be deleted.
Construct G(1) {~(1,0,0)1~ > a}. Then we have:
4
~
o 3Al + AZ + "3
where from the second equation it is seen that the system is not satisfied
for Al > 0, AZ > 0, A3 > O. The fourth column cannot be deleted. We end
up with the reduced matrix N(l) with columns 3 and 4 left. Construct N(Z):
r
renumbered
columns 1
1(2) {3,4}
r
x
1
1 Since ni 2) = 1, we would delete
2
x 0 the first column and end up with
3
x 0 a vacuous matrix. The x 1 is assumed
4
x 0 to be nondominated. The algorithm
5
x 0 ends.
L 1
<
The final partition:
2 3 4 5
{x ,x ,x ,x }
X
* 1 2 345
{x ,x ,x ,x ,x }
123
{x,x,x},
2345 123
meaning that C[x ,x ,x ,x ] and C[x ,x ,x ] c N.
146
and E).
J
= Min
r
Yrj > o}
z~t)
J
than Nex but all points of Nex are included in it. The problem remains
in Figure 5.1.1.
e\<.}
:;
"1=
t=
<=
E
E
lillllilWULllll1
:;2 \0
y <=
Eo
E
illllJUIl
y~
J't
o
Figure 5.1.1.
Observe in Figure 5.1.1. that yl, ./, y3 and y4 are all non
W1·th the se t {I
y, y2 , y 3 , y4 ,y 6} Wh·1Ch conta1ns
. Nex = {yI , y2 , y3 ,
6
Notice that y cannot be discarded by direct comparison of
y6, i.e., those which are noncomparable with others but dominated only
There are some properties of e[xJ for linear case which allow us
to work with e[X] or the image instead of with X directly. For example,
...... , of X which have the property that they are all noncomparable
with each other and also with some nondominated extreme point (and
reaches its unique maximum at one of these points. Points x4 and x 5
are not dominated by any of these points. But they can still be dom
inated or nondominated. 4
Like, for example, x is dominated by a convex
combination of xl and x 3 with ~ = t, i.e.,
graphically since the image e[X] c £2. So we can plot all non
1 2
comparable vectors (e (x), e (x)) for all noncomparable extreme
points {xl, x 2 , ..• , xk} and delete those which are dominated, by
simple inspection.
150
Let (xl' Yl) and (x 2 ' Y2) be two distinct points in E2. Then
c Y2 Y 1
,
x x
2 l
Yl    ~        
I
I
o x
Figure 5.1. 2.
Since we consider only two objectives we may further simplify
1 2
our notation. Let for any x e X, ~ = a (x), v = a (x).
1 2 1
Let x ,x e X be unique (i.e. nondominated) maxima of e (x)
and a 2 (x) respectively.
151
v :gl(X)
o
Figure 5.1.3.
> {
L= = (u,u)lu
_ 2 l) ~ (U U
u u u ul _ U l ) UJ ,
(Uu2 u (513)
2 1 2 1
and
L< its complement.
152
Define,
and
(514)
at Figure 5.1.3.
(U2 ' u 2) ] i.e. closed line segment connecting (Ul ' u l ) and
22,
~ e 1 (x 2) = 25, and u2 = e2 (x 2) = 35.
153
3522 22 _ 3522 26
u  2526 . u 2526 .
8\x)
35
Figure 5.1.3.
are that though the concept of the cutting hyperplane may be gen
The above theory can be used safely only in two dimensions, i.e.,
for two objective functions. In higher dimensions the problems of
uniqueness and degeneracy of the cutting hyperplane make its use
questionable at this point. A couple of graphical examples should
clarify the difficulties.
alex) + 3e 2 (x) 40
Substituting x 3
16.6 37
o ,
<10
I I
I
I I
, I
 '  ~I ,  (
 ..,
I

I
I
10
Figure 5.1.4.
156
Substituting xl
1
16.6 + 3.(3.3) 26.5 ~ x 'Nex '
x
2 8 9 8 Minima: x 4 , x 1 , x 2 x3
3 9 See Figure 5.1.5.
x 8 8
4
x 0 8.8 8.8
get
Substituting x4 we get
'I
1 /
g ;
    .£.   (
/ I I'
' 6(x1) /
/
1 /
I /
9 
'/
• 1;0
Figure 5.1.5.
158
ld
4 1 41 2 4 3
W (x) + 49 a (x) + W (x) = 396.
i, j = 1, ... , t. Let there exist some xeX such that Sex) ~ S(x*).
This implies that also ei(x) ~ ei(x*) and sj(x) ~ ej(x*). We have
a contradiction. Q.E.D.
consider:
'.lel(xr)
A + ,~2e 2( x r) + ... + ,1,
~ e 1,(xr),
... , k; j .;, r
... , k
1, " ' , k
J
E X. 1, ... , t.
i=l ].
1, ... , t
k
E L 1, L > 0, j 1, ... , k •
j=l J J
161
extreme point. Any of the above approaches may be used after other,
heuristic ways of screening have been used and some of the points
still stay undetermined with respect to nondominance.
To demonstrate, let us return to the example in Section 3.3.
on Multicriteria Simplex Method. Suppose that we did not use the
nondominance subroutine and ended with all the noncomparable extreme
points:
x3 @ @ 16
x4 16 0 . @
x5 0 8 @
x6 5.33 21.33 5.33
x7 5.33 21.33 5.33
8
x 12 16 8
9  6.4 9.6 6.4
x
problems.
The methods introduced in this study are applicable only when the
o
Figure 5.2.1.
163
Sex) (al(x), a 2 Cx)) and let S[X] be non convex with a "gap" as in
Figure 5.2.2.
e\x)
t.
t::
c
t
t: e[N1
c
c L 
L
~ t
C != E [

L
C
c E '= E
t:: t: t c
I
E 1= C
I='
E r=c
'\
l f t: i=
0 e1 c2 c~ c~ c' e" etcx)
Figure 5.2.2.
164
Fix alex) = cl and solve Max a 2 (x) subject to alex) ~ c l and a[XJ.
The solution is denoted as yl. Next solve Max e 2 (x) subject to el(x)
>= c 2 , get t.1ng y1 agaln.
. Solving the same problem consecutively for
234 5
c 3 , c 4 , c S and c 6 we get Y , Y ,y and y. Notice that by solving
f or a 11 ce[cl, c6J we may compu t e the woe
h 1 se t N wh·l . ltaneous 1 y
1 e S1mu
ei+l(x), ... , al,(x)) which is derived from e(x) by deleting its ith
component ei(x). e(i) (x) :En~El,l.
where "'c(i) = (1
c, ... ,cil c i+l , •.• , ct)
and c(i)eC = {c(i)lc(i)cE t  l ,  ~< ci < ~}. Using Fritz John stationary
point necessary optimality theorem [see Mangasarian, 1969] we can state
the following necessary conditions for x*eX to be nondominated.
*
(~i' ~ , * r*) ~ 0
*
(~i' ~ *, r*, s*) f o.
Theorem 5.2.3. N ~ [.
Proof. Suppose x*eN and x*~[. Then there exists some Xl€x*
such that for any c, ei(xl) > ei(x*) and e(i) (xl) ~ c. We can,
however, make
1 >
c = c(x*) c.
section 3.4. In this case the final decision might still be hard to
achieve.
Ultimately the decision maker must choose a single nondominated
This 1\ (xi) represents a set of optimal weights for xi. So, we have
Ibj ectives.
!omplete uncertainty as what the actual weights should be. i.e. all
.E:fL are considered equally plausible. then the entire set N is the re
:ult. If the decision maker could limit the choice of weights. the
.f the decision maker can express reliably his preference between say
. k . k .
J. x E: N. then if x J is preferred to x we can remove x J from further
:onsideration and thus reduce N. The resulting set of compromise solutions
han one. More often the decision maker cannot compare, does not want
:0 compare or does not know how to compare any two alternatives with
final
subset
Figure 5.3.1
a .2 .5 .6
171
i  .
Max 8. (x) 8. (x ) = 8., ~ 1, . . . , J,. (531)
x~ X ~
~ ~,
d.
~
(x j ) , 0 ~ d.
~
(x j ) ~ 1, (532)
i=l, . . . , J" j= 1 • . • • , k.
172
=ormal functions:
. 9. (x j )
d. (x J )
~
= ~
 e. ~
(5331
9i
or more complicated
8i (x j )  ~
(534)
8i  ~i
(535)
(539)
J.
[I A~ ~(xj)J lip, 1 ~ p ~ =. (5310)
i=l
J.
=
\'
L Ai
P
wr01 d i (x j ). (5311)
i=l
insight into relation between C and N can be gained from Figure 5.3.2.
' , _ 91 (x)
x
I
I
/
/
N _J.. /___
c
_
~
Figure 5.3.2. 1
x
l~
d i (x j )
x
1
x
2
x
3
x
~
x
5
x
6
~
d i (xJ) x
1
x
2
x
3
x
~
x
5
x
6
L1 (x j ) 1.~17
CD 1.667 1.75 1.557 1.557
~
.th x
cri~erion x
1
. x
k
I
1 d ll
· d lk Dl
.
. · . (5315)
.
R, di,t . · d.ek DR,
177
(5316)
. .. , d ) (5317)
P,
To each d.
1
E d we assign a measure of its contrast intensity
or entropy, denoted by e(d i ).
In (5315) observe that
i 1, (5318)
178
L D.
k
d ••
K ..l:J.. (5319)
j=l J..
(5321)
N l, ...
)". = A.. w. / !: )"i w. i 1, ... , J, • (5322)
J.. J.. J.. i=l J..
179
xj
1 2 3 4
wi S.1 (x j ) x x x X
.8 91 (x j ) 7 8 8.5 0)
.1 92 (x j ) @ 60 20 80 (5323)
.1 9 (x j )
3
4 4 CD 2
1 2 3 4 "
x x x x L
1 .778 .889 .944 1 3.611
2 1 .6 .2 .8 2.6 (5324)
3 .667 .667 1 .334 2.668
We get
t
I
i=l
A. d ..
1. 1.J
(§) .664 .547 .695
Ior ..
i d 1.J @ .628 .486 .647
i=l
4
So the traditional approach would recommend x while our
method indicates xl to be the best solution.
/
/
""
""
/
/
"
/
/
/
/
/
e(X)
" '\.
""
Figure 5.3.3.
"
BIBLIOGRAPHY
13. Manas, J. and Nedoma, J., "Finding All Verti ces 0 f a Convex
Polyhedron," Numerische Mathematik, Vol. 12 (1968), pp. 226229.
20. Von Neumann, J. and Morgenstern, 0., Theory of Games and Economic
Behavior, Princeton University Press, 1953.
23. Yu, P.L., "Cone Convexity, Cone Extreme Points and Nondominated
Solutions in Decision Problems with Mul tiobj ectives," Center for
System Science, 7202, University of Rochester, Rochester, New York,
1972.
24. Yu, P. L. and Zeleny, M., "The Set of All Nondominated Solu
tions in Linear Cases and A Multicriteria Simplex Method.,"
Center for System Science, CSS 7303, University of Rochester,
1973.
25. Yu, P. L. and Zeleny, 1'1., "On Some NultiParametric Programs,"
Center for Systems Science, CSS 7305, University of Rochester,
1973.
25. Zadeh, L. A., "Optimality and Nonscalar  Valued Performance
Criteria," IEEE Transactions on Automatic Control, AC8 (1953)1,
5950.
185
29. Zeleny, M. and Cochrane, J.L., "A Priori and a Posteriori Goals
in Macroeconomic Policy Making," In: Multiple Criteria Decision
Making, Columbia: USC Press, 1973.
30. Zeleny, M., "Compromise Programming," In: Multiple Criteria De
cision Making, Columbia: USC Press, 1973.
31. Zeleny, M., "A Selected Bibliography of Works Related to the
Multiple Criteria Decision Making," In: Multiple Criteria Deci
sion Making, Columbia: USC Press, 1973.
186
APPENDIX
187
Dl. A constraint. say constraint 1, is inessential if, and only if, for
all x such that
2 •.•.• m
D"2. A constraint, say constraint 1. is strongly redundant if. and only if,
for all x such that
A.x.$.b
1  i
.• i 2•...• m
D3. Constraint 1 is strongly redundant if, and only if, there exists no
vector x such that
Ai X ~ bi , 2, ... , m
and Alx ~ bl .
Ai X ~ bi ' 2, ... , m
and Alx = bl .
D~. A constraint, say constraint 1, is weakly redundant if, and only if,
there exists x such that
Ai X ~ bi , Alx = bl
Ai X ~ bi , Alx > bl
Note. The first part of this definition assures no sredundancy; the second
part the inessentiality.
In the traditional LP analysis all redundant constraints are dispensable.
For the purpose of our problem only sredundant constraints are superfluous.
All essential and wredundant constraints are operative in our computations.
190
essenti a1
essenti al essenti al
191
D5. Constraint 1 is essential if, and only if, there is some x for which
Note. If the set {XIXEEn; A.x ~ b., 2, ... , m} ¢, then the constraint
1  1
= b1
s. t.
2, ... , m.
If there are some constraints of the form ~ or = among the i=2, ... , m
constraints, then corresponding artificial variables must be appended.
192
dominant nondominant
sredundant wredundant
dominant
sredundant
essential dominant
)nstraints sredundant
wredundant
194
Example. Consider the following set of constraints and discard all
sredundant constraints.
l. 2xl + 2x 2 ~ 4
2. 3x l + x2 ~ 3
3. 5 3 7
Yl + 2 x2 ~ 2
4. lx
2 1
+ x2 ~ 3
5. 5 3
2 x l + 2x2 ~ 4
1 1 < 1
2 x l + x2
2 
1
1xl + "Ix2 ~ 1
5 3
JX 1 + x <
7 2=
1
1 1
2 x l + "Ix2 ~ 1
ix
8 1
+ 1<2 < 1
8 =
We can see that (i. t) and (~. ~) have both components smaller than (~. t)
so the last two constraints may be deleted.
Consider
2x l + 2x2 ~ 4
3xl + x2 ~ 3
5 3 7
2 x l + 2 x 2 ;;"2·
195
 2 2 4 x2 1 2
Yl 0 0 0 0
"2
3 0 0 3 2 0 _1 0
Y2 Y2 "2
5 3 0 0 7 0 3 0
1
Y3 "2 "2 2 Y3 4" "2
2 2 0 0 0 4 0 0 0 0 0
Yl 2 2 0 0 4 4 _2 0 2
Yl 0
3 3
 3 0 0 3 1 0 1 0
Y2 xl 3 3
5 3 7 2
Y3 "2 "2 0 0
2 Y3 0
3 0
%
3 1 0 0 0 3 0 0 0 0 0
2 2 0 0 4 4 2 0 2
Yl Yl 0
3 3
3 0 0 3 1 0 1 0
Y2 xl 3 3

Y3
5 3 0 0 7 Y3 0 2 0 5
6"
"2 2 2 3
5 3 7 2 0 5 0 1
"2 "2 0 0 0 2 0 3 6"
196
Problem (1 l.
X3 + 3x 4 + 2xS
vMax x3 + 2x 4 + 3x S + x6
subject to:
x.J. ~ 0, i=l , , 7
_________________ EXECUTION
43
******
BASIS VALUE
r; 0.79999998E 01
9 0.00000000E38
10 O.OOOOOOOOE38
11 O.H999998E 01
BASIS VALUE
1 0.16000000E 02
9 O.48000000E 02
10 O.32000000E 02
11 o .160Q0000E 02
0.32000000£ OZ
tlASIS VALUE
1 0.79999998E 01
'1 O.32000000E 02
10 0.16000000E 02
3 0.7'l999998E 01
E!ASIS VALUE
4 0.53333:::;3E 01
9 o .10U:66C7E 02
10 o .533>33333E 01
3 0.10666667E 02
C:.32000000E 02
BASIS VALUf
1 0.00000000E38
9 0.5J33B32E 01
.5 O.5333::333E 01
3 o .53;33330E 01
199
fASIS VALUE
1 0.00000000E38
9 0.00000000E38
5 0.79999998E 01
11 0.79999995E 01
EASI!) VALUE
10, o .16000000E 02
3 0.00000000E38
4 0.16000000E 02
11 0.319<;9999E Of
BASIS VA LlJE
io 0.16000000E 02
1 0.00000000E38
4 0.16000000E 02
11 0.319~9999E 02
Problem ( 2) •
r
+ SX 2 + x3  x 4 + 6xS + BX 6 + 3x 7  2x B
vMax SX l  2x 2 + SX 3 + 6xS + 7x 6 + 2x 7 + 6x B
Xl + x2 + x3 + x4 + Xs + x6 + x7 + xB
subject to:
Xl + 3x 2  4x 3 + x 4 Xs + x6 + x7 + xB t,; 40
SX l + 2x 2 + 4x 3 x 4 + 3x S + 7x 6 + 2x 7 + 7x B ~ B4
4x 2 x3  x4  3x S xB f 1B
12x 1 + BX 2 x) + 4x4 x6 + x7 ~ 40
Xl + x2 + x3 + x4 + Xs + x6 + x7 + x B ?; 12
X. ~ 0 , i 1, ••• , 8
~
 EXECUTION
0200
BASIS VALUE
9 O.64444444E 02
14 O.231111llE 02
11 o .ll266H7E 03
4 o .53333333E 01
13 O.lSe:666f6E 02
5 O.29177718E 02
15 0.38444444E 02
16 o .94Ht665E 02
BASIS VALUE
9 O.58tllllOE 02
14 o .26t l1lllE 02
11 0.l1033333E O!
4 O.76666665f 01
2 O.lH66667E 01
5 O.291771l7E 02
15 0.43111110E 02
16 0.99333330E 02
BASIS VALUE
9 O.575t02E2E 02
14 O.27347517E 02
11 O.1l30t383E 03
4 0.97021274E 01
6 0.1l914894E 01
5 O.28453900E 02
15 O.19645390E 02
16 O.90297869E 02
Definitions.
List of Subroutines.
(1) PRINIT
This prints out the simplex tableau and Z array i f re
quested.
(2) OBJVAL
This computes the value of each objective function.
(3) CHCOMP
This computes each element of the objective functions
array, Z.
(4) REMOVE
This performes the pivot transformation and changes
the basis.
(S) TEST
This tests whether any objective function is at its
maximum and whether ,alternative maxima exist which
would dominate the current basis.
(6) TEST 2
This tests whether a proposed basis has been used pre
viously, and stores it in the list of bases previously
searched, i f so instructed.
(7) TEST 3
This tests whether a proposed basis has been placed in
the auxiliary list previously and, i f not, stores it.
(8) fORBAS
This selects the auxiliary basis "closest" to the curr
ent basis.
(9) DROP
This is used by fORBAS to delete a chosen auxiliary ba
sis from the auxiliary list.
(10)TEST 4
This is used by fORBAS to test whether the chosen auxi
liary basis has been used previously.
204
(11) [HOS
This is used by FORBAS to select the auxiliary basis
"closest" to the current basis.
(12) FEASBL
This tests if the current basis is nondominated and,
if it is, stores it in the file of nondominated bases.
(13) REMOV
This is used by FEASBL to pivot in the subproblem ta
bleau, used in FEASBL
205
C THIS IS THE ~AIN SUBROUTINE FOR A ~uLf!OBJECTIVE LINEAR
C PROGRA~"ING PROBLE~.
C THE NU~BER OF CONSTRAINTS IS LIMITED TO 12, THE NUMBER OF
C OBJECTIVE FUNCTIONS TO 7, AND THE NUMBER OF VARIABLES
C INCLUDING ALL SLACK AND ARTIFICIAL VARIELES TO 40
C NOVI NUMBER OF VARIABLES
C NOCI NUMBER OF CONSTRAINTS
C NOBI NUMBER OF OBJECTIVE FUNCTIONS
C NOBZt COMPOSITE OBJECTIVE FUNCTION NUMBER'NOB & 1
C
C WHEN THE CONSTRAINT IS LESS THAN OR EQUAL, IEQ%J< I 0
C WHEN THE CONSTRAINT IS EQUAL, IEQIJ< I 1
C WHEN THE CONSTRAINT IS GREATER THAN OR EQUAL, IEQ%J<t 2
C
COMMON NOV,NOC.NOB,KAUX,KUSE,NOBZ,NOV1,TAB(12,40),V(12).
1C(8,40, ,Z(8,40) ,X(8) ,10(12) ,IB(40) .IEQ(12) ,T(12,40) ,INN(300,3),
2IAUX(200,3) ,FIIT(12)
COMMON/NDOMP/ NONDP
DIIIENSION VC(8)
DIMENSION 111(40)
REWIND 9
NONDP=O
READ (5,900) NOV,NOC,NO'B
900 FORMAT (212,11)
WRITE (6,990) NOV,NOC,NOB
990 FORMAT (5X,2I2,I1) ,
NOV1=NOV
DO 5 1=1,8
DO 5 J=1, 40
5 C{I,J)=O
DO 6 1=1,12
10(1)=0
X(I)=O.O
DO 6 J=1,40
IB (J) =0
Z (I, J) =0. 0
T(I,J)=O.O
6 TAB(I,J)=O.O
KAUX=O
KUSE=O
NOBZ=NOB+1
READ (5,901) rEQ
901 FORMAT (2011)
WRITE (6,991) IEQ
991 FORfHT (SX, 2011)
READ (5,902) FMT
902 PORMAT(18A4)
WRITE (6,993) PMT
993 FORMAT(SX,18A4)
DO 9 I=1,NOC
REAl2&<A5,FIIT) (TAB (I,J) ,J=1,NOV), V (I)
9 WRITE (6,992) (TAB(1.J),J=1,NOV),V(I)
9n PORIIAT (5X, 8P10. 5)
DO 10 I=1,NOB
READ (5,FMT) (C(I,J),J=1,NOV)
10 WRITE (6,992) (C (I,J) ,J::1 ,NOV)
DO 11 I=1,NOV
206
DO 11 J=1,lfOB
11 C(NOBZ,I)=C(NOB~,I}+C(J,I)
IfOV1=lfOV
NOA=O
K=1
CALL PRINIT
DO 15 J=1,NOC
IF(IEQ(J).EQ.1) GO TO 15
IF(IEQ(J).EQ.O) GO TO 13
NOV1=NOV1+1
TAB(J,NOV1)=1.0
GO TO 15
13 NOV1=NOV1+1
TAB(J.NOV1)=1.0
IBCNOV1)=1
ID (J) =NOv1
K=K+1
15 CONTINUE
NOV2=NOV1
NOV=NOV1
DO 20 I=1,NOC
IF(IEQ(I).EQ.O) GO TO 20
NOV1=NOV1+1
IB(NOV1,=3
'l'AB(I,lfOV1) =1.0
ID(I)=NOV1
K=K+1
20 CONTINUE
IF(NOV1.EQ.NOV2) GO TO 66
NOV3=NOV2+1
NOM=NOV1NOV2
NOC1=NOCNOM
NOC2=NOC+1
DO 60 I=1,NOM
DO 35 J=1,NOV2
DO 35 K=1,NOB
DO 30 11=1,NOC
IF{ID{I1).GT.NOV2) GO TO 25
IDI1=ID(I1)
T(K,J)=T(K,J)+V(I1)*C(I{,IDI1)
GO TO 30
25 Z(I{"J)=Z(K,J)+V(I1)
30 CONTINUE
T (K ,J) =T (I{,J) C (I{ ,J)
35 CONTINUE
IJ1=0
DO 45 J=1,NOV2
IF (IJ1.EQ.1) GO TO 45
IF(IB(J).NE.O)GO TO 45
IK1=0
DO 40 K= 1.lfOB
IF (IK1. NE. 0) GO TO 40
IF (Z(K,J).GT.O.O) GO TO 40
IF (Z(K,J).NE.O.O) GO TO 38
IF (T(K,J).GT.O.O) GO TO 40
38 IK1=1
40 CONTINUE
IF (IK1. NE. 0) GO TO 45
207
JJ=O
AMIN=100000
DO 43 K= 1 , NOC
IF (TAB(K,J).LE.O.O) GO TO 43
VV=V (K) /11B (K,J)
IF (VV.GT.AMIN) GO TO 43
Al'IIN=VV
JJ=K
43 CONTINUE
IF (JJ. EQ. 0) GO TO 45
IDJ=ID(JJ)
IF (IDJ.LE.NOV2) GO TO 45
IJ=J
IJ1=1
45 CONTINUE
IF (IJ1.NE.0) GO TO 46
WRITE (6,904)
904 FORMAT (5X,30H THERE IS NO FEASIBLE SOLUTION)
STOP
46 ICOL=IJ
IROW=JJ
DO 50 J=1,NOC
IF (ID(J) .LE. NOV2) G,O TO 50
IF (TAB{J,I~ .LE.O.O) GO TO 50
IROW=J
50 CONTINUE
CALL REMOVE (IROW,ICOl)
ID(IROW)=ICOL
IBCICOL)=1
DO 51 K= 1,8
DO 51 J=1,40
51 T(K,J)=O.O
60 CONTINUE
C AT THIS POINT ALL ARTIFICIAL VARIABLES HAVE BEEN RF.MOVED
NOl'l=O
DO 65 I=l,NOC
DO 65 J=NOV3,NOV1
65 TAB(I,J)=O.O
66 CONTINUE
NOV1=NOV2
CALL OBJVAL
CALL CHCOMP
CALL PRINIT
CALL TEST'
fi7 11=0
IJ=O
IMIN=O.O
DO 75 I=l,NOV
IF (IB(I) .GE.1) GO TO 75
IJ=O
IF (Z (NOBZ, I) • LE. O. 0) GO TO 75
IF (Z (NOBZ,I) • LT. AMIN) GO TO 75
DO 70 J=l,NOB
IF (IJ. NE.O) GO TO 70
IF (Z(J,I).GE.O.O) GO TO 70
IJ=1
70 CONTINUE
IF (IJ. NE.O) GO TO 75
208
AKIN=Z (NOBZ,I)
11=1
75 CONTINUE
IF (II.EQ.O) GO TO 85
AKIN=.1E10
IJ=1
DO 80 I=1,NOC
IF (TAB(I,II» 80,80,78
78 AK=V(I)/TAB(I,II)
IF (AK.GE.AKIN) GO TO 80
IJ=I
AKIN=AK
80 CONTINUE
IF (IJ) 81,81,82
81 WRITE (6,905)
905 FORMAT (11sX,19H UNBOUNDED SOLUTION)
STOP
82 IROif=IJ
ICOL=II
CALL TEST2(IROW,ICOI,O)
IF (IB(ICOL).EQ.5) GO TO 67
CALL REMOVE (IROW,ICOl)
GO TO 66
85 IF (NOB. EQ.1) STOP
GO TO 186
86 CONTINUE
C AT THIS POINT THERE IS NO VECTOR THAT IMPROVES ALL OBJ. FUNC.
DO 110 I=l,NOV
IF (IB (I) • EQ. 1) GO TO 110
AKIN=.lEl0
11=0
DO 100 J=l,NOC
IF (TAB(J,I}.LE.O.O) GO TO 100
VV=V (3) ITAB (J ,I)
IF (VV.LT.O.O) GO TO 100
IF (VV.GT.AKIN) GO TO 100
AKIN=VV
11=3
100 CONTINUE
IF (II. EQ. 0) IB (I) =2
110IK(I)=II
DO 115 I=l,NOV
IF (IB(I).GT.O) GO TO 115
JJ=O
DO 114 J=l,NOB
IF (Z (J,I» 114,114,113
113 JJ=1
114 CONTINUE
IF (JJ.EQ.O) IB(1)=3
115 CONTI NUE
NOVV=NOV1
DO 150 I=1,NOVV
IF (IB(I) .NE.O) GO TO 150
12=1+1
DO 140 3=I2,NOV
IF (IB(I).NE.O.OR.IB(J).NE.O) GO TO 140
31=IIt (I)
J2=IK (J)
209~
TH1=V(J1)/TlB(J1,I)
IF (TH1.NE.0.0) GO TO 118
IB (I) =3
GO TO 140
11A TH2=V(J2)/TAB(J2,J)
IF (TH2.NE.0.0) GO TO 119
IB (J) =3
GO TO 140
119 DO 120 K=1,NOB
120 VC(K)~Z(K.I)*TH1Z(K,J)*TH2
IJ=O
IK=O
DO 125 K=1,NOB
IF (VC(K).GT.O.O) IJ=1
125 IF (VC(K) .LT.O.O) IK=1
IF (IJ.EQ.0.AND.IK.EQ.1) IB(I)=3
IF (IJ.EQ.1.lND.IK.EQ.0) IB(J)=3
140 CONTINUE
150 CONTINUE
DO 151 I=1,NOV
Il' (IB{I) .NE. 3) GO TO 151
CALL TEST2 (IM(1),I,2)
151 CONTINUE
155 DO 165 I=1,NOV
IF (lB{l) .GT.O) GO TO 165
CALL TEST2 (11'1(1) ,1,1)
165 CONTINUE
J1=0
DO 175 I=1,NOV
IF(J1.EQ.O)GO TO 170
Il' (NONDP. EQ. 0) GO TO 175
IF (IBCI) .GT.O) GO TO 175
CALL TEST3(nI(I) ,I)
GO TO 175
170 IF (IB (I) .GT. 0) GO TO 175
IROW=IM (I)
IC=I
J1=1
175 CONTINUE
DO 177 1=1, !lOV
177 Il' (IBCI).GT.1) IB(I)=O
IF (IROW.EQ.O.OR.IC.EQ.O) GO TO 210
176 CALL TEST2 (IROll,IC,O)
IF (IB(IC).EQ.5) GO TO 190
180 CALL REMOVE(IROW,IC)
185 CAT.L OBJVAL
CALL CHCOMP
CALL TEST1
186 CALL FEASBL (K)
IF (K.EQ.O.lND.NONDP.GT.O) GO TO 190
GO TO 86
190 IF(KAUX.EQ.O) GO TO 220
CALL FORBlS (KKU)
IF (KKU. EQ.1) GO TO 220
GO TO 185
195 BIG=.1E10
J1=0
DO 200 K=1,NOC
210
IF(TAB(K,IC).LE.O.O)GO TO 200
IF(V(K)/TAB(K,IC) .GE.BIG)GO TO 200
BIG=V(K,/TAB(K,IC)
J1=K
200 CONTINUE
IF(J1.NE.0)GO TO 205
WRITE(6,906)
906 FORHAT(SX,25H SOLUTION IS NOT FEASIBLE)
STOP
205 IROW=J1
CALL TEST2(IROW,IC,0)
IF(IB(IC).EQ.~)GO TO 190
GO TO 180
210 WRITE (6,1003) nOll,IC
1003 FORMAT (//SX,5HIROWt,IS,10X,3HICt,I5//)
220 IF (NONOP.EQ.O) GO TO 999
REWIND 9
WRITE (6,1011)
1011 FORMAT (1 H1)
DO 240 I=1,NONOP
READ (9,1005) (ID(.l),J=1,NOC)
1005 FORI'IAT (812)
READ (9,1006) (V(J) ,J=1,NOC)
1006 FORMAT (RE15.8)
READ (9,1006) (X (J) ,J=1,NOBZ)
WRITE (6,1007) .
1007 FOR!'!AT ( ///SX,SHBASIS,10X,5HVALUE)
DO 230 J=1,NOC
230 WRITE (6,1008) IO(J) ,V(J)
1008 FOP!'!AT (SX,lS,5X,E20.8)
WRITE (6,1009)
100<:J FORI'1AT C///5X,33HVALUES OF THE OBJECTIVE FUNCTIONS
241) WRITE (6,1010) (X (J) ,J= 1, NOBZ)
1010 FORMAT (SX,6E20.8)
<:J99 STOP
END
SUBROUTINE OBJVAL
C THIS SUBROUTINE COMPUTES THE VALUE OF EACH OBJ~CTIVE FUNCTION
COMMON NOV,NOC,NOB,KAUX,KUSE,NOBZ,NOV1,TAB(12,40),V{12),
1C(8,40} ,Z(8,40) ,X(8) ,ID(12) ,IB(40} ,IEO(12) ,T(12,40) ,INN(300,3),
2IAUX(200,3) ,FI'IT(12)
DO 20 I=1,NOBZ
SUI'I=O.O
DO 15 J=1,NOC
ID1=ID(J)
15 SU~=SUI'I.C(I,ID1)*V(J)
20 =SU!I
X (1)
RETURN
END
211
SUBROUTINE PRINIT
COMMON NOV,NOC,NOB,KAUX,KUSE,NOB~,NOV1,TAB(12,40),V(12),
lC{8,40) .Z(8,40) ,X(8) ,ID(12) ,IB(40) ,IEQ(12) ,T(12,40) ,INN(300,3),
2IAUI (200,3) , PitT (12)
WRITE (6,900)
900 PORMAT (115X,16H SIMPLEX TABLEAUII)
WRITE (6,901)
901 FORMAT (1X,6H BASIS,21,6H VALOE,7X,10H VARIABLESII)
DO 100 I=1,NOV1,1
IL=I+6
IF (IL. GT. NOV 1) I L=NOV 1
DO 90 J=1,NOC
90 WRITE (6,902) ID{J},V(J),(TAB(J,K),K=I,IL)
902 FORMAT (1X,I6,8E15.8)
WRITE (6,903)
903 FORUT (1X)
100 CONTINUE
WRITE (6,904)
904 PORMAT (1115X,20H OBJECTIVE FUNCTIONS II)
WRITE (6,905)
905 FORMAT (1I,7HOBJ. F.,2X,6H VALUE,1X,10H VARIABLES II)
DO 200 I=1,NOV1,7
IL=I+6
IF (IL.GT.NOV1) IL=NOVl
DO 190 J=l,NOB
190 WRITE (6,902) J,X(J), (Z(J,K) ,K=I,IL)
WRITE (6,903)
200 CONTINUE
WRITE (6,906)
906 FORMAT (/5X,19H COMPOSITE FU~CTION)
DO 210 I=l,NOV1,7
IL=I+6
IF (IL.GT.NOV1) IL=NOV1
210 WRITE (6,902) NOBZ,X(NOBZ), (Z(NOBZ,K) ,K=I,IL)
WRITE (6,901) KAUX
901 FORMAT (5X,5HKAUXI,I8)
RETURN
END
SUBROUTINE CHCOMP
C THIS SUBROUTINE COMPUTES TRE VALOE OF EACH ELEMENT OF THE
C OBJECTIVE FUNCTION ARRAY
COKMON NOV,NOC,NOB,KAUX,KUSE,NOB~,NOV1,TAB(12,40),V(12),
1C(8,40) ,7.(8,40) ,X(8) ,ID(12) ,IB(40) ,tEQ(12) ,T(12,40) ,INN(300,3),
2IAUX(200,3) ,FMT(12)
DO 20 I=l,NOBZ
DO 20 J=l,NOVl
5UK=0.0
DO 15 K=l,NOC
ID1=ID (K)
15 SUM=SUM+TAB(K,J)*C{I,ID1)
20 Z(I,J)=SUMC{I,J)
RETURN
END
212
SUBROUTINE REMOVE(IROW,ICCL)
THIS SUBROUTINE TRANSFORMS THE TABLEAUS AND ENTERS THE ICOL
ELEMENT IN PLACE OF THE IROW EtE~ENT IN THE BASIS
COMMON NOV,NOC,NOB,KAUX,KUSE,NOBZ,NOV1,TAB(12,40),V(12),
1 C (8 ,40) , Z (8 ,40) , X (a" 10 (12) , I B (40) , TEQ (12) , T (12,40) , IN N (300, 3) ,
2IAUX(200,3) ,Fl1T (12)
01 = TAB (IROW, ICOL)
DO 10 I=1,NOV1
10 TAB(IROW,I) =TAB{IROW,I) 101
V(IROW)=V(IROW)/DI
DO 20 I=1,NOC
IF (I.EQ.IROW) GO TO 20
V(I)=V(I)V(IROW)*TAB(I,ICOL)
IF (V(II.LT.0.000001) V(I)=O.o
20 CONTINUE
DO 30 I=1,NOC
IF(I.EQ.IROW)GO TO 30
DO 25 J=1,NOV1
IF(J.EQ.ICOL)GO TO 25
TAB (I,J) =T.1B (I,J) TAB (I,ICCL) *TAB (IROW ,J)
25 CONTINUE
30 CONTINUE
DO 40 I=1,NOC
IF (I.EQ.TROW) GO TO 40
TAB (I,ICOL)=O.O
40 CONTINUE
DO 45 1=1,8
DO 45 J=1,40
45 Z(I,J)=O.O
RETURN
END
SUBROUTINE TEST1 .
THIS SUBROUTINE TESTS TO SEE IF ANY 3BJ. FUNCTION IS AT ITS
KAXI~UM AND IF IT IS , TESTS TO SEE IF ANY ALTERNATIVE KAXIMA
IMPROVES ANOTHER FUNCTION
COMMON NOV,NOC,NOB,KAUX,KUSE,NOBZ,NOV1,TAB(12,40),V(12),
1C(8,40) ,Z(8,40I,X(8) ,ID(12) ,IB(40) ,IEQ(12) ,T(12,40) ,INN(300,3),
2IAUX(200,3) ,nT(12)
DO 100 I=1,NOBZ
I'lAX=O
DO 10 J=1,NOV1
IF (IB(J}.EQ.1) GO TO 10
IF (lUX. LT. 0) GO TO 10
IF (Z(I,J}) 10,7,8
8 MAX=1
GO TO 10
7 MAX=1
10 CONTINUE
IF (lUX) 100,11,11
11 WRITE (6,906) I
906 FORKAT (1ISX,4H THE,I4,3X,3SH OBJECTIVE FUNC~ION IS AT A MINIMUM)
I F (II AX • EQ. 1) GO TO 12
WRITE (6, 907)
907 FORMAT ( SX,32H THERE IS NO ALTERN'TIVE MINIMUM II)
GO TO 50
12 WRITE (6,908)
908 FORMAT ( SX,29H THERE ARE ALTERNATIVE MINIMAll)
DO 30 J=1,NOV1
IF (IB(J).EQ.1) GO TO 30
IF (Z(I,J)) 30,13,30
13 DO 25 K=1,NOBZ
IF (Z(K,J)) 25,25,14
14 MAX=2
909 FORMAT (11SX,14H OBJ. FONCTION,I4,2X,12H IS IMPROVED II)
25 CONTINUE
30 CONTINUE
IF (I'IAX.EQ.2) GO TO 100
50 CONTINUE
910 FORMAT (1ISX,33H BUT NO OBJ. FUNCTION IS IKPROVED II)
100 CONTINUE
RETURN
END
214
ATAB(I,NV)=1.0
IC(I)=NV
IG(NV)=1
30 CI(NV)=1.0
31 II=O
AMAX=O
DO 35 I=1,NV
IF (ZI(I).GE.AMAX) GO TO 35
II=I
AMAX=ZI (I)
35 CONTINUE
IF (II.NE.O) GO TO 36
C THE BASIS IS NCNDOMINATED IP lItO
GO TO 70
36 CONTINUE
III=O
A!'IIN=.1E10
DO 50 J=1,NOB
IF (ATAB(J,II) .LE.O.O.OR.ATAB(J.II).GE.AMIN) GO TO 50
III=J .'
AMIN=ATAB (J, II)
~O CONTINUE .
C THE BASIS IS DOMINATED IF 11110
IF (III.EQ.O) GO TO 51
CALL REMOV1 (III, II)
GO TO 31
51 CONTINUE
DO 60 1=1, NOC
IF (TAB(I,II).LE.O.O) GO TO 60
III=1
60 CONTINUE
IF (III) 61,62.62
61 WRITE (6.901)
301 FORMAT ( SX,2QH SUBPROBLEM IS UNBOUNDEDII)
GO TO 99
62 WRITE (6,902)
902 FORHAT (115X,30H ORIGINAL PROBLEM IS UNBOUNDED II)
GO TO 99
70 NONDP=NONDP+1
CALL PRINIT
K=1
WRITE (9,911) (ID(J) ,J=1,NOC)
911 FORMAT (812)
WRITE (9,912) (V (.1) ,J=1, MOC)
912 FORHAT (8E1S.8)
WRITE (9,912) (X(J),J=1,NOBZ)
99 RETURN
END
Vol. 59: J. A Hanson, Growth in Open Economics. IV, 127 pages. Vol. 86: Symposium on the Theory of Scheduling and Its Ap
4°.1971. OM 16, plications. Edited by S. E. Elmaghraby. VIII, 437 pages. 4°. 1973.
Vol. 60: H. Hauptmann, Schiltz· und Kontrolltheorie in stetigen OM 32,
dynamischen Wirtschaftsmodellen. V, 104 Seiten. 4°. 1971. Vol. 87: G. F. Newell, Approximate Stochastic Behavior of nServer
OM 16, Service Systems with Large n. VIII, 118 pages. 4°. 1973. OM 16,
Vol. 61 : K. H. F. Meyer, Wartesysteme mit variabler Bearbeitungs· Vol. 88: H. Steckhan, GOterstrllme in Netzen. VII, 134 Seiten. 4°.
rate. VII, 314 Seiten. 4°.1971. OM 24, 1973. OM 16,
Vol. 62: W. Krelle u. G. Gabisch unter Mitarbeit von J. Burger Vol. 89: J. P. Wallace and A. Sherret, Estimation of Product
meister, Wachstumstheorie. VII, 223 Seiten. 4°. 1972. OM 20, Attributes and Their Importances. V, 94 pages. 4°. 1973. OM 16,
Vol. 63: J. Kohlas, Monte Carlo Simulation im Operations Re Vol. 90: J.F. Richard, Posterior and Predictive Densities for
search. VI,162 Seiten. 4°.1972. OM 16, Simultaneous Equation Models. VI, 226 pages. 4°. 1973. OM 20,
Vol. 64: P. Gessner u. K. Spremann, Optimierung in Funktionen Vol. 91: Th. Marschak and R. Selten, General Equilibrium with
riiumen.IV, 120 Seiten. 4°. 1972. OM 16, PriceMaking Firms. XI, 246 pages. 4°.1974. OM 22,
Vol. 65: W. Everling, Exercises in Computer Systams Analysis. Vol. 92: E. Dierker, Topological Methods in Walrasian Economics.
VIII, 184 pages. 4°. 1972. OM 18, IV, 130 pages. 4°. 1974. OM 16,
Vol. 66: F. Bauer, P. Garabedian and D. Korn, Supercritical Wing Vol. 93: 4th IFAC/IFIP International Conference on Digital Com
Sections. V, 211 pages. 4°.1972. OM 20, puter Applications to Process Control, ZUrich/Switzerland, March
Vol. 67: I. V. Girsanov, Lectures on Mathematical Theory of 1922, 1974. Edited by M. Mansour and W. Schaufelberger.
Extremum Problems. V, 136 pages. 4°. 1972. OM 16, XVIII, 544 pages. 4°. 1974. OM 36,
Vol. 68: J. Loeckx, Computability and Decidability. An Introduction Vol. 94: 4th IFAC/IFlP International Conference on Digital Com
for Students of Computer Science. VI, 76 pages. 4°. 1972. OM 16, puterApplications to Process Control, ZUrich/Switzerland, March
1922, 1974. Edited by M. Mansour and W. Schaufelberger.
Vol. 69: S. Ashour, Sequencing Theory. V, 133 pages. 4°. 1972. XVIII, 546 pages. 4°. 1974. OM 36,
OM 16,
Vol. 95: M. Zeleny, linear Multiobjective Programming. XII, 220
Vol. 70: J. P. Brown, The Economic Effects of Floods. Investiga pages. 4°. 1974. OM 20,
tions of a Stochastic Model of Rational Investment Behavior in the
Face of Floods. V, 87 pages. 4°. 1972. OM 16,
Vol. 71: R. Henn und O. Opitz, Konsum und Produktionsthllorie II.
V, 134 Seiten. 4°.1972. OM 16,
Vol. 72: T. P. Bagchi and J. G. C. Templeton, Numerical Methods in
Markov Chains and Bulk Queues. XI, 89 pages. 4°.1972. OM 16,
Vol. 73: H. Kiendl, Suboptimale Regier mit abschnittweise linearer
Struktur. VI, 146 Seiten. 4°.1972. OM 16,
Vol. 74: F. Pokropp, Aggregation von Produktionsfunktionen. VI,
107 Seiten. 4°. 1972. OM 16,
Vol. 75: GIGesellschaft fUr Informatik _. V. Bericht Nr. 3. 1. Fach
tagung Uber Programmiersprachen· MUnchen, 911. MlIrz 1971.
Herausgegeben im Auftag der Gesellschaft fUr Informatik von H.
Langmaack und M. Paul. VII, 280 Seiten. 4°. 1972. OM 24,
Vol. 76: G. Fandel, Optimale Entscheidung bei mehrfacher Ziel
setzung. 121 Seilen. 4°. 1972. OM 16,
Vol. 77: A Auslender, Problemes de Minimax via l'Analyse Con
vexe et les Inegalites Variationnelles: Theorie et Algorithmes. VII,
132 pages. 4°.1972. OM 16,
Vol. 78: GIGesellschaft fOr Informatik e. V. 2. Jahrestagung, Karls
ruhe, 2.4. Oktober 1972. Herausgegeben im Auftrag der Gesell
schaft fUr Informatik von P. Deussen. XI, 576 Seiten. 4°. 1973.
OM 36,
Vol. 79: A Berman, Cones, Matrices and Mathematical Program
ming. V, 96 pages. 4°.1973. OM 16,