17.

Nonlinear Programming
17.1 Introduction 450
72Lagrangean Method 450
7 3 Kuhn-Tucker Conditions 456
174 Quadratic Programming 458
17.5 Separable Programming 463
17.6 Chance-constrained Programming or Stc
Questions 470
 Nonin Programming
can method

,r1ms tthe
o m s
h e Lagrangean funct

stcfP
1his

X-
17) irst
partial derivative of L with
partial,

rewith
Lol8{X,. X
varies fromto X,\ 1s Xx
respect
the
respect to .
Here
and also with
where ii
as shown below

SL
obtained. where
0 m.
Then
x 0 varvesof
equate tach
j

Nonlinear Programming j 1.2. n from\s

them s
dL
0,
i= \.2. m
nsists
consists
ot n + m
the
system

unknown
rst order partial derivatives.
17.1 INTRODUCTION So.
of the tirst variables wth
n
Nonlinear programming is an extension ot linear step. the solution to the above +m
programming. In many real-life problems. the
objective function may be nonlinear but the set of constraints may be linear or system of simultaneons sgquatons
nonlinear. Such
problems are called as nonlinear problems. A set ot nonlinear programming problems is as
follows: y T h e n
ordered Hessian sq
matrix H") ot cquations is
foun
ALnd o
Nonlinear programming problem of general nature Table 17.1. size n+
m is
Quadratic programming problems hle 17.1 General Form
formed as
Separable programming problems
of
Bordered Hessian show
Each of these problems requires different solution m 2 Mat
procedure. In this chapter, the procedures for the
above topics are demonstrated with examples:
0
Coefticients of
17.2 LAGRANGEAN METHOD SL
Let us consider the following form of nonl1near
programming.
Si= 1. 2.
0 0 0 wTitten row-wise
Maximize or minimize Z =f(X1. X2, .. X,.X,)
subject to
GX1, Xz X,) Coefticients of
X =
b,. i= 1, 2.
.. m
X,20. j= 1, 2, 3, . SL i , 2. m
A modified form of the above model is shown below: For
writen column-wise
Maximize or minimize Z=f(X1. X2,. X, .. X,) i 1. 2 and
subject to j -1.2.n
8{X1. X2. . X X) = 0. i = 1. 2, m
idelines for testing the stationary point (Xi,
where,
X,20. j= 1, 2, 3, ... n
sinimization type obJective runction
X, x,. are
x f
presented below umizadon
&X X X . X,) = G(X, X X X,) - b hiective
Marimization ype function. The stationary point will gve the
This model consists of n variables and m constraints. The objective function is nonlinear and valueifthe sign each
of each of the last (n m) prineipal minor
of - mavimum objeive function
he constraints are in linear form. of (-1"*, ending with the (2m + Ith
determinants of the bordered
Hessian matrin
Let. L be the
is same
principal minor determinant
Lagrangean function and ó be the Lagrangean multiplier of the ith constraint.
en ene objective function. The
stationary pomt will give the minimum
objetive
f the sign of each of the last (n m) principal minor
determinants of the bordered Hessianfunctiom
-

L
=fX1, X. XX,) -

2 ols.(X1, X. X, X,)] eas that of (-)", ending with the (2m + Ith
principal minor determinant.
mai

450
 Nonlinear Programmir
the above
System
system Ot
equations is
rning
eSOnon (XX d) =(94, 13
452 Te and T.
Operations Research secondorder
partial
derivatives
summariz.ed n Table =2 9
a n example. the last threc principal minor determinants of the bordered Hessian matrix wh:. .
Table 17.2 12
s Shown in the Fig. 17. 1(a) are presented from Fig. 17.1(b) to Fig. 17.1(d). Sccond Order
Partial
35
-0.04 Dervatives
5X, OX,
35 5
öX, ôx,, 9
2 SL
(a) Bordered Hessian matrix (b) First principal
minor determinant
OX, öx,-904
matrix ot the given
35
h o r d e r e dH e s s i a n problem
2 is
shown in
: bora
Table 17.3 Bordered Table 17.3.
Hessian Marix of
(c) Sccond principal (d) Third principal 2
Example 17
minor determinant minor determinant
0.04
Fig. 17.1 Sample of bordered Hessian matrix.
2 -0.04
Erample 17.1 Solve the following nonlinear programming using Lagrangean method:
Maximize Z 4X1 0.02Xf
= -

+ X2 -0.02X3 is 1 and the corresponding last one

isI
and

subject to of n -
m
17.3. The value
of the determinant in Cipal minor
113
detsignermmnant
value
X+2X = 120 is
ds to0.2.
Theb solution (X.
the solution(Xi. X2, ¢)«
the The Sign
Hence, of is same as
oi thame
X1 and X2 2 0 13011 results are nds the this stoan
of
T h e r e t o r e , t h e

he optimal
o p t i m
presented below. maximum obyecivevalue is same as
Solution The general form of the given nonlinear
programming problem is presented X,X)= (94, 13) and function value.
below
Maximize Z =
4X, -

0.02X +X2 -0.02X Z(maximum) =2089

subject to Solve the llowing nonlinear
17.2 programming using
Minimize Z= 2X; 3X} + Lagrangean methoxd:
X + 2X2 120 = 0 Exap
le

X, X2 2 0 18X,
Here, the number of variables, 2 and, the number of constraints, subject t o
n
2X+X=8
=
m =
1. Then
L =
4X -

0.02X 2X2 + X2 -

0.02X2 -X + -

120) X and X 20
Ihe first order partial derivatives with respect to X1, Xz and o are: The eral form of the given nonlinear
Solution programming problem is
oL 4-0.04X - = 0, or 0.04X, + d =4
MinimizeZ= 2X -3X 18X +

(17.1) ubject to

2X+X-8 =0
SL
= 1 -

0.04X, -

20 =
0, or 0.04X2 + 20 =1 X1 and X2 20
(17.2)
number of variables, n 2 and the number
=
of constraints, m =1. Then
SL
- ( X , + 2X2 - 120) = 0, or X, + 2X, = 120
(17.3) L 2X-3X+ 18X- 0(2X +X2-8)
 2X$
2X X, 2X, - 30 -0
A54 Operations Research
X and
X, 20
n =
S and the
HerenC numbe of variables,
n e i r s t order partial derivatives with respect to X- X2 and o are
number
L X+2X+ X$ ot
constraints, m-
X, =4X
-

24 =0 .uith respect to
-0(2x, +

Xi X X3 andX,
(17.4)
partial
derivatives
d e r i

+2x, -30) Then

SL 6X2 18 = 0. or 6X2 +
o = 18 oL o are
SX =-6X2
+ -

tirst

(17.5
1he
OX 2X 20 -0
SL
=-(2X, +
X2 8) =O, or 2X, +
X2 = 8
(17.6)
The solution of the above system of equations
is: öX, 4X- =0
(X. X2. ) (3, 2, 6) and z 42
8L -2X
= =

The second order partial derivatives of the given problem are summarized in Table 17.4 OX, 20 =0
Table 17.4 Second Order Partial Derivatives
-(2X, +
X2 *
2X, -

30) =
0 or
4 SL dL 2X, +X2+
=

8X, X,
of
the above s
the
system
above
ysten of equations 2X 30
s oJ
he luu tti o n
oL
SX Ox, x,
=
-6
(X,X',Xi.) =\17 30 120

derivatis
11 and T
=105 83
The bordered Hessian matrix of the given problem is shown in Table 17.5.
cond
order
partial
summar1zed in Table 17.6.
Table 17.5 Bordered Hessian Matrix of Example 17.2 Table 17.6 Second Order Partial
Derivatives
o 2 1 2
8L SL
8X, 8X, 8x, 8x,0
I0-6| X, 5 X 0
The value of n -

m is 1 and the corresponding last one principal minor determinant is

the same as
bordered Hessian matrix as shown in the Table 17.5. The value of determinant this
principal minor SL L
s-28 which is negative. The sign of this value is same as that of (-1)'. Hence, the solution 8X, 8X, X, 3X,0 oL L
, X2, ) corresponds to the minimunm objective function value. The optimal results are: 8x, 8x -2

(X',X2) =(3, 2) and Z(minimum) =42 eeian

Hessian matrix
matrix of th
of the given problem is
The
bordered shown in Table 17a,
ample 17.3 Solve the following nonlinear programming problem using Lagrangean method:
Table 17.7 Bordered Hessian Matrix of
Maximize Z Xf Example 17.3
ject to
= +
2X + X}
o22
2X + X2 +2X3 30 =
|2 2 0 0
X and X 20 20 02
Solution The general form of the
given nonlinear programming problem is presented below:
 Nonlinear Programming
X Xj A,)+S} =0
X,2 0. =\.2.3 n
456 the ith ot
=\,2, m
Operations Research ntary slack constr
be the Lagrangean function
Nion,
ain.
and This mOdel
he value of n - m is 2 and the coresponding last 2 principal minor determinants of
oordered Hessian matrix are shown in Tables 17.8 and 17.9, respectively. he
i sa
straints. Let

be he cons1Sts of
Table 17.8 Third Principal Minor Determinant vslrunt TheL(X1, X2 .
X). (S,. S,. Lagrangean mniltiplhot
n
m

er
fX1. X2 X,)-
Value = 14

ation proble with concave

above maximization
convex type straints), the object iunction
traints (convex

of o and alue
h twh
establishedshould
Table 17.9 Fourth (last) Principal Minor Determinant
theabe

of . Kuhn-lucker has estahl.should h all

al
p econstra r e l a t i o n

be \ess
nis
he Yeater han than or
o 2 1 2
ioNOWIng
o m a n n a n t h i

1, 2,
3, . m
or ea
2 2 0o Value = 4 0. i
=
eaual 0
neessary onditons 15

( a )

2, 3,n \irst partial derivatives

0. j
minor
SX,
Xn) =
0, i=
1, 2, 3
The
sign the value
of
that of(-1)1.
of
cach ofthe(A,Az,
Hence, the solution
last 2principal determinants is positive which is sa (X.X2
m

as

optimal results
A3 P) corresponds to the
maximum objective functi (
2
X,)S0,
is1, 2, 3, .
m
value. Therefore, the are (d)&A1 X2.
with convex object
120 30 120
ization probler

m ( o n c a v e type constraints), alue of

function and with all
and
2(maximum) = 105.88 o should be greater than
17 17 17 For
a
Lonsiran ease of maximization
tbye
problem and c
nvex in the greater than or
or
eaqual to
caual 0. {L's
munimization toproblem.
are presented below: case ot
Concave DOssibilities
of Pi
the
bjective function and with tliary
type constraints,
For maximization

17.3 KUHN-TUCKER cONDITIONS

( )
nization objective fun ion and
F o r m a x i m i z
0,
with 2 type constraints, 0,i= 1,2.3,
Consider the following general form of nonlinear programming problem which is having a maximization
objective function with all less than or equal to type constraints. For
objective function and
maximization with= type
constraints. 0S0.i=1,2.3..
o, is
Maximize
1)
for i=
1, 2, 3, .. m.
unrestricted in sjan
Z =f(X1, X2, ..,
Xj, X,) tion objective funciuon and with 's
subject to Forminimization
type constraints, o, s0,
i 1.2 1
G,(1, Az,-, , ) s bj, i=1, 2, .., m minimization objectiv function and with '2 ype m.
=

constraints, , 20, i
(v) hiective function and with'= 1,1,3, m. =

X20, j= 1, 2, 3, .., n minimization

type constraints, ois

m. unrestricted in sien
(vi)
, 2, 3, ..,

A modified form of the above model is for i=

Example1 7 . 4 S o lt
vhe e following nonlinear programming problem using
Maximize Z=fX1, X2, .,X Kuhn-Tucker conditions
subject to Maximize Z 3X; =
+
14X,X-8X
g(1, 2 . , X , ) S 0, i =1, 2, .., m
subject to
3X + 6X2S12
X,2 0, j=1, 2, 3,.., n
where X and X 20
& ( X ,X, Xp s Xa) = G,(X1, X, Xp X,) - b;
is modified as:
The given problem
Again, the above model is modified as: Solution
Maximize Z = 3Xj +14X,X, -8K
Maximize Z=f(X1, X2, .., X X)
 d

458
Operations Hesearch7
d
subject to
72 s 0
3A, +6.X2
-

and A2 > 0
A 22
function:
Then. we have the Lagrangean
¢(3X, +6X2 72)
8X -
-

14X1X2
=3X7
-

+
L 8
conditions are as given below.
of Kuhn-Tucker
sets
The four (17.11)
(a) 0 m2 n
173
+ 14X2
-

3¢ = 0, or 6X, +
14X2 =
3¢ (17.12)
(b)
-

6X, the coefficient

represents the term X;
dj presents the coeffficient of in the 1ance
the
16X2 60 =0, or 14X1 16X2 =60 (17.13)
I nm a t r i x[ D )

to j)
Í)
:

is
term
further modified by bala X,X, in objective function
he
14.Y, and
-

objective
-

model

function. d,lhen
cqual
The above ine
constraints
1 Sn o t c

o(3X, +
62 -

72) =
0 (17.14) Maximize Z-CX+ XDx as
shown
(c)
72 s 0 (17.15) 'below,
(d) 3X +6X2
-

Subject to
AX +S=B, i= 1,2, 3,
From Eqs. (17.12) and (17.13), we get (Type I constraints)
-2X + 44X2 =
0 (17.1
J 1, 2, 3, . , n;
-X,0, j=1,2,3, ,
equated to 0, X and X2 should be equal to 0, which is not true.
Therefore, X z0, multiplier of the (Type Il constraints
From Eq. (17.14), if o is
he the Lagrangean ith constraint
Letd multiplier with the h in Type I set of
etraint oi offconstraint
3X +6X 72
-0, or
3X +6X-72 (17.17) associated
Tyne nI. in TType Il set of constraints, u,
-

Lagrangean
By solving Eqs. (1 7.16) and (17.17), we
get
the c o m p l e m e n t a r y s l a c k
followino
Type Application of constraints, andbeS}the
results
into the
into the following system of constraints. Kuhn-Tucker conditions
Z (maximum) = 1,752
X= 22, X2= 1 theodel o

17.4 QUADRATIC PROGRAMMING

in which the highest order of
Ct
Let
the
us consider the
polynomial of the
following form of nonlinear programming problem
objective function is restricted to at most 2.
MS
Maximize Z =ciX + c2X + + CyX
HX=0, j=1,2,3,
+C+X + Cn+2*2 + . +C2,X

+C2n+1A2 + C2n+213 + + C2n+ n e n - 1 n

S 0, i=1,2,3, *m
identity matrix, and
an
subject to where,
I is
b, i= 1, 2,.,m
a; +az2 +. +
aiy S
S
X, Xz, X , 20 P2
ne above quadratic programming model may be represented in short, follows: Q M-
Maximize Z= CX +XDx
subject to
AXSB, i= 1, 2, 3,., m
X20
set of slack variables.
where S is
a
 f c o n s t r a i n t s, , r
r e
e s
sp e c

a the NIh
Lagrangean
AppiicatioConst
nrmltiiKihn
pamler *VLIaen
460 o fthe systcn
constraints,

following
Tespectively
ystem oi constraints.
Operations Research into
the

of andd
Wolfe has
developed a method to solve the above system o1 constamis. 1ne
resuS
2 ie smettonn 1

method

F'rocedure
are listed

to
as

solve
foilows:
steps ot Wolfe'
e l

4
4
6
2-1 01
30 -
whseeatets
quadratic programming problem
Step The system of constraints is to be formed first.
2 3 0 o0
Srep 2 Next. add R, as the artificial variable
tor the Jh consraint or u e tirst n
cach of these first n constraints has no basic variable in it.
constraints, since

Srep 3 A minimization type objective function is then formed by summing the artificial variahi..
Srep 4
Solve the model consisting of the above objective function and system of constraints e ing
the two-phase simplex method.
HXHX=0
S S= 0
Example 175 Solve the following quadratic programming problem using Wolfe's method. of
aations is written as
follows
hea0ove SYster s t e m

subject to
Maximize Z 6X, + 3X
=

2X}- 3X; -4X,X

-

4X+4X2 +o1 2 H= 6
4X +6X2 * +
X, + X2 S 1
30-
2,+3X2 s4 X+X + S, 3

X and X, 2 0 2.X+3X2+ S
Solution The given problem is written as follows:
Maximize 2 cX + c2X + c,Xf + caX} + csXX,
=

6X, +
3X2 -2X -3X4X1X I of the hase
subject to modelfor
r
Phase
simplex
of the type of the method is
(17.18) lunction
function
iTre
irTespective

dbjective presened below

with
Minimize Z -R,+R function in îhe onmgjnalminimization The
a
2X, +3X2 s4 (17.19)
objective

- X 0
of proble-
0dlem.
(17.20) Subyect t o
4X, +4X2 + 0i* 20- 4+ R =6
(17.21)
All the matrices that are required to generate the system of constraints except Matrix D can be 4X,+6X* * 3- *1+
R =3
obtained from the given problem easily. The guideline for obtaining the Matrix D are as
follows X+X2+ S
2X,+3X+ S2 (724
(1725
(17 26

1S 0 (172
The model of the (17 23)
given problem is:
S=0
(17 29)
X1, Xz, R1 and R,20
toximize2-[6 3 variables R1, R2 are ncluded in the constraints (1722) and (17.23), tespectiv
subject to Note: The a
hle in each of them. Table 17.10 represents the initial table of Phase 1of the t
basi
have a
to
phase s i m p l e x method.

where X, s0, -X s0
 in
Table entering varizable
e n t e r i n g

le. The
per
corteipmding e
U h e

462 u2
15
variabie
and leaviing
Operations Research entering
.

**70on
or
Table
Table 17.10 Initial Table 17.13 eration 3
0
B Basic P2 1 2 R2
S2Solution B a s i c

-\
6 -I

0
1
10

9
-8 -10

In Table 17.10, the non-basic variable X^ has the maximum

negative criterion value. 13. the non-basic variable h a s the
is not in the basis,
X2 is the entering variable. [In this iteration, both u2 and X2 should Since
in the basis as
per the constraint (T7.2)]. The not be nree 2 the basis. So, the
ThoOnsiderdo,
corresponding leaving variable is Rz. As
nsidered as the non-basic
Sent
c a n n o t

combination of entering variable and leaving variable, pivot operations are per thinis
Table
basis,
o is valu varahle TTn
entering variabl.hich h w
n

shown in Table
r o w

17.11 the
ing leaving vara
Table 17.11 Iteration 1
11. ; in
criteriOn
variable. The corTesponding
variable.

leaving vanable
variable, Pivot operations are
v a r i -

s
rive e n t e r i n g leaving
R. AS
aitering
t h e variab,
and

shown in t is
G variable

Table 11.14 Table 11A4. omibman i

CB Basic X X Iteration 4
Ri R2
4/3 S S | Solution
R 1/3 2/3 -2/3
2 2/3 1/6 1/2 -1/6 1/6
S 1/3 -1/6 -1/2 0 1/2 X
0
1/6 0 -1/6 Basic 2
R
S2 -1/2 -3/2 1/2 -1/2
1/2 -I

4/3 S/2
1/3 -1 2/3 -2/3 0 X
C-2,4/3 0 -1/3 2/3 S/3

In Table 17.11, the non-basic variable

X, has the maximum negative criterion value. Since -2 0
not in the basis, X, is the entering variable. a, is
and
[In this iteration, as per the constraint (17.26), both
X should not be present in the basis. The ju,
combination of entering variable and leaving corresponding leaving variable is X2. As per
this i n the
Criterion r o w of Tah
11.14 are all
variable, pivot operations are shown in Table 17.12. ace
Since
the values
of the problem is: X =
1, =0 nonnegative, the optmality
and the is

Table 17.12 Iteration 2

Theoptimals o l u t i o n
function
is 4. corresponding maximum value -

objective
0 the
0
C Basic 17.5 SEPARABL PROGRAMMIN
2 R S2 Solution
R 0 2 -I -1 the following nonlinear programming probler
3
0
X 3/2 1/4 3/4 0 -1/4 1/4 3/4
Let us consider

0 -1/2 -1/4 3/4 Maximize Z=fX%, X X, X X

2 0
0 74 0 -1/4 0 1/4
-1/2 3/2 1/2 -1/2 0 5/2 subject to
2 -| -I
8:X,X. X, X.,X,) sb,i=1,2.3.
3
-z,0 2
-1 0 2 0
X,20 1,2,3,.